generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
cbd1315ac4
vpx/vp9/encoder/vp9_rdopt.c

4922 lines
177 KiB
C
Raw Normal View History

Initial WebM release
2010-05-18 17:58:33 +02:00
/*
Use WebM in copyright notice for consistency Changes 'The VP8 project' to 'The WebM project', for consistency with other webmproject.org repositories. Fixes issue #97. Change-Id: I37c13ed5fbdb9d334ceef71c6350e9febed9bbba
2010-09-09 14:16:39 +02:00
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
Initial WebM release
2010-05-18 17:58:33 +02:00
*
cosmetics: trim trailing whitespace When the license headers were updated, they accidentally contained trailing whitespace, so unfortunately we have to touch all the files again. Change-Id: I236c05fade06589e417179c0444cb39b09e4200d
2010-06-18 18:39:21 +02:00
* Use of this source code is governed by a BSD-style license
LICENSE: update with latest text Change-Id: Ieebea089095d9073b3a94932791099f614ce120c
2010-06-04 22:19:40 +02:00
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
cosmetics: trim trailing whitespace When the license headers were updated, they accidentally contained trailing whitespace, so unfortunately we have to touch all the files again. Change-Id: I236c05fade06589e417179c0444cb39b09e4200d
2010-06-18 18:39:21 +02:00
* in the file PATENTS. All contributing project authors may
LICENSE: update with latest text Change-Id: Ieebea089095d9073b3a94932791099f614ce120c
2010-06-04 22:19:40 +02:00
* be found in the AUTHORS file in the root of the source tree.
Initial WebM release
2010-05-18 17:58:33 +02:00
*/
#include <stdio.h>
#include <math.h>
#include <limits.h>
#include <assert.h>
Add vp9_ prefix to all vp9 files Support for gyp which doesn't support multiple objects in the same static library having the same basename. Change-Id: Ib947eefbaf68f8b177a796d23f875ccdfa6bc9dc
2012-11-27 22:59:17 +01:00
#include "vp9/common/vp9_pragmas.h"
fixed includes to be fully specified Change-Id: Ia1cce221f8511561b9cbd8edb7726fbc286ff243
2012-11-28 19:41:40 +01:00
#include "vp9/encoder/vp9_tokenize.h"
#include "vp9/encoder/vp9_treewriter.h"
#include "vp9/encoder/vp9_onyx_int.h"
#include "vp9/encoder/vp9_modecosts.h"
#include "vp9/encoder/vp9_encodeintra.h"
Add vp9_ prefix to all vp9 files Support for gyp which doesn't support multiple objects in the same static library having the same basename. Change-Id: Ib947eefbaf68f8b177a796d23f875ccdfa6bc9dc
2012-11-27 22:59:17 +01:00
#include "vp9/common/vp9_entropymode.h"
#include "vp9/common/vp9_reconinter.h"
#include "vp9/common/vp9_reconintra.h"
#include "vp9/common/vp9_findnearmv.h"
#include "vp9/common/vp9_quant_common.h"
fixed includes to be fully specified Change-Id: Ia1cce221f8511561b9cbd8edb7726fbc286ff243
2012-11-28 19:41:40 +01:00
#include "vp9/encoder/vp9_encodemb.h"
#include "vp9/encoder/vp9_quantize.h"
#include "vp9/encoder/vp9_variance.h"
#include "vp9/encoder/vp9_mcomp.h"
#include "vp9/encoder/vp9_rdopt.h"
#include "vp9/encoder/vp9_ratectrl.h"
Initial WebM release
2010-05-18 17:58:33 +02:00
#include "vpx_mem/vpx_mem.h"
Add vp9_ prefix to all vp9 files Support for gyp which doesn't support multiple objects in the same static library having the same basename. Change-Id: Ib947eefbaf68f8b177a796d23f875ccdfa6bc9dc
2012-11-27 22:59:17 +01:00
#include "vp9/common/vp9_systemdependent.h"
#include "vp9/encoder/vp9_encodemv.h"
Initial WebM release
2010-05-18 17:58:33 +02:00
Add vp9_ prefix to all vp9 files Support for gyp which doesn't support multiple objects in the same static library having the same basename. Change-Id: Ib947eefbaf68f8b177a796d23f875ccdfa6bc9dc
2012-11-27 22:59:17 +01:00
#include "vp9/common/vp9_seg_common.h"
#include "vp9/common/vp9_pred_common.h"
#include "vp9/common/vp9_entropy.h"
support building vp8 and vp9 into a single lib Change-Id: Ib8f8a66c9fd31e508cdc9caa662192f38433aa3d
2012-11-09 02:09:30 +01:00
#include "vp9_rtcd.h"
Add vp9_ prefix to all vp9 files Support for gyp which doesn't support multiple objects in the same static library having the same basename. Change-Id: Ib947eefbaf68f8b177a796d23f875ccdfa6bc9dc
2012-11-27 22:59:17 +01:00
#include "vp9/common/vp9_mvref_common.h"
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
#include "vp9/common/vp9_common.h"
New Motion Reference Search Alternative strategy for finding a list of candidate motion vectors to use as reference values in mv coding and as nearest and near. Sort by sad in vp8_find_best_ref_mvs() rather than just pick the best. Allow 0,0 as a best ref option but not a nearest or near unless there are no alternatives. Encode/Decode verified on at least some clips. Some commented out experimental and stats code still in place. Gain over existing code averages about 1% on derf (alll metrics) with improvement on all clips. Other test results pending. The entropy coding of the mode (nearest/near etc) still depends upon and requires the old "findnear" code so this needs looking at and may provide room for further gains. Change-Id: I871d7cba1d1c379c4bad9bcccce1fb19c46b8247
2012-08-24 16:44:01 +02:00
Initial WebM release
2010-05-18 17:58:33 +02:00
#define MAXF(a,b) (((a) > (b)) ? (a) : (b))
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
#define INVALID_MV 0x80008000
Adds support for switchable interpolation filters. Allows for swtiching/setting interpolation filters at the MB level. A frame level flag indicates whether to use a specifc filter for the entire frame or to signal the interpolation filter for each MB. When switchable filters are used, the encoder chooses between 8-tap and 8-tap sharp filters. The code currently has options to explore other variations as well, which will be cleaned up subsequently. One issue with the framework is that encoding is slow. I tried to do some tricks to speed things up but it is still slow. Decoding speed should not be affected since the number of filter taps remain unchanged. With the current version, we are up 0.5% on derf on average but some videos city/mobile improve by close to 4 and 2% respectively. If we did a full-search by turning the SEARCH_BEST_FILTER flag on, the results are somewhat better. The framework can be combined with filtered prediction, and I seek feedback regarding that. Rebased. Change-Id: I8f632cb2c111e76284140a2bd480945d6d42b77a
2012-07-18 22:43:01 +02:00
/* Factor to weigh the rate for switchable interp filters */
#define SWITCHABLE_INTERP_RATE_FACTOR 1
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
static const int auto_speed_thresh[17] = {
1000,
200,
150,
130,
150,
125,
120,
115,
115,
115,
115,
115,
115,
115,
115,
115,
105
Initial WebM release
2010-05-18 17:58:33 +02:00
};
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
const MODE_DEFINITION vp9_mode_order[MAX_MODES] = {
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
{ZEROMV, LAST_FRAME, NONE},
{DC_PRED, INTRA_FRAME, NONE},
Initial WebM release
2010-05-18 17:58:33 +02:00
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
{NEARESTMV, LAST_FRAME, NONE},
{NEARMV, LAST_FRAME, NONE},
Initial WebM release
2010-05-18 17:58:33 +02:00
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
{ZEROMV, GOLDEN_FRAME, NONE},
{NEARESTMV, GOLDEN_FRAME, NONE},
Initial WebM release
2010-05-18 17:58:33 +02:00
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
{ZEROMV, ALTREF_FRAME, NONE},
{NEARESTMV, ALTREF_FRAME, NONE},
Initial WebM release
2010-05-18 17:58:33 +02:00
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
{NEARMV, GOLDEN_FRAME, NONE},
{NEARMV, ALTREF_FRAME, NONE},
Initial WebM release
2010-05-18 17:58:33 +02:00
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
{V_PRED, INTRA_FRAME, NONE},
{H_PRED, INTRA_FRAME, NONE},
{D45_PRED, INTRA_FRAME, NONE},
{D135_PRED, INTRA_FRAME, NONE},
{D117_PRED, INTRA_FRAME, NONE},
{D153_PRED, INTRA_FRAME, NONE},
{D27_PRED, INTRA_FRAME, NONE},
{D63_PRED, INTRA_FRAME, NONE},
Initial WebM release
2010-05-18 17:58:33 +02:00
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
{TM_PRED, INTRA_FRAME, NONE},
Initial WebM release
2010-05-18 17:58:33 +02:00
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
{NEWMV, LAST_FRAME, NONE},
{NEWMV, GOLDEN_FRAME, NONE},
{NEWMV, ALTREF_FRAME, NONE},
Initial WebM release
2010-05-18 17:58:33 +02:00
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
{SPLITMV, LAST_FRAME, NONE},
{SPLITMV, GOLDEN_FRAME, NONE},
{SPLITMV, ALTREF_FRAME, NONE},
Dual 16x16 inter prediction. This patch introduces the concept of dual inter16x16 prediction. A 16x16 inter-predicted macroblock can use 2 references instead of 1, where both references use the same mvmode (new, near/est, zero). In the case of newmv, this means that two MVs are coded instead of one. The frame can be encoded in 3 ways: all MBs single-prediction, all MBs dual prediction, or per-MB single/dual prediction selection ("hybrid"), in which case a single bit is coded per-MB to indicate whether the MB uses single or dual inter prediction. In the future, we can (maybe?) get further gains by mixing this with Adrian's 32x32 work, per-segment dual prediction settings, or adding support for dual splitmv/8x8mv inter prediction. Gain (on derf-set, CQ mode) is ~2.8% (SSIM) or ~3.6% (glb PSNR). Most gain is at medium/high bitrates, but there's minor gains at low bitrates also. Output was confirmed to match between encoder and decoder. Note for optimization people: this patch introduces a 2nd version of 16x16/8x8 sixtap/bilin functions, which does an avg instead of a store. They may want to look and make sure this is implemented to their satisfaction so we can optimize it best in the future. Change-ID: I59dc84b07cbb3ccf073ac0f756d03d294cb19281
2011-12-06 20:53:02 +01:00
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
{I4X4_PRED, INTRA_FRAME, NONE},
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
{I8X8_PRED, INTRA_FRAME, NONE},
Dual 16x16 inter prediction. This patch introduces the concept of dual inter16x16 prediction. A 16x16 inter-predicted macroblock can use 2 references instead of 1, where both references use the same mvmode (new, near/est, zero). In the case of newmv, this means that two MVs are coded instead of one. The frame can be encoded in 3 ways: all MBs single-prediction, all MBs dual prediction, or per-MB single/dual prediction selection ("hybrid"), in which case a single bit is coded per-MB to indicate whether the MB uses single or dual inter prediction. In the future, we can (maybe?) get further gains by mixing this with Adrian's 32x32 work, per-segment dual prediction settings, or adding support for dual splitmv/8x8mv inter prediction. Gain (on derf-set, CQ mode) is ~2.8% (SSIM) or ~3.6% (glb PSNR). Most gain is at medium/high bitrates, but there's minor gains at low bitrates also. Output was confirmed to match between encoder and decoder. Note for optimization people: this patch introduces a 2nd version of 16x16/8x8 sixtap/bilin functions, which does an avg instead of a store. They may want to look and make sure this is implemented to their satisfaction so we can optimize it best in the future. Change-ID: I59dc84b07cbb3ccf073ac0f756d03d294cb19281
2011-12-06 20:53:02 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/* compound prediction modes */
{ZEROMV, LAST_FRAME, GOLDEN_FRAME},
{NEARESTMV, LAST_FRAME, GOLDEN_FRAME},
{NEARMV, LAST_FRAME, GOLDEN_FRAME},
Dual 16x16 inter prediction. This patch introduces the concept of dual inter16x16 prediction. A 16x16 inter-predicted macroblock can use 2 references instead of 1, where both references use the same mvmode (new, near/est, zero). In the case of newmv, this means that two MVs are coded instead of one. The frame can be encoded in 3 ways: all MBs single-prediction, all MBs dual prediction, or per-MB single/dual prediction selection ("hybrid"), in which case a single bit is coded per-MB to indicate whether the MB uses single or dual inter prediction. In the future, we can (maybe?) get further gains by mixing this with Adrian's 32x32 work, per-segment dual prediction settings, or adding support for dual splitmv/8x8mv inter prediction. Gain (on derf-set, CQ mode) is ~2.8% (SSIM) or ~3.6% (glb PSNR). Most gain is at medium/high bitrates, but there's minor gains at low bitrates also. Output was confirmed to match between encoder and decoder. Note for optimization people: this patch introduces a 2nd version of 16x16/8x8 sixtap/bilin functions, which does an avg instead of a store. They may want to look and make sure this is implemented to their satisfaction so we can optimize it best in the future. Change-ID: I59dc84b07cbb3ccf073ac0f756d03d294cb19281
2011-12-06 20:53:02 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
{ZEROMV, ALTREF_FRAME, LAST_FRAME},
{NEARESTMV, ALTREF_FRAME, LAST_FRAME},
{NEARMV, ALTREF_FRAME, LAST_FRAME},
Dual 16x16 inter prediction. This patch introduces the concept of dual inter16x16 prediction. A 16x16 inter-predicted macroblock can use 2 references instead of 1, where both references use the same mvmode (new, near/est, zero). In the case of newmv, this means that two MVs are coded instead of one. The frame can be encoded in 3 ways: all MBs single-prediction, all MBs dual prediction, or per-MB single/dual prediction selection ("hybrid"), in which case a single bit is coded per-MB to indicate whether the MB uses single or dual inter prediction. In the future, we can (maybe?) get further gains by mixing this with Adrian's 32x32 work, per-segment dual prediction settings, or adding support for dual splitmv/8x8mv inter prediction. Gain (on derf-set, CQ mode) is ~2.8% (SSIM) or ~3.6% (glb PSNR). Most gain is at medium/high bitrates, but there's minor gains at low bitrates also. Output was confirmed to match between encoder and decoder. Note for optimization people: this patch introduces a 2nd version of 16x16/8x8 sixtap/bilin functions, which does an avg instead of a store. They may want to look and make sure this is implemented to their satisfaction so we can optimize it best in the future. Change-ID: I59dc84b07cbb3ccf073ac0f756d03d294cb19281
2011-12-06 20:53:02 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
{ZEROMV, GOLDEN_FRAME, ALTREF_FRAME},
{NEARESTMV, GOLDEN_FRAME, ALTREF_FRAME},
{NEARMV, GOLDEN_FRAME, ALTREF_FRAME},
Dual 16x16 inter prediction. This patch introduces the concept of dual inter16x16 prediction. A 16x16 inter-predicted macroblock can use 2 references instead of 1, where both references use the same mvmode (new, near/est, zero). In the case of newmv, this means that two MVs are coded instead of one. The frame can be encoded in 3 ways: all MBs single-prediction, all MBs dual prediction, or per-MB single/dual prediction selection ("hybrid"), in which case a single bit is coded per-MB to indicate whether the MB uses single or dual inter prediction. In the future, we can (maybe?) get further gains by mixing this with Adrian's 32x32 work, per-segment dual prediction settings, or adding support for dual splitmv/8x8mv inter prediction. Gain (on derf-set, CQ mode) is ~2.8% (SSIM) or ~3.6% (glb PSNR). Most gain is at medium/high bitrates, but there's minor gains at low bitrates also. Output was confirmed to match between encoder and decoder. Note for optimization people: this patch introduces a 2nd version of 16x16/8x8 sixtap/bilin functions, which does an avg instead of a store. They may want to look and make sure this is implemented to their satisfaction so we can optimize it best in the future. Change-ID: I59dc84b07cbb3ccf073ac0f756d03d294cb19281
2011-12-06 20:53:02 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
{NEWMV, LAST_FRAME, GOLDEN_FRAME},
{NEWMV, ALTREF_FRAME, LAST_FRAME },
{NEWMV, GOLDEN_FRAME, ALTREF_FRAME},
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
{SPLITMV, LAST_FRAME, GOLDEN_FRAME},
{SPLITMV, ALTREF_FRAME, LAST_FRAME },
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
{SPLITMV, GOLDEN_FRAME, ALTREF_FRAME},
#if CONFIG_COMP_INTERINTRA_PRED
/* compound inter-intra prediction */
{ZEROMV, LAST_FRAME, INTRA_FRAME},
{NEARESTMV, LAST_FRAME, INTRA_FRAME},
{NEARMV, LAST_FRAME, INTRA_FRAME},
{NEWMV, LAST_FRAME, INTRA_FRAME},
{ZEROMV, GOLDEN_FRAME, INTRA_FRAME},
{NEARESTMV, GOLDEN_FRAME, INTRA_FRAME},
{NEARMV, GOLDEN_FRAME, INTRA_FRAME},
{NEWMV, GOLDEN_FRAME, INTRA_FRAME},
{ZEROMV, ALTREF_FRAME, INTRA_FRAME},
{NEARESTMV, ALTREF_FRAME, INTRA_FRAME},
{NEARMV, ALTREF_FRAME, INTRA_FRAME},
{NEWMV, ALTREF_FRAME, INTRA_FRAME},
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
};
Introduce vp9_coeff_probs/counts/stats/accum types. Use these, instead of the 4/5-dimensional arrays, to hold statistics, counts, accumulations and probabilities for coefficient tokens. This commit also re-allows ENTROPY_STATS to compile. Change-Id: If441ffac936f52a3af91d8f2922ea8a0ceabdaa5
2012-12-08 01:09:59 +01:00
static void fill_token_costs(vp9_coeff_count *c,
vp9_coeff_probs *p,
Framework changes in nzc to allow more flexibility The patch adds the flexibility to use standard EOB based coding on smaller block sizes and nzc based coding on larger blocksizes. The tx-sizes that use nzc based coding and those that use EOB based coding are controlled by a function get_nzc_used(). By default, this function uses nzc based coding for 16x16 and 32x32 transform blocks, which seem to bridge the performance gap substantially. All sets are now lower by 0.5% to 0.7%, as opposed to ~1.8% before. Change-Id: I06abed3df57b52d241ea1f51b0d571c71e38fd0b
2013-03-26 23:23:30 +01:00
TX_SIZE tx_size) {
Split coefficient token tables intra vs. inter. Change-Id: I5416455f8f129ca0f450d00e48358d2012605072
2013-02-19 22:36:38 +01:00
int i, j, k, l;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Framework changes in nzc to allow more flexibility The patch adds the flexibility to use standard EOB based coding on smaller block sizes and nzc based coding on larger blocksizes. The tx-sizes that use nzc based coding and those that use EOB based coding are controlled by a function get_nzc_used(). By default, this function uses nzc based coding for 16x16 and 32x32 transform blocks, which seem to bridge the performance gap substantially. All sets are now lower by 0.5% to 0.7%, as opposed to ~1.8% before. Change-Id: I06abed3df57b52d241ea1f51b0d571c71e38fd0b
2013-03-26 23:23:30 +01:00
for (i = 0; i < BLOCK_TYPES; i++)
Split coefficient token tables intra vs. inter. Change-Id: I5416455f8f129ca0f450d00e48358d2012605072
2013-02-19 22:36:38 +01:00
for (j = 0; j < REF_TYPES; j++)
for (k = 0; k < COEF_BANDS; k++)
for (l = 0; l < PREV_COEF_CONTEXTS; l++) {
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
vp9_cost_tokens_skip((int *)(c[i][j][k][l]),
p[i][j][k][l],
vp9_coef_tree);
}
}
Adjust style to match Google Coding Style a little more closely. Most of these were picked up in the previous commit (prefix change from vp8_ to vp9_), but I'm pushing this separately so that it's easier to review. Change-Id: I91e959895778b8632d7d33375523df8a7568a490
2012-10-30 22:25:33 +01:00
static int rd_iifactor[32] = { 4, 4, 3, 2, 1, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, };
Initial WebM release
2010-05-18 17:58:33 +02:00
experiment extending the quantizer range Prior to this change, VP8 min quantizer is 4, which caps the highest quality around 51DB. This experimental change extends the min quantizer to 1, removes the cap and allows the highest quality to be around ~73DB, consistent with the fdct/idct round trip error. To test this change, at configure time use options: --enable-experimental --enable-extend_qrange The following is a brief log of changes in each of the patch sets patch set 1: In this commit, the quantization/dequantization constants are kept unchanged, instead scaling factor 4 is rolled into fdct/idct. Fixed Q0 encoding tests on mobile: Before: 9560.567kbps Overall PSNR:50.255DB VPXSSIM:98.288 Now: 18035.774kbps Overall PSNR:73.022DB VPXSSIM:99.991 patch set 2: regenerated dc/ac quantizer lookup tables based on the scaling factor rolled in the fdct/idct. Also slightly extended the range towards the high quantizer end. patch set 3: slightly tweaked the quantizer tables and generated bits_per_mb table based on Paul's suggestions. patch set 4: fix a typo in idct, re-calculated tables relating active max Q to active min Q patch set 5: added rdmult lookup table based on Q patch set 6: fix rdmult scale: dct coefficient has scaled up by 4 patch set 7: make transform coefficients to be within 16bits patch set 8: normalize 2nd order quantizers patch set 9: fix mis-spellings patch set 10: change the configure script and macros to allow experimental code to be enabled at configure time with --enable-extend_qrange patch set 11: rebase for merge Change-Id: Ib50641ddd44aba2a52ed890222c309faa31cc59c
2010-12-02 00:50:14 +01:00
// 3* dc_qlookup[Q]*dc_qlookup[Q];
Further resolution of QIndex LUTS; This commit resolves further QIndex look up tables to facilitate experimentation with the quantizer range. In some cases rather than remove the look up tables completely I have created functions that are called once to populate them using a formulaic approach base on the actual quantizer. The use of these functions based on best fit of data from the original tables does affect the results on some clips but across the derf test set the effect was broadly neutral. Change-Id: I8baa61c97ce87dc09a6340d56fdeb681b9345793
2011-11-23 12:32:20 +01:00
adjusted sad_per_bit to correlate with quantizer Re-calibrated sad_per_bit16 and sad_per_bit4 tables to linearly correlated to quantizer values, these two variables are used in motion search for costing motion vectors. This change has an small positive effect on compression. Change-Id: Ic9b5ea6fb8d5078ef663ba4899db019cc51f4166
2010-12-24 04:59:12 +01:00
/* values are now correlated to quantizer */
Removal of Qindex LUTS. One of the problems arising when tweaking or adjusting the quantizer tables is that there are a lot of look up tables that depend on the QINDEX. Any adjustment to the link between QINDEX and real quantizer therefore tends to break aspects of for example the rate control. In this check in I have replaced several of the look up tables with functions that approximate the same results as the old Q luts but use a formulaic approach based on real Q values rather than QIndex. This should hopefully make it easier to experiment with changes to the Q tables without always having to go through and hand optimize a set of look up tables. Once things stabilize we may choose to re-instate luts for the sake of performance. Patch 2: Addressed Ronald's comments. vp8_init_me_luts() Added so luts only initialized once. Change-Id: Ic80db2212d2fd01e08e8cb5c7dca1fda1102be57
2011-11-21 16:45:10 +01:00
static int sad_per_bit16lut[QINDEX_RANGE];
static int sad_per_bit4lut[QINDEX_RANGE];
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
void vp9_init_me_luts() {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int i;
// Initialize the sad lut tables using a formulaic calculation for now
// This is to make it easier to resolve the impact of experimental changes
// to the quantizer tables.
for (i = 0; i < QINDEX_RANGE; i++) {
sad_per_bit16lut[i] =
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
(int)((0.0418 * vp9_convert_qindex_to_q(i)) + 2.4107);
sad_per_bit4lut[i] = (int)((0.063 * vp9_convert_qindex_to_q(i)) + 2.742);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Removal of Qindex LUTS. One of the problems arising when tweaking or adjusting the quantizer tables is that there are a lot of look up tables that depend on the QINDEX. Any adjustment to the link between QINDEX and real quantizer therefore tends to break aspects of for example the rate control. In this check in I have replaced several of the look up tables with functions that approximate the same results as the old Q luts but use a formulaic approach based on real Q values rather than QIndex. This should hopefully make it easier to experiment with changes to the Q tables without always having to go through and hand optimize a set of look up tables. Once things stabilize we may choose to re-instate luts for the sake of performance. Patch 2: Addressed Ronald's comments. vp8_init_me_luts() Added so luts only initialized once. Change-Id: Ic80db2212d2fd01e08e8cb5c7dca1fda1102be57
2011-11-21 16:45:10 +01:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
rdopt: make locally used symbols static. Change-Id: Id33c448084ca5c1c0525a05c3c4f35d6b0805fbe
2012-10-30 02:04:33 +01:00
static int compute_rd_mult(int qindex) {
Code cleanup. Removing redundant 'extern' keyword, lowercase variable names. Change-Id: I608e8d8579aba8981f5fac3493f77b4481b13808
2013-03-02 02:39:31 +01:00
int q = vp9_dc_quant(qindex, 0);
Changes to rd error_per_bit calculation. Specifically changes to retain more precision especially at low Q through to the point of use. Change-Id: Ief5f010f2ca4daaabef49520e7edb46c35daf397
2013-03-07 16:02:57 +01:00
return (11 * q * q) >> 2;
Further resolution of QIndex LUTS; This commit resolves further QIndex look up tables to facilitate experimentation with the quantizer range. In some cases rather than remove the look up tables completely I have created functions that are called once to populate them using a formulaic approach base on the actual quantizer. The use of these functions based on best fit of data from the original tables does affect the results on some clips but across the derf test set the effect was broadly neutral. Change-Id: I8baa61c97ce87dc09a6340d56fdeb681b9345793
2011-11-23 12:32:20 +01:00
}
Code cleanup. Removing redundant 'extern' keyword, lowercase variable names. Change-Id: I608e8d8579aba8981f5fac3493f77b4481b13808
2013-03-02 02:39:31 +01:00
void vp9_initialize_me_consts(VP9_COMP *cpi, int qindex) {
cpi->mb.sadperbit16 = sad_per_bit16lut[qindex];
cpi->mb.sadperbit4 = sad_per_bit4lut[qindex];
Initial WebM release
2010-05-18 17:58:33 +02:00
}
experiment extending the quantizer range Prior to this change, VP8 min quantizer is 4, which caps the highest quality around 51DB. This experimental change extends the min quantizer to 1, removes the cap and allows the highest quality to be around ~73DB, consistent with the fdct/idct round trip error. To test this change, at configure time use options: --enable-experimental --enable-extend_qrange The following is a brief log of changes in each of the patch sets patch set 1: In this commit, the quantization/dequantization constants are kept unchanged, instead scaling factor 4 is rolled into fdct/idct. Fixed Q0 encoding tests on mobile: Before: 9560.567kbps Overall PSNR:50.255DB VPXSSIM:98.288 Now: 18035.774kbps Overall PSNR:73.022DB VPXSSIM:99.991 patch set 2: regenerated dc/ac quantizer lookup tables based on the scaling factor rolled in the fdct/idct. Also slightly extended the range towards the high quantizer end. patch set 3: slightly tweaked the quantizer tables and generated bits_per_mb table based on Paul's suggestions. patch set 4: fix a typo in idct, re-calculated tables relating active max Q to active min Q patch set 5: added rdmult lookup table based on Q patch set 6: fix rdmult scale: dct coefficient has scaled up by 4 patch set 7: make transform coefficients to be within 16bits patch set 8: normalize 2nd order quantizers patch set 9: fix mis-spellings patch set 10: change the configure script and macros to allow experimental code to be enabled at configure time with --enable-extend_qrange patch set 11: rebase for merge Change-Id: Ib50641ddd44aba2a52ed890222c309faa31cc59c
2010-12-02 00:50:14 +01:00
Code cleanup. Removing redundant 'extern' keyword, lowercase variable names. Change-Id: I608e8d8579aba8981f5fac3493f77b4481b13808
2013-03-02 02:39:31 +01:00
void vp9_initialize_rd_consts(VP9_COMP *cpi, int qindex) {
Combine vp8_block_error_c and vp8_submb_error_c. Change-Id: I58792bac6c7e5a524a6dcdb5241ef18bcc0c3a45
2012-08-07 00:03:04 +02:00
int q, i;
Initial WebM release
2010-05-18 17:58:33 +02:00
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
vp9_clear_system_state(); // __asm emms;
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// Further tests required to see if optimum is different
// for key frames, golden frames and arf frames.
// if (cpi->common.refresh_golden_frame ||
// cpi->common.refresh_alt_ref_frame)
Code cleanup. Removing redundant 'extern' keyword, lowercase variable names. Change-Id: I608e8d8579aba8981f5fac3493f77b4481b13808
2013-03-02 02:39:31 +01:00
qindex = (qindex < 0) ? 0 : ((qindex > MAXQ) ? MAXQ : qindex);
Further resolution of QIndex LUTS; This commit resolves further QIndex look up tables to facilitate experimentation with the quantizer range. In some cases rather than remove the look up tables completely I have created functions that are called once to populate them using a formulaic approach base on the actual quantizer. The use of these functions based on best fit of data from the original tables does affect the results on some clips but across the derf test set the effect was broadly neutral. Change-Id: I8baa61c97ce87dc09a6340d56fdeb681b9345793
2011-11-23 12:32:20 +01:00
Code cleanup. Removing redundant 'extern' keyword, lowercase variable names. Change-Id: I608e8d8579aba8981f5fac3493f77b4481b13808
2013-03-02 02:39:31 +01:00
cpi->RDMULT = compute_rd_mult(qindex);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (cpi->pass == 2 && (cpi->common.frame_type != KEY_FRAME)) {
if (cpi->twopass.next_iiratio > 31)
cpi->RDMULT += (cpi->RDMULT * rd_iifactor[31]) >> 4;
else
cpi->RDMULT +=
Code cleanup. Removing redundant 'extern' keyword, lowercase variable names. Change-Id: I608e8d8579aba8981f5fac3493f77b4481b13808
2013-03-02 02:39:31 +01:00
(cpi->RDMULT * rd_iifactor[cpi->twopass.next_iiratio]) >> 4;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Changes to rd error_per_bit calculation. Specifically changes to retain more precision especially at low Q through to the point of use. Change-Id: Ief5f010f2ca4daaabef49520e7edb46c35daf397
2013-03-07 16:02:57 +01:00
cpi->mb.errorperbit = cpi->RDMULT >> 6;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
cpi->mb.errorperbit += (cpi->mb.errorperbit == 0);
cap the best quantizer for 2nd order DC This commit also removes artificial RDMULT cap for low quantizers. The intention is to address some abnormal behavior of mode selections at the low quantizer end, where many macroblocks were coded with SPLITMV with all partitions using same motion vector including (0,0). This change improves the compression quality substantially for high quality encodings in both PSNR and SSIM terms. Overall effect on mid/low rate range is also positive for all metrics, but smaller in magnitude. Change-Id: I864b29c4bd9ff610d2545fa94a19cc7e80c02667
2011-01-26 07:24:22 +01:00
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
vp9_set_speed_features(cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
Code cleanup. Removing redundant 'extern' keyword, lowercase variable names. Change-Id: I608e8d8579aba8981f5fac3493f77b4481b13808
2013-03-02 02:39:31 +01:00
q = (int)pow(vp9_dc_quant(qindex, 0) >> 2, 1.25);
q <<= 2;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (q < 8)
q = 8;
experiment extending the quantizer range Prior to this change, VP8 min quantizer is 4, which caps the highest quality around 51DB. This experimental change extends the min quantizer to 1, removes the cap and allows the highest quality to be around ~73DB, consistent with the fdct/idct round trip error. To test this change, at configure time use options: --enable-experimental --enable-extend_qrange The following is a brief log of changes in each of the patch sets patch set 1: In this commit, the quantization/dequantization constants are kept unchanged, instead scaling factor 4 is rolled into fdct/idct. Fixed Q0 encoding tests on mobile: Before: 9560.567kbps Overall PSNR:50.255DB VPXSSIM:98.288 Now: 18035.774kbps Overall PSNR:73.022DB VPXSSIM:99.991 patch set 2: regenerated dc/ac quantizer lookup tables based on the scaling factor rolled in the fdct/idct. Also slightly extended the range towards the high quantizer end. patch set 3: slightly tweaked the quantizer tables and generated bits_per_mb table based on Paul's suggestions. patch set 4: fix a typo in idct, re-calculated tables relating active max Q to active min Q patch set 5: added rdmult lookup table based on Q patch set 6: fix rdmult scale: dct coefficient has scaled up by 4 patch set 7: make transform coefficients to be within 16bits patch set 8: normalize 2nd order quantizers patch set 9: fix mis-spellings patch set 10: change the configure script and macros to allow experimental code to be enabled at configure time with --enable-extend_qrange patch set 11: rebase for merge Change-Id: Ib50641ddd44aba2a52ed890222c309faa31cc59c
2010-12-02 00:50:14 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (cpi->RDMULT > 1000) {
cpi->RDDIV = 1;
cpi->RDMULT /= 100;
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
for (i = 0; i < MAX_MODES; i++) {
if (cpi->sf.thresh_mult[i] < INT_MAX) {
cpi->rd_threshes[i] = cpi->sf.thresh_mult[i] * q / 100;
} else {
cpi->rd_threshes[i] = INT_MAX;
}
cpi->rd_baseline_thresh[i] = cpi->rd_threshes[i];
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
} else {
cpi->RDDIV = 100;
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
for (i = 0; i < MAX_MODES; i++) {
if (cpi->sf.thresh_mult[i] < (INT_MAX / q)) {
cpi->rd_threshes[i] = cpi->sf.thresh_mult[i] * q;
} else {
cpi->rd_threshes[i] = INT_MAX;
}
cpi->rd_baseline_thresh[i] = cpi->rd_threshes[i];
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Introduce vp9_coeff_probs/counts/stats/accum types. Use these, instead of the 4/5-dimensional arrays, to hold statistics, counts, accumulations and probabilities for coefficient tokens. This commit also re-allows ENTROPY_STATS to compile. Change-Id: If441ffac936f52a3af91d8f2922ea8a0ceabdaa5
2012-12-08 01:09:59 +01:00
fill_token_costs(cpi->mb.token_costs[TX_4X4],
Framework changes in nzc to allow more flexibility The patch adds the flexibility to use standard EOB based coding on smaller block sizes and nzc based coding on larger blocksizes. The tx-sizes that use nzc based coding and those that use EOB based coding are controlled by a function get_nzc_used(). By default, this function uses nzc based coding for 16x16 and 32x32 transform blocks, which seem to bridge the performance gap substantially. All sets are now lower by 0.5% to 0.7%, as opposed to ~1.8% before. Change-Id: I06abed3df57b52d241ea1f51b0d571c71e38fd0b
2013-03-26 23:23:30 +01:00
cpi->common.fc.coef_probs_4x4, TX_4X4);
Introduce vp9_coeff_probs/counts/stats/accum types. Use these, instead of the 4/5-dimensional arrays, to hold statistics, counts, accumulations and probabilities for coefficient tokens. This commit also re-allows ENTROPY_STATS to compile. Change-Id: If441ffac936f52a3af91d8f2922ea8a0ceabdaa5
2012-12-08 01:09:59 +01:00
fill_token_costs(cpi->mb.token_costs[TX_8X8],
Framework changes in nzc to allow more flexibility The patch adds the flexibility to use standard EOB based coding on smaller block sizes and nzc based coding on larger blocksizes. The tx-sizes that use nzc based coding and those that use EOB based coding are controlled by a function get_nzc_used(). By default, this function uses nzc based coding for 16x16 and 32x32 transform blocks, which seem to bridge the performance gap substantially. All sets are now lower by 0.5% to 0.7%, as opposed to ~1.8% before. Change-Id: I06abed3df57b52d241ea1f51b0d571c71e38fd0b
2013-03-26 23:23:30 +01:00
cpi->common.fc.coef_probs_8x8, TX_8X8);
Introduce vp9_coeff_probs/counts/stats/accum types. Use these, instead of the 4/5-dimensional arrays, to hold statistics, counts, accumulations and probabilities for coefficient tokens. This commit also re-allows ENTROPY_STATS to compile. Change-Id: If441ffac936f52a3af91d8f2922ea8a0ceabdaa5
2012-12-08 01:09:59 +01:00
fill_token_costs(cpi->mb.token_costs[TX_16X16],
Framework changes in nzc to allow more flexibility The patch adds the flexibility to use standard EOB based coding on smaller block sizes and nzc based coding on larger blocksizes. The tx-sizes that use nzc based coding and those that use EOB based coding are controlled by a function get_nzc_used(). By default, this function uses nzc based coding for 16x16 and 32x32 transform blocks, which seem to bridge the performance gap substantially. All sets are now lower by 0.5% to 0.7%, as opposed to ~1.8% before. Change-Id: I06abed3df57b52d241ea1f51b0d571c71e38fd0b
2013-03-26 23:23:30 +01:00
cpi->common.fc.coef_probs_16x16, TX_16X16);
Introduce vp9_coeff_probs/counts/stats/accum types. Use these, instead of the 4/5-dimensional arrays, to hold statistics, counts, accumulations and probabilities for coefficient tokens. This commit also re-allows ENTROPY_STATS to compile. Change-Id: If441ffac936f52a3af91d8f2922ea8a0ceabdaa5
2012-12-08 01:09:59 +01:00
fill_token_costs(cpi->mb.token_costs[TX_32X32],
Framework changes in nzc to allow more flexibility The patch adds the flexibility to use standard EOB based coding on smaller block sizes and nzc based coding on larger blocksizes. The tx-sizes that use nzc based coding and those that use EOB based coding are controlled by a function get_nzc_used(). By default, this function uses nzc based coding for 16x16 and 32x32 transform blocks, which seem to bridge the performance gap substantially. All sets are now lower by 0.5% to 0.7%, as opposed to ~1.8% before. Change-Id: I06abed3df57b52d241ea1f51b0d571c71e38fd0b
2013-03-26 23:23:30 +01:00
cpi->common.fc.coef_probs_32x32, TX_32X32);
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
Fairly basic integration of rectangular blocks in encoding RD loop. Adds RD integration for 32x16, 16x32, 64x32 and 32x64 rectangular blocks. Derf almost +0.6%, HD a little over +1.0%, STDHD +1.3%. Change-Id: Id651fdb6a655fdbb5c47009757e63317acfb88a5
2013-04-17 18:25:06 +02:00
for (i = 0; i < 2; i++)
vp9_cost_tokens(cpi->mb.partition_cost[i],
cpi->common.fc.partition_prob[i],
vp9_partition_tree);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/*rough estimate for costing*/
cpi->common.kf_ymode_probs_index = cpi->common.base_qindex >> 4;
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
vp9_init_mode_costs(cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
if (cpi->common.frame_type != KEY_FRAME) {
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
vp9_build_nmv_cost_table(
New motion vector entropy coding Adds a new experiment with redesigned/refactored motion vector entropy coding. The patch also takes a first step towards separating the integer and fractional pel components of a MV. However the fractional pel encoding still depends on the integer pel part and so they are not fully independent. Further experiments are in progress to see how much they can be decoupled without affecting performance. All components including entropy coding/decoding, costing for MV search, forward updates and backward updates to probability tables, have been implemented. Results so far: derf: +0.19% std-hd: +0.28% yt: +0.80% hd: +1.15% Patch: Simplifies the fractional pel models: derf: +0.284% std-hd: +0.289% yt: +0.849% hd: +1.254% Patch: Some changes in the models, rebased. derf: +0.330% std-hd: +0.306% yt: +0.816% hd: +1.225% Change-Id: I646b3c48f3587f4cc909639b78c3798da6402678
2012-07-26 22:42:07 +02:00
cpi->mb.nmvjointcost,
cpi->mb.e_mbd.allow_high_precision_mv ?
cpi->mb.nmvcost_hp : cpi->mb.nmvcost,
&cpi->common.fc.nmvc,
cpi->mb.e_mbd.allow_high_precision_mv, 1, 1);
}
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Use standard integer types for pixel values and coefficients. For coefficients, use int16_t (instead of short); for pixel values in 16-bit intermediates, use uint16_t (instead of unsigned short); for all others, use uint8_t (instead of unsigned char). Change-Id: I3619cd9abf106c3742eccc2e2f5e89a62774f7da
2012-12-19 00:31:19 +01:00
int vp9_block_error_c(int16_t *coeff, int16_t *dqcoeff, int block_size) {
Combine vp8_block_error_c and vp8_submb_error_c. Change-Id: I58792bac6c7e5a524a6dcdb5241ef18bcc0c3a45
2012-08-07 00:03:04 +02:00
int i, error = 0;
Initial WebM release
2010-05-18 17:58:33 +02:00
Combine vp8_block_error_c and vp8_submb_error_c. Change-Id: I58792bac6c7e5a524a6dcdb5241ef18bcc0c3a45
2012-08-07 00:03:04 +02:00
for (i = 0; i < block_size; i++) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int this_diff = coeff[i] - dqcoeff[i];
error += this_diff * this_diff;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
return error;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
static INLINE int cost_coeffs(VP9_COMMON *const cm, MACROBLOCK *mb,
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
int ib, PLANE_TYPE type,
Make cost_coeffs() more efficient. Cache the constant offset in one variable to prevent re-loading that in each loop iteration, and mark the function as inline so we can use the fact that the transform size is always known in the caller. Almost 1% faster encoding overall. Change-Id: Id78325a60b025057d8f4ecd9003a74086ccbf85a
2013-02-07 02:16:36 +01:00
ENTROPY_CONTEXT *a,
ENTROPY_CONTEXT *l,
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
TX_SIZE tx_size,
int y_blocks) {
Move eob from BLOCKD to MACROBLOCKD. Consistent with VP8. Change-Id: I8c316ee49f072e15abbb033a80e9c36617891f07
2013-02-27 19:00:24 +01:00
MACROBLOCKD *const xd = &mb->e_mbd;
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
int pt;
Remove some Y2-related code. Change-Id: I4f46d142c2a8d1e8a880cfac63702dcbfb999b78
2013-02-15 19:15:42 +01:00
int c = 0;
Use above/left (instead of previous in scan-order) as token context. Pearson correlation for above or left is significantly higher than for previous-in-scan-order (absolute values depend on position in scan, but in general, we gain about 0.1-0.2 by using either above or left; using both basically just makes this even better). For eob branch skipping, we continue to use the previous token in scan order. This helps about 0.9% on derf after re-training on a limited data set. Full re-training and results on larger-resolution clips are pending. Note that this commit breaks trellis, so we can probably get further gains out of it by fixing trellis at some later point. Change-Id: Iead68e296fc3a105cca746b5e3da9555d6010cfe
2013-03-27 00:46:09 +01:00
int cost = 0, pad;
const int *scan, *nb;
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
const struct plane_block_idx pb_idx = plane_block_idx(y_blocks, ib);
const int eob = xd->plane[pb_idx.plane].eobs[pb_idx.block];
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
const int16_t *qcoeff_ptr = BLOCK_OFFSET(xd->plane[pb_idx.plane].qcoeff,
pb_idx.block, 16);
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
const int ref = mbmi->ref_frame != INTRA_FRAME;
Make cost_coeffs() more efficient. Cache the constant offset in one variable to prevent re-loading that in each loop iteration, and mark the function as inline so we can use the fact that the transform size is always known in the caller. Almost 1% faster encoding overall. Change-Id: Id78325a60b025057d8f4ecd9003a74086ccbf85a
2013-02-07 02:16:36 +01:00
unsigned int (*token_costs)[PREV_COEF_CONTEXTS][MAX_ENTROPY_TOKENS] =
Split coefficient token tables intra vs. inter. Change-Id: I5416455f8f129ca0f450d00e48358d2012605072
2013-02-19 22:36:38 +01:00
mb->token_costs[tx_size][type][ref];
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
ENTROPY_CONTEXT a_ec, l_ec;
make cost_coeffs to use combined context Change-Id: Ia15f4244595fab49bffda0c651a750a8a9481d28
2013-02-26 02:01:30 +01:00
ENTROPY_CONTEXT *const a1 = a +
sizeof(ENTROPY_CONTEXT_PLANES)/sizeof(ENTROPY_CONTEXT);
ENTROPY_CONTEXT *const l1 = l +
sizeof(ENTROPY_CONTEXT_PLANES)/sizeof(ENTROPY_CONTEXT);
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
TX_TYPE tx_type = DCT_DCT;
Change the RD cost functions to take into account the seg_eob. Change-Id: I2f96e5b205f70fdb559f01637c4b5730246c38a8
2012-08-07 03:29:59 +02:00
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
#if CONFIG_CODE_ZEROGROUP
int last_nz_pos[3] = {-1, -1, -1}; // Encoder only
int is_eoo_list[3] = {0, 0, 0};
int is_eoo_negative[3] = {0, 0, 0};
int is_last_zero[3] = {0, 0, 0};
int o, rc, skip_coef_val;
vp9_zpc_probs *zpc_probs;
uint8_t token_cache_full[1024];
Framework changes in nzc to allow more flexibility The patch adds the flexibility to use standard EOB based coding on smaller block sizes and nzc based coding on larger blocksizes. The tx-sizes that use nzc based coding and those that use EOB based coding are controlled by a function get_nzc_used(). By default, this function uses nzc based coding for 16x16 and 32x32 transform blocks, which seem to bridge the performance gap substantially. All sets are now lower by 0.5% to 0.7%, as opposed to ~1.8% before. Change-Id: I06abed3df57b52d241ea1f51b0d571c71e38fd0b
2013-03-26 23:23:30 +01:00
#endif
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
const int segment_id = xd->mode_info_context->mbmi.segment_id;
Use above/left (instead of previous in scan-order) as token context. Pearson correlation for above or left is significantly higher than for previous-in-scan-order (absolute values depend on position in scan, but in general, we gain about 0.1-0.2 by using either above or left; using both basically just makes this even better). For eob branch skipping, we continue to use the previous token in scan order. This helps about 0.9% on derf after re-training on a limited data set. Full re-training and results on larger-resolution clips are pending. Note that this commit breaks trellis, so we can probably get further gains out of it by fixing trellis at some later point. Change-Id: Iead68e296fc3a105cca746b5e3da9555d6010cfe
2013-03-27 00:46:09 +01:00
vp9_prob (*coef_probs)[REF_TYPES][COEF_BANDS][PREV_COEF_CONTEXTS]
[ENTROPY_NODES];
int seg_eob, default_eob;
uint8_t token_cache[1024];
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
#if CONFIG_CODE_ZEROGROUP
vpx_memset(token_cache, UNKNOWN_TOKEN, sizeof(token_cache));
#endif
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
// Check for consistency of tx_size with mode info
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
assert((!type && !pb_idx.plane) || (type && pb_idx.plane));
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
if (type == PLANE_TYPE_Y_WITH_DC) {
assert(xd->mode_info_context->mbmi.txfm_size == tx_size);
} else {
TX_SIZE tx_size_uv = get_uv_tx_size(xd);
assert(tx_size == tx_size_uv);
}
Entropy coding for hybrid transform Separates the entropy coding context models for 4x4, 8x8 and 16x16 ADST variants. There is a small improvement for HD (hd/std-hd) by about 0.1-0.2%. Results on derf/yt are about the same, probably because there is not enough statistics. Results may improve somewhat once the initial probability tables are updated for the hybrid transforms which is coming soon. Change-Id: Ic7c0c62dacc68ef551054fdb575be8b8507d32a8
2012-09-10 07:42:35 +02:00
switch (tx_size) {
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
case TX_4X4: {
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
get_tx_type_4x4(xd, ib) : DCT_DCT;
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
a_ec = *a;
l_ec = *l;
Framework changes in nzc to allow more flexibility The patch adds the flexibility to use standard EOB based coding on smaller block sizes and nzc based coding on larger blocksizes. The tx-sizes that use nzc based coding and those that use EOB based coding are controlled by a function get_nzc_used(). By default, this function uses nzc based coding for 16x16 and 32x32 transform blocks, which seem to bridge the performance gap substantially. All sets are now lower by 0.5% to 0.7%, as opposed to ~1.8% before. Change-Id: I06abed3df57b52d241ea1f51b0d571c71e38fd0b
2013-03-26 23:23:30 +01:00
coef_probs = cm->fc.coef_probs_4x4;
Use above/left (instead of previous in scan-order) as token context. Pearson correlation for above or left is significantly higher than for previous-in-scan-order (absolute values depend on position in scan, but in general, we gain about 0.1-0.2 by using either above or left; using both basically just makes this even better). For eob branch skipping, we continue to use the previous token in scan order. This helps about 0.9% on derf after re-training on a limited data set. Full re-training and results on larger-resolution clips are pending. Note that this commit breaks trellis, so we can probably get further gains out of it by fixing trellis at some later point. Change-Id: Iead68e296fc3a105cca746b5e3da9555d6010cfe
2013-03-27 00:46:09 +01:00
seg_eob = 16;
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
if (tx_type == ADST_DCT) {
scan = vp9_row_scan_4x4;
} else if (tx_type == DCT_ADST) {
scan = vp9_col_scan_4x4;
} else {
scan = vp9_default_zig_zag1d_4x4;
Combine cost_coeffs functions. Change-Id: I3421d07fdcf2148d75c8dbfc8c44bc2d96a3a8f6
2012-08-06 21:15:24 +02:00
}
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
#if CONFIG_CODE_ZEROGROUP
zpc_probs = &cm->fc.zpc_probs_4x4;
#endif
Combine cost_coeffs functions. Change-Id: I3421d07fdcf2148d75c8dbfc8c44bc2d96a3a8f6
2012-08-06 21:15:24 +02:00
break;
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
}
Add col/row-based coefficient scanning patterns for 1D 8x8/16x16 ADSTs. These are mostly just for experimental purposes. I saw small gains (in the 0.1% range) when playing with this on derf. Change-Id: Ib21eed477bbb46bddcd73b21c5c708a5b46abedc
2013-03-25 20:30:00 +01:00
case TX_8X8: {
const BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;
Make usage of sb_type independent of literal values. Change-Id: I0d12f9ef9d960df0172a1377f8e5236eb6d90492
2013-04-11 01:50:01 +02:00
const int sz = 3 + mb_width_log2(sb_type);
const int x = ib & ((1 << sz) - 1), y = ib - x;
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
TX_TYPE tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
get_tx_type_8x8(xd, y + (x >> 1)) : DCT_DCT;
make cost_coeffs to use combined context Change-Id: Ia15f4244595fab49bffda0c651a750a8a9481d28
2013-02-26 02:01:30 +01:00
a_ec = (a[0] + a[1]) != 0;
l_ec = (l[0] + l[1]) != 0;
Add col/row-based coefficient scanning patterns for 1D 8x8/16x16 ADSTs. These are mostly just for experimental purposes. I saw small gains (in the 0.1% range) when playing with this on derf. Change-Id: Ib21eed477bbb46bddcd73b21c5c708a5b46abedc
2013-03-25 20:30:00 +01:00
if (tx_type == ADST_DCT) {
scan = vp9_row_scan_8x8;
} else if (tx_type == DCT_ADST) {
scan = vp9_col_scan_8x8;
} else {
scan = vp9_default_zig_zag1d_8x8;
}
Framework changes in nzc to allow more flexibility The patch adds the flexibility to use standard EOB based coding on smaller block sizes and nzc based coding on larger blocksizes. The tx-sizes that use nzc based coding and those that use EOB based coding are controlled by a function get_nzc_used(). By default, this function uses nzc based coding for 16x16 and 32x32 transform blocks, which seem to bridge the performance gap substantially. All sets are now lower by 0.5% to 0.7%, as opposed to ~1.8% before. Change-Id: I06abed3df57b52d241ea1f51b0d571c71e38fd0b
2013-03-26 23:23:30 +01:00
coef_probs = cm->fc.coef_probs_8x8;
Use above/left (instead of previous in scan-order) as token context. Pearson correlation for above or left is significantly higher than for previous-in-scan-order (absolute values depend on position in scan, but in general, we gain about 0.1-0.2 by using either above or left; using both basically just makes this even better). For eob branch skipping, we continue to use the previous token in scan order. This helps about 0.9% on derf after re-training on a limited data set. Full re-training and results on larger-resolution clips are pending. Note that this commit breaks trellis, so we can probably get further gains out of it by fixing trellis at some later point. Change-Id: Iead68e296fc3a105cca746b5e3da9555d6010cfe
2013-03-27 00:46:09 +01:00
seg_eob = 64;
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
#if CONFIG_CODE_ZEROGROUP
zpc_probs = &cm->fc.zpc_probs_8x8;
#endif
Combine cost_coeffs functions. Change-Id: I3421d07fdcf2148d75c8dbfc8c44bc2d96a3a8f6
2012-08-06 21:15:24 +02:00
break;
Add col/row-based coefficient scanning patterns for 1D 8x8/16x16 ADSTs. These are mostly just for experimental purposes. I saw small gains (in the 0.1% range) when playing with this on derf. Change-Id: Ib21eed477bbb46bddcd73b21c5c708a5b46abedc
2013-03-25 20:30:00 +01:00
}
case TX_16X16: {
const BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;
Make usage of sb_type independent of literal values. Change-Id: I0d12f9ef9d960df0172a1377f8e5236eb6d90492
2013-04-11 01:50:01 +02:00
const int sz = 4 + mb_width_log2(sb_type);
const int x = ib & ((1 << sz) - 1), y = ib - x;
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
TX_TYPE tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
get_tx_type_16x16(xd, y + (x >> 2)) : DCT_DCT;
Add col/row-based coefficient scanning patterns for 1D 8x8/16x16 ADSTs. These are mostly just for experimental purposes. I saw small gains (in the 0.1% range) when playing with this on derf. Change-Id: Ib21eed477bbb46bddcd73b21c5c708a5b46abedc
2013-03-25 20:30:00 +01:00
if (tx_type == ADST_DCT) {
scan = vp9_row_scan_16x16;
} else if (tx_type == DCT_ADST) {
scan = vp9_col_scan_16x16;
} else {
scan = vp9_default_zig_zag1d_16x16;
}
Framework changes in nzc to allow more flexibility The patch adds the flexibility to use standard EOB based coding on smaller block sizes and nzc based coding on larger blocksizes. The tx-sizes that use nzc based coding and those that use EOB based coding are controlled by a function get_nzc_used(). By default, this function uses nzc based coding for 16x16 and 32x32 transform blocks, which seem to bridge the performance gap substantially. All sets are now lower by 0.5% to 0.7%, as opposed to ~1.8% before. Change-Id: I06abed3df57b52d241ea1f51b0d571c71e38fd0b
2013-03-26 23:23:30 +01:00
coef_probs = cm->fc.coef_probs_16x16;
Use above/left (instead of previous in scan-order) as token context. Pearson correlation for above or left is significantly higher than for previous-in-scan-order (absolute values depend on position in scan, but in general, we gain about 0.1-0.2 by using either above or left; using both basically just makes this even better). For eob branch skipping, we continue to use the previous token in scan order. This helps about 0.9% on derf after re-training on a limited data set. Full re-training and results on larger-resolution clips are pending. Note that this commit breaks trellis, so we can probably get further gains out of it by fixing trellis at some later point. Change-Id: Iead68e296fc3a105cca746b5e3da9555d6010cfe
2013-03-27 00:46:09 +01:00
seg_eob = 256;
make cost_coeffs() and tokenize_b() consistent Change-Id: I7cdb5c32a1400f88ec36d08ea982e38b77731602
2013-01-03 18:00:30 +01:00
if (type == PLANE_TYPE_UV) {
make cost_coeffs to use combined context Change-Id: Ia15f4244595fab49bffda0c651a750a8a9481d28
2013-02-26 02:01:30 +01:00
a_ec = (a[0] + a[1] + a1[0] + a1[1]) != 0;
l_ec = (l[0] + l[1] + l1[0] + l1[1]) != 0;
} else {
a_ec = (a[0] + a[1] + a[2] + a[3]) != 0;
l_ec = (l[0] + l[1] + l[2] + l[3]) != 0;
Some cleanups and fixes. Separates the logic on transform type selection previously spread out over a number of files into a separate function. Currently the tx_type field in b_mode_info is not used, but still left in there to eventually use for signaling the transform type in the bitstream. Also, now for tx_type = DCT_DCT, the regular integer DCT is used, as opposed to the floating point DCT used in conjuction with hybrid transform. Results change somewhat due to the transform change, but are within reasonable limits. The hd/std-hd sets are slightly up, while derf/yt are slightly down. Change-Id: I5776840c2239ca2da31ca6cfd7fd1148dc5f9e0f
2012-10-16 01:41:41 +02:00
}
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
#if CONFIG_CODE_ZEROGROUP
zpc_probs = &cm->fc.zpc_probs_16x16;
#endif
Combine cost_coeffs functions. Change-Id: I3421d07fdcf2148d75c8dbfc8c44bc2d96a3a8f6
2012-08-06 21:15:24 +02:00
break;
Add col/row-based coefficient scanning patterns for 1D 8x8/16x16 ADSTs. These are mostly just for experimental purposes. I saw small gains (in the 0.1% range) when playing with this on derf. Change-Id: Ib21eed477bbb46bddcd73b21c5c708a5b46abedc
2013-03-25 20:30:00 +01:00
}
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
case TX_32X32:
scan = vp9_default_zig_zag1d_32x32;
Framework changes in nzc to allow more flexibility The patch adds the flexibility to use standard EOB based coding on smaller block sizes and nzc based coding on larger blocksizes. The tx-sizes that use nzc based coding and those that use EOB based coding are controlled by a function get_nzc_used(). By default, this function uses nzc based coding for 16x16 and 32x32 transform blocks, which seem to bridge the performance gap substantially. All sets are now lower by 0.5% to 0.7%, as opposed to ~1.8% before. Change-Id: I06abed3df57b52d241ea1f51b0d571c71e38fd0b
2013-03-26 23:23:30 +01:00
coef_probs = cm->fc.coef_probs_32x32;
Use above/left (instead of previous in scan-order) as token context. Pearson correlation for above or left is significantly higher than for previous-in-scan-order (absolute values depend on position in scan, but in general, we gain about 0.1-0.2 by using either above or left; using both basically just makes this even better). For eob branch skipping, we continue to use the previous token in scan order. This helps about 0.9% on derf after re-training on a limited data set. Full re-training and results on larger-resolution clips are pending. Note that this commit breaks trellis, so we can probably get further gains out of it by fixing trellis at some later point. Change-Id: Iead68e296fc3a105cca746b5e3da9555d6010cfe
2013-03-27 00:46:09 +01:00
seg_eob = 1024;
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
if (type == PLANE_TYPE_UV) {
ENTROPY_CONTEXT *a2, *a3, *l2, *l3;
a2 = a1 + sizeof(ENTROPY_CONTEXT_PLANES) / sizeof(ENTROPY_CONTEXT);
a3 = a2 + sizeof(ENTROPY_CONTEXT_PLANES) / sizeof(ENTROPY_CONTEXT);
l2 = l1 + sizeof(ENTROPY_CONTEXT_PLANES) / sizeof(ENTROPY_CONTEXT);
l3 = l2 + sizeof(ENTROPY_CONTEXT_PLANES) / sizeof(ENTROPY_CONTEXT);
a_ec = (a[0] + a[1] + a1[0] + a1[1] +
a2[0] + a2[1] + a3[0] + a3[1]) != 0;
l_ec = (l[0] + l[1] + l1[0] + l1[1] +
l2[0] + l2[1] + l3[0] + l3[1]) != 0;
} else {
a_ec = (a[0] + a[1] + a[2] + a[3] +
a1[0] + a1[1] + a1[2] + a1[3]) != 0;
l_ec = (l[0] + l[1] + l[2] + l[3] +
l1[0] + l1[1] + l1[2] + l1[3]) != 0;
}
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
#if CONFIG_CODE_ZEROGROUP
zpc_probs = &cm->fc.zpc_probs_32x32;
#endif
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
break;
Combine cost_coeffs functions. Change-Id: I3421d07fdcf2148d75c8dbfc8c44bc2d96a3a8f6
2012-08-06 21:15:24 +02:00
default:
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
abort();
Combine cost_coeffs functions. Change-Id: I3421d07fdcf2148d75c8dbfc8c44bc2d96a3a8f6
2012-08-06 21:15:24 +02:00
break;
}
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
assert(eob <= seg_eob);
More refactoring of RD cost to make later changes easier. Change-Id: Ia6a290a6a53fdb88c971bea2de47ac73dcb7a364
2012-08-06 20:22:28 +02:00
Replacing VP9_COMBINEENTROPYCONTEXTS macro with function. Change-Id: I3bbc31840af69481e1d9bb4427c9ee25abf82946
2013-04-17 00:30:28 +02:00
pt = combine_entropy_contexts(a_ec, l_ec);
Use above/left (instead of previous in scan-order) as token context. Pearson correlation for above or left is significantly higher than for previous-in-scan-order (absolute values depend on position in scan, but in general, we gain about 0.1-0.2 by using either above or left; using both basically just makes this even better). For eob branch skipping, we continue to use the previous token in scan order. This helps about 0.9% on derf after re-training on a limited data set. Full re-training and results on larger-resolution clips are pending. Note that this commit breaks trellis, so we can probably get further gains out of it by fixing trellis at some later point. Change-Id: Iead68e296fc3a105cca746b5e3da9555d6010cfe
2013-03-27 00:46:09 +01:00
nb = vp9_get_coef_neighbors_handle(scan, &pad);
default_eob = seg_eob;
make cost_coeffs() and tokenize_b() consistent Change-Id: I7cdb5c32a1400f88ec36d08ea982e38b77731602
2013-01-03 18:00:30 +01:00
Removes the code_nonzerocount experiment This patch does not seem to give any benefits. Change-Id: I9d2b4091d6af3dfc0875f24db86c01e2de57f8db
2013-04-22 19:58:49 +02:00
if (vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP))
seg_eob = 0;
make cost_coeffs() and tokenize_b() consistent Change-Id: I7cdb5c32a1400f88ec36d08ea982e38b77731602
2013-01-03 18:00:30 +01:00
Bugfix in encode_inter_mb_segment_8x8 Fixes an indexing bug. Looks like the bug has been there for a while. Change-Id: I9fc04b0c30754bcb47366ad94a08112925600c4d
2013-04-04 20:07:19 +02:00
/* sanity check to ensure that we do not have spurious non-zero q values */
if (eob < seg_eob)
assert(qcoeff_ptr[scan[eob]] == 0);
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
#if CONFIG_CODE_ZEROGROUP
vpx_memset(token_cache_full, ZERO_TOKEN, sizeof(token_cache_full));
for (c = 0; c < eob; ++c) {
rc = scan[c];
token_cache_full[rc] = vp9_dct_value_tokens_ptr[qcoeff_ptr[rc]].token;
o = vp9_get_orientation(rc, tx_size);
if (qcoeff_ptr[rc] != 0)
last_nz_pos[o] = c;
}
#endif
Split coefficient token tables intra vs. inter. Change-Id: I5416455f8f129ca0f450d00e48358d2012605072
2013-02-19 22:36:38 +01:00
{
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
for (c = 0; c < eob; c++) {
Entropy coding for hybrid transform Separates the entropy coding context models for 4x4, 8x8 and 16x16 ADST variants. There is a small improvement for HD (hd/std-hd) by about 0.1-0.2%. Results on derf/yt are about the same, probably because there is not enough statistics. Results may improve somewhat once the initial probability tables are updated for the hybrid transforms which is coming soon. Change-Id: Ic7c0c62dacc68ef551054fdb575be8b8507d32a8
2012-09-10 07:42:35 +02:00
int v = qcoeff_ptr[scan[c]];
Fixing member names inside TOKENVALUE and TOKENEXTRA structs. Change-Id: I183ec5819d4d80966c92db36db75b8c3be0d381d
2013-04-19 01:18:08 +02:00
int t = vp9_dct_value_tokens_ptr[v].token;
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
int band = get_coef_band(scan, tx_size, c);
if (c)
pt = vp9_get_coef_context(scan, nb, pad, token_cache, c, default_eob);
#if CONFIG_CODE_ZEROGROUP
rc = scan[c];
o = vp9_get_orientation(rc, tx_size);
skip_coef_val = (token_cache[rc] == ZERO_TOKEN || is_eoo_list[o]);
if (!skip_coef_val) {
cost += token_costs[band][pt][t] + vp9_dct_value_cost_ptr[v];
} else {
assert(v == 0);
}
#else
cost += token_costs[band][pt][t] + vp9_dct_value_cost_ptr[v];
#endif
Removes the code_nonzerocount experiment This patch does not seem to give any benefits. Change-Id: I9d2b4091d6af3dfc0875f24db86c01e2de57f8db
2013-04-22 19:58:49 +02:00
if (!c || token_cache[scan[c - 1]])
cost += vp9_cost_bit(coef_probs[type][ref][band][pt][0], 1);
End of orientation zero group experiment Adds an experiment that codes an end-of-orientation symbol for every eligible zero encountered in scan order. This cleans out various other sub-experiments that were part of the origiinal patch, which will be later included if found useful. Results are slightly positive on all sets (0.1 - 0.2% range). Change-Id: I57765c605fefc7fb9d1b57f1b356843602abefaf
2013-03-28 18:42:23 +01:00
token_cache[scan[c]] = t;
#if CONFIG_CODE_ZEROGROUP
if (t == ZERO_TOKEN && !skip_coef_val) {
int eoo = 0, use_eoo;
#if USE_ZPC_EOORIENT == 1
use_eoo = vp9_use_eoo(c, seg_eob, scan, tx_size,
is_last_zero, is_eoo_list);
#else
use_eoo = 0;
#endif
if (use_eoo) {
eoo = vp9_is_eoo(c, eob, scan, tx_size, qcoeff_ptr, last_nz_pos);
if (eoo && is_eoo_negative[o]) eoo = 0;
if (eoo) {
int c_;
int savings = 0;
int zsaved = 0;
savings = vp9_cost_bit((*zpc_probs)[ref]
[coef_to_zpc_band(band)]
[coef_to_zpc_ptok(pt)][0], 1) -
vp9_cost_bit((*zpc_probs)[ref]
[coef_to_zpc_band(band)]
[coef_to_zpc_ptok(pt)][0], 0);
for (c_ = c + 1; c_ < eob; ++c_) {
if (o == vp9_get_orientation(scan[c_], tx_size)) {
int pt_ = vp9_get_coef_context(scan, nb, pad,
token_cache_full, c_,
default_eob);
int band_ = get_coef_band(scan, tx_size, c_);
assert(token_cache_full[scan[c_]] == ZERO_TOKEN);
if (!c_ || token_cache_full[scan[c_ - 1]])
savings += vp9_cost_bit(
coef_probs[type][ref][band_][pt_][0], 1);
savings += vp9_cost_bit(
coef_probs[type][ref][band_][pt_][1], 0);
zsaved++;
}
}
if (savings < 0) {
// if (zsaved < ZPC_ZEROSSAVED_EOO) {
eoo = 0;
is_eoo_negative[o] = 1;
}
}
}
if (use_eoo) {
cost += vp9_cost_bit((*zpc_probs)[ref]
[coef_to_zpc_band(band)]
[coef_to_zpc_ptok(pt)][0], !eoo);
if (eoo) {
assert(is_eoo_list[o] == 0);
is_eoo_list[o] = 1;
}
}
}
is_last_zero[o] = (t == ZERO_TOKEN);
Use above/left (instead of previous in scan-order) as token context. Pearson correlation for above or left is significantly higher than for previous-in-scan-order (absolute values depend on position in scan, but in general, we gain about 0.1-0.2 by using either above or left; using both basically just makes this even better). For eob branch skipping, we continue to use the previous token in scan order. This helps about 0.9% on derf after re-training on a limited data set. Full re-training and results on larger-resolution clips are pending. Note that this commit breaks trellis, so we can probably get further gains out of it by fixing trellis at some later point. Change-Id: Iead68e296fc3a105cca746b5e3da9555d6010cfe
2013-03-27 00:46:09 +01:00
#endif
Entropy coding for hybrid transform Separates the entropy coding context models for 4x4, 8x8 and 16x16 ADST variants. There is a small improvement for HD (hd/std-hd) by about 0.1-0.2%. Results on derf/yt are about the same, probably because there is not enough statistics. Results may improve somewhat once the initial probability tables are updated for the hybrid transforms which is coming soon. Change-Id: Ic7c0c62dacc68ef551054fdb575be8b8507d32a8
2012-09-10 07:42:35 +02:00
}
Removes the code_nonzerocount experiment This patch does not seem to give any benefits. Change-Id: I9d2b4091d6af3dfc0875f24db86c01e2de57f8db
2013-04-22 19:58:49 +02:00
if (c < seg_eob) {
if (c)
pt = vp9_get_coef_context(scan, nb, pad, token_cache, c, default_eob);
cost += mb->token_costs[tx_size][type][ref]
[get_coef_band(scan, tx_size, c)]
[pt][DCT_EOB_TOKEN];
}
More refactoring of RD cost to make later changes easier. Change-Id: Ia6a290a6a53fdb88c971bea2de47ac73dcb7a364
2012-08-06 20:22:28 +02:00
}
Removes the code_nonzerocount experiment This patch does not seem to give any benefits. Change-Id: I9d2b4091d6af3dfc0875f24db86c01e2de57f8db
2013-04-22 19:58:49 +02:00
// is eob first coefficient;
pt = (c > 0);
More refactoring of RD cost to make later changes easier. Change-Id: Ia6a290a6a53fdb88c971bea2de47ac73dcb7a364
2012-08-06 20:22:28 +02:00
*a = *l = pt;
make cost_coeffs to use combined context Change-Id: Ia15f4244595fab49bffda0c651a750a8a9481d28
2013-02-26 02:01:30 +01:00
if (tx_size >= TX_8X8) {
a[1] = l[1] = pt;
if (tx_size >= TX_16X16) {
if (type == PLANE_TYPE_UV) {
a1[0] = a1[1] = l1[0] = l1[1] = pt;
} else {
a[2] = a[3] = l[2] = l[3] = pt;
if (tx_size >= TX_32X32) {
a1[0] = a1[1] = a1[2] = a1[3] = pt;
l1[0] = l1[1] = l1[2] = l1[3] = pt;
}
}
}
}
More refactoring of RD cost to make later changes easier. Change-Id: Ia6a290a6a53fdb88c971bea2de47ac73dcb7a364
2012-08-06 20:22:28 +02:00
return cost;
}
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
static void choose_txfm_size_from_rd(VP9_COMP *cpi, MACROBLOCK *x,
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
int (*r)[2], int *rate,
int *d, int *distortion,
int *s, int *skip,
int64_t txfm_cache[NB_TXFM_MODES],
TX_SIZE max_txfm_size) {
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
vp9_prob skip_prob = vp9_get_pred_prob(cm, xd, PRED_MBSKIP);
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
int64_t rd[TX_SIZE_MAX_SB][2];
int n, m;
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
int s0, s1;
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
for (n = TX_4X4; n <= max_txfm_size; n++) {
r[n][1] = r[n][0];
for (m = 0; m <= n - (n == max_txfm_size); m++) {
if (m == n)
r[n][1] += vp9_cost_zero(cm->prob_tx[m]);
else
r[n][1] += vp9_cost_one(cm->prob_tx[m]);
}
}
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
assert(skip_prob > 0);
s0 = vp9_cost_bit(skip_prob, 0);
s1 = vp9_cost_bit(skip_prob, 1);
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
for (n = TX_4X4; n <= max_txfm_size; n++) {
if (s[n]) {
rd[n][0] = rd[n][1] = RDCOST(x->rdmult, x->rddiv, s1, d[n]);
} else {
rd[n][0] = RDCOST(x->rdmult, x->rddiv, r[n][0] + s0, d[n]);
rd[n][1] = RDCOST(x->rdmult, x->rddiv, r[n][1] + s0, d[n]);
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
}
}
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
if (max_txfm_size == TX_32X32 &&
(cm->txfm_mode == ALLOW_32X32 ||
(cm->txfm_mode == TX_MODE_SELECT &&
rd[TX_32X32][1] < rd[TX_16X16][1] && rd[TX_32X32][1] < rd[TX_8X8][1] &&
rd[TX_32X32][1] < rd[TX_4X4][1]))) {
mbmi->txfm_size = TX_32X32;
Merge tx32x32 experiment. Change-Id: I615651e4c7b09e576a341ad425cf80c393637833
2013-01-10 17:23:59 +01:00
} else if ( cm->txfm_mode == ALLOW_16X16 ||
(max_txfm_size == TX_16X16 && cm->txfm_mode == ALLOW_32X32) ||
(cm->txfm_mode == TX_MODE_SELECT &&
rd[TX_16X16][1] < rd[TX_8X8][1] &&
rd[TX_16X16][1] < rd[TX_4X4][1])) {
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
mbmi->txfm_size = TX_16X16;
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
} else if (cm->txfm_mode == ALLOW_8X8 ||
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
(cm->txfm_mode == TX_MODE_SELECT && rd[TX_8X8][1] < rd[TX_4X4][1])) {
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
mbmi->txfm_size = TX_8X8;
} else {
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
assert(cm->txfm_mode == ONLY_4X4 || cm->txfm_mode == TX_MODE_SELECT);
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
mbmi->txfm_size = TX_4X4;
}
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
*distortion = d[mbmi->txfm_size];
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
*rate = r[mbmi->txfm_size][cm->txfm_mode == TX_MODE_SELECT];
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
*skip = s[mbmi->txfm_size];
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
txfm_cache[ONLY_4X4] = rd[TX_4X4][0];
txfm_cache[ALLOW_8X8] = rd[TX_8X8][0];
txfm_cache[ALLOW_16X16] = rd[TX_16X16][0];
txfm_cache[ALLOW_32X32] = rd[max_txfm_size][0];
if (max_txfm_size == TX_32X32 &&
rd[TX_32X32][1] < rd[TX_16X16][1] && rd[TX_32X32][1] < rd[TX_8X8][1] &&
rd[TX_32X32][1] < rd[TX_4X4][1])
txfm_cache[TX_MODE_SELECT] = rd[TX_32X32][1];
Merge tx32x32 experiment. Change-Id: I615651e4c7b09e576a341ad425cf80c393637833
2013-01-10 17:23:59 +01:00
else if (rd[TX_16X16][1] < rd[TX_8X8][1] && rd[TX_16X16][1] < rd[TX_4X4][1])
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
txfm_cache[TX_MODE_SELECT] = rd[TX_16X16][1];
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
else
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
txfm_cache[TX_MODE_SELECT] = rd[TX_4X4][1] < rd[TX_8X8][1] ?
rd[TX_4X4][1] : rd[TX_8X8][1];
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
}
Use standard integer types for pixel values and coefficients. For coefficients, use int16_t (instead of short); for pixel values in 16-bit intermediates, use uint16_t (instead of unsigned short); for all others, use uint8_t (instead of unsigned char). Change-Id: I3619cd9abf106c3742eccc2e2f5e89a62774f7da
2012-12-19 00:31:19 +01:00
static int vp9_sb_block_error_c(int16_t *coeff, int16_t *dqcoeff,
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
int block_size, int shift) {
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
int i;
int64_t error = 0;
for (i = 0; i < block_size; i++) {
unsigned int this_diff = coeff[i] - dqcoeff[i];
error += this_diff * this_diff;
}
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
error >>= shift;
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
libvpx: Fix some warnings. Change-Id: If8be8b9d28a29631f29c46daea8a226ab3580610
2013-01-18 18:44:23 +01:00
return error > INT_MAX ? INT_MAX : (int)error;
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
}
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
static int vp9_sb_uv_block_error_c(int16_t *coeff,
int16_t *dqcoeff0, int16_t *dqcoeff1,
int block_size, int shift) {
int i;
int64_t error = 0;
for (i = 0; i < block_size / 2; i++) {
unsigned int this_diff = coeff[i] - dqcoeff0[i];
error += this_diff * this_diff;
}
coeff += block_size / 2;
for (i = 0; i < block_size / 2; i++) {
unsigned int this_diff = coeff[i] - dqcoeff1[i];
error += this_diff * this_diff;
}
error >>= shift;
return error > INT_MAX ? INT_MAX : (int)error;
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static int rdcost_sby_4x4(VP9_COMMON *const cm, MACROBLOCK *x,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) + 2, bw = 1 << bwl;
const int bh = 1 << (mb_height_log2(bsize) + 2);
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
int cost = 0, b;
MACROBLOCKD *const xd = &x->e_mbd;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
ENTROPY_CONTEXT_PLANES t_above[4], t_left[4];
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vpx_memcpy(&t_above, xd->above_context,
(sizeof(ENTROPY_CONTEXT_PLANES) * bw) >> 2);
vpx_memcpy(&t_left, xd->left_context,
(sizeof(ENTROPY_CONTEXT_PLANES) * bh) >> 2);
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
for (b = 0; b < bw * bh; b++) {
const int x_idx = b & (bw - 1), y_idx = b >> bwl;
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
cost += cost_coeffs(cm, x, b, PLANE_TYPE_Y_WITH_DC,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
((ENTROPY_CONTEXT *) &t_above[x_idx >> 2]) + (x_idx & 3),
((ENTROPY_CONTEXT *) &t_left[y_idx >> 2]) + (y_idx & 3),
TX_4X4, bw * bh);
}
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
return cost;
}
static void super_block_yrd_4x4(VP9_COMMON *const cm, MACROBLOCK *x,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
int *rate, int *distortion, int *skippable,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) + 2, bhl = mb_height_log2(bsize) + 2;
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
MACROBLOCKD *const xd = &x->e_mbd;
xd->mode_info_context->mbmi.txfm_size = TX_4X4;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vp9_transform_sby_4x4(x, bsize);
vp9_quantize_sby_4x4(x, bsize);
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
*distortion = vp9_sb_block_error_c(x->coeff, xd->plane[0].dqcoeff,
16 << (bwl + bhl), 2);
*rate = rdcost_sby_4x4(cm, x, bsize);
tokenize: convert skippable functions Use the common block walker to calculate skippability. Change-Id: I6721e42f065df237426c91c1d871ec226ba7cdcb
2013-04-11 20:14:31 +02:00
*skippable = vp9_sby_is_skippable(xd, bsize);
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static int rdcost_sby_8x8(VP9_COMMON *const cm, MACROBLOCK *x,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) + 1, bw = 1 << bwl;
const int bh = 1 << (mb_height_log2(bsize) + 1);
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
int cost = 0, b;
MACROBLOCKD *const xd = &x->e_mbd;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
ENTROPY_CONTEXT_PLANES t_above[4], t_left[4];
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vpx_memcpy(&t_above, xd->above_context,
(sizeof(ENTROPY_CONTEXT_PLANES) * bw) >> 1);
vpx_memcpy(&t_left, xd->left_context,
(sizeof(ENTROPY_CONTEXT_PLANES) * bh) >> 1);
for (b = 0; b < bw * bh; b++) {
const int x_idx = b & (bw - 1), y_idx = b >> bwl;
cost += cost_coeffs(cm, x, b * 4, PLANE_TYPE_Y_WITH_DC,
((ENTROPY_CONTEXT *) &t_above[x_idx >> 1]) + ((x_idx & 1) << 1),
((ENTROPY_CONTEXT *) &t_left[y_idx >> 1]) + ((y_idx & 1) << 1),
TX_8X8, 4 * bw * bh);
}
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
return cost;
}
static void super_block_yrd_8x8(VP9_COMMON *const cm, MACROBLOCK *x,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
int *rate, int *distortion, int *skippable,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) + 1, bhl = mb_height_log2(bsize) + 1;
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
MACROBLOCKD *const xd = &x->e_mbd;
xd->mode_info_context->mbmi.txfm_size = TX_8X8;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vp9_transform_sby_8x8(x, bsize);
vp9_quantize_sby_8x8(x, bsize);
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
*distortion = vp9_sb_block_error_c(x->coeff, xd->plane[0].dqcoeff,
64 << (bhl + bwl), 2);
*rate = rdcost_sby_8x8(cm, x, bsize);
tokenize: convert skippable functions Use the common block walker to calculate skippability. Change-Id: I6721e42f065df237426c91c1d871ec226ba7cdcb
2013-04-11 20:14:31 +02:00
*skippable = vp9_sby_is_skippable(xd, bsize);
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static int rdcost_sby_16x16(VP9_COMMON *const cm, MACROBLOCK *x,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize), bw = 1 << bwl;
const int bh = 1 << mb_height_log2(bsize);
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
int cost = 0, b;
MACROBLOCKD *const xd = &x->e_mbd;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
ENTROPY_CONTEXT_PLANES t_above[4], t_left[4];
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vpx_memcpy(&t_above, xd->above_context, sizeof(ENTROPY_CONTEXT_PLANES) * bw);
vpx_memcpy(&t_left, xd->left_context, sizeof(ENTROPY_CONTEXT_PLANES) * bh);
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
for (b = 0; b < bw * bh; b++) {
const int x_idx = b & (bw - 1), y_idx = b >> bwl;
cost += cost_coeffs(cm, x, b * 16, PLANE_TYPE_Y_WITH_DC,
(ENTROPY_CONTEXT *) &t_above[x_idx],
(ENTROPY_CONTEXT *) &t_left[y_idx],
TX_16X16, bw * bh * 16);
}
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
return cost;
}
static void super_block_yrd_16x16(VP9_COMMON *const cm, MACROBLOCK *x,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
int *rate, int *distortion, int *skippable,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize), bhl = mb_height_log2(bsize);
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
MACROBLOCKD *const xd = &x->e_mbd;
xd->mode_info_context->mbmi.txfm_size = TX_16X16;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vp9_transform_sby_16x16(x, bsize);
vp9_quantize_sby_16x16(x, bsize);
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
*distortion = vp9_sb_block_error_c(x->coeff, xd->plane[0].dqcoeff,
256 << (bwl + bhl), 2);
*rate = rdcost_sby_16x16(cm, x, bsize);
tokenize: convert skippable functions Use the common block walker to calculate skippability. Change-Id: I6721e42f065df237426c91c1d871ec226ba7cdcb
2013-04-11 20:14:31 +02:00
*skippable = vp9_sby_is_skippable(xd, bsize);
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static int rdcost_sby_32x32(VP9_COMMON *const cm, MACROBLOCK *x,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) - 1, bw = 1 << bwl;
const int bh = 1 << (mb_height_log2(bsize) - 1);
int cost = 0, b;
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
MACROBLOCKD * const xd = &x->e_mbd;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
ENTROPY_CONTEXT_PLANES t_above[4], t_left[4];
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vpx_memcpy(&t_above, xd->above_context,
sizeof(ENTROPY_CONTEXT_PLANES) * bw * 2);
vpx_memcpy(&t_left, xd->left_context,
sizeof(ENTROPY_CONTEXT_PLANES) * bh * 2);
for (b = 0; b < bw * bh; b++) {
const int x_idx = b & (bw - 1), y_idx = b >> bwl;
cost += cost_coeffs(cm, x, b * 64, PLANE_TYPE_Y_WITH_DC,
(ENTROPY_CONTEXT *) &t_above[x_idx * 2],
(ENTROPY_CONTEXT *) &t_left[y_idx * 2],
TX_32X32, bw * bh * 64);
}
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
return cost;
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
}
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
static void super_block_yrd_32x32(VP9_COMMON *const cm, MACROBLOCK *x,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
int *rate, int *distortion, int *skippable,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) - 1, bhl = mb_height_log2(bsize) - 1;
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
MACROBLOCKD *const xd = &x->e_mbd;
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
xd->mode_info_context->mbmi.txfm_size = TX_32X32;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vp9_transform_sby_32x32(x, bsize);
vp9_quantize_sby_32x32(x, bsize);
Removal of Hybrid DWT/DCT experiment. Removal of experiment to simplify code base for other changes. Change-Id: If0a33952504558511926ad212bc311fc2bffb19a
2013-02-11 13:35:28 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
*distortion = vp9_sb_block_error_c(x->coeff, xd->plane[0].dqcoeff,
1024 << (bwl + bhl), 0);
*rate = rdcost_sby_32x32(cm, x, bsize);
tokenize: convert skippable functions Use the common block walker to calculate skippability. Change-Id: I6721e42f065df237426c91c1d871ec226ba7cdcb
2013-04-11 20:14:31 +02:00
*skippable = vp9_sby_is_skippable(xd, bsize);
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
}
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
static void super_block_yrd(VP9_COMP *cpi,
MACROBLOCK *x, int *rate, int *distortion,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
int *skip, BLOCK_SIZE_TYPE bs,
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
int64_t txfm_cache[NB_TXFM_MODES]) {
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
VP9_COMMON *const cm = &cpi->common;
int r[TX_SIZE_MAX_SB][2], d[TX_SIZE_MAX_SB], s[TX_SIZE_MAX_SB];
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
vp9_subtract_sby(x, bs);
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
if (bs >= BLOCK_SIZE_SB32X32)
super_block_yrd_32x32(cm, x, &r[TX_32X32][0], &d[TX_32X32], &s[TX_32X32],
bs);
super_block_yrd_16x16(cm, x, &r[TX_16X16][0], &d[TX_16X16], &s[TX_16X16], bs);
super_block_yrd_8x8(cm, x, &r[TX_8X8][0], &d[TX_8X8], &s[TX_8X8], bs);
super_block_yrd_4x4(cm, x, &r[TX_4X4][0], &d[TX_4X4], &s[TX_4X4], bs);
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
choose_txfm_size_from_rd(cpi, x, r, rate, d, distortion, s, skip, txfm_cache,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
TX_32X32 - (bs < BLOCK_SIZE_SB32X32));
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
}
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
B_PREDICTION_MODE *best_mode,
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
int *bmode_costs,
ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l,
int *bestrate, int *bestratey,
int *bestdistortion) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
B_PREDICTION_MODE mode;
Some cleanups and fixes. Separates the logic on transform type selection previously spread out over a number of files into a separate function. Currently the tx_type field in b_mode_info is not used, but still left in there to eventually use for signaling the transform type in the bitstream. Also, now for tx_type = DCT_DCT, the regular integer DCT is used, as opposed to the floating point DCT used in conjuction with hybrid transform. Results change somewhat due to the transform change, but are within reasonable limits. The hd/std-hd sets are slightly up, while derf/yt are slightly down. Change-Id: I5776840c2239ca2da31ca6cfd7fd1148dc5f9e0f
2012-10-16 01:41:41 +02:00
MACROBLOCKD *xd = &x->e_mbd;
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t best_rd = INT64_MAX;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int rate = 0;
int distortion;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
VP9_COMMON *const cm = &cpi->common;
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
BLOCK *be = x->block + ib;
BLOCKD *b = xd->block + ib;
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
int16_t* const src_diff =
raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, ib,
x->plane[0].src_diff);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
ENTROPY_CONTEXT ta = *a, tempa = *a;
ENTROPY_CONTEXT tl = *l, templ = *l;
Some cleanups and fixes. Separates the logic on transform type selection previously spread out over a number of files into a separate function. Currently the tx_type field in b_mode_info is not used, but still left in there to eventually use for signaling the transform type in the bitstream. Also, now for tx_type = DCT_DCT, the regular integer DCT is used, as opposed to the floating point DCT used in conjuction with hybrid transform. Results change somewhat due to the transform change, but are within reasonable limits. The hd/std-hd sets are slightly up, while derf/yt are slightly down. Change-Id: I5776840c2239ca2da31ca6cfd7fd1148dc5f9e0f
2012-10-16 01:41:41 +02:00
TX_TYPE tx_type = DCT_DCT;
TX_TYPE best_tx_type = DCT_DCT;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/*
* The predictor buffer is a 2d buffer with a stride of 16. Create
* a temp buffer that meets the stride requirements, but we are only
* interested in the left 4x4 block
* */
Use standard integer types for pixel values and coefficients. For coefficients, use int16_t (instead of short); for pixel values in 16-bit intermediates, use uint16_t (instead of unsigned short); for all others, use uint8_t (instead of unsigned char). Change-Id: I3619cd9abf106c3742eccc2e2f5e89a62774f7da
2012-12-19 00:31:19 +01:00
DECLARE_ALIGNED_ARRAY(16, int16_t, best_dqcoeff, 16);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
assert(ib < 16);
New b-intra mode where direction is contextual Preliminary patch on a new 4x4 intra mode B_CONTEXT_PRED where the dominant direction from the context is used to encode. Various decoder changes are needed to support decoding of B_CONTEXT_PRED in conjunction with hybrid transforms since the scan order and tokenization depends on the actual direction of prediction obtained from the context. Currently the traditional directional modes are used in conjunction with the B_CONTEXT_PRED, which also seems to provide the best results. The gains are small - in the 0.1% range. Change-Id: I5a7ea80b5218f42a9c0dfb42d3f79a68c7f0cdc2
2012-10-09 22:19:15 +02:00
#if CONFIG_NEWBINTRAMODES
Context-pred fix to not use top/left on edges This fix resolves some of the mismatch issues being seen recently. While this is the right thing to do when tiling is used for this experiment, it is not the underlying cause of the the mismatches. Something else is causing writing outside of the allowable frame area in the encoder leading to this mismatch. Change-Id: If52c6f67555aa18ab8762865384e323b47237277
2013-03-16 17:26:52 +01:00
b->bmi.as_mode.context = vp9_find_bpred_context(xd, b);
New b-intra mode where direction is contextual Preliminary patch on a new 4x4 intra mode B_CONTEXT_PRED where the dominant direction from the context is used to encode. Various decoder changes are needed to support decoding of B_CONTEXT_PRED in conjunction with hybrid transforms since the scan order and tokenization depends on the actual direction of prediction obtained from the context. Currently the traditional directional modes are used in conjunction with the B_CONTEXT_PRED, which also seems to provide the best results. The gains are small - in the 0.1% range. Change-Id: I5a7ea80b5218f42a9c0dfb42d3f79a68c7f0cdc2
2012-10-09 22:19:15 +02:00
#endif
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
xd->mode_info_context->mbmi.txfm_size = TX_4X4;
New b-intra mode where direction is contextual Preliminary patch on a new 4x4 intra mode B_CONTEXT_PRED where the dominant direction from the context is used to encode. Various decoder changes are needed to support decoding of B_CONTEXT_PRED in conjunction with hybrid transforms since the scan order and tokenization depends on the actual direction of prediction obtained from the context. Currently the traditional directional modes are used in conjunction with the B_CONTEXT_PRED, which also seems to provide the best results. The gains are small - in the 0.1% range. Change-Id: I5a7ea80b5218f42a9c0dfb42d3f79a68c7f0cdc2
2012-10-09 22:19:15 +02:00
for (mode = B_DC_PRED; mode < LEFT4X4; mode++) {
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
int64_t this_rd;
int ratey;
New b-intra mode where direction is contextual Preliminary patch on a new 4x4 intra mode B_CONTEXT_PRED where the dominant direction from the context is used to encode. Various decoder changes are needed to support decoding of B_CONTEXT_PRED in conjunction with hybrid transforms since the scan order and tokenization depends on the actual direction of prediction obtained from the context. Currently the traditional directional modes are used in conjunction with the B_CONTEXT_PRED, which also seems to provide the best results. The gains are small - in the 0.1% range. Change-Id: I5a7ea80b5218f42a9c0dfb42d3f79a68c7f0cdc2
2012-10-09 22:19:15 +02:00
#if CONFIG_NEWBINTRAMODES
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
if (xd->frame_type == KEY_FRAME) {
if (mode == B_CONTEXT_PRED) continue;
} else {
if (mode >= B_CONTEXT_PRED - CONTEXT_PRED_REPLACEMENTS &&
mode < B_CONTEXT_PRED)
continue;
}
New b-intra mode where direction is contextual Preliminary patch on a new 4x4 intra mode B_CONTEXT_PRED where the dominant direction from the context is used to encode. Various decoder changes are needed to support decoding of B_CONTEXT_PRED in conjunction with hybrid transforms since the scan order and tokenization depends on the actual direction of prediction obtained from the context. Currently the traditional directional modes are used in conjunction with the B_CONTEXT_PRED, which also seems to provide the best results. The gains are small - in the 0.1% range. Change-Id: I5a7ea80b5218f42a9c0dfb42d3f79a68c7f0cdc2
2012-10-09 22:19:15 +02:00
#endif
Improved rd_pick_intra4x4block Eliminated unnecessary calculations. Improved performance by 10% on keyframes and 1.6% overall for the test clip used. Change-Id: I87671b26af5e2cc439e81d0fee3b15c7cd2a3309
2011-02-14 19:32:58 +01:00
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
b->bmi.as_mode.first = mode;
New b-intra mode where direction is contextual Preliminary patch on a new 4x4 intra mode B_CONTEXT_PRED where the dominant direction from the context is used to encode. Various decoder changes are needed to support decoding of B_CONTEXT_PRED in conjunction with hybrid transforms since the scan order and tokenization depends on the actual direction of prediction obtained from the context. Currently the traditional directional modes are used in conjunction with the B_CONTEXT_PRED, which also seems to provide the best results. The gains are small - in the 0.1% range. Change-Id: I5a7ea80b5218f42a9c0dfb42d3f79a68c7f0cdc2
2012-10-09 22:19:15 +02:00
#if CONFIG_NEWBINTRAMODES
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
rate = bmode_costs[
mode == B_CONTEXT_PRED ? mode - CONTEXT_PRED_REPLACEMENTS : mode];
New b-intra mode where direction is contextual Preliminary patch on a new 4x4 intra mode B_CONTEXT_PRED where the dominant direction from the context is used to encode. Various decoder changes are needed to support decoding of B_CONTEXT_PRED in conjunction with hybrid transforms since the scan order and tokenization depends on the actual direction of prediction obtained from the context. Currently the traditional directional modes are used in conjunction with the B_CONTEXT_PRED, which also seems to provide the best results. The gains are small - in the 0.1% range. Change-Id: I5a7ea80b5218f42a9c0dfb42d3f79a68c7f0cdc2
2012-10-09 22:19:15 +02:00
#else
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
rate = bmode_costs[mode];
New b-intra mode where direction is contextual Preliminary patch on a new 4x4 intra mode B_CONTEXT_PRED where the dominant direction from the context is used to encode. Various decoder changes are needed to support decoding of B_CONTEXT_PRED in conjunction with hybrid transforms since the scan order and tokenization depends on the actual direction of prediction obtained from the context. Currently the traditional directional modes are used in conjunction with the B_CONTEXT_PRED, which also seems to provide the best results. The gains are small - in the 0.1% range. Change-Id: I5a7ea80b5218f42a9c0dfb42d3f79a68c7f0cdc2
2012-10-09 22:19:15 +02:00
#endif
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
vp9_intra4x4_predict(xd, b, mode, *(b->base_dst) + b->dst, b->dst_stride);
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
vp9_subtract_block(4, 4, src_diff, 16,
*(be->base_src) + be->src, be->src_stride,
*(b->base_dst) + b->dst, b->dst_stride);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
b->bmi.as_mode.first = mode;
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
tx_type = get_tx_type_4x4(xd, be - x->block);
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
if (tx_type != DCT_DCT) {
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
vp9_short_fht4x4(src_diff, be->coeff, 16, tx_type);
Move eob from BLOCKD to MACROBLOCKD. Consistent with VP8. Change-Id: I8c316ee49f072e15abbb033a80e9c36617891f07
2013-02-27 19:00:24 +01:00
vp9_ht_quantize_b_4x4(x, be - x->block, tx_type);
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
} else {
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
x->fwd_txm4x4(src_diff, be->coeff, 32);
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
x->quantize_b_4x4(x, be - x->block, 16);
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
tempa = ta;
templ = tl;
Adds hybrid transform Adds ADST/DCT hybrid transform coding for Intra4x4 mode. The ADST is applied to directions in which the boundary pixels are used for prediction, while DCT applied to directions without corresponding boundary prediction. Adds enum TX_TYPE in b_mode_infor to indicate the transform type used. Make coding style consistent with google style. Fixed the commented issues. Experimental results in terms of bit-rate reduction: derf: 0.731% yt: 0.982% std-hd: 0.459% hd: 0.725% Will be looking at 8x8 transforms next. Change-Id: I46dbd7b80dbb3e8856e9c34fbc58cb3764a12fcf
2012-06-25 21:26:09 +02:00
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
ratey = cost_coeffs(cm, x, b - xd->block,
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
PLANE_TYPE_Y_WITH_DC, &tempa, &templ, TX_4X4, 16);
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
rate += ratey;
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
distortion = vp9_block_error(be->coeff,
BLOCK_OFFSET(xd->plane[0].dqcoeff, ib, 16),
16) >> 2;
Adds hybrid transform Adds ADST/DCT hybrid transform coding for Intra4x4 mode. The ADST is applied to directions in which the boundary pixels are used for prediction, while DCT applied to directions without corresponding boundary prediction. Adds enum TX_TYPE in b_mode_infor to indicate the transform type used. Make coding style consistent with google style. Fixed the commented issues. Experimental results in terms of bit-rate reduction: derf: 0.731% yt: 0.982% std-hd: 0.459% hd: 0.725% Will be looking at 8x8 transforms next. Change-Id: I46dbd7b80dbb3e8856e9c34fbc58cb3764a12fcf
2012-06-25 21:26:09 +02:00
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
this_rd = RDCOST(x->rdmult, x->rddiv, rate, distortion);
Adds hybrid transform Adds ADST/DCT hybrid transform coding for Intra4x4 mode. The ADST is applied to directions in which the boundary pixels are used for prediction, while DCT applied to directions without corresponding boundary prediction. Adds enum TX_TYPE in b_mode_infor to indicate the transform type used. Make coding style consistent with google style. Fixed the commented issues. Experimental results in terms of bit-rate reduction: derf: 0.731% yt: 0.982% std-hd: 0.459% hd: 0.725% Will be looking at 8x8 transforms next. Change-Id: I46dbd7b80dbb3e8856e9c34fbc58cb3764a12fcf
2012-06-25 21:26:09 +02:00
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
if (this_rd < best_rd) {
*bestrate = rate;
*bestratey = ratey;
*bestdistortion = distortion;
best_rd = this_rd;
*best_mode = mode;
best_tx_type = tx_type;
*a = tempa;
*l = templ;
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
vpx_memcpy(best_dqcoeff, BLOCK_OFFSET(xd->plane[0].dqcoeff, ib, 16), 32);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
b->bmi.as_mode.first = (B_PREDICTION_MODE)(*best_mode);
Improved rd_pick_intra4x4block Eliminated unnecessary calculations. Improved performance by 10% on keyframes and 1.6% overall for the test clip used. Change-Id: I87671b26af5e2cc439e81d0fee3b15c7cd2a3309
2011-02-14 19:32:58 +01:00
Adds hybrid transform Adds ADST/DCT hybrid transform coding for Intra4x4 mode. The ADST is applied to directions in which the boundary pixels are used for prediction, while DCT applied to directions without corresponding boundary prediction. Adds enum TX_TYPE in b_mode_infor to indicate the transform type used. Make coding style consistent with google style. Fixed the commented issues. Experimental results in terms of bit-rate reduction: derf: 0.731% yt: 0.982% std-hd: 0.459% hd: 0.725% Will be looking at 8x8 transforms next. Change-Id: I46dbd7b80dbb3e8856e9c34fbc58cb3764a12fcf
2012-06-25 21:26:09 +02:00
// inverse transform
Some cleanups and fixes. Separates the logic on transform type selection previously spread out over a number of files into a separate function. Currently the tx_type field in b_mode_info is not used, but still left in there to eventually use for signaling the transform type in the bitstream. Also, now for tx_type = DCT_DCT, the regular integer DCT is used, as opposed to the floating point DCT used in conjuction with hybrid transform. Results change somewhat due to the transform change, but are within reasonable limits. The hd/std-hd sets are slightly up, while derf/yt are slightly down. Change-Id: I5776840c2239ca2da31ca6cfd7fd1148dc5f9e0f
2012-10-16 01:41:41 +02:00
if (best_tx_type != DCT_DCT)
Changing pitch value meaning for fht and iht transforms. Pitch now means the number of elements, not the number of bytes. Change-Id: Idb9f2f012e39b09d596a3cc1802305a80b7c13af
2013-02-26 03:19:55 +01:00
vp9_short_iht4x4(best_dqcoeff, b->diff, 16, best_tx_type);
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
else
enable bitstream lossless support 1. Added a bit in frame header to to indicate if a frame is encoded in lossless mode, so decoder does not make the decision based on Q0 2. Minor changes to make sure that lossy coding works same as when the lossless experiment is not enabled. 3. Renamed function pointers for transforms to be consistent, using prefix fwd_txm and inv_txm for forward and inverse respectively To encode in lossless mode, using "--lossless=1 --min-q=0 --max-q=0" with vpxenc. Change-Id: Ifae53b26d2ffbe378d707e29d96817b8a5e6c068
2013-02-12 06:14:46 +01:00
xd->inv_txm4x4(best_dqcoeff, b->diff, 32);
Adds hybrid transform Adds ADST/DCT hybrid transform coding for Intra4x4 mode. The ADST is applied to directions in which the boundary pixels are used for prediction, while DCT applied to directions without corresponding boundary prediction. Adds enum TX_TYPE in b_mode_infor to indicate the transform type used. Make coding style consistent with google style. Fixed the commented issues. Experimental results in terms of bit-rate reduction: derf: 0.731% yt: 0.982% std-hd: 0.459% hd: 0.725% Will be looking at 8x8 transforms next. Change-Id: I46dbd7b80dbb3e8856e9c34fbc58cb3764a12fcf
2012-06-25 21:26:09 +02:00
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
vp9_intra4x4_predict(xd, b, *best_mode,
*(b->base_dst) + b->dst, b->dst_stride);
vp9_recon_b(*(b->base_dst) + b->dst, b->diff,
*(b->base_dst) + b->dst, b->dst_stride);
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
return best_rd;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
static int64_t rd_pick_intra4x4mby_modes(VP9_COMP *cpi, MACROBLOCK *mb,
int *Rate, int *rate_y,
Remove "update_context" variable from VP9_COMP context. The variable is always zero. Change-Id: Id5cdbecad543bca465a5b1d471badaec7e112c8d
2013-01-25 01:28:53 +01:00
int *Distortion, int64_t best_rd) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int i;
MACROBLOCKD *const xd = &mb->e_mbd;
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
int cost = mb->mbmode_cost[xd->frame_type][I4X4_PRED];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int distortion = 0;
int tot_rate_y = 0;
int64_t total_rd = 0;
ENTROPY_CONTEXT_PLANES t_above, t_left;
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
ENTROPY_CONTEXT *ta, *tl;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int *bmode_costs;
Remove "update_context" variable from VP9_COMP context. The variable is always zero. Change-Id: Id5cdbecad543bca465a5b1d471badaec7e112c8d
2013-01-25 01:28:53 +01:00
vpx_memcpy(&t_above, xd->above_context,
sizeof(ENTROPY_CONTEXT_PLANES));
vpx_memcpy(&t_left, xd->left_context,
sizeof(ENTROPY_CONTEXT_PLANES));
Changed above and left context data layout The main reason for the change was to reduce cycles in the token decoder. (~1.5% gain for 32 bit) This layout should be more cache friendly. As a result of this change, the encoder had to be updated. Change-Id: Id5e804169d8889da0378b3a519ac04dabd28c837 Note: dixie uses a similar layout
2010-08-31 16:49:57 +02:00
Remove "update_context" variable from VP9_COMP context. The variable is always zero. Change-Id: Id5cdbecad543bca465a5b1d471badaec7e112c8d
2013-01-25 01:28:53 +01:00
ta = (ENTROPY_CONTEXT *)&t_above;
tl = (ENTROPY_CONTEXT *)&t_left;
Initial WebM release
2010-05-18 17:58:33 +02:00
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
xd->mode_info_context->mbmi.mode = I4X4_PRED;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
bmode_costs = mb->inter_bmode_costs;
Improved rd_pick_intra4x4block Eliminated unnecessary calculations. Improved performance by 10% on keyframes and 1.6% overall for the test clip used. Change-Id: I87671b26af5e2cc439e81d0fee3b15c7cd2a3309
2011-02-14 19:32:58 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
for (i = 0; i < 16; i++) {
MODE_INFO *const mic = xd->mode_info_context;
const int mis = xd->mode_info_stride;
B_PREDICTION_MODE UNINITIALIZED_IS_SAFE(best_mode);
int UNINITIALIZED_IS_SAFE(r), UNINITIALIZED_IS_SAFE(ry), UNINITIALIZED_IS_SAFE(d);
Initial WebM release
2010-05-18 17:58:33 +02:00
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
if (xd->frame_type == KEY_FRAME) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
const B_PREDICTION_MODE A = above_block_mode(mic, i, mis);
const B_PREDICTION_MODE L = left_block_mode(mic, i);
Improved rd_pick_intra4x4block Eliminated unnecessary calculations. Improved performance by 10% on keyframes and 1.6% overall for the test clip used. Change-Id: I87671b26af5e2cc439e81d0fee3b15c7cd2a3309
2011-02-14 19:32:58 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
bmode_costs = mb->bmode_costs[A][L];
}
New b-intra mode where direction is contextual Preliminary patch on a new 4x4 intra mode B_CONTEXT_PRED where the dominant direction from the context is used to encode. Various decoder changes are needed to support decoding of B_CONTEXT_PRED in conjunction with hybrid transforms since the scan order and tokenization depends on the actual direction of prediction obtained from the context. Currently the traditional directional modes are used in conjunction with the B_CONTEXT_PRED, which also seems to provide the best results. The gains are small - in the 0.1% range. Change-Id: I5a7ea80b5218f42a9c0dfb42d3f79a68c7f0cdc2
2012-10-09 22:19:15 +02:00
#if CONFIG_NEWBINTRAMODES
Context-pred fix to not use top/left on edges This fix resolves some of the mismatch issues being seen recently. While this is the right thing to do when tiling is used for this experiment, it is not the underlying cause of the the mismatches. Something else is causing writing outside of the allowable frame area in the encoder leading to this mismatch. Change-Id: If52c6f67555aa18ab8762865384e323b47237277
2013-03-16 17:26:52 +01:00
mic->bmi[i].as_mode.context = vp9_find_bpred_context(xd, xd->block + i);
New b-intra mode where direction is contextual Preliminary patch on a new 4x4 intra mode B_CONTEXT_PRED where the dominant direction from the context is used to encode. Various decoder changes are needed to support decoding of B_CONTEXT_PRED in conjunction with hybrid transforms since the scan order and tokenization depends on the actual direction of prediction obtained from the context. Currently the traditional directional modes are used in conjunction with the B_CONTEXT_PRED, which also seems to provide the best results. The gains are small - in the 0.1% range. Change-Id: I5a7ea80b5218f42a9c0dfb42d3f79a68c7f0cdc2
2012-10-09 22:19:15 +02:00
#endif
Improved rd_pick_intra4x4block Eliminated unnecessary calculations. Improved performance by 10% on keyframes and 1.6% overall for the test clip used. Change-Id: I87671b26af5e2cc439e81d0fee3b15c7cd2a3309
2011-02-14 19:32:58 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
total_rd += rd_pick_intra4x4block(
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
cpi, mb, i, &best_mode,
clean up tokenize_b() and stuff_b() Change-Id: I0c1be01aae933243311ad321b6c456adaec1a0f5
2012-12-06 21:40:57 +01:00
bmode_costs, ta + vp9_block2above[TX_4X4][i],
tl + vp9_block2left[TX_4X4][i], &r, &ry, &d);
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
cost += r;
distortion += d;
tot_rate_y += ry;
MODE_INFO size reduction Declared the bmi in MODE_INFO as a union instead of B_MODE_INFO. This reduced the memory footprint by 518,400 bytes for 1080 resolutions. The decoder performance improved by ~4% for the clip used and the encoder showed very small improvements. (0.5%) This reduction was first mentioned to me by John K. and in a later discussion by Yaowu. This is WIP. Change-Id: I8e175fdbc46d28c35277302a04bee4540efc8d29
2011-05-24 19:24:52 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
mic->bmi[i].as_mode.first = best_mode;
Added early breakout for vp8_rd_pick_intra4x4mby_modes Improved performance of good quality, speed 0 (3% average) with no average quality loss. Change-Id: Ica34473f99bd74260eaebde6b132185e09e3c09d
2011-02-08 22:50:43 +01:00
New b-intra mode where direction is contextual Preliminary patch on a new 4x4 intra mode B_CONTEXT_PRED where the dominant direction from the context is used to encode. Various decoder changes are needed to support decoding of B_CONTEXT_PRED in conjunction with hybrid transforms since the scan order and tokenization depends on the actual direction of prediction obtained from the context. Currently the traditional directional modes are used in conjunction with the B_CONTEXT_PRED, which also seems to provide the best results. The gains are small - in the 0.1% range. Change-Id: I5a7ea80b5218f42a9c0dfb42d3f79a68c7f0cdc2
2012-10-09 22:19:15 +02:00
#if 0 // CONFIG_NEWBINTRAMODES
printf("%d %d\n", mic->bmi[i].as_mode.first, mic->bmi[i].as_mode.context);
#endif
Make RDCOST() results 64-bit. This prevents various integer overflows that several of us have been seeing in the past few days. Change-Id: If0020613a608e13d311b7838796c7ad53cb97255
2012-08-02 19:07:33 +02:00
if (total_rd >= best_rd)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
break;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Make RDCOST() results 64-bit. This prevents various integer overflows that several of us have been seeing in the past few days. Change-Id: If0020613a608e13d311b7838796c7ad53cb97255
2012-08-02 19:07:33 +02:00
if (total_rd >= best_rd)
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
return INT64_MAX;
Added early breakout for vp8_rd_pick_intra4x4mby_modes Improved performance of good quality, speed 0 (3% average) with no average quality loss. Change-Id: Ica34473f99bd74260eaebde6b132185e09e3c09d
2011-02-08 22:50:43 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
*Rate = cost;
rdopt: fix use of uninitialized value in addition rd_pick_intra4x4mby_modes / rd_pick_intra8x8mby_modes would both use the input value of 'rate_y' in the return calculation. In many places this value is uninitialized. Remove the unneeded sum. Change-Id: Icbd3df685303000301e69291c0ebc06f74bd548d
2012-11-05 21:50:16 +01:00
*rate_y = tot_rate_y;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
*Distortion = distortion;
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
return RDCOST(mb->rdmult, mb->rddiv, cost, distortion);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Move RD intra block mode selection to rdopt.c This change is analogous to I0b67dae1f8a74902378da7bdf565e39ab832dda7, which made the move for the non-RD path. Change-Id: If63fc1b0cd1eb7f932e710f83ff24d91454f8ed1
2011-06-08 18:05:05 +02:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static int64_t rd_pick_intra_sby_mode(VP9_COMP *cpi, MACROBLOCK *x,
int *rate, int *rate_tokenonly,
int *distortion, int *skippable,
BLOCK_SIZE_TYPE bsize,
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
int64_t txfm_cache[NB_TXFM_MODES]) {
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
MB_PREDICTION_MODE mode;
MB_PREDICTION_MODE UNINITIALIZED_IS_SAFE(mode_selected);
int this_rate, this_rate_tokenonly;
Add proper skip support to intra frames. Change-Id: I7524883fb29f42303fb46a5bc6772fbcf8781d1d
2012-08-29 19:43:20 +02:00
int this_distortion, s;
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t best_rd = INT64_MAX, this_rd;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
TX_SIZE UNINITIALIZED_IS_SAFE(best_tx);
int i;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
for (i = 0; i < NB_TXFM_MODES; i++)
txfm_cache[i] = INT64_MAX;
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
/* Y Search for 32x32 intra prediction mode */
for (mode = DC_PRED; mode <= TM_PRED; mode++) {
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
int64_t local_txfm_cache[NB_TXFM_MODES];
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
x->e_mbd.mode_info_context->mbmi.mode = mode;
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
vp9_build_intra_predictors_sby_s(&x->e_mbd, bsize);
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
super_block_yrd(cpi, x, &this_rate_tokenonly, &this_distortion, &s,
bsize, local_txfm_cache);
this_rate = this_rate_tokenonly + x->mbmode_cost[x->e_mbd.frame_type][mode];
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
this_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_distortion);
if (this_rd < best_rd) {
mode_selected = mode;
best_rd = this_rd;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
best_tx = x->e_mbd.mode_info_context->mbmi.txfm_size;
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
*rate = this_rate;
*rate_tokenonly = this_rate_tokenonly;
*distortion = this_distortion;
*skippable = s;
}
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
for (i = 0; i < NB_TXFM_MODES; i++) {
int64_t adj_rd = this_rd + local_txfm_cache[i] -
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
local_txfm_cache[cpi->common.txfm_mode];
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
if (adj_rd < txfm_cache[i]) {
txfm_cache[i] = adj_rd;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
x->e_mbd.mode_info_context->mbmi.mode = mode_selected;
x->e_mbd.mode_info_context->mbmi.txfm_size = best_tx;
hybrid transform of 16x16 dimension Enable ADST/DCT of dimension 16x16 for I16X16 modes. This change provides benefits mostly for hd sequences. Set up the framework for selectable transform dimension. Also allowing quantization parameter threshold to control the use of hybrid transform (This is currently disabled by setting threshold always above the quantization parameter. Adaptive thresholding can be built upon this, which will further improve the coding performance.) The coding performance gains (with respect to the codec that has all other configuration settings turned on) are derf: 0.013 yt: 0.086 hd: 0.198 std-hd: 0.501 Change-Id: Ibb4263a61fc74e0b3c345f54d73e8c73552bf926
2012-08-29 20:25:38 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
return best_rd;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Adds hybrid transform Adds ADST/DCT hybrid transform coding for Intra4x4 mode. The ADST is applied to directions in which the boundary pixels are used for prediction, while DCT applied to directions without corresponding boundary prediction. Adds enum TX_TYPE in b_mode_infor to indicate the transform type used. Make coding style consistent with google style. Fixed the commented issues. Experimental results in terms of bit-rate reduction: derf: 0.731% yt: 0.982% std-hd: 0.459% hd: 0.725% Will be looking at 8x8 transforms next. Change-Id: I46dbd7b80dbb3e8856e9c34fbc58cb3764a12fcf
2012-06-25 21:26:09 +02:00
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
static int64_t rd_pick_intra8x8block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
B_PREDICTION_MODE *best_mode,
int *mode_costs,
ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l,
int *bestrate, int *bestratey,
int *bestdistortion) {
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
VP9_COMMON *const cm = &cpi->common;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
MB_PREDICTION_MODE mode;
MACROBLOCKD *xd = &x->e_mbd;
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t best_rd = INT64_MAX;
Fixing 8x8/4x4 ADST for intra modes with tx select This patch allows use of 8x8 and 4x4 ADST correctly for Intra 16x16 modes and Intra 8x8 modes when the block size selected is smaller than the prediction mode. Also includes some cleanups and refactoring. Rebase. Change-Id: Ie3257bdf07bdb9c6e9476915e3a80183c8fa005a
2012-11-16 00:14:38 +01:00
int distortion = 0, rate = 0;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
BLOCK *be = x->block + ib;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
BLOCKD *b = xd->block + ib;
make cost_coeffs to use combined context Change-Id: Ia15f4244595fab49bffda0c651a750a8a9481d28
2013-02-26 02:01:30 +01:00
ENTROPY_CONTEXT_PLANES ta, tl;
ENTROPY_CONTEXT *ta0, *ta1, besta0 = 0, besta1 = 0;
ENTROPY_CONTEXT *tl0, *tl1, bestl0 = 0, bestl1 = 0;
Use 8x8 DCT transform for I8X8 prediction mode Apply 2D-DCT transform of dimension 8x8 to encode prediction residuals of I8X8 mode. Brought back block type 3 probability context model for 8x8 tokens, which is used for the coefficients of Y blocks in I8x8 modes. The coefficient costs estimate of I8X8 mode in rate-distortion is also changed appropriately. Performance results: derf: 0.246 yt: 0.114 std-hd: 0.730 hd: 0.670 Change-Id: If1d970eeb4e1827c9f0d2c5b27d33089b347ea27
2012-08-01 19:18:25 +02:00
// perform transformation of dimension 8x8
// note the input and output index mapping
int idx = (ib & 0x02) ? (ib + 2) : ib;
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
int16_t* const src_diff =
raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, ib,
x->plane[0].src_diff);
Use 8x8 DCT transform for I8X8 prediction mode Apply 2D-DCT transform of dimension 8x8 to encode prediction residuals of I8X8 mode. Brought back block type 3 probability context model for 8x8 tokens, which is used for the coefficients of Y blocks in I8x8 modes. The coefficient costs estimate of I8X8 mode in rate-distortion is also changed appropriately. Performance results: derf: 0.246 yt: 0.114 std-hd: 0.730 hd: 0.670 Change-Id: If1d970eeb4e1827c9f0d2c5b27d33089b347ea27
2012-08-01 19:18:25 +02:00
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
assert(ib < 16);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
for (mode = DC_PRED; mode <= TM_PRED; mode++) {
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
int64_t this_rd;
int rate_t = 0;
add 8x8 intra prediction modes Patch 1 to Patch 3 is an initial implementation of 8x8 intra prediction modes, here are with the following assumptions: a. 8x8 has 4 prediction modes DC, H, V and TM b. UV 4x4 block use the same mode as corresponding 8x8 area c. i8x8 modes are enabled for key frame only for now Patch 4: d. removed debug code from previous patches Patch 5: e. added stats code to collect entropy stats and further cleaned up Patch 6: f. changed mode stats code to collect finer stats of modes Patch 7: g. normalized i8x8 modes distribution to total at 256 (8bits). Patch 8: h. fixed a bug in decoder and removed debug printf output. Patch 9: i. more cleanups to address paul's comment Patch 10: j. messy rebase/merges to bring the commit up to date. Tests on HD clips encoded with all key frame showing consistent gain on all clips and all metrics:~0.5%(psnr) and 0.6%(ssim): http://www.corp.google.com/~yaowu/no_crawl/i8x8hd_allkey_fixedq.html To build and test, configure with: --enable-experimental --enable-i8x8 Change-Id: I9813fe07ae48cab5fdb5d904bca022514ad01e7f
2011-08-05 01:30:27 +02:00
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
// FIXME rate for compound mode and second intrapred mode
rate = mode_costs[mode];
b->bmi.as_mode.first = mode;
add 8x8 intra prediction modes Patch 1 to Patch 3 is an initial implementation of 8x8 intra prediction modes, here are with the following assumptions: a. 8x8 has 4 prediction modes DC, H, V and TM b. UV 4x4 block use the same mode as corresponding 8x8 area c. i8x8 modes are enabled for key frame only for now Patch 4: d. removed debug code from previous patches Patch 5: e. added stats code to collect entropy stats and further cleaned up Patch 6: f. changed mode stats code to collect finer stats of modes Patch 7: g. normalized i8x8 modes distribution to total at 256 (8bits). Patch 8: h. fixed a bug in decoder and removed debug printf output. Patch 9: i. more cleanups to address paul's comment Patch 10: j. messy rebase/merges to bring the commit up to date. Tests on HD clips encoded with all key frame showing consistent gain on all clips and all metrics:~0.5%(psnr) and 0.6%(ssim): http://www.corp.google.com/~yaowu/no_crawl/i8x8hd_allkey_fixedq.html To build and test, configure with: --enable-experimental --enable-i8x8 Change-Id: I9813fe07ae48cab5fdb5d904bca022514ad01e7f
2011-08-05 01:30:27 +02:00
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
vp9_intra8x8_predict(xd, b, mode, *(b->base_dst) + b->dst, b->dst_stride);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
vp9_subtract_block(8, 8, src_diff, 16,
*(be->base_src) + be->src, be->src_stride,
*(b->base_dst) + b->dst, b->dst_stride);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
if (xd->mode_info_context->mbmi.txfm_size == TX_8X8) {
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
TX_TYPE tx_type = get_tx_type_8x8(xd, ib);
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
if (tx_type != DCT_DCT)
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
vp9_short_fht8x8(src_diff, (x->block + idx)->coeff, 16, tx_type);
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
else
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
x->fwd_txm8x8(src_diff, (x->block + idx)->coeff, 32);
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
x->quantize_b_8x8(x, idx, tx_type, 16);
Separating 8x8DCT use from hybridtransform8x8 With this change, even if hybridtransform8x8 experiment is off, 8x8 dct is used for the I8x8 mode. However note that the gains observed with the hybridtransform8x8 experiment will now be less, since part of the gain is now merged in. Change-Id: I9afb3880906fd0a1368a374041fc08efcf060c54
2012-09-21 23:20:15 +02:00
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
// compute quantization mse of 8x8 block
distortion = vp9_block_error_c((x->block + idx)->coeff,
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16), 64);
make cost_coeffs to use combined context Change-Id: Ia15f4244595fab49bffda0c651a750a8a9481d28
2013-02-26 02:01:30 +01:00
vpx_memcpy(&ta, a, sizeof(ENTROPY_CONTEXT_PLANES));
vpx_memcpy(&tl, l, sizeof(ENTROPY_CONTEXT_PLANES));
ta0 = ((ENTROPY_CONTEXT*)&ta) + vp9_block2above[TX_8X8][idx];
tl0 = ((ENTROPY_CONTEXT*)&tl) + vp9_block2left[TX_8X8][idx];
ta1 = ta0 + 1;
tl1 = tl0 + 1;
Use 8x8 DCT transform for I8X8 prediction mode Apply 2D-DCT transform of dimension 8x8 to encode prediction residuals of I8X8 mode. Brought back block type 3 probability context model for 8x8 tokens, which is used for the coefficients of Y blocks in I8x8 modes. The coefficient costs estimate of I8X8 mode in rate-distortion is also changed appropriately. Performance results: derf: 0.246 yt: 0.114 std-hd: 0.730 hd: 0.670 Change-Id: If1d970eeb4e1827c9f0d2c5b27d33089b347ea27
2012-08-01 19:18:25 +02:00
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
rate_t = cost_coeffs(cm, x, idx, PLANE_TYPE_Y_WITH_DC,
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
ta0, tl0, TX_8X8, 16);
Use 8x8 DCT transform for I8X8 prediction mode Apply 2D-DCT transform of dimension 8x8 to encode prediction residuals of I8X8 mode. Brought back block type 3 probability context model for 8x8 tokens, which is used for the coefficients of Y blocks in I8x8 modes. The coefficient costs estimate of I8X8 mode in rate-distortion is also changed appropriately. Performance results: derf: 0.246 yt: 0.114 std-hd: 0.730 hd: 0.670 Change-Id: If1d970eeb4e1827c9f0d2c5b27d33089b347ea27
2012-08-01 19:18:25 +02:00
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
rate += rate_t;
} else {
static const int iblock[4] = {0, 1, 4, 5};
TX_TYPE tx_type;
int i;
make cost_coeffs to use combined context Change-Id: Ia15f4244595fab49bffda0c651a750a8a9481d28
2013-02-26 02:01:30 +01:00
vpx_memcpy(&ta, a, sizeof(ENTROPY_CONTEXT_PLANES));
vpx_memcpy(&tl, l, sizeof(ENTROPY_CONTEXT_PLANES));
ta0 = ((ENTROPY_CONTEXT*)&ta) + vp9_block2above[TX_4X4][ib];
tl0 = ((ENTROPY_CONTEXT*)&tl) + vp9_block2left[TX_4X4][ib];
ta1 = ta0 + 1;
tl1 = tl0 + 1;
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
distortion = 0;
rate_t = 0;
for (i = 0; i < 4; ++i) {
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
int16_t* const src_diff =
raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16,
0, ib + iblock[i],
x->plane[0].src_diff);
Use fdct8x4 instead of fdct4x4 where the block size allows it. This allows for faster SIMD implementations in the future (currently there is no speed impact). Change-Id: I732647e9148b5dcb44e6bc8728138f0141218329
2013-02-06 23:13:05 +01:00
int do_two = 0;
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
b = &xd->block[ib + iblock[i]];
be = &x->block[ib + iblock[i]];
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
tx_type = get_tx_type_4x4(xd, ib + iblock[i]);
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
if (tx_type != DCT_DCT) {
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
vp9_short_fht4x4(src_diff, be->coeff, 16, tx_type);
Move eob from BLOCKD to MACROBLOCKD. Consistent with VP8. Change-Id: I8c316ee49f072e15abbb033a80e9c36617891f07
2013-02-27 19:00:24 +01:00
vp9_ht_quantize_b_4x4(x, ib + iblock[i], tx_type);
Re-add support for ADST in superblocks. This also changes the RD search to take account of the correct block index when searching (this is required for ADST positioning to work correctly in combination with tx_select). Change-Id: Ie50d05b3a024a64ecd0b376887aa38ac5f7b6af6
2013-03-06 00:18:06 +01:00
} else if (!(i & 1) &&
get_tx_type_4x4(xd, ib + iblock[i] + 1) == DCT_DCT) {
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
x->fwd_txm8x4(src_diff, be->coeff, 32);
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
x->quantize_b_4x4_pair(x, ib + iblock[i], ib + iblock[i] + 1, 16);
Use fdct8x4 instead of fdct4x4 where the block size allows it. This allows for faster SIMD implementations in the future (currently there is no speed impact). Change-Id: I732647e9148b5dcb44e6bc8728138f0141218329
2013-02-06 23:13:05 +01:00
do_two = 1;
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
} else {
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
x->fwd_txm4x4(src_diff, be->coeff, 32);
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
x->quantize_b_4x4(x, ib + iblock[i], 16);
Fixing 8x8/4x4 ADST for intra modes with tx select This patch allows use of 8x8 and 4x4 ADST correctly for Intra 16x16 modes and Intra 8x8 modes when the block size selected is smaller than the prediction mode. Also includes some cleanups and refactoring. Rebase. Change-Id: Ie3257bdf07bdb9c6e9476915e3a80183c8fa005a
2012-11-16 00:14:38 +01:00
}
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
distortion += vp9_block_error_c(be->coeff,
BLOCK_OFFSET(xd->plane[0].dqcoeff, ib + iblock[i], 16),
16 << do_two);
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
rate_t += cost_coeffs(cm, x, ib + iblock[i], PLANE_TYPE_Y_WITH_DC,
make cost_coeffs to use combined context Change-Id: Ia15f4244595fab49bffda0c651a750a8a9481d28
2013-02-26 02:01:30 +01:00
i&1 ? ta1 : ta0, i&2 ? tl1 : tl0,
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
TX_4X4, 16);
Use fdct8x4 instead of fdct4x4 where the block size allows it. This allows for faster SIMD implementations in the future (currently there is no speed impact). Change-Id: I732647e9148b5dcb44e6bc8728138f0141218329
2013-02-06 23:13:05 +01:00
if (do_two) {
make cost_coeffs to use combined context Change-Id: Ia15f4244595fab49bffda0c651a750a8a9481d28
2013-02-26 02:01:30 +01:00
i++;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
rate_t += cost_coeffs(cm, x, ib + iblock[i], PLANE_TYPE_Y_WITH_DC,
make cost_coeffs to use combined context Change-Id: Ia15f4244595fab49bffda0c651a750a8a9481d28
2013-02-26 02:01:30 +01:00
i&1 ? ta1 : ta0, i&2 ? tl1 : tl0,
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
TX_4X4, 16);
Use fdct8x4 instead of fdct4x4 where the block size allows it. This allows for faster SIMD implementations in the future (currently there is no speed impact). Change-Id: I732647e9148b5dcb44e6bc8728138f0141218329
2013-02-06 23:13:05 +01:00
}
Use transform-size selection for i8x8_pred also. Change-Id: Iecb282fc89f9b5145ef31c5eda294ad42bc32a5d
2012-10-11 02:18:22 +02:00
}
Fix block pointer corruption in intra8x8 prediction with 4x4 transform. The RD loop would change the pointer after the first mode (DC) was tested, leading to corrupt block objects being provided for the others. This would essentially render the i8x8 predictor useless. Change-Id: I16c5906ca64fb34878ac32ce59af8974e4582bb8
2013-01-26 01:08:19 +01:00
b = &xd->block[ib];
be = &x->block[ib];
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
rate += rate_t;
}
Use 8x8 DCT transform for I8X8 prediction mode Apply 2D-DCT transform of dimension 8x8 to encode prediction residuals of I8X8 mode. Brought back block type 3 probability context model for 8x8 tokens, which is used for the coefficients of Y blocks in I8x8 modes. The coefficient costs estimate of I8X8 mode in rate-distortion is also changed appropriately. Performance results: derf: 0.246 yt: 0.114 std-hd: 0.730 hd: 0.670 Change-Id: If1d970eeb4e1827c9f0d2c5b27d33089b347ea27
2012-08-01 19:18:25 +02:00
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
distortion >>= 2;
this_rd = RDCOST(x->rdmult, x->rddiv, rate, distortion);
if (this_rd < best_rd) {
*bestrate = rate;
*bestratey = rate_t;
*bestdistortion = distortion;
make cost_coeffs to use combined context Change-Id: Ia15f4244595fab49bffda0c651a750a8a9481d28
2013-02-26 02:01:30 +01:00
besta0 = *ta0;
besta1 = *ta1;
bestl0 = *tl0;
bestl1 = *tl1;
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
best_rd = this_rd;
*best_mode = mode;
add 8x8 intra prediction modes Patch 1 to Patch 3 is an initial implementation of 8x8 intra prediction modes, here are with the following assumptions: a. 8x8 has 4 prediction modes DC, H, V and TM b. UV 4x4 block use the same mode as corresponding 8x8 area c. i8x8 modes are enabled for key frame only for now Patch 4: d. removed debug code from previous patches Patch 5: e. added stats code to collect entropy stats and further cleaned up Patch 6: f. changed mode stats code to collect finer stats of modes Patch 7: g. normalized i8x8 modes distribution to total at 256 (8bits). Patch 8: h. fixed a bug in decoder and removed debug printf output. Patch 9: i. more cleanups to address paul's comment Patch 10: j. messy rebase/merges to bring the commit up to date. Tests on HD clips encoded with all key frame showing consistent gain on all clips and all metrics:~0.5%(psnr) and 0.6%(ssim): http://www.corp.google.com/~yaowu/no_crawl/i8x8hd_allkey_fixedq.html To build and test, configure with: --enable-experimental --enable-i8x8 Change-Id: I9813fe07ae48cab5fdb5d904bca022514ad01e7f
2011-08-05 01:30:27 +02:00
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
b->bmi.as_mode.first = (*best_mode);
removed the idct rtcd idct calls More cleanup to do after this, but this is a good chunk of removing rtcd. Change-Id: I551db75e341a0a85c3ad650df1e9a60dc305681a
2012-11-25 04:33:58 +01:00
vp9_encode_intra8x8(x, ib);
Use 8x8 DCT transform for I8X8 prediction mode Apply 2D-DCT transform of dimension 8x8 to encode prediction residuals of I8X8 mode. Brought back block type 3 probability context model for 8x8 tokens, which is used for the coefficients of Y blocks in I8x8 modes. The coefficient costs estimate of I8X8 mode in rate-distortion is also changed appropriately. Performance results: derf: 0.246 yt: 0.114 std-hd: 0.730 hd: 0.670 Change-Id: If1d970eeb4e1827c9f0d2c5b27d33089b347ea27
2012-08-01 19:18:25 +02:00
Use transform-size selection for i8x8_pred also. Change-Id: Iecb282fc89f9b5145ef31c5eda294ad42bc32a5d
2012-10-11 02:18:22 +02:00
if (xd->mode_info_context->mbmi.txfm_size == TX_8X8) {
clean up tokenize_b() and stuff_b() Change-Id: I0c1be01aae933243311ad321b6c456adaec1a0f5
2012-12-06 21:40:57 +01:00
a[vp9_block2above[TX_8X8][idx]] = besta0;
a[vp9_block2above[TX_8X8][idx] + 1] = besta1;
l[vp9_block2left[TX_8X8][idx]] = bestl0;
l[vp9_block2left[TX_8X8][idx] + 1] = bestl1;
Use transform-size selection for i8x8_pred also. Change-Id: Iecb282fc89f9b5145ef31c5eda294ad42bc32a5d
2012-10-11 02:18:22 +02:00
} else {
clean up tokenize_b() and stuff_b() Change-Id: I0c1be01aae933243311ad321b6c456adaec1a0f5
2012-12-06 21:40:57 +01:00
a[vp9_block2above[TX_4X4][ib]] = besta0;
a[vp9_block2above[TX_4X4][ib + 1]] = besta1;
l[vp9_block2left[TX_4X4][ib]] = bestl0;
l[vp9_block2left[TX_4X4][ib + 4]] = bestl1;
Use transform-size selection for i8x8_pred also. Change-Id: Iecb282fc89f9b5145ef31c5eda294ad42bc32a5d
2012-10-11 02:18:22 +02:00
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
return best_rd;
add 8x8 intra prediction modes Patch 1 to Patch 3 is an initial implementation of 8x8 intra prediction modes, here are with the following assumptions: a. 8x8 has 4 prediction modes DC, H, V and TM b. UV 4x4 block use the same mode as corresponding 8x8 area c. i8x8 modes are enabled for key frame only for now Patch 4: d. removed debug code from previous patches Patch 5: e. added stats code to collect entropy stats and further cleaned up Patch 6: f. changed mode stats code to collect finer stats of modes Patch 7: g. normalized i8x8 modes distribution to total at 256 (8bits). Patch 8: h. fixed a bug in decoder and removed debug printf output. Patch 9: i. more cleanups to address paul's comment Patch 10: j. messy rebase/merges to bring the commit up to date. Tests on HD clips encoded with all key frame showing consistent gain on all clips and all metrics:~0.5%(psnr) and 0.6%(ssim): http://www.corp.google.com/~yaowu/no_crawl/i8x8hd_allkey_fixedq.html To build and test, configure with: --enable-experimental --enable-i8x8 Change-Id: I9813fe07ae48cab5fdb5d904bca022514ad01e7f
2011-08-05 01:30:27 +02:00
}
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
static int64_t rd_pick_intra8x8mby_modes(VP9_COMP *cpi, MACROBLOCK *mb,
rdopt: make locally used symbols static. Change-Id: Id33c448084ca5c1c0525a05c3c4f35d6b0805fbe
2012-10-30 02:04:33 +01:00
int *Rate, int *rate_y,
int *Distortion, int64_t best_rd) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
MACROBLOCKD *const xd = &mb->e_mbd;
int i, ib;
int cost = mb->mbmode_cost [xd->frame_type] [I8X8_PRED];
int distortion = 0;
int tot_rate_y = 0;
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t total_rd = 0;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
ENTROPY_CONTEXT_PLANES t_above, t_left;
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
ENTROPY_CONTEXT *ta, *tl;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int *i8x8mode_costs;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
vpx_memcpy(&t_above, xd->above_context, sizeof(ENTROPY_CONTEXT_PLANES));
vpx_memcpy(&t_left, xd->left_context, sizeof(ENTROPY_CONTEXT_PLANES));
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
ta = (ENTROPY_CONTEXT *)&t_above;
tl = (ENTROPY_CONTEXT *)&t_left;
Remove mode_rdopt from MB_MODE_INFO. The variable is essentially a duplicate of mode for RD-only purposes. Removing it gives identical results, and saves 4 bytes per macroblock (i.e. 32.5kB for a 1080p HD video encode). Change-Id: I22d5058fdb80ab0b69862caee825e9d86bb148b3
2012-10-15 22:49:45 +02:00
xd->mode_info_context->mbmi.mode = I8X8_PRED;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
i8x8mode_costs = mb->i8x8_mode_costs;
for (i = 0; i < 4; i++) {
MODE_INFO *const mic = xd->mode_info_context;
B_PREDICTION_MODE UNINITIALIZED_IS_SAFE(best_mode);
int UNINITIALIZED_IS_SAFE(r), UNINITIALIZED_IS_SAFE(ry), UNINITIALIZED_IS_SAFE(d);
add 8x8 intra prediction modes Patch 1 to Patch 3 is an initial implementation of 8x8 intra prediction modes, here are with the following assumptions: a. 8x8 has 4 prediction modes DC, H, V and TM b. UV 4x4 block use the same mode as corresponding 8x8 area c. i8x8 modes are enabled for key frame only for now Patch 4: d. removed debug code from previous patches Patch 5: e. added stats code to collect entropy stats and further cleaned up Patch 6: f. changed mode stats code to collect finer stats of modes Patch 7: g. normalized i8x8 modes distribution to total at 256 (8bits). Patch 8: h. fixed a bug in decoder and removed debug printf output. Patch 9: i. more cleanups to address paul's comment Patch 10: j. messy rebase/merges to bring the commit up to date. Tests on HD clips encoded with all key frame showing consistent gain on all clips and all metrics:~0.5%(psnr) and 0.6%(ssim): http://www.corp.google.com/~yaowu/no_crawl/i8x8hd_allkey_fixedq.html To build and test, configure with: --enable-experimental --enable-i8x8 Change-Id: I9813fe07ae48cab5fdb5d904bca022514ad01e7f
2011-08-05 01:30:27 +02:00
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
ib = vp9_i8x8_block[i];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
total_rd += rd_pick_intra8x8block(
cpi, mb, ib, &best_mode,
i8x8mode_costs, ta, tl, &r, &ry, &d);
cost += r;
distortion += d;
tot_rate_y += ry;
mic->bmi[ib].as_mode.first = best_mode;
}
Problem of over smoothing with intra modes. In some cases intra modes in inter frames give an over smoothed appearance. Especially with noisy but flat content. Also in some cases there were problems with key frame sizing again with very flat but noisy content. These are temporary changes to help alleviate the visual problems but will almost certainly hurt metric results especially at the very low data rate end. Change-Id: I11549179a19277ffc283d9788bc70168f2a8bdc9
2012-12-14 18:49:46 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
*Rate = cost;
rdopt: fix use of uninitialized value in addition rd_pick_intra4x4mby_modes / rd_pick_intra8x8mby_modes would both use the input value of 'rate_y' in the return calculation. In many places this value is uninitialized. Remove the unneeded sum. Change-Id: Icbd3df685303000301e69291c0ebc06f74bd548d
2012-11-05 21:50:16 +01:00
*rate_y = tot_rate_y;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
*Distortion = distortion;
return RDCOST(mb->rdmult, mb->rddiv, cost, distortion);
add 8x8 intra prediction modes Patch 1 to Patch 3 is an initial implementation of 8x8 intra prediction modes, here are with the following assumptions: a. 8x8 has 4 prediction modes DC, H, V and TM b. UV 4x4 block use the same mode as corresponding 8x8 area c. i8x8 modes are enabled for key frame only for now Patch 4: d. removed debug code from previous patches Patch 5: e. added stats code to collect entropy stats and further cleaned up Patch 6: f. changed mode stats code to collect finer stats of modes Patch 7: g. normalized i8x8 modes distribution to total at 256 (8bits). Patch 8: h. fixed a bug in decoder and removed debug printf output. Patch 9: i. more cleanups to address paul's comment Patch 10: j. messy rebase/merges to bring the commit up to date. Tests on HD clips encoded with all key frame showing consistent gain on all clips and all metrics:~0.5%(psnr) and 0.6%(ssim): http://www.corp.google.com/~yaowu/no_crawl/i8x8hd_allkey_fixedq.html To build and test, configure with: --enable-experimental --enable-i8x8 Change-Id: I9813fe07ae48cab5fdb5d904bca022514ad01e7f
2011-08-05 01:30:27 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Add support for tx_select in i8x8 encoding in keyframes. Also enable tx_select for keyframes. Change-Id: Iadb1231d9fa7af0c8dce3d9b41830b93a302479e
2013-03-07 19:45:25 +01:00
static int64_t rd_pick_intra8x8mby_modes_and_txsz(VP9_COMP *cpi, MACROBLOCK *x,
int *rate, int *rate_y,
int *distortion,
int *mode8x8,
int64_t best_yrd,
int64_t *txfm_cache) {
VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
int cost0 = vp9_cost_bit(cm->prob_tx[0], 0);
int cost1 = vp9_cost_bit(cm->prob_tx[0], 1);
int64_t tmp_rd_4x4s, tmp_rd_8x8s;
int64_t tmp_rd_4x4, tmp_rd_8x8, tmp_rd;
int r4x4, tok4x4, d4x4, r8x8, tok8x8, d8x8;
mbmi->txfm_size = TX_4X4;
tmp_rd_4x4 = rd_pick_intra8x8mby_modes(cpi, x, &r4x4, &tok4x4,
&d4x4, best_yrd);
mode8x8[0] = xd->mode_info_context->bmi[0].as_mode.first;
mode8x8[1] = xd->mode_info_context->bmi[2].as_mode.first;
mode8x8[2] = xd->mode_info_context->bmi[8].as_mode.first;
mode8x8[3] = xd->mode_info_context->bmi[10].as_mode.first;
mbmi->txfm_size = TX_8X8;
tmp_rd_8x8 = rd_pick_intra8x8mby_modes(cpi, x, &r8x8, &tok8x8,
&d8x8, best_yrd);
txfm_cache[ONLY_4X4] = tmp_rd_4x4;
txfm_cache[ALLOW_8X8] = tmp_rd_8x8;
txfm_cache[ALLOW_16X16] = tmp_rd_8x8;
tmp_rd_4x4s = tmp_rd_4x4 + RDCOST(x->rdmult, x->rddiv, cost0, 0);
tmp_rd_8x8s = tmp_rd_8x8 + RDCOST(x->rdmult, x->rddiv, cost1, 0);
txfm_cache[TX_MODE_SELECT] = tmp_rd_4x4s < tmp_rd_8x8s ?
tmp_rd_4x4s : tmp_rd_8x8s;
if (cm->txfm_mode == TX_MODE_SELECT) {
if (tmp_rd_4x4s < tmp_rd_8x8s) {
*rate = r4x4 + cost0;
*rate_y = tok4x4 + cost0;
*distortion = d4x4;
mbmi->txfm_size = TX_4X4;
tmp_rd = tmp_rd_4x4s;
} else {
*rate = r8x8 + cost1;
*rate_y = tok8x8 + cost1;
*distortion = d8x8;
mbmi->txfm_size = TX_8X8;
tmp_rd = tmp_rd_8x8s;
mode8x8[0] = xd->mode_info_context->bmi[0].as_mode.first;
mode8x8[1] = xd->mode_info_context->bmi[2].as_mode.first;
mode8x8[2] = xd->mode_info_context->bmi[8].as_mode.first;
mode8x8[3] = xd->mode_info_context->bmi[10].as_mode.first;
}
} else if (cm->txfm_mode == ONLY_4X4) {
*rate = r4x4;
*rate_y = tok4x4;
*distortion = d4x4;
mbmi->txfm_size = TX_4X4;
tmp_rd = tmp_rd_4x4;
} else {
*rate = r8x8;
*rate_y = tok8x8;
*distortion = d8x8;
mbmi->txfm_size = TX_8X8;
tmp_rd = tmp_rd_8x8;
mode8x8[0] = xd->mode_info_context->bmi[0].as_mode.first;
mode8x8[1] = xd->mode_info_context->bmi[2].as_mode.first;
mode8x8[2] = xd->mode_info_context->bmi[8].as_mode.first;
mode8x8[3] = xd->mode_info_context->bmi[10].as_mode.first;
}
return tmp_rd;
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
#define UVCTX(c, p) ((p) ? (c).v : (c).u)
static int rd_cost_sbuv_4x4(VP9_COMMON *const cm, MACROBLOCK *x,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) + 1, bw = 1 << bwl;
const int bh = 1 << (mb_height_log2(bsize) + 1);
int yoff = 4 * bw * bh;
int p, b, cost = 0;
MACROBLOCKD *const xd = &x->e_mbd;
ENTROPY_CONTEXT_PLANES t_above[4], t_left[4];
revised the rate distortion computation for UV this commit changed the UV r/d calculation in the mode decision process to properly account for the rate of 8x8 transform coefficients. Change-Id: I485f8f35f2b61db0b6539beb32e83481b1cf083b
2012-01-21 00:30:31 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vpx_memcpy(&t_above, xd->above_context,
(sizeof(ENTROPY_CONTEXT_PLANES) * bw) >> 1);
vpx_memcpy(&t_left, xd->left_context,
(sizeof(ENTROPY_CONTEXT_PLANES) * bh) >> 1);
for (p = 0; p < 2; p++) {
for (b = 0; b < bw * bh; b++) {
const int x_idx = b & (bw - 1), y_idx = b >> bwl;
cost += cost_coeffs(cm, x, yoff + b, PLANE_TYPE_UV,
UVCTX(t_above[x_idx >> 1], p) + (x_idx & 1),
UVCTX(t_left[y_idx >> 1], p) + (y_idx & 1),
TX_4X4, bw * bh * 4);
}
yoff = (yoff * 5) >> 2; // u -> v
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
revised the rate distortion computation for UV this commit changed the UV r/d calculation in the mode decision process to properly account for the rate of 8x8 transform coefficients. Change-Id: I485f8f35f2b61db0b6539beb32e83481b1cf083b
2012-01-21 00:30:31 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
return cost;
revised the rate distortion computation for UV this commit changed the UV r/d calculation in the mode decision process to properly account for the rate of 8x8 transform coefficients. Change-Id: I485f8f35f2b61db0b6539beb32e83481b1cf083b
2012-01-21 00:30:31 +01:00
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static void super_block_uvrd_4x4(VP9_COMMON *const cm, MACROBLOCK *x,
int *rate, int *distortion, int *skip,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) + 2, bhl = mb_height_log2(bsize) + 2;
MACROBLOCKD *const xd = &x->e_mbd;
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vp9_transform_sbuv_4x4(x, bsize);
vp9_quantize_sbuv_4x4(x, bsize);
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
*rate = rd_cost_sbuv_4x4(cm, x, bsize);
*distortion = vp9_sb_uv_block_error_c(x->coeff + (16 << (bwl + bhl)),
xd->plane[1].dqcoeff,
xd->plane[2].dqcoeff,
32 << (bwl + bhl - 2), 2);
tokenize: convert skippable functions Use the common block walker to calculate skippability. Change-Id: I6721e42f065df237426c91c1d871ec226ba7cdcb
2013-04-11 20:14:31 +02:00
*skip = vp9_sbuv_is_skippable(xd, bsize);
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static int rd_cost_sbuv_8x8(VP9_COMMON *const cm, MACROBLOCK *x,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize), bw = 1 << bwl;
const int bh = 1 << mb_height_log2(bsize);
int yoff = 16 * bw * bh;
int p, b, cost = 0;
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
MACROBLOCKD *const xd = &x->e_mbd;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
ENTROPY_CONTEXT_PLANES t_above[4], t_left[4];
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vpx_memcpy(&t_above, xd->above_context,
sizeof(ENTROPY_CONTEXT_PLANES) * bw);
vpx_memcpy(&t_left, xd->left_context,
sizeof(ENTROPY_CONTEXT_PLANES) * bh);
for (p = 0; p < 2; p++) {
for (b = 0; b < bw * bh; b++) {
const int x_idx = b & (bw - 1), y_idx = b >> bwl;
cost += cost_coeffs(cm, x, yoff + b * 4, PLANE_TYPE_UV,
UVCTX(t_above[x_idx], p),
UVCTX(t_left[y_idx], p),
TX_8X8, bw * bh * 16);
}
yoff = (yoff * 5) >> 2; // u -> v
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
}
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
return cost;
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static void super_block_uvrd_8x8(VP9_COMMON *const cm, MACROBLOCK *x,
int *rate, int *distortion, int *skip,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) + 1, bhl = mb_height_log2(bsize) + 1;
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
MACROBLOCKD *const xd = &x->e_mbd;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vp9_transform_sbuv_8x8(x, bsize);
vp9_quantize_sbuv_8x8(x, bsize);
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
*rate = rd_cost_sbuv_8x8(cm, x, bsize);
*distortion = vp9_sb_uv_block_error_c(x->coeff + (64 << (bwl + bhl)),
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
xd->plane[1].dqcoeff,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
xd->plane[2].dqcoeff,
128 << (bwl + bhl - 2), 2);
tokenize: convert skippable functions Use the common block walker to calculate skippability. Change-Id: I6721e42f065df237426c91c1d871ec226ba7cdcb
2013-04-11 20:14:31 +02:00
*skip = vp9_sbuv_is_skippable(xd, bsize);
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
}
revised the rate distortion computation for UV this commit changed the UV r/d calculation in the mode decision process to properly account for the rate of 8x8 transform coefficients. Change-Id: I485f8f35f2b61db0b6539beb32e83481b1cf083b
2012-01-21 00:30:31 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static int rd_cost_sbuv_16x16(VP9_COMMON *const cm, MACROBLOCK *x,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) - 1, bw = 1 << bwl;
const int bh = 1 << (mb_height_log2(bsize) - 1);
int yoff = 64 * bw * bh;
int p, b, cost = 0;
MACROBLOCKD *const xd = &x->e_mbd;
ENTROPY_CONTEXT_PLANES t_above[4], t_left[4];
Initial WebM release
2010-05-18 17:58:33 +02:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vpx_memcpy(&t_above, xd->above_context,
sizeof(ENTROPY_CONTEXT_PLANES) * 2 * bw);
vpx_memcpy(&t_left, xd->left_context,
sizeof(ENTROPY_CONTEXT_PLANES) * 2 * bh);
for (p = 0; p < 2; p++) {
for (b = 0; b < bw * bh; b++) {
const int x_idx = b & (bw - 1), y_idx = b >> bwl;
cost += cost_coeffs(cm, x, yoff + b * 16, PLANE_TYPE_UV,
UVCTX(t_above[x_idx * 2], p),
UVCTX(t_left[y_idx * 2], p),
TX_16X16, bw * bh * 64);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
yoff = (yoff * 5) >> 2; // u -> v
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
return cost;
Changed how UV r/d estimates are done for Intra Modes The commit changed to compute UV intra RD estimates for 4x4 and 8x8 separately to be used in mode decision for MB modes associated with the appropriate transform size respectively. Now finally after many other changes related 8x8 quantizer zbin boost and zbin_mode_boost, this change overall helps the HD(with 8x8) by around ~.13%. (avg .13% glb .13% ssim .17%) The commit also has a few changes for eliminating compiler warnings. Change-Id: Ibab35dad44820c87e6b44799c66f8d519cc37344
2012-02-22 03:10:18 +01:00
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static void super_block_uvrd_16x16(VP9_COMMON *const cm, MACROBLOCK *x,
int *rate, int *distortion, int *skip,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize), bhl = mb_height_log2(bsize);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
MACROBLOCKD *const xd = &x->e_mbd;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vp9_transform_sbuv_16x16(x, bsize);
vp9_quantize_sbuv_16x16(x, bsize);
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
*rate = rd_cost_sbuv_16x16(cm, x, bsize);
*distortion = vp9_sb_uv_block_error_c(x->coeff + (256 << (bwl + bhl)),
xd->plane[1].dqcoeff,
xd->plane[2].dqcoeff,
512 << (bwl + bhl - 2), 2);
tokenize: convert skippable functions Use the common block walker to calculate skippability. Change-Id: I6721e42f065df237426c91c1d871ec226ba7cdcb
2013-04-11 20:14:31 +02:00
*skip = vp9_sbuv_is_skippable(xd, bsize);
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
}
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static int rd_cost_sbuv_32x32(VP9_COMMON *const cm, MACROBLOCK *x,
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) - 2, bw = 1 << bwl;
const int bh = 1 << (mb_height_log2(bsize) - 2);
int yoff = 256 * bh * bw;
int p, b, cost = 0;
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
MACROBLOCKD *const xd = &x->e_mbd;
ENTROPY_CONTEXT_PLANES t_above[4], t_left[4];
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vpx_memcpy(&t_above, xd->above_context,
sizeof(ENTROPY_CONTEXT_PLANES) * 4 * bw);
vpx_memcpy(&t_left, xd->left_context,
sizeof(ENTROPY_CONTEXT_PLANES) * 4 * bh);
for (p = 0; p < 2; p++) {
for (b = 0; b < bw * bh; b++) {
const int x_idx = b * (bw - 1), y_idx = b >> bwl;
cost += cost_coeffs(cm, x, yoff + b * 64, PLANE_TYPE_UV,
UVCTX(t_above[x_idx * 4], p),
UVCTX(t_left[y_idx * 4], p),
TX_32X32, 256 * bh * bw);
}
yoff = (yoff * 5) >> 2; // u -> v
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
}
return cost;
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
#undef UVCTX
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static void super_block_uvrd_32x32(VP9_COMMON *const cm, MACROBLOCK *x,
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
int *rate, int *distortion, int *skip,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) - 1, bhl = mb_height_log2(bsize) - 1;
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
MACROBLOCKD *const xd = &x->e_mbd;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
vp9_transform_sbuv_32x32(x, bsize);
vp9_quantize_sbuv_32x32(x, bsize);
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
*rate = rd_cost_sbuv_32x32(cm, x, bsize);
*distortion = vp9_sb_uv_block_error_c(x->coeff + (1024 << (bwl + bhl)),
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
xd->plane[1].dqcoeff,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
xd->plane[2].dqcoeff,
2048 << (bwl + bhl - 2), 0);
tokenize: convert skippable functions Use the common block walker to calculate skippability. Change-Id: I6721e42f065df237426c91c1d871ec226ba7cdcb
2013-04-11 20:14:31 +02:00
*skip = vp9_sbuv_is_skippable(xd, bsize);
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static void super_block_uvrd(VP9_COMMON *const cm, MACROBLOCK *x,
int *rate, int *distortion, int *skippable,
BLOCK_SIZE_TYPE bsize) {
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
vp9_subtract_sbuv(x, bsize);
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
if (mbmi->txfm_size >= TX_32X32 && bsize >= BLOCK_SIZE_SB64X64) {
super_block_uvrd_32x32(cm, x, rate, distortion, skippable, bsize);
} else if (mbmi->txfm_size >= TX_16X16 && bsize >= BLOCK_SIZE_SB32X32) {
super_block_uvrd_16x16(cm, x, rate, distortion, skippable, bsize);
} else if (mbmi->txfm_size >= TX_8X8) {
super_block_uvrd_8x8(cm, x, rate, distortion, skippable, bsize);
} else {
assert(mbmi->txfm_size == TX_4X4);
super_block_uvrd_4x4(cm, x, rate, distortion, skippable, bsize);
}
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
static int64_t rd_pick_intra_sbuv_mode(VP9_COMP *cpi, MACROBLOCK *x,
int *rate, int *rate_tokenonly,
int *distortion, int *skippable,
BLOCK_SIZE_TYPE bsize) {
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
MB_PREDICTION_MODE mode;
MB_PREDICTION_MODE UNINITIALIZED_IS_SAFE(mode_selected);
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t best_rd = INT64_MAX, this_rd;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
int this_rate_tokenonly, this_rate;
Add proper skip support to intra frames. Change-Id: I7524883fb29f42303fb46a5bc6772fbcf8781d1d
2012-08-29 19:43:20 +02:00
int this_distortion, s;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
for (mode = DC_PRED; mode <= TM_PRED; mode++) {
x->e_mbd.mode_info_context->mbmi.uv_mode = mode;
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
vp9_build_intra_predictors_sbuv_s(&x->e_mbd, bsize);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
super_block_uvrd(&cpi->common, x, &this_rate_tokenonly,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
&this_distortion, &s, bsize);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
this_rate = this_rate_tokenonly +
fix costing bug in pick_uv_sb_mode. Change-Id: Ia24e0fddcca9125f8e41e95dbb22444dc51767c7
2012-11-15 00:03:39 +01:00
x->intra_uv_mode_cost[x->e_mbd.frame_type][mode];
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
this_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_distortion);
if (this_rd < best_rd) {
mode_selected = mode;
best_rd = this_rd;
*rate = this_rate;
*rate_tokenonly = this_rate_tokenonly;
*distortion = this_distortion;
Add proper skip support to intra frames. Change-Id: I7524883fb29f42303fb46a5bc6772fbcf8781d1d
2012-08-29 19:43:20 +02:00
*skippable = s;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
}
x->e_mbd.mode_info_context->mbmi.uv_mode = mode_selected;
return best_rd;
}
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
int vp9_cost_mv_ref(VP9_COMP *cpi,
made vp8_mode_context adaptive vp8_mode_contexts[] is an entropy table used to code inter mode choices. It was a fixed constant table. This commit made the entropy context adaptive. Tests on derf set showed very good consistent gains on all metrics: avg psnr .47%, overall psnr .46% and ssim .40%. http://www.corp.google.com/~yaowu/no_crawl/newModeContext.html Change-Id: Ia62b14485c948e2b74586118619c5eb2068b43b2
2011-12-06 21:03:42 +01:00
MB_PREDICTION_MODE m,
New inter mode context This change is a fix / extension of the newbestrefmv experiment. As such it is presented without IFDEF. The change creates a new context for coding inter modes in vp9_find_mv_refs(). This replaces the context that was previously calculated in vp9_find_near_mvs(). The new context is unoptimized and not necessarily any better at this stage (results pending), but eliminates the need for a legacy call to vp9_find_near_mvs(). Based on numbers from Scott, this could help decode speed by several %. In a later patch I will add support for forward update of context (assuming this helps) and refine the context as necessary. Change-Id: I1cd991b82c8df86cc02237a34185e6d67510698a
2012-11-12 16:09:25 +01:00
const int mode_context) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
MACROBLOCKD *xd = &cpi->mb.e_mbd;
int segment_id = xd->mode_info_context->mbmi.segment_id;
Segment Skip Flag First step in simplifying the segment mode and segment EOB flags into a simpler segment skip flag that implies 0,0 mv and EOB at position 0. Change-Id: Ib750cac31a7a02dc21082580498efd9f7d8d72a5
2013-01-28 16:22:53 +01:00
// Dont account for mode here if segment skip is enabled.
if (!vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP)) {
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
VP9_COMMON *pc = &cpi->common;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
vp9_prob p [VP9_MVREFS - 1];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
assert(NEARESTMV <= m && m <= SPLITMV);
New inter mode context This change is a fix / extension of the newbestrefmv experiment. As such it is presented without IFDEF. The change creates a new context for coding inter modes in vp9_find_mv_refs(). This replaces the context that was previously calculated in vp9_find_near_mvs(). The new context is unoptimized and not necessarily any better at this stage (results pending), but eliminates the need for a legacy call to vp9_find_near_mvs(). Based on numbers from Scott, this could help decode speed by several %. In a later patch I will add support for forward update of context (assuming this helps) and refine the context as necessary. Change-Id: I1cd991b82c8df86cc02237a34185e6d67510698a
2012-11-12 16:09:25 +01:00
vp9_mv_ref_probs(pc, p, mode_context);
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
return cost_token(vp9_mv_ref_tree, p,
vp9_mv_ref_encoding_array - NEARESTMV + m);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
} else
return 0;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
void vp9_set_mbmode_and_mvs(MACROBLOCK *x, MB_PREDICTION_MODE mb, int_mv *mv) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
x->e_mbd.mode_info_context->mbmi.mode = mb;
Refactor RD to take same codepath for single and comp pred Change-Id: Id38baf1b89648ef534e28be72f583137871f920c
2012-08-10 01:07:41 +02:00
x->e_mbd.mode_info_context->mbmi.mv[0].as_int = mv->as_int;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
New motion vector entropy coding Adds a new experiment with redesigned/refactored motion vector entropy coding. The patch also takes a first step towards separating the integer and fractional pel components of a MV. However the fractional pel encoding still depends on the integer pel part and so they are not fully independent. Further experiments are in progress to see how much they can be decoupled without affecting performance. All components including entropy coding/decoding, costing for MV search, forward updates and backward updates to probability tables, have been implemented. Results so far: derf: +0.19% std-hd: +0.28% yt: +0.80% hd: +1.15% Patch: Simplifies the fractional pel models: derf: +0.284% std-hd: +0.289% yt: +0.849% hd: +1.254% Patch: Some changes in the models, rebased. derf: +0.330% std-hd: +0.306% yt: +0.816% hd: +1.225% Change-Id: I646b3c48f3587f4cc909639b78c3798da6402678
2012-07-26 22:42:07 +02:00
static int labels2mode(
MACROBLOCK *x,
int const *labelings, int which_label,
B_PREDICTION_MODE this_mode,
int_mv *this_mv, int_mv *this_second_mv,
int_mv seg_mvs[MAX_REF_FRAMES - 1],
int_mv *best_ref_mv,
int_mv *second_best_ref_mv,
remove macros obfuscating mv costing cleanup Change-Id: I565eee40d900e0441ad211b65ac829fc5b93d94a
2012-11-09 00:44:39 +01:00
int *mvjcost, int *mvcost[2]) {
cosmetic: fix '= & ' construct remove useless space after address-of operator. Change-Id: I1fb9e82e8d6cf87558fbd454fb5c0f87599ca2ab
2012-10-17 23:51:27 +02:00
MACROBLOCKD *const xd = &x->e_mbd;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
MODE_INFO *const mic = xd->mode_info_context;
Code clean up. References to MACROBLOCKD that use "x" changed to "xd" to comply with convention elsewhere that x = MACROBLOCK and xd = MACROBLOCKD. Simplify some repeat references using local variables. Change-Id: I0ba2e79536add08140a6c8b19698fcf5077246bc
2012-08-14 12:32:29 +02:00
MB_MODE_INFO * mbmi = &mic->mbmi;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
const int mis = xd->mode_info_stride;
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
int i, cost = 0, thismvcost = 0;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/* We have to be careful retrieving previously-encoded motion vectors.
Ones from this macroblock have to be pulled from the BLOCKD array
as they have not yet made it to the bmi array in our MB_MODE_INFO. */
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
for (i = 0; i < 16; ++i) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
BLOCKD *const d = xd->block + i;
const int row = i >> 2, col = i & 3;
B_PREDICTION_MODE m;
if (labelings[i] != which_label)
continue;
if (col && labelings[i] == labelings[i - 1])
m = LEFT4X4;
else if (row && labelings[i] == labelings[i - 4])
m = ABOVE4X4;
else {
// the only time we should do costing for new motion vector or mode
// is when we are on a new label (jbb May 08, 2007)
switch (m = this_mode) {
case NEW4X4 :
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0) {
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
this_mv->as_int = seg_mvs[mbmi->ref_frame - 1].as_int;
this_second_mv->as_int =
seg_mvs[mbmi->second_ref_frame - 1].as_int;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
remove macros obfuscating mv costing cleanup Change-Id: I565eee40d900e0441ad211b65ac829fc5b93d94a
2012-11-09 00:44:39 +01:00
thismvcost = vp9_mv_bit_cost(this_mv, best_ref_mv, mvjcost, mvcost,
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
102, xd->allow_high_precision_mv);
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0) {
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
thismvcost += vp9_mv_bit_cost(this_second_mv, second_best_ref_mv,
remove macros obfuscating mv costing cleanup Change-Id: I565eee40d900e0441ad211b65ac829fc5b93d94a
2012-11-09 00:44:39 +01:00
mvjcost, mvcost, 102,
Merging in high_precision_mv experiment Merged in the high_precision_mv experiment to make it easier to work on new mv encoding strategies. Also removed coef_update_probs3(). Change-Id: I82d3b0bb642419fe05dba82528bc9ba010e90924
2012-08-03 21:17:18 +02:00
xd->allow_high_precision_mv);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
break;
case LEFT4X4:
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
this_mv->as_int = col ? d[-1].bmi.as_mv[0].as_int :
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
left_block_mv(xd, mic, i);
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0)
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
this_second_mv->as_int = col ? d[-1].bmi.as_mv[1].as_int :
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
left_block_second_mv(xd, mic, i);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
break;
case ABOVE4X4:
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
this_mv->as_int = row ? d[-4].bmi.as_mv[0].as_int :
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
above_block_mv(mic, i, mis);
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0)
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
this_second_mv->as_int = row ? d[-4].bmi.as_mv[1].as_int :
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
above_block_second_mv(mic, i, mis);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
break;
case ZERO4X4:
this_mv->as_int = 0;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
this_second_mv->as_int = 0;
break;
default:
break;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (m == ABOVE4X4) { // replace above with left if same
int_mv left_mv, left_second_mv;
MODE_INFO size reduction Declared the bmi in MODE_INFO as a union instead of B_MODE_INFO. This reduced the memory footprint by 518,400 bytes for 1080 resolutions. The decoder performance improved by ~4% for the clip used and the encoder showed very small improvements. (0.5%) This reduction was first mentioned to me by John K. and in a later discussion by Yaowu. This is WIP. Change-Id: I8e175fdbc46d28c35277302a04bee4540efc8d29
2011-05-24 19:24:52 +02:00
Shut up some compiler warnings on gcc Change-Id: I8548aebd3c39797b253eef472bfe1c40f2f7ec59
2012-08-14 01:20:20 +02:00
left_second_mv.as_int = 0;
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
left_mv.as_int = col ? d[-1].bmi.as_mv[0].as_int :
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
left_block_mv(xd, mic, i);
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0)
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
left_second_mv.as_int = col ? d[-1].bmi.as_mv[1].as_int :
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
left_block_second_mv(xd, mic, i);
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (left_mv.as_int == this_mv->as_int &&
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
(mbmi->second_ref_frame <= 0 ||
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
left_second_mv.as_int == this_second_mv->as_int))
m = LEFT4X4;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
New b-intra mode where direction is contextual Preliminary patch on a new 4x4 intra mode B_CONTEXT_PRED where the dominant direction from the context is used to encode. Various decoder changes are needed to support decoding of B_CONTEXT_PRED in conjunction with hybrid transforms since the scan order and tokenization depends on the actual direction of prediction obtained from the context. Currently the traditional directional modes are used in conjunction with the B_CONTEXT_PRED, which also seems to provide the best results. The gains are small - in the 0.1% range. Change-Id: I5a7ea80b5218f42a9c0dfb42d3f79a68c7f0cdc2
2012-10-09 22:19:15 +02:00
#if CONFIG_NEWBINTRAMODES
cost = x->inter_bmode_costs[
m == B_CONTEXT_PRED ? m - CONTEXT_PRED_REPLACEMENTS : m];
#else
cost = x->inter_bmode_costs[m];
#endif
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
d->bmi.as_mv[0].as_int = this_mv->as_int;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0)
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
d->bmi.as_mv[1].as_int = this_second_mv->as_int;
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
x->partition_info->bmi[i].mode = m;
x->partition_info->bmi[i].mv.as_int = this_mv->as_int;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
x->partition_info->bmi[i].second_mv.as_int = this_second_mv->as_int;
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
cost += thismvcost;
return cost;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
static int64_t encode_inter_mb_segment(VP9_COMMON *const cm,
MACROBLOCK *x,
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
int const *labels,
int which_label,
int *labelyrate,
int *distortion,
ENTROPY_CONTEXT *ta,
removed the idct rtcd idct calls More cleanup to do after this, but this is a good chunk of removing rtcd. Change-Id: I551db75e341a0a85c3ad650df1e9a60dc305681a
2012-11-25 04:33:58 +01:00
ENTROPY_CONTEXT *tl) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int i;
Code clean up. References to MACROBLOCKD that use "x" changed to "xd" to comply with convention elsewhere that x = MACROBLOCK and xd = MACROBLOCKD. Simplify some repeat references using local variables. Change-Id: I0ba2e79536add08140a6c8b19698fcf5077246bc
2012-08-14 12:32:29 +02:00
MACROBLOCKD *xd = &x->e_mbd;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
*labelyrate = 0;
*distortion = 0;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
for (i = 0; i < 16; i++) {
if (labels[i] == which_label) {
BLOCKD *bd = &x->e_mbd.block[i];
BLOCK *be = &x->block[i];
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
int16_t* const src_diff =
raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, i,
x->plane[0].src_diff);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int thisdistortion;
Refactor inter recon functions to support scaling Ensure that all inter prediction goes through a common code path that takes scaling into account. Removes a bunch of duplicate 1st/2nd predictor code. Also introduces a 16x8 mode for 8x8 MVs, similar to the 8x4 trick we were doing before. This has an unexpected effect with EIGHTTAP_SMOOTH, so it's disabled in that case for now. Change-Id: Ia053e823a8bc616a988a0af30452e1e75a739cba
2013-02-09 02:49:44 +01:00
vp9_build_inter_predictor(*(bd->base_pre) + bd->pre,
bd->pre_stride,
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
*(bd->base_dst) + bd->dst,
bd->dst_stride,
Refactor inter recon functions to support scaling Ensure that all inter prediction goes through a common code path that takes scaling into account. Removes a bunch of duplicate 1st/2nd predictor code. Also introduces a 16x8 mode for 8x8 MVs, similar to the 8x4 trick we were doing before. This has an unexpected effect with EIGHTTAP_SMOOTH, so it's disabled in that case for now. Change-Id: Ia053e823a8bc616a988a0af30452e1e75a739cba
2013-02-09 02:49:44 +01:00
&bd->bmi.as_mv[0],
&xd->scale_factor[0],
4, 4, 0 /* no avg */, &xd->subpix);
Implicit weighted prediction experiment Adds an experiment to use a weighted prediction of two INTER predictors, where the weight is one of (1/4, 3/4), (3/8, 5/8), (1/2, 1/2), (5/8, 3/8) or (3/4, 1/4), and is chosen implicitly based on consistency of the predictors to the already reconstructed pixels to the top and left of the current macroblock or superblock. Currently the weighting is not applied to SPLITMV modes, which default to the usual (1/2, 1/2) weighting. However the code is in place controlled by a macro. The same weighting is used for Y and UV components, where the weight is derived from analyzing the Y component only. Results (over compound inter-intra experiment) derf: +0.18% yt: +0.34% hd: +0.49% stdhd: +0.23% The experiment suggests bigger benefit for explicitly signaled weights. Change-Id: I5438539ff4485c5752874cd1eb078ff14bf5235a
2013-03-12 22:21:08 +01:00
// TODO(debargha): Make this work properly with the
// implicit-compoundinter-weight experiment when implicit
// weighting for splitmv modes is turned on.
Refactor inter recon functions to support scaling Ensure that all inter prediction goes through a common code path that takes scaling into account. Removes a bunch of duplicate 1st/2nd predictor code. Also introduces a 16x8 mode for 8x8 MVs, similar to the 8x4 trick we were doing before. This has an unexpected effect with EIGHTTAP_SMOOTH, so it's disabled in that case for now. Change-Id: Ia053e823a8bc616a988a0af30452e1e75a739cba
2013-02-09 02:49:44 +01:00
if (xd->mode_info_context->mbmi.second_ref_frame > 0) {
Implicit weighted prediction experiment Adds an experiment to use a weighted prediction of two INTER predictors, where the weight is one of (1/4, 3/4), (3/8, 5/8), (1/2, 1/2), (5/8, 3/8) or (3/4, 1/4), and is chosen implicitly based on consistency of the predictors to the already reconstructed pixels to the top and left of the current macroblock or superblock. Currently the weighting is not applied to SPLITMV modes, which default to the usual (1/2, 1/2) weighting. However the code is in place controlled by a macro. The same weighting is used for Y and UV components, where the weight is derived from analyzing the Y component only. Results (over compound inter-intra experiment) derf: +0.18% yt: +0.34% hd: +0.49% stdhd: +0.23% The experiment suggests bigger benefit for explicitly signaled weights. Change-Id: I5438539ff4485c5752874cd1eb078ff14bf5235a
2013-03-12 22:21:08 +01:00
vp9_build_inter_predictor(
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
*(bd->base_second_pre) + bd->pre, bd->pre_stride,
*(bd->base_dst) + bd->dst, bd->dst_stride,
Implicit weighted prediction experiment Adds an experiment to use a weighted prediction of two INTER predictors, where the weight is one of (1/4, 3/4), (3/8, 5/8), (1/2, 1/2), (5/8, 3/8) or (3/4, 1/4), and is chosen implicitly based on consistency of the predictors to the already reconstructed pixels to the top and left of the current macroblock or superblock. Currently the weighting is not applied to SPLITMV modes, which default to the usual (1/2, 1/2) weighting. However the code is in place controlled by a macro. The same weighting is used for Y and UV components, where the weight is derived from analyzing the Y component only. Results (over compound inter-intra experiment) derf: +0.18% yt: +0.34% hd: +0.49% stdhd: +0.23% The experiment suggests bigger benefit for explicitly signaled weights. Change-Id: I5438539ff4485c5752874cd1eb078ff14bf5235a
2013-03-12 22:21:08 +01:00
&bd->bmi.as_mv[1], &xd->scale_factor[1], 4, 4,
1 << (2 * CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT) /* avg */,
&xd->subpix);
Refactor inter recon functions to support scaling Ensure that all inter prediction goes through a common code path that takes scaling into account. Removes a bunch of duplicate 1st/2nd predictor code. Also introduces a 16x8 mode for 8x8 MVs, similar to the 8x4 trick we were doing before. This has an unexpected effect with EIGHTTAP_SMOOTH, so it's disabled in that case for now. Change-Id: Ia053e823a8bc616a988a0af30452e1e75a739cba
2013-02-09 02:49:44 +01:00
}
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
vp9_subtract_block(4, 4, src_diff, 16,
*(be->base_src) + be->src, be->src_stride,
*(bd->base_dst) + bd->dst, bd->dst_stride);
x->fwd_txm4x4(src_diff, be->coeff, 32);
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
x->quantize_b_4x4(x, i, 16);
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
thisdistortion = vp9_block_error(be->coeff,
BLOCK_OFFSET(xd->plane[0].dqcoeff, i, 16), 16);
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
*distortion += thisdistortion;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
*labelyrate += cost_coeffs(cm, x, i, PLANE_TYPE_Y_WITH_DC,
clean up tokenize_b() and stuff_b() Change-Id: I0c1be01aae933243311ad321b6c456adaec1a0f5
2012-12-06 21:40:57 +01:00
ta + vp9_block2above[TX_4X4][i],
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
tl + vp9_block2left[TX_4X4][i], TX_4X4, 16);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
}
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
*distortion >>= 2;
return RDCOST(x->rdmult, x->rddiv, *labelyrate, *distortion);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm,
MACROBLOCK *x,
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
int const *labels,
int which_label,
int *labelyrate,
int *distortion,
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
int64_t *otherrd,
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
ENTROPY_CONTEXT *ta,
removed the idct rtcd idct calls More cleanup to do after this, but this is a good chunk of removing rtcd. Change-Id: I551db75e341a0a85c3ad650df1e9a60dc305681a
2012-11-25 04:33:58 +01:00
ENTROPY_CONTEXT *tl) {
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
int i, j;
MACROBLOCKD *xd = &x->e_mbd;
const int iblock[4] = { 0, 1, 4, 5 };
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
int othercost = 0, otherdist = 0;
ENTROPY_CONTEXT_PLANES tac, tlc;
ENTROPY_CONTEXT *tacp = (ENTROPY_CONTEXT *) &tac,
*tlcp = (ENTROPY_CONTEXT *) &tlc;
if (otherrd) {
memcpy(&tac, ta, sizeof(ENTROPY_CONTEXT_PLANES));
memcpy(&tlc, tl, sizeof(ENTROPY_CONTEXT_PLANES));
}
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
*distortion = 0;
*labelyrate = 0;
for (i = 0; i < 4; i++) {
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
int ib = vp9_i8x8_block[i];
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
if (labels[ib] == which_label) {
Refactor inter recon functions to support scaling Ensure that all inter prediction goes through a common code path that takes scaling into account. Removes a bunch of duplicate 1st/2nd predictor code. Also introduces a 16x8 mode for 8x8 MVs, similar to the 8x4 trick we were doing before. This has an unexpected effect with EIGHTTAP_SMOOTH, so it's disabled in that case for now. Change-Id: Ia053e823a8bc616a988a0af30452e1e75a739cba
2013-02-09 02:49:44 +01:00
const int use_second_ref =
xd->mode_info_context->mbmi.second_ref_frame > 0;
int which_mv;
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
const int idx = (ib & 8) + ((ib & 2) << 1);
BLOCKD *bd = &xd->block[ib];
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
BLOCK *be = &x->block[ib], *be2 = &x->block[idx];
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
int16_t* const src_diff =
raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, ib,
x->plane[0].src_diff);
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
int thisdistortion;
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
assert(idx < 16);
Refactor inter recon functions to support scaling Ensure that all inter prediction goes through a common code path that takes scaling into account. Removes a bunch of duplicate 1st/2nd predictor code. Also introduces a 16x8 mode for 8x8 MVs, similar to the 8x4 trick we were doing before. This has an unexpected effect with EIGHTTAP_SMOOTH, so it's disabled in that case for now. Change-Id: Ia053e823a8bc616a988a0af30452e1e75a739cba
2013-02-09 02:49:44 +01:00
for (which_mv = 0; which_mv < 1 + use_second_ref; ++which_mv) {
uint8_t **base_pre = which_mv ? bd->base_second_pre : bd->base_pre;
Implicit weighted prediction experiment Adds an experiment to use a weighted prediction of two INTER predictors, where the weight is one of (1/4, 3/4), (3/8, 5/8), (1/2, 1/2), (5/8, 3/8) or (3/4, 1/4), and is chosen implicitly based on consistency of the predictors to the already reconstructed pixels to the top and left of the current macroblock or superblock. Currently the weighting is not applied to SPLITMV modes, which default to the usual (1/2, 1/2) weighting. However the code is in place controlled by a macro. The same weighting is used for Y and UV components, where the weight is derived from analyzing the Y component only. Results (over compound inter-intra experiment) derf: +0.18% yt: +0.34% hd: +0.49% stdhd: +0.23% The experiment suggests bigger benefit for explicitly signaled weights. Change-Id: I5438539ff4485c5752874cd1eb078ff14bf5235a
2013-03-12 22:21:08 +01:00
// TODO(debargha): Make this work properly with the
// implicit-compoundinter-weight experiment when implicit
// weighting for splitmv modes is turned on.
vp9_build_inter_predictor(
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
*base_pre + bd->pre, bd->pre_stride,
*(bd->base_dst) + bd->dst, bd->dst_stride,
Implicit weighted prediction experiment Adds an experiment to use a weighted prediction of two INTER predictors, where the weight is one of (1/4, 3/4), (3/8, 5/8), (1/2, 1/2), (5/8, 3/8) or (3/4, 1/4), and is chosen implicitly based on consistency of the predictors to the already reconstructed pixels to the top and left of the current macroblock or superblock. Currently the weighting is not applied to SPLITMV modes, which default to the usual (1/2, 1/2) weighting. However the code is in place controlled by a macro. The same weighting is used for Y and UV components, where the weight is derived from analyzing the Y component only. Results (over compound inter-intra experiment) derf: +0.18% yt: +0.34% hd: +0.49% stdhd: +0.23% The experiment suggests bigger benefit for explicitly signaled weights. Change-Id: I5438539ff4485c5752874cd1eb078ff14bf5235a
2013-03-12 22:21:08 +01:00
&bd->bmi.as_mv[which_mv], &xd->scale_factor[which_mv], 8, 8,
which_mv << (2 * CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT),
&xd->subpix);
Refactor inter recon functions to support scaling Ensure that all inter prediction goes through a common code path that takes scaling into account. Removes a bunch of duplicate 1st/2nd predictor code. Also introduces a 16x8 mode for 8x8 MVs, similar to the 8x4 trick we were doing before. This has an unexpected effect with EIGHTTAP_SMOOTH, so it's disabled in that case for now. Change-Id: Ia053e823a8bc616a988a0af30452e1e75a739cba
2013-02-09 02:49:44 +01:00
}
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
vp9_subtract_block(8, 8, src_diff, 16,
*(be->base_src) + be->src, be->src_stride,
*(bd->base_dst) + bd->dst, bd->dst_stride);
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
if (xd->mode_info_context->mbmi.txfm_size == TX_4X4) {
if (otherrd) {
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
x->fwd_txm8x8(src_diff, be2->coeff, 32);
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
x->quantize_b_8x8(x, idx, DCT_DCT, 16);
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
thisdistortion = vp9_block_error_c(be2->coeff,
BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16), 64);
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
otherdist += thisdistortion;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
xd->mode_info_context->mbmi.txfm_size = TX_8X8;
othercost += cost_coeffs(cm, x, idx, PLANE_TYPE_Y_WITH_DC,
Make superblocks independent of macroblock code and data. Split macroblock and superblock tokenization and detokenization functions and coefficient-related data structs so that the bitstream layout and related code of superblock coefficients looks less like it's a hack to fit macroblocks in superblocks. In addition, unify chroma transform size selection from luma transform size (i.e. always use the same size, as long as it fits the predictor); in practice, this means 32x32 and 64x64 superblocks using the 16x16 luma transform will now use the 16x16 (instead of the 8x8) chroma transform, and 64x64 superblocks using the 32x32 luma transform will now use the 32x32 (instead of the 16x16) chroma transform. Lastly, add a trellis optimize function for 32x32 transform blocks. HD gains about 0.3%, STDHD about 0.15% and derf about 0.1%. There's a few negative points here and there that I might want to analyze a little closer. Change-Id: Ibad7c3ddfe1acfc52771dfc27c03e9783e054430
2013-03-04 23:12:17 +01:00
tacp + vp9_block2above[TX_8X8][idx],
tlcp + vp9_block2left[TX_8X8][idx],
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
TX_8X8, 16);
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
xd->mode_info_context->mbmi.txfm_size = TX_4X4;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
}
for (j = 0; j < 4; j += 2) {
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
int16_t* const src_diff =
raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16,
0, ib + iblock[j],
x->plane[0].src_diff);
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
bd = &xd->block[ib + iblock[j]];
be = &x->block[ib + iblock[j]];
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
x->fwd_txm8x4(src_diff, be->coeff, 32);
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
x->quantize_b_4x4_pair(x, ib + iblock[j], ib + iblock[j] + 1, 16);
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
thisdistortion = vp9_block_error_c(be->coeff,
BLOCK_OFFSET(xd->plane[0].dqcoeff, ib + iblock[j], 16), 32);
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
*distortion += thisdistortion;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
*labelyrate +=
cost_coeffs(cm, x, ib + iblock[j], PLANE_TYPE_Y_WITH_DC,
ta + vp9_block2above[TX_4X4][ib + iblock[j]],
tl + vp9_block2left[TX_4X4][ib + iblock[j]],
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
TX_4X4, 16);
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
*labelyrate +=
cost_coeffs(cm, x, ib + iblock[j] + 1,
PLANE_TYPE_Y_WITH_DC,
ta + vp9_block2above[TX_4X4][ib + iblock[j] + 1],
tl + vp9_block2left[TX_4X4][ib + iblock[j]],
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
TX_4X4, 16);
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
}
} else /* 8x8 */ {
if (otherrd) {
for (j = 0; j < 4; j += 2) {
Fixing 8x8/4x4 ADST for intra modes with tx select This patch allows use of 8x8 and 4x4 ADST correctly for Intra 16x16 modes and Intra 8x8 modes when the block size selected is smaller than the prediction mode. Also includes some cleanups and refactoring. Rebase. Change-Id: Ie3257bdf07bdb9c6e9476915e3a80183c8fa005a
2012-11-16 00:14:38 +01:00
BLOCK *be = &x->block[ib + iblock[j]];
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
int16_t* const src_diff =
raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16,
0, ib + iblock[j],
x->plane[0].src_diff);
x->fwd_txm8x4(src_diff, be->coeff, 32);
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
x->quantize_b_4x4_pair(x, ib + iblock[j], ib + iblock[j] + 1, 16);
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
thisdistortion = vp9_block_error_c(be->coeff,
BLOCK_OFFSET(xd->plane[0].dqcoeff, ib + iblock[j], 16), 32);
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
otherdist += thisdistortion;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
xd->mode_info_context->mbmi.txfm_size = TX_4X4;
othercost +=
cost_coeffs(cm, x, ib + iblock[j], PLANE_TYPE_Y_WITH_DC,
tacp + vp9_block2above[TX_4X4][ib + iblock[j]],
tlcp + vp9_block2left[TX_4X4][ib + iblock[j]],
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
TX_4X4, 16);
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
othercost +=
cost_coeffs(cm, x, ib + iblock[j] + 1,
PLANE_TYPE_Y_WITH_DC,
tacp + vp9_block2above[TX_4X4][ib + iblock[j] + 1],
tlcp + vp9_block2left[TX_4X4][ib + iblock[j]],
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
TX_4X4, 16);
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
xd->mode_info_context->mbmi.txfm_size = TX_8X8;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
}
}
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
x->fwd_txm8x8(src_diff, be2->coeff, 32);
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
x->quantize_b_8x8(x, idx, DCT_DCT, 16);
Move qcoeff, dqcoeff from BLOCKD to per-plane data Start grouping data per-plane, as part of refactoring to support additional planes, and chroma planes with other-than 4:2:0 subsampling. Change-Id: Idb76a0e23ab239180c818025bae1f36f1608bb23
2013-04-02 23:50:40 +02:00
thisdistortion = vp9_block_error_c(be2->coeff,
BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16), 64);
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
*distortion += thisdistortion;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
*labelyrate += cost_coeffs(cm, x, idx, PLANE_TYPE_Y_WITH_DC,
clean up tokenize_b() and stuff_b() Change-Id: I0c1be01aae933243311ad321b6c456adaec1a0f5
2012-12-06 21:40:57 +01:00
ta + vp9_block2above[TX_8X8][idx],
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
tl + vp9_block2left[TX_8X8][idx], TX_8X8,
16);
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
}
}
}
*distortion >>= 2;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
if (otherrd) {
Fix yet another typo in splitmv/tx_select code. Change-Id: I6a28cc87af275fc267b3cc8d90d642dcc870c249
2012-10-25 18:17:47 +02:00
otherdist >>= 2;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
*otherrd = RDCOST(x->rdmult, x->rddiv, othercost, otherdist);
}
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
return RDCOST(x->rdmult, x->rddiv, *labelyrate, *distortion);
}
fix build issues for experimental branch experimental branch build was broken from some merge artifacts, this commit fixes those issues to enable the experimental branch to build. Change-Id: Ic52b2d2f1d1b80abb7ecaa4c0927bcf887ac0c2a
2011-06-23 00:07:04 +02:00
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
static const unsigned int segmentation_to_sseshift[4] = {3, 3, 2, 0};
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
typedef struct {
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
int_mv *ref_mv, *second_ref_mv;
Using int_mv instead of MV The compiler produces better assembly when using int_mv for assignments. The compiler shifts and ors the two 16bit values when assigning MV. Change-Id: I52ce4bc2bfbfaf3f1151204b2f21e1e0654f960f
2011-05-12 16:50:16 +02:00
int_mv mvp;
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
Make RDCOST() results 64-bit. This prevents various integer overflows that several of us have been seeing in the past few days. Change-Id: If0020613a608e13d311b7838796c7ad53cb97255
2012-08-02 19:07:33 +02:00
int64_t segment_rd;
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
SPLITMV_PARTITIONING_TYPE segment_num;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
TX_SIZE txfm_size;
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
int r;
int d;
int segment_yrate;
B_PREDICTION_MODE modes[16];
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
int_mv mvs[16], second_mvs[16];
Change eob[] array type in splitmv RD loop to a regular integer. Change-Id: I240d6b50069fd3f35cc4fed2f4507796f0ef25e9
2012-10-23 18:23:23 +02:00
int eobs[16];
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
int mvthresh;
int *mdcounts;
Using int_mv instead of MV The compiler produces better assembly when using int_mv for assignments. The compiler shifts and ors the two 16bit values when assigning MV. Change-Id: I52ce4bc2bfbfaf3f1151204b2f21e1e0654f960f
2011-05-12 16:50:16 +02:00
int_mv sv_mvp[4]; // save 4 mvp from 8x8
Modify motion estimation for SPLITMV mode 1. Search for block8x16/block16x8 uses block8x8's search results. 2. Check block4x4 only if block8x8 is chosen. (This hurts quality, which will be improved in another check-in.) 3. In block4x4 search, the previous block's result is used as MV predictor for next block. This change improves performance. Change-Id: I9dc089007ca08129fb6c11fe7692777ebb8647b0
2010-12-23 17:23:03 +01:00
int sv_istep[2]; // save 2 initial step_param for 16x8/8x16
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
} BEST_SEG_INFO;
Use configure checks for various inline keywords. Change-Id: I8508f1a3d3430f998bb9295f849e88e626a52a24
2013-02-06 21:45:28 +01:00
static INLINE int mv_check_bounds(MACROBLOCK *x, int_mv *mv) {
Begin refactoring to help merge parts of rd inter. Also refactor some RD code, remove some comments, and cosmetic changes. Change-Id: I43806ae573ab5a5036602ac4425d750bbfa1e812
2012-08-08 20:13:53 +02:00
int r = 0;
r |= (mv->as_mv.row >> 3) < x->mv_row_min;
r |= (mv->as_mv.row >> 3) > x->mv_row_max;
r |= (mv->as_mv.col >> 3) < x->mv_col_min;
r |= (mv->as_mv.col >> 3) > x->mv_col_max;
return r;
}
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
static void rd_check_segment_txsize(VP9_COMP *cpi, MACROBLOCK *x,
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
BEST_SEG_INFO *bsi,
SPLITMV_PARTITIONING_TYPE segmentation,
TX_SIZE tx_size, int64_t *otherrds,
int64_t *rds, int *completed,
/* 16 = n_blocks */
int_mv seg_mvs[16 /* n_blocks */]
[MAX_REF_FRAMES - 1]) {
Detect skippable blocks in splitmv RD loop case. Change-Id: Id8a7265dc721e5cdcaa144e0041beeb32a98f51c
2012-10-17 23:32:17 +02:00
int i, j;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int const *labels;
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
int br = 0, bd = 0;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
B_PREDICTION_MODE this_mode;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
MB_MODE_INFO * mbmi = &x->e_mbd.mode_info_context->mbmi;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int label_count;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
int64_t this_segment_rd = 0, other_segment_rd;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int label_mv_thresh;
int rate = 0;
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
int sbr = 0, sbd = 0;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int segmentyrate = 0;
Change eob[] array type in splitmv RD loop to a regular integer. Change-Id: I240d6b50069fd3f35cc4fed2f4507796f0ef25e9
2012-10-23 18:23:23 +02:00
int best_eobs[16] = { 0 };
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
vp9_variance_fn_ptr_t *v_fn_ptr;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
ENTROPY_CONTEXT_PLANES t_above, t_left;
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
ENTROPY_CONTEXT *ta, *tl;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
ENTROPY_CONTEXT_PLANES t_above_b, t_left_b;
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
ENTROPY_CONTEXT *ta_b, *tl_b;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
vpx_memcpy(&t_above, x->e_mbd.above_context, sizeof(ENTROPY_CONTEXT_PLANES));
vpx_memcpy(&t_left, x->e_mbd.left_context, sizeof(ENTROPY_CONTEXT_PLANES));
ta = (ENTROPY_CONTEXT *)&t_above;
tl = (ENTROPY_CONTEXT *)&t_left;
ta_b = (ENTROPY_CONTEXT *)&t_above_b;
tl_b = (ENTROPY_CONTEXT *)&t_left_b;
v_fn_ptr = &cpi->fn_ptr[segmentation];
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
labels = vp9_mbsplits[segmentation];
label_count = vp9_mbsplit_count[segmentation];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// 64 makes this threshold really big effectively
// making it so that we very rarely check mvs on
// segments. setting this to 1 would make mv thresh
// roughly equal to what it is for macroblocks
label_mv_thresh = 1 * bsi->mvthresh / label_count;
// Segmentation method overheads
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
rate = cost_token(vp9_mbsplit_tree, vp9_mbsplit_probs,
vp9_mbsplit_encodings + segmentation);
New inter mode context This change is a fix / extension of the newbestrefmv experiment. As such it is presented without IFDEF. The change creates a new context for coding inter modes in vp9_find_mv_refs(). This replaces the context that was previously calculated in vp9_find_near_mvs(). The new context is unoptimized and not necessarily any better at this stage (results pending), but eliminates the need for a legacy call to vp9_find_near_mvs(). Based on numbers from Scott, this could help decode speed by several %. In a later patch I will add support for forward update of context (assuming this helps) and refine the context as necessary. Change-Id: I1cd991b82c8df86cc02237a34185e6d67510698a
2012-11-12 16:09:25 +01:00
rate += vp9_cost_mv_ref(cpi, SPLITMV,
mbmi->mb_mode_context[mbmi->ref_frame]);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
this_segment_rd += RDCOST(x->rdmult, x->rddiv, rate, 0);
br += rate;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
other_segment_rd = this_segment_rd;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
mbmi->txfm_size = tx_size;
for (i = 0; i < label_count && this_segment_rd < bsi->segment_rd; i++) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int_mv mode_mv[B_MODE_COUNT], second_mode_mv[B_MODE_COUNT];
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t best_label_rd = INT64_MAX, best_other_rd = INT64_MAX;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
B_PREDICTION_MODE mode_selected = ZERO4X4;
int bestlabelyrate = 0;
// search for the best motion vector on this segment
for (this_mode = LEFT4X4; this_mode <= NEW4X4; this_mode ++) {
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
int64_t this_rd, other_rd;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int distortion;
int labelyrate;
ENTROPY_CONTEXT_PLANES t_above_s, t_left_s;
ENTROPY_CONTEXT *ta_s;
ENTROPY_CONTEXT *tl_s;
vpx_memcpy(&t_above_s, &t_above, sizeof(ENTROPY_CONTEXT_PLANES));
vpx_memcpy(&t_left_s, &t_left, sizeof(ENTROPY_CONTEXT_PLANES));
ta_s = (ENTROPY_CONTEXT *)&t_above_s;
tl_s = (ENTROPY_CONTEXT *)&t_left_s;
// motion search for newmv (single predictor case only)
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame <= 0 && this_mode == NEW4X4) {
Some refactoring of mcomp functions. Change-Id: Ic7a7cb1199b085e98ede0e634619b3077c348d57
2012-08-08 01:44:26 +02:00
int sseshift, n;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int step_param = 0;
int further_steps;
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
int thissme, bestsme = INT_MAX;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
BLOCK *c;
BLOCKD *e;
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
/* Is the best so far sufficiently good that we cant justify doing
* and new motion search. */
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (best_label_rd < label_mv_thresh)
break;
if (cpi->compressor_speed) {
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
if (segmentation == PARTITIONING_8X16 ||
segmentation == PARTITIONING_16X8) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
bsi->mvp.as_int = bsi->sv_mvp[i].as_int;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
if (i == 1 && segmentation == PARTITIONING_16X8)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
bsi->mvp.as_int = bsi->sv_mvp[2].as_int;
step_param = bsi->sv_istep[i];
}
// use previous block's result as next block's MV predictor.
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
if (segmentation == PARTITIONING_4X4 && i > 0) {
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
bsi->mvp.as_int = x->e_mbd.block[i - 1].bmi.as_mv[0].as_int;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (i == 4 || i == 8 || i == 12)
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
bsi->mvp.as_int = x->e_mbd.block[i - 4].bmi.as_mv[0].as_int;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
step_param = 2;
}
}
further_steps = (MAX_MVSEARCH_STEPS - 1) - step_param;
{
int sadpb = x->sadperbit4;
int_mv mvp_full;
mvp_full.as_mv.row = bsi->mvp.as_mv.row >> 3;
mvp_full.as_mv.col = bsi->mvp.as_mv.col >> 3;
// find first label
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
n = vp9_mbsplit_offset[segmentation][i];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
c = &x->block[n];
e = &x->e_mbd.block[n];
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
bestsme = vp9_full_pixel_diamond(cpi, x, c, e, &mvp_full, step_param,
More refactoring of diamond search. This should make merging inter code paths easier. Change-Id: I5cb81f25d56fa4790b4e9cfa4bc32b9062c2cfdf
2012-08-08 20:52:22 +02:00
sadpb, further_steps, 0, v_fn_ptr,
bsi->ref_mv, &mode_mv[NEW4X4]);
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
sseshift = segmentation_to_sseshift[segmentation];
Add half-pixel variance RTCD functions NEON has optimized 16x16 half-pixel variance functions, but they were not part of the RTCD framework. Add these functions to RTCD, so that other platforms can make use of this optimization in the future and special-case ARM code can be removed. A number of functions were taking two variance functions as parameters. These functions were changed to take a single parameter, a pointer to a struct containing all the variance functions for that block size. This provides additional flexibility for calling additional variance functions (the half-pixel special case, for example) and by initializing the table for all block sizes, we don't have to construct this function pointer table for each macroblock. Change-Id: I78289ff36b2715f9a7aa04d5f6fbe3d23acdc29c
2010-10-26 21:34:16 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// Should we do a full search (best quality only)
if ((cpi->compressor_speed == 0) && (bestsme >> sseshift) > 4000) {
/* Check if mvp_full is within the range. */
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
clamp_mv(&mvp_full, x->mv_col_min, x->mv_col_max,
x->mv_row_min, x->mv_row_max);
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
thissme = cpi->full_search_sad(x, c, e, &mvp_full,
sadpb, 16, v_fn_ptr,
remove macros obfuscating mv costing cleanup Change-Id: I565eee40d900e0441ad211b65ac829fc5b93d94a
2012-11-09 00:44:39 +01:00
x->nmvjointcost, x->mvcost,
bsi->ref_mv);
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (thissme < bestsme) {
bestsme = thissme;
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
mode_mv[NEW4X4].as_int = e->bmi.as_mv[0].as_int;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
} else {
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
/* The full search result is actually worse so re-instate the
* previous best vector */
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
e->bmi.as_mv[0].as_int = mode_mv[NEW4X4].as_int;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
}
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (bestsme < INT_MAX) {
int distortion;
unsigned int sse;
cpi->find_fractional_mv_step(x, c, e, &mode_mv[NEW4X4],
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
bsi->ref_mv, x->errorperbit, v_fn_ptr,
remove macros obfuscating mv costing cleanup Change-Id: I565eee40d900e0441ad211b65ac829fc5b93d94a
2012-11-09 00:44:39 +01:00
x->nmvjointcost, x->mvcost,
&distortion, &sse);
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// safe motion search result for use in compound prediction
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
seg_mvs[i][mbmi->ref_frame - 1].as_int = mode_mv[NEW4X4].as_int;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
} else if (mbmi->second_ref_frame > 0 && this_mode == NEW4X4) {
/* NEW4X4 */
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
/* motion search not completed? Then skip newmv for this block with
* comppred */
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
if (seg_mvs[i][mbmi->second_ref_frame - 1].as_int == INVALID_MV ||
seg_mvs[i][mbmi->ref_frame - 1].as_int == INVALID_MV) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
continue;
}
}
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
rate = labels2mode(x, labels, i, this_mode, &mode_mv[this_mode],
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
&second_mode_mv[this_mode], seg_mvs[i],
remove macros obfuscating mv costing cleanup Change-Id: I565eee40d900e0441ad211b65ac829fc5b93d94a
2012-11-09 00:44:39 +01:00
bsi->ref_mv, bsi->second_ref_mv, x->nmvjointcost,
x->mvcost);
Fix compiler warnings Change-Id: I2a97f08cc3c7808ce5be39e910cc5147ecf03a1d
2010-06-11 20:33:49 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// Trap vectors that reach beyond the UMV borders
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
if (((mode_mv[this_mode].as_mv.row >> 3) < x->mv_row_min) ||
((mode_mv[this_mode].as_mv.row >> 3) > x->mv_row_max) ||
((mode_mv[this_mode].as_mv.col >> 3) < x->mv_col_min) ||
((mode_mv[this_mode].as_mv.col >> 3) > x->mv_col_max)) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
continue;
}
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0 &&
Begin refactoring to help merge parts of rd inter. Also refactor some RD code, remove some comments, and cosmetic changes. Change-Id: I43806ae573ab5a5036602ac4425d750bbfa1e812
2012-08-08 20:13:53 +02:00
mv_check_bounds(x, &second_mode_mv[this_mode]))
continue;
Fix compiler warnings Change-Id: I2a97f08cc3c7808ce5be39e910cc5147ecf03a1d
2010-06-11 20:33:49 +02:00
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
if (segmentation == PARTITIONING_4X4) {
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
this_rd = encode_inter_mb_segment(&cpi->common,
x, labels, i, &labelyrate,
removed the idct rtcd idct calls More cleanup to do after this, but this is a good chunk of removing rtcd. Change-Id: I551db75e341a0a85c3ad650df1e9a60dc305681a
2012-11-25 04:33:58 +01:00
&distortion, ta_s, tl_s);
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
other_rd = this_rd;
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
} else {
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
this_rd = encode_inter_mb_segment_8x8(&cpi->common,
x, labels, i, &labelyrate,
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
&distortion, &other_rd,
removed the idct rtcd idct calls More cleanup to do after this, but this is a good chunk of removing rtcd. Change-Id: I551db75e341a0a85c3ad650df1e9a60dc305681a
2012-11-25 04:33:58 +01:00
ta_s, tl_s);
Use 8x8 block bases for non-4x4 splitmv RD. Change-Id: I76095d5a02edcc5d404ea830157780b24fd509d7
2012-10-17 20:40:00 +02:00
}
this_rd += RDCOST(x->rdmult, x->rddiv, rate, 0);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
rate += labelyrate;
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (this_rd < best_label_rd) {
sbr = rate;
sbd = distortion;
bestlabelyrate = labelyrate;
mode_selected = this_mode;
best_label_rd = this_rd;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
if (x->e_mbd.mode_info_context->mbmi.txfm_size == TX_4X4) {
for (j = 0; j < 16; j++)
if (labels[j] == i)
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
best_eobs[j] = x->e_mbd.plane[0].eobs[j];
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
} else {
for (j = 0; j < 4; j++) {
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
int ib = vp9_i8x8_block[j], idx = j * 4;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
if (labels[ib] == i)
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
best_eobs[idx] = x->e_mbd.plane[0].eobs[idx];
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
}
}
if (other_rd < best_other_rd)
best_other_rd = other_rd;
Modify motion estimation for SPLITMV mode 1. Search for block8x16/block16x8 uses block8x8's search results. 2. Check block4x4 only if block8x8 is chosen. (This hurts quality, which will be improved in another check-in.) 3. In block4x4 search, the previous block's result is used as MV predictor for next block. This change improves performance. Change-Id: I9dc089007ca08129fb6c11fe7692777ebb8647b0
2010-12-23 17:23:03 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
vpx_memcpy(ta_b, ta_s, sizeof(ENTROPY_CONTEXT_PLANES));
vpx_memcpy(tl_b, tl_s, sizeof(ENTROPY_CONTEXT_PLANES));
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
} /*for each 4x4 mode*/
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
vpx_memcpy(ta, ta_b, sizeof(ENTROPY_CONTEXT_PLANES));
vpx_memcpy(tl, tl_b, sizeof(ENTROPY_CONTEXT_PLANES));
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
labels2mode(x, labels, i, mode_selected, &mode_mv[mode_selected],
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
&second_mode_mv[mode_selected], seg_mvs[i],
remove macros obfuscating mv costing cleanup Change-Id: I565eee40d900e0441ad211b65ac829fc5b93d94a
2012-11-09 00:44:39 +01:00
bsi->ref_mv, bsi->second_ref_mv, x->nmvjointcost, x->mvcost);
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
br += sbr;
bd += sbd;
segmentyrate += bestlabelyrate;
this_segment_rd += best_label_rd;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
other_segment_rd += best_other_rd;
if (rds)
rds[i] = this_segment_rd;
if (otherrds)
Fix typo in splitmv/tx_select code. Change-Id: I2823043634b0fa617c2715d63ff18595b7209287
2012-10-24 22:03:51 +02:00
otherrds[i] = other_segment_rd;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
} /* for each label */
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (this_segment_rd < bsi->segment_rd) {
bsi->r = br;
bsi->d = bd;
bsi->segment_yrate = segmentyrate;
bsi->segment_rd = this_segment_rd;
bsi->segment_num = segmentation;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
bsi->txfm_size = mbmi->txfm_size;
Modify motion estimation for SPLITMV mode 1. Search for block8x16/block16x8 uses block8x8's search results. 2. Check block4x4 only if block8x8 is chosen. (This hurts quality, which will be improved in another check-in.) 3. In block4x4 search, the previous block's result is used as MV predictor for next block. This change improves performance. Change-Id: I9dc089007ca08129fb6c11fe7692777ebb8647b0
2010-12-23 17:23:03 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// store everything needed to come back to this!!
for (i = 0; i < 16; i++) {
bsi->mvs[i].as_mv = x->partition_info->bmi[i].mv.as_mv;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
bsi->second_mvs[i].as_mv = x->partition_info->bmi[i].second_mv.as_mv;
bsi->modes[i] = x->partition_info->bmi[i].mode;
Detect skippable blocks in splitmv RD loop case. Change-Id: Id8a7265dc721e5cdcaa144e0041beeb32a98f51c
2012-10-17 23:32:17 +02:00
bsi->eobs[i] = best_eobs[i];
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
if (completed) {
*completed = i;
}
}
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
static void rd_check_segment(VP9_COMP *cpi, MACROBLOCK *x,
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
BEST_SEG_INFO *bsi,
unsigned int segmentation,
/* 16 = n_blocks */
int_mv seg_mvs[16][MAX_REF_FRAMES - 1],
int64_t txfm_cache[NB_TXFM_MODES]) {
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
int i, n, c = vp9_mbsplit_count[segmentation];
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
if (segmentation == PARTITIONING_4X4) {
int64_t rd[16];
rd_check_segment_txsize(cpi, x, bsi, segmentation, TX_4X4, NULL,
rd, &n, seg_mvs);
if (n == c) {
for (i = 0; i < NB_TXFM_MODES; i++) {
if (rd[c - 1] < txfm_cache[i])
txfm_cache[i] = rd[c - 1];
}
}
} else {
int64_t diff, base_rd;
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
int cost4x4 = vp9_cost_bit(cpi->common.prob_tx[0], 0);
int cost8x8 = vp9_cost_bit(cpi->common.prob_tx[0], 1);
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
if (cpi->common.txfm_mode == TX_MODE_SELECT) {
int64_t rd4x4[4], rd8x8[4];
int n4x4, n8x8, nmin;
BEST_SEG_INFO bsi4x4, bsi8x8;
/* factor in cost of cost4x4/8x8 in decision */
vpx_memcpy(&bsi4x4, bsi, sizeof(*bsi));
vpx_memcpy(&bsi8x8, bsi, sizeof(*bsi));
rd_check_segment_txsize(cpi, x, &bsi4x4, segmentation,
TX_4X4, NULL, rd4x4, &n4x4, seg_mvs);
rd_check_segment_txsize(cpi, x, &bsi8x8, segmentation,
TX_8X8, NULL, rd8x8, &n8x8, seg_mvs);
if (bsi4x4.segment_num == segmentation) {
bsi4x4.segment_rd += RDCOST(x->rdmult, x->rddiv, cost4x4, 0);
if (bsi4x4.segment_rd < bsi->segment_rd)
vpx_memcpy(bsi, &bsi4x4, sizeof(*bsi));
}
if (bsi8x8.segment_num == segmentation) {
bsi8x8.segment_rd += RDCOST(x->rdmult, x->rddiv, cost8x8, 0);
if (bsi8x8.segment_rd < bsi->segment_rd)
vpx_memcpy(bsi, &bsi8x8, sizeof(*bsi));
}
n = n4x4 > n8x8 ? n4x4 : n8x8;
if (n == c) {
nmin = n4x4 < n8x8 ? n4x4 : n8x8;
diff = rd8x8[nmin - 1] - rd4x4[nmin - 1];
if (n == n4x4) {
base_rd = rd4x4[c - 1];
} else {
base_rd = rd8x8[c - 1] - diff;
}
}
} else {
int64_t rd[4], otherrd[4];
if (cpi->common.txfm_mode == ONLY_4X4) {
rd_check_segment_txsize(cpi, x, bsi, segmentation, TX_4X4, otherrd,
rd, &n, seg_mvs);
if (n == c) {
base_rd = rd[c - 1];
diff = otherrd[c - 1] - rd[c - 1];
}
} else /* use 8x8 transform */ {
rd_check_segment_txsize(cpi, x, bsi, segmentation, TX_8X8, otherrd,
rd, &n, seg_mvs);
if (n == c) {
diff = rd[c - 1] - otherrd[c - 1];
base_rd = otherrd[c - 1];
}
}
}
if (n == c) {
if (base_rd < txfm_cache[ONLY_4X4]) {
txfm_cache[ONLY_4X4] = base_rd;
}
Minor cosmetic cleanups. Change-Id: I13d8ae754827368755575dd699a087b3b11f5b16
2013-02-16 01:39:22 +01:00
if (base_rd + diff < txfm_cache[ALLOW_8X8]) {
Prevent filling transform size cache with uninitialized values. The 32x32 value in case of splitmv was uninitialized. this leads to all kind of erratic behaviour down the line. Also fill in dummy values for superblocks in keyframes (the values are currently unused, but we run into integer overflows anyway, which makes detecting bad cases harder). Lastly, in case we did not find any RD value at all, don't set tx_diff to INT_MIN, but instead set it to zero (since if we couldn't find a mode, it's unlikely that any particular transform would have made that worse or better; rather, it's likely equally bad for all tx_sizes). Change-Id: If236fd3aa2037e5b398d03f3b1978fbbc5ce740e
2013-02-16 00:55:31 +01:00
txfm_cache[ALLOW_8X8] = txfm_cache[ALLOW_16X16] =
txfm_cache[ALLOW_32X32] = base_rd + diff;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
}
if (diff < 0) {
base_rd += diff + RDCOST(x->rdmult, x->rddiv, cost8x8, 0);
} else {
base_rd += RDCOST(x->rdmult, x->rddiv, cost4x4, 0);
}
if (base_rd < txfm_cache[TX_MODE_SELECT]) {
txfm_cache[TX_MODE_SELECT] = base_rd;
}
}
}
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
}
Modify motion estimation for SPLITMV mode 1. Search for block8x16/block16x8 uses block8x8's search results. 2. Check block4x4 only if block8x8 is chosen. (This hurts quality, which will be improved in another check-in.) 3. In block4x4 search, the previous block's result is used as MV predictor for next block. This change improves performance. Change-Id: I9dc089007ca08129fb6c11fe7692777ebb8647b0
2010-12-23 17:23:03 +01:00
Use configure checks for various inline keywords. Change-Id: I8508f1a3d3430f998bb9295f849e88e626a52a24
2013-02-06 21:45:28 +01:00
static INLINE void cal_step_param(int sr, int *sp) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int step = 0;
Modify motion estimation for SPLITMV mode 1. Search for block8x16/block16x8 uses block8x8's search results. 2. Check block4x4 only if block8x8 is chosen. (This hurts quality, which will be improved in another check-in.) 3. In block4x4 search, the previous block's result is used as MV predictor for next block. This change improves performance. Change-Id: I9dc089007ca08129fb6c11fe7692777ebb8647b0
2010-12-23 17:23:03 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (sr > MAX_FIRST_STEP) sr = MAX_FIRST_STEP;
else if (sr < 1) sr = 1;
Modify motion estimation for SPLITMV mode 1. Search for block8x16/block16x8 uses block8x8's search results. 2. Check block4x4 only if block8x8 is chosen. (This hurts quality, which will be improved in another check-in.) 3. In block4x4 search, the previous block's result is used as MV predictor for next block. This change improves performance. Change-Id: I9dc089007ca08129fb6c11fe7692777ebb8647b0
2010-12-23 17:23:03 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
while (sr >>= 1)
step++;
Modify motion estimation for SPLITMV mode 1. Search for block8x16/block16x8 uses block8x8's search results. 2. Check block4x4 only if block8x8 is chosen. (This hurts quality, which will be improved in another check-in.) 3. In block4x4 search, the previous block's result is used as MV predictor for next block. This change improves performance. Change-Id: I9dc089007ca08129fb6c11fe7692777ebb8647b0
2010-12-23 17:23:03 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
*sp = MAX_MVSEARCH_STEPS - 1 - step;
Modify motion estimation for SPLITMV mode 1. Search for block8x16/block16x8 uses block8x8's search results. 2. Check block4x4 only if block8x8 is chosen. (This hurts quality, which will be improved in another check-in.) 3. In block4x4 search, the previous block's result is used as MV predictor for next block. This change improves performance. Change-Id: I9dc089007ca08129fb6c11fe7692777ebb8647b0
2010-12-23 17:23:03 +01:00
}
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
static int rd_pick_best_mbsegmentation(VP9_COMP *cpi, MACROBLOCK *x,
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
int_mv *best_ref_mv,
int_mv *second_best_ref_mv,
int64_t best_rd,
int *mdcounts,
int *returntotrate,
int *returnyrate,
int *returndistortion,
int *skippable, int mvthresh,
int_mv seg_mvs[NB_PARTITIONINGS]
[16 /* n_blocks */]
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
[MAX_REF_FRAMES - 1],
int64_t txfm_cache[NB_TXFM_MODES]) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int i;
BEST_SEG_INFO bsi;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
MB_MODE_INFO * mbmi = &x->e_mbd.mode_info_context->mbmi;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
vpx_memset(&bsi, 0, sizeof(bsi));
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
for (i = 0; i < NB_TXFM_MODES; i++)
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
txfm_cache[i] = INT64_MAX;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
bsi.segment_rd = best_rd;
bsi.ref_mv = best_ref_mv;
bsi.second_ref_mv = second_best_ref_mv;
bsi.mvp.as_int = best_ref_mv->as_int;
bsi.mvthresh = mvthresh;
bsi.mdcounts = mdcounts;
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
bsi.txfm_size = TX_4X4;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
for (i = 0; i < 16; i++)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
bsi.modes[i] = ZERO4X4;
if (cpi->compressor_speed == 0) {
/* for now, we will keep the original segmentation order
when in best quality mode */
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
rd_check_segment(cpi, x, &bsi, PARTITIONING_16X8,
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
seg_mvs[PARTITIONING_16X8], txfm_cache);
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
rd_check_segment(cpi, x, &bsi, PARTITIONING_8X16,
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
seg_mvs[PARTITIONING_8X16], txfm_cache);
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
rd_check_segment(cpi, x, &bsi, PARTITIONING_8X8,
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
seg_mvs[PARTITIONING_8X8], txfm_cache);
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
rd_check_segment(cpi, x, &bsi, PARTITIONING_4X4,
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
seg_mvs[PARTITIONING_4X4], txfm_cache);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
} else {
int sr;
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
rd_check_segment(cpi, x, &bsi, PARTITIONING_8X8,
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
seg_mvs[PARTITIONING_8X8], txfm_cache);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (bsi.segment_rd < best_rd) {
int tmp_col_min = x->mv_col_min;
int tmp_col_max = x->mv_col_max;
int tmp_row_min = x->mv_row_min;
int tmp_row_max = x->mv_row_max;
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
vp9_clamp_mv_min_max(x, best_ref_mv);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/* Get 8x8 result */
bsi.sv_mvp[0].as_int = bsi.mvs[0].as_int;
bsi.sv_mvp[1].as_int = bsi.mvs[2].as_int;
bsi.sv_mvp[2].as_int = bsi.mvs[8].as_int;
bsi.sv_mvp[3].as_int = bsi.mvs[10].as_int;
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
/* Use 8x8 result as 16x8/8x16's predictor MV. Adjust search range
* according to the closeness of 2 MV. */
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/* block 8X16 */
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
sr = MAXF((abs(bsi.sv_mvp[0].as_mv.row - bsi.sv_mvp[2].as_mv.row)) >> 3,
(abs(bsi.sv_mvp[0].as_mv.col - bsi.sv_mvp[2].as_mv.col)) >> 3);
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
cal_step_param(sr, &bsi.sv_istep[0]);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
sr = MAXF((abs(bsi.sv_mvp[1].as_mv.row - bsi.sv_mvp[3].as_mv.row)) >> 3,
(abs(bsi.sv_mvp[1].as_mv.col - bsi.sv_mvp[3].as_mv.col)) >> 3);
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
cal_step_param(sr, &bsi.sv_istep[1]);
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
rd_check_segment(cpi, x, &bsi, PARTITIONING_8X16,
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
seg_mvs[PARTITIONING_8X16], txfm_cache);
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/* block 16X8 */
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
sr = MAXF((abs(bsi.sv_mvp[0].as_mv.row - bsi.sv_mvp[1].as_mv.row)) >> 3,
(abs(bsi.sv_mvp[0].as_mv.col - bsi.sv_mvp[1].as_mv.col)) >> 3);
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
cal_step_param(sr, &bsi.sv_istep[0]);
Modify motion estimation for SPLITMV mode 1. Search for block8x16/block16x8 uses block8x8's search results. 2. Check block4x4 only if block8x8 is chosen. (This hurts quality, which will be improved in another check-in.) 3. In block4x4 search, the previous block's result is used as MV predictor for next block. This change improves performance. Change-Id: I9dc089007ca08129fb6c11fe7692777ebb8647b0
2010-12-23 17:23:03 +01:00
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
sr = MAXF((abs(bsi.sv_mvp[2].as_mv.row - bsi.sv_mvp[3].as_mv.row)) >> 3,
(abs(bsi.sv_mvp[2].as_mv.col - bsi.sv_mvp[3].as_mv.col)) >> 3);
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
cal_step_param(sr, &bsi.sv_istep[1]);
Modify motion estimation for SPLITMV mode 1. Search for block8x16/block16x8 uses block8x8's search results. 2. Check block4x4 only if block8x8 is chosen. (This hurts quality, which will be improved in another check-in.) 3. In block4x4 search, the previous block's result is used as MV predictor for next block. This change improves performance. Change-Id: I9dc089007ca08129fb6c11fe7692777ebb8647b0
2010-12-23 17:23:03 +01:00
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
rd_check_segment(cpi, x, &bsi, PARTITIONING_16X8,
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
seg_mvs[PARTITIONING_16X8], txfm_cache);
Modify motion estimation for SPLITMV mode 1. Search for block8x16/block16x8 uses block8x8's search results. 2. Check block4x4 only if block8x8 is chosen. (This hurts quality, which will be improved in another check-in.) 3. In block4x4 search, the previous block's result is used as MV predictor for next block. This change improves performance. Change-Id: I9dc089007ca08129fb6c11fe7692777ebb8647b0
2010-12-23 17:23:03 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/* If 8x8 is better than 16x8/8x16, then do 4x4 search */
/* Not skip 4x4 if speed=0 (good quality) */
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
if (cpi->sf.no_skip_block4x4_search ||
bsi.segment_num == PARTITIONING_8X8) {
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
/* || (sv_segment_rd8x8-bsi.segment_rd) < sv_segment_rd8x8>>5) */
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
bsi.mvp.as_int = bsi.sv_mvp[0].as_int;
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
rd_check_segment(cpi, x, &bsi, PARTITIONING_4X4,
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
seg_mvs[PARTITIONING_4X4], txfm_cache);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Adjust MV borders for SPLITMV mode Add limits to avoid MV going out of range. Change-Id: I8a5deb40bf393488d29f694b5a56804d578e68b5
2010-12-28 19:23:07 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/* restore UMV window */
x->mv_col_min = tmp_col_min;
x->mv_col_max = tmp_col_max;
x->mv_row_min = tmp_row_min;
x->mv_row_max = tmp_row_max;
Changed segmentation check order In SPLITMV, the 8x8 segment will be checked first. If the 8x8 rd is better than the best, we check the other segments. Otherwise bail. Adjustments to the thresh_mult were necessary to make up for the initial quality loss. The performance improved by 20% (average) for good quality, speed 0 and speed 1, while the overall quality remained the same. Change-Id: I717aef401323c8a254fba3e9777d2a316c774cc3
2010-12-16 23:01:27 +01:00
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
vp8_rd_pick_best_mbsegmentation code restructure Moved the code from the segmentation loop into a function which is now called for each segment. This will allow us to change the segment order checking more easily. Change-Id: I9510d26f0acae5a73043fcca8f1984b121d3e052
2010-12-06 22:42:52 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/* set it to the best */
for (i = 0; i < 16; i++) {
BLOCKD *bd = &x->e_mbd.block[i];
Initial WebM release
2010-05-18 17:58:33 +02:00
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
bd->bmi.as_mv[0].as_int = bsi.mvs[i].as_int;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0)
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
bd->bmi.as_mv[1].as_int = bsi.second_mvs[i].as_int;
Move EOB to per-plane data Continue migrating data from BLOCKD/MACROBLOCKD to the per-plane structures. Change-Id: Ibbfa68d6da438d32dcbe8df68245ee28b0a2fa2c
2013-04-04 21:03:27 +02:00
x->e_mbd.plane[0].eobs[i] = bsi.eobs[i];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/* save partitions */
8x8 transform support in splitmv. For splitmv, where partitioning is 8x16, 16x8 or 8x8, this patch uses the 8x8 transform (instead of the 4x4) if txfm_mode is ALLOW_8X8 or ALLOW_16X16. For TX_MODE_SELECT, splitmv can indicate which of the 2 transform sizes (4x4 or 8x8) it wants to use. Gains (with hybridtx4x4/8x8/16x16 and tx_select experiments enabled) on derf: +0.9%, HD: +0.4%, STD/HD: +0.8% (SSIM or overall PSNR, both metrics show similar improvements). Change-Id: Ide954b000b415548ed92a7ac78e24f36e60fcf06
2012-10-22 20:49:00 +02:00
mbmi->txfm_size = bsi.txfm_size;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
mbmi->partitioning = bsi.segment_num;
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
x->partition_info->count = vp9_mbsplit_count[bsi.segment_num];
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
for (i = 0; i < x->partition_info->count; i++) {
int j;
Initial WebM release
2010-05-18 17:58:33 +02:00
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
j = vp9_mbsplit_offset[bsi.segment_num][i];
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
x->partition_info->bmi[i].mode = bsi.modes[j];
x->partition_info->bmi[i].mv.as_mv = bsi.mvs[j].as_mv;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
x->partition_info->bmi[i].second_mv.as_mv = bsi.second_mvs[j].as_mv;
}
/*
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
* used to set mbmi->mv.as_int
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
*/
x->partition_info->bmi[15].mv.as_int = bsi.mvs[15].as_int;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
x->partition_info->bmi[15].second_mv.as_int = bsi.second_mvs[15].as_int;
tokenize: convert skippable functions Use the common block walker to calculate skippability. Change-Id: I6721e42f065df237426c91c1d871ec226ba7cdcb
2013-04-11 20:14:31 +02:00
*returntotrate = bsi.r;
*returndistortion = bsi.d;
*returnyrate = bsi.segment_yrate;
*skippable = vp9_sby_is_skippable(&x->e_mbd, BLOCK_SIZE_MB16X16);
silent a lot of MSVC compiler warnings there are still a couple type of warning left, which are related to double constants assigned to float type. As those would be addressed by the conversion of transforms into integer version. This commit has left those un-dealt with. Change-Id: I48fd9b489c0c27ad6b543f4177423419f929f2bb
2012-11-05 23:22:59 +01:00
return (int)(bsi.segment_rd);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
static void mv_pred(VP9_COMP *cpi, MACROBLOCK *x,
Use standard integer types for pixel values and coefficients. For coefficients, use int16_t (instead of short); for pixel values in 16-bit intermediates, use uint16_t (instead of unsigned short); for all others, use uint8_t (instead of unsigned char). Change-Id: I3619cd9abf106c3742eccc2e2f5e89a62774f7da
2012-12-19 00:31:19 +01:00
uint8_t *ref_y_buffer, int ref_y_stride,
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
int ref_frame, enum BlockSize block_size ) {
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
MACROBLOCKD *xd = &x->e_mbd;
MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
int_mv this_mv;
int i;
Removing TRUE and FALSE macro definitions. Using regular 0 and 1 constants now. Change-Id: Ie763503cbb727847cc8f1d6506cd6f2ee607f056
2013-04-16 00:24:39 +02:00
int zero_seen = 0;
unused variables & warnings Change-Id: Ibea3bc80eb26a975faaa60268bbc93237f82bc57
2012-11-29 18:02:47 +01:00
int best_index = 0;
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
int best_sad = INT_MAX;
int this_sad = INT_MAX;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
BLOCK *b = &x->block[0];
Use standard integer types for pixel values and coefficients. For coefficients, use int16_t (instead of short); for pixel values in 16-bit intermediates, use uint16_t (instead of unsigned short); for all others, use uint8_t (instead of unsigned char). Change-Id: I3619cd9abf106c3742eccc2e2f5e89a62774f7da
2012-12-19 00:31:19 +01:00
uint8_t *src_y_ptr = *(b->base_src);
uint8_t *ref_y_ptr;
unused variables & warnings Change-Id: Ibea3bc80eb26a975faaa60268bbc93237f82bc57
2012-11-29 18:02:47 +01:00
int row_offset, col_offset;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
// Get the sad for each candidate reference mv
Simplification of MVref search. As we are no longer able to sort the candidate mvrefs in both encoder and decode and given that the cost of explicit signalling has proved prohibitive, it no longer makes sense to find more than 2 candidates. This patch: Modifies and simplifies add_candidate_mv() Removes the forced addition of a 0 vector in the MAX_MV_REF_CANDIDATES-1 position (in preparation to reducing MAX_MV_REF_CANDIDATES to 2). Re-orders the addition of candidates slightly. This actually gives small gains (circa 0.2% on std-hd) A subsequent patch will remove NEW_MVREF experiment, reduce MAX_MV_REF_CANDIDATES to 2 and remove distance weights as these are implicit now in the order. Change-Id: I3dbe1a6f8a1a18b3c108257069c22a1141a207a4
2013-04-19 12:10:16 +02:00
for (i = 0; i < MAX_MV_REF_CANDIDATES; i++) {
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
this_mv.as_int = mbmi->ref_mvs[ref_frame][i].as_int;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
// The list is at an end if we see 0 for a second time.
if (!this_mv.as_int && zero_seen)
break;
zero_seen = zero_seen || !this_mv.as_int;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
row_offset = this_mv.as_mv.row >> 3;
col_offset = this_mv.as_mv.col >> 3;
ref_y_ptr = ref_y_buffer + (ref_y_stride * row_offset) + col_offset;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
// Find sad for current vector.
this_sad = cpi->fn_ptr[block_size].sdf(src_y_ptr, b->src_stride,
ref_y_ptr, ref_y_stride,
0x7fffffff);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
// Note if it is the best so far.
if (this_sad < best_sad) {
best_sad = this_sad;
best_index = i;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
}
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
// Note the index of the mv that worked best in the reference list.
x->mv_best_ref_index[ref_frame] = best_index;
Improve MV prediction accuracy to achieve performance gain Add vp8_mv_pred() to better predict starting MV for NEWMV mode in vp8_rd_pick_inter_mode(). Set different search ranges according to MV prediction accuracy, which improves encoder performance without hurting the quality. Also, as Yaowu suggested, using diamond search result as full search starting point and therefore adjusting(reducing) full search range helps the performance. Change-Id: Ie4a3c8df87e697c1f4f6e2ddb693766bba1b77b6
2010-12-03 17:26:21 +01:00
}
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
static void set_i8x8_block_modes(MACROBLOCK *x, int modes[4]) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int i;
MACROBLOCKD *xd = &x->e_mbd;
for (i = 0; i < 4; i++) {
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
int ib = vp9_i8x8_block[i];
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
xd->mode_info_context->bmi[ib + 0].as_mode.first = modes[i];
xd->mode_info_context->bmi[ib + 1].as_mode.first = modes[i];
xd->mode_info_context->bmi[ib + 4].as_mode.first = modes[i];
xd->mode_info_context->bmi[ib + 5].as_mode.first = modes[i];
// printf("%d,%d,%d,%d\n",
// modes[0], modes[1], modes[2], modes[3]);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
for (i = 0; i < 16; i++) {
xd->block[i].bmi = xd->mode_info_context->bmi[i];
}
enabled 8x8 intra prediction modes on inter frames This commit enabled the usage of 8x8 intra prediction modes on inter frames. There are a few TODO items related to this: 1)baseline entropy need be calibrated; 2)cost of UV need to be done more properly rather than using decision only relying on Y; 3)Threshold for allowing picking 8x8 intra prediction should be lowered to lower than the B_PRED. Even with all the TODOs, tests showed consistent gain on derf set ~0.1% (PSNR:0.08% and SSIM:0.14%). It is assumed that 8x8 intra prediction will help more on large resolution clips, especially with above TODOs addressed. Change-Id: I398ada49dfc32575cfab962a569c2885111ae3ba
2011-12-01 01:25:00 +01:00
}
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
extern void vp9_calc_ref_probs(int *count, vp9_prob *probs);
static void estimate_curframe_refprobs(VP9_COMP *cpi, vp9_prob mod_refprobs[3], int pred_ref) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int norm_cnt[MAX_REF_FRAMES];
const int *const rfct = cpi->count_mb_ref_frame_usage;
int intra_count = rfct[INTRA_FRAME];
int last_count = rfct[LAST_FRAME];
int gf_count = rfct[GOLDEN_FRAME];
int arf_count = rfct[ALTREF_FRAME];
// Work out modified reference frame probabilities to use where prediction
// of the reference frame fails
if (pred_ref == INTRA_FRAME) {
norm_cnt[0] = 0;
norm_cnt[1] = last_count;
norm_cnt[2] = gf_count;
norm_cnt[3] = arf_count;
Name space clean up. Preparation for project restructuring. Added vp9_ prefix on some function names that have global scope. Added static declaration on some that dont. Change-Id: If072f78b4300e8c17cfeed82c5d17b59946dcc5e
2012-10-29 14:44:18 +01:00
vp9_calc_ref_probs(norm_cnt, mod_refprobs);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
mod_refprobs[0] = 0; // This branch implicit
} else if (pred_ref == LAST_FRAME) {
norm_cnt[0] = intra_count;
norm_cnt[1] = 0;
norm_cnt[2] = gf_count;
norm_cnt[3] = arf_count;
Name space clean up. Preparation for project restructuring. Added vp9_ prefix on some function names that have global scope. Added static declaration on some that dont. Change-Id: If072f78b4300e8c17cfeed82c5d17b59946dcc5e
2012-10-29 14:44:18 +01:00
vp9_calc_ref_probs(norm_cnt, mod_refprobs);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
mod_refprobs[1] = 0; // This branch implicit
} else if (pred_ref == GOLDEN_FRAME) {
norm_cnt[0] = intra_count;
norm_cnt[1] = last_count;
norm_cnt[2] = 0;
norm_cnt[3] = arf_count;
Name space clean up. Preparation for project restructuring. Added vp9_ prefix on some function names that have global scope. Added static declaration on some that dont. Change-Id: If072f78b4300e8c17cfeed82c5d17b59946dcc5e
2012-10-29 14:44:18 +01:00
vp9_calc_ref_probs(norm_cnt, mod_refprobs);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
mod_refprobs[2] = 0; // This branch implicit
} else {
norm_cnt[0] = intra_count;
norm_cnt[1] = last_count;
norm_cnt[2] = gf_count;
norm_cnt[3] = 0;
Name space clean up. Preparation for project restructuring. Added vp9_ prefix on some function names that have global scope. Added static declaration on some that dont. Change-Id: If072f78b4300e8c17cfeed82c5d17b59946dcc5e
2012-10-29 14:44:18 +01:00
vp9_calc_ref_probs(norm_cnt, mod_refprobs);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
mod_refprobs[2] = 0; // This branch implicit
}
Rewrite reference frame costing in the RD loop. I now see I didn't write a very long description, so let's do it here then. We took a pretty big quality hit (0.1-0.2%) from my recent fix of the inversion of arguments to vp8_cost_bit() in the RD reference frame costing. I looked into it and basically the costing prevented us from switching reference frames. This is of course silly, since each frame codes its own prob_intra_coded, so using last frame cost indications as a limiting factor can never be right. Here, I've rewritten that code to estimate costings based partially on statistics from progress on current frame encoding. Overall, this gives us a ~0.2%-0.3% improvement over what we had previously before my argument-inversion-fix, and thus about ~0.4% over current git (on derf-set), and a little more (0.5-1.0%) on HD/STD-HD/YT. Change-Id: I79ebd4ccec4d6edbf0e152d9590d103ba2747775
2012-05-15 02:39:42 +02:00
}
Use configure checks for various inline keywords. Change-Id: I8508f1a3d3430f998bb9295f849e88e626a52a24
2013-02-06 21:45:28 +01:00
static INLINE unsigned weighted_cost(vp9_prob *tab0, vp9_prob *tab1,
int idx, int val, int weight) {
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
unsigned cost0 = tab0[idx] ? vp9_cost_bit(tab0[idx], val) : 0;
unsigned cost1 = tab1[idx] ? vp9_cost_bit(tab1[idx], val) : 0;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// weight is 16-bit fixed point, so this basically calculates:
// 0.5 + weight * cost1 + (1.0 - weight) * cost0
return (0x8000 + weight * cost1 + (0x10000 - weight) * cost0) >> 16;
Rewrite reference frame costing in the RD loop. I now see I didn't write a very long description, so let's do it here then. We took a pretty big quality hit (0.1-0.2%) from my recent fix of the inversion of arguments to vp8_cost_bit() in the RD reference frame costing. I looked into it and basically the costing prevented us from switching reference frames. This is of course silly, since each frame codes its own prob_intra_coded, so using last frame cost indications as a limiting factor can never be right. Here, I've rewritten that code to estimate costings based partially on statistics from progress on current frame encoding. Overall, this gives us a ~0.2%-0.3% improvement over what we had previously before my argument-inversion-fix, and thus about ~0.4% over current git (on derf-set), and a little more (0.5-1.0%) on HD/STD-HD/YT. Change-Id: I79ebd4ccec4d6edbf0e152d9590d103ba2747775
2012-05-15 02:39:42 +02:00
}
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
static void estimate_ref_frame_costs(VP9_COMP *cpi, int segment_id, unsigned int *ref_costs) {
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
VP9_COMMON *cm = &cpi->common;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
MACROBLOCKD *xd = &cpi->mb.e_mbd;
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
vp9_prob *mod_refprobs;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
unsigned int cost;
int pred_ref;
int pred_flag;
int pred_ctx;
int i;
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
vp9_prob pred_prob, new_pred_prob;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int seg_ref_active;
int seg_ref_count = 0;
seg_common: give all exported symbols a vp9_ prefix. Change-Id: Ie8ba934a757acec1c80ac37ab9307c9a2783254e
2012-10-30 06:15:27 +01:00
seg_ref_active = vp9_segfeature_active(xd,
segment_id,
SEG_LVL_REF_FRAME);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (seg_ref_active) {
seg_common: give all exported symbols a vp9_ prefix. Change-Id: Ie8ba934a757acec1c80ac37ab9307c9a2783254e
2012-10-30 06:15:27 +01:00
seg_ref_count = vp9_check_segref(xd, segment_id, INTRA_FRAME) +
vp9_check_segref(xd, segment_id, LAST_FRAME) +
vp9_check_segref(xd, segment_id, GOLDEN_FRAME) +
vp9_check_segref(xd, segment_id, ALTREF_FRAME);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
// Get the predicted reference for this mb
Name space clean up. Preparation for project restructuring. Added vp9_ prefix on some function names that have global scope. Added static declaration on some that dont. Change-Id: If072f78b4300e8c17cfeed82c5d17b59946dcc5e
2012-10-29 14:44:18 +01:00
pred_ref = vp9_get_pred_ref(cm, xd);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// Get the context probability for the prediction flag (based on last frame)
Name space clean up. Preparation for project restructuring. Added vp9_ prefix on some function names that have global scope. Added static declaration on some that dont. Change-Id: If072f78b4300e8c17cfeed82c5d17b59946dcc5e
2012-10-29 14:44:18 +01:00
pred_prob = vp9_get_pred_prob(cm, xd, PRED_REF);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// Predict probability for current frame based on stats so far
Name space clean up. Preparation for project restructuring. Added vp9_ prefix on some function names that have global scope. Added static declaration on some that dont. Change-Id: If072f78b4300e8c17cfeed82c5d17b59946dcc5e
2012-10-29 14:44:18 +01:00
pred_ctx = vp9_get_pred_context(cm, xd, PRED_REF);
Consistently use get_prob(), clip_prob() and newly added clip_pixel(). Add a function clip_pixel() to clip a pixel value to the [0,255] range of allowed values, and use this where-ever appropriate (e.g. prediction, reconstruction). Likewise, consistently use the recently added function clip_prob(), which calculates a binary probability in the [1,255] range. If possible, try to use get_prob() or its sister get_binary_prob() to calculate binary probabilities, for consistency. Since in some places, this means that binary probability calculations are changed (we use {255,256}*count0/(total) in a range of places, and all of these are now changed to use 256*count0+(total>>1)/total), this changes the encoding result, so this patch warrants some extensive testing. Change-Id: Ibeeff8d886496839b8e0c0ace9ccc552351f7628
2012-12-10 21:09:07 +01:00
new_pred_prob = get_binary_prob(cpi->ref_pred_count[pred_ctx][0],
cpi->ref_pred_count[pred_ctx][1]);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// Get the set of probabilities to use if prediction fails
mod_refprobs = cm->mod_refprobs[pred_ref];
// For each possible selected reference frame work out a cost.
for (i = 0; i < MAX_REF_FRAMES; i++) {
if (seg_ref_active && seg_ref_count == 1) {
cost = 0;
} else {
pred_flag = (i == pred_ref);
// Get the prediction for the current mb
cost = weighted_cost(&pred_prob, &new_pred_prob, 0,
pred_flag, cpi->seg0_progress);
if (cost > 1024) cost = 768; // i.e. account for 4 bits max.
// for incorrectly predicted cases
if (! pred_flag) {
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
vp9_prob curframe_mod_refprobs[3];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (cpi->seg0_progress) {
estimate_curframe_refprobs(cpi, curframe_mod_refprobs, pred_ref);
} else {
vpx_memset(curframe_mod_refprobs, 0, sizeof(curframe_mod_refprobs));
}
cost += weighted_cost(mod_refprobs, curframe_mod_refprobs, 0,
(i != INTRA_FRAME), cpi->seg0_progress);
if (i != INTRA_FRAME) {
cost += weighted_cost(mod_refprobs, curframe_mod_refprobs, 1,
(i != LAST_FRAME), cpi->seg0_progress);
if (i != LAST_FRAME) {
cost += weighted_cost(mod_refprobs, curframe_mod_refprobs, 2,
(i != GOLDEN_FRAME), cpi->seg0_progress);
}
}
}
}
ref_costs[i] = cost;
}
Implementation of new prediction model for reference frame coding. This check in uses the common prediction interface functions to code reference frame. Some updates made regarding the impact of the new code in rd loop but there remain TODOs in this regard. Change-Id: I9da3ed5dfdaa489e0903ab33258b0767a585567f
2012-01-28 13:20:14 +01:00
}
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
static void store_coding_context(MACROBLOCK *x, PICK_MODE_CONTEXT *ctx,
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int mode_index,
PARTITION_INFO *partition,
int_mv *ref_mv,
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
int_mv *second_ref_mv,
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
int64_t comp_pred_diff[NB_PREDICTION_TYPES],
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
int64_t txfm_size_diff[NB_TXFM_MODES]) {
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
MACROBLOCKD *const xd = &x->e_mbd;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// Take a snapshot of the coding context so it can be
// restored if we decide to encode this way
Re-enable support for static_threshold (encode_breakout). Change-Id: Ibd7380f478d3127f9db91d0a4fd2fd0dfde961ab
2013-01-07 20:02:14 +01:00
ctx->skip = x->skip;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
ctx->best_mode_index = mode_index;
vpx_memcpy(&ctx->mic, xd->mode_info_context,
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
sizeof(MODE_INFO));
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
if (partition)
vpx_memcpy(&ctx->partition_info, partition,
sizeof(PARTITION_INFO));
ctx->best_ref_mv.as_int = ref_mv->as_int;
ctx->second_best_ref_mv.as_int = second_ref_mv->as_int;
libvpx: Fix some warnings. Change-Id: If8be8b9d28a29631f29c46daea8a226ab3580610
2013-01-18 18:44:23 +01:00
ctx->single_pred_diff = (int)comp_pred_diff[SINGLE_PREDICTION_ONLY];
ctx->comp_pred_diff = (int)comp_pred_diff[COMP_PREDICTION_ONLY];
ctx->hybrid_pred_diff = (int)comp_pred_diff[HYBRID_PREDICTION];
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
memcpy(ctx->txfm_rd_diff, txfm_size_diff, sizeof(ctx->txfm_rd_diff));
Added save coding context & modified MV bounds Added code to save the coding context in vp8_rd_pick_inter_mode when the coding mode is forced to ARF(0,0). Also, modified the MV bounds computation to comply with the change in MV border from 32 to 64 pixels. Change-Id: I96963a6f5f4d04ce84c807ae11e0635177c3ad6c
2012-04-13 19:26:49 +02:00
}
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
static void setup_buffer_inter(VP9_COMP *cpi, MACROBLOCK *x,
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
int idx, MV_REFERENCE_FRAME frame_type,
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
enum BlockSize block_size,
Pass macroblock index to pick inter functions Pass the current mb row and column around rather than the recon_yoffset and recon_uvoffset, since those offsets will change from predictor to predictor, based on the reference frame selection. Change-Id: If3f9df059e00f5048ca729d3d083ff428e1859c1
2013-02-07 19:09:05 +01:00
int mb_row, int mb_col,
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
int_mv frame_nearest_mv[MAX_REF_FRAMES],
int_mv frame_near_mv[MAX_REF_FRAMES],
rdopt: make locally used symbols static. Change-Id: Id33c448084ca5c1c0525a05c3c4f35d6b0805fbe
2012-10-30 02:04:33 +01:00
int frame_mdcounts[4][4],
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
YV12_BUFFER_CONFIG yv12_mb[4],
struct scale_factors scale[MAX_REF_FRAMES]) {
VP9_COMMON *cm = &cpi->common;
Use ref_frame_map vice active_ref_idx on the encoder This patch makes the encoder's use of ref_frame_map and active_ref_idx consistent with the decoder. ref_frame_map[] maps a reference buffer index to its actual location in the yv12_fb array, since many references may share an underlying buffer. active_ref_idx[] mirrors cpi->{lst,gld,alt}_fb_idx, holding the active references in each slot. This also fixes a bug in setup_buffer_inter() where the incorrect reference was used to populate the scaling factors. Change-Id: Id3728f6d77cffcd27c248903bf51f9c3e594287e
2013-02-22 20:22:03 +01:00
YV12_BUFFER_CONFIG *yv12 = &cm->yv12_fb[cpi->common.ref_frame_map[idx]];
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
Mv ref candidates cut to 2. Further simplification of mvref search to return only the top two candidates. Distance weights removed as the test order reflects distance anyway. Change-Id: I0518cab7280258fec2058670add4f853fab7b855
2013-04-19 16:40:36 +02:00
int use_prev_in_find_mv_refs;
MV reference changes Extend experiment to use both vectors from MBs coded using compound prediction as candidates. In final sort only consider best 4 candidates for now but make sure 0,0 is always one of them. Other minor changes to new MV reference code. Pass in Mv list to vp8_find_best_ref_mvs(). Change-Id: Ib96220c33c6b80bd1d5e0fbe8b68121be7997095
2012-09-07 13:46:41 +02:00
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
// set up scaling factors
scale[frame_type] = cpi->common.active_ref_scale[frame_type - 1];
scale[frame_type].x_offset_q4 =
(mb_col * 16 * scale[frame_type].x_num / scale[frame_type].x_den) & 0xf;
scale[frame_type].y_offset_q4 =
(mb_row * 16 * scale[frame_type].y_num / scale[frame_type].y_den) & 0xf;
// TODO(jkoleszar): Is the UV buffer ever used here? If so, need to make this
// use the UV scaling factors.
setup_pred_block(&yv12_mb[frame_type], yv12, mb_row, mb_col,
&scale[frame_type], &scale[frame_type]);
New Motion Reference Search Alternative strategy for finding a list of candidate motion vectors to use as reference values in mv coding and as nearest and near. Sort by sad in vp8_find_best_ref_mvs() rather than just pick the best. Allow 0,0 as a best ref option but not a nearest or near unless there are no alternatives. Encode/Decode verified on at least some clips. Some commented out experimental and stats code still in place. Gain over existing code averages about 1% on derf (alll metrics) with improvement on all clips. Other test results pending. The entropy coding of the mode (nearest/near etc) still depends upon and requires the old "findnear" code so this needs looking at and may provide room for further gains. Change-Id: I871d7cba1d1c379c4bad9bcccce1fb19c46b8247
2012-08-24 16:44:01 +02:00
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
// Gets an initial list of candidate vectors from neighbours and orders them
Code cleanup: lower case variable names. Renaming Width to width, Height to height and Version to version in several structs and function signatures. Change-Id: I084c3f7e747cb2ce3345aff27a3dff9b13a87543
2013-03-21 00:41:30 +01:00
use_prev_in_find_mv_refs = cm->width == cm->last_width &&
cm->height == cm->last_height &&
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
!cpi->common.error_resilient_mode;
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
vp9_find_mv_refs(&cpi->common, xd, xd->mode_info_context,
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
use_prev_in_find_mv_refs ? xd->prev_mode_info_context : NULL,
Name space clean up. Preparation for project restructuring. Added vp9_ prefix on some function names that have global scope. Added static declaration on some that dont. Change-Id: If072f78b4300e8c17cfeed82c5d17b59946dcc5e
2012-10-29 14:44:18 +01:00
frame_type,
mbmi->ref_mvs[frame_type],
cpi->common.ref_frame_sign_bias);
New Motion Reference Search Alternative strategy for finding a list of candidate motion vectors to use as reference values in mv coding and as nearest and near. Sort by sad in vp8_find_best_ref_mvs() rather than just pick the best. Allow 0,0 as a best ref option but not a nearest or near unless there are no alternatives. Encode/Decode verified on at least some clips. Some commented out experimental and stats code still in place. Gain over existing code averages about 1% on derf (alll metrics) with improvement on all clips. Other test results pending. The entropy coding of the mode (nearest/near etc) still depends upon and requires the old "findnear" code so this needs looking at and may provide room for further gains. Change-Id: I871d7cba1d1c379c4bad9bcccce1fb19c46b8247
2012-08-24 16:44:01 +02:00
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
// Candidate refinement carried out at encoder and decoder
Adds an error-resilient mode with test Adds an error-resilient mode where frames can be continued to be decoded even when there are errors (due to network losses) on a prior frame. Specifically, backward updates are turned off and probabilities of various symbols are reset to defaults at the beginning of each frame. Further, the last frame's mvs are not used for the mv reference list, and the sorting of the initial list based on search on previous frames is turned off as well. Also adds a test where an arbitrary set of frames are skipped from decoding to simulate errors. The test verifies (1) that if the error frames are droppable - i.e. frame buffer updates have been turned off - there are no mismatch errors for the remaining frames after the error frames; and (2) if the error-frames are non droppable, there are not only no decoding errors but the mismatch PSNR between the decoder's version of the post-error frames and the encoder's version is at least 20 dB. Change-Id: Ie6e2bcd436b1e8643270356d3a930e8989ff52a5
2013-01-15 15:43:35 +01:00
vp9_find_best_ref_mvs(xd,
MV reference changes Extend experiment to use both vectors from MBs coded using compound prediction as candidates. In final sort only consider best 4 candidates for now but make sure 0,0 is always one of them. Other minor changes to new MV reference code. Pass in Mv list to vp8_find_best_ref_mvs(). Change-Id: Ib96220c33c6b80bd1d5e0fbe8b68121be7997095
2012-09-07 13:46:41 +02:00
mbmi->ref_mvs[frame_type],
Changed to use reference mv as nearest mv The reference motion vector selected by surrounding pixels that has the best matching score is used as nearest motion vector. The change has shown consistent gain on all test sets, compression gains range from .2% to .6%. The variation is largely dependent on various other experiments on or off. Change-Id: I5552e1c2f6fc57c3e8818a5ee41ffda89af05e75
2012-08-06 19:51:20 +02:00
&frame_nearest_mv[frame_type],
&frame_near_mv[frame_type]);
Modified mv prediction. Modified the mv_pred() fuunction that chooses a centre point from which to start step searches to use the top candidate vectors chosen previously. Some gains (mainly on HD and tested with SB off). Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036% Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
2012-11-26 12:29:08 +01:00
// Further refinement that is encode side only to test the top few candidates
// in full and choose the best as the centre point for subsequent searches.
Fix pulsing issue with scaling Updates the YV12_BUFFER_CONFIG structure to be crop-aware. The exiting width/height parameters are left unchanged, storing the width and height algined to a 16 byte boundary. The cropped dimensions are added as new fields. This fixes a nasty visual pulse when switching between scaled and unscaled frame dimensions due to a mismatch between the scaling ratio and the 16-byte aligned sizes. Change-Id: Id4a3f6aea6b9b9ae38bdfa1b87b7eb2cfcdd57b6
2013-03-14 01:09:05 +01:00
// The current implementation doesn't support scaling.
if (scale[frame_type].x_num == scale[frame_type].x_den &&
scale[frame_type].y_num == scale[frame_type].y_den)
mv_pred(cpi, x, yv12_mb[frame_type].y_buffer, yv12->y_stride,
frame_type, block_size);
Refactor setting up MVs and buffers in rdopt.c Change-Id: I94f08f2cfb37d11d71050dca3200f1d391269a6a
2012-08-08 03:55:28 +02:00
}
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
static void model_rd_from_var_lapndz(int var, int n, int qstep,
int *rate, int *dist) {
// This function models the rate and distortion for a Laplacian
// source with given variance when quantized with a uniform quantizer
// with given stepsize. The closed form expressions are in:
// Hang and Chen, "Source Model for transform video coder and its
// application - Part I: Fundamental Theory", IEEE Trans. Circ.
// Sys. for Video Tech., April 1997.
// The function is implemented as piecewise approximation to the
// exact computation.
// TODO(debargha): Implement the functions by interpolating from a
// look-up table
vp9_clear_system_state();
{
double D, R;
double s2 = (double) var / n;
double s = sqrt(s2);
double x = qstep / s;
if (x > 1.0) {
double y = exp(-x / 2);
double y2 = y * y;
D = 2.069981728764738 * y2 - 2.764286806516079 * y + 1.003956960819275;
R = 0.924056758535089 * y2 + 2.738636469814024 * y - 0.005169662030017;
} else {
double x2 = x * x;
D = 0.075303187668830 * x2 + 0.004296954321112 * x - 0.000413209252807;
if (x > 0.125)
R = 1 / (-0.03459733614226 * x2 + 0.36561675733603 * x +
0.1626989668625);
else
R = -1.442252874826093 * log(x) + 1.944647760719664;
}
if (R < 0) {
*rate = 0;
*dist = var;
} else {
*rate = (n * R * 256 + 0.5);
*dist = (n * D * s2 + 0.5);
}
}
vp9_clear_system_state();
}
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
static enum BlockSize y_to_uv_block_size(enum BlockSize bs) {
switch (bs) {
case BLOCK_64X64: return BLOCK_32X32;
Add rectangular block size variance/sad functions. With this, the RD loop properly supports rectangular blocks. Change-Id: Iece79048fb4e84741ee1ada982da129a7bf00470
2013-04-15 19:00:34 +02:00
#if CONFIG_SBSEGMENT
case BLOCK_64X32: return BLOCK_32X16;
case BLOCK_32X64: return BLOCK_16X32;
#endif
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
case BLOCK_32X32: return BLOCK_16X16;
Add rectangular block size variance/sad functions. With this, the RD loop properly supports rectangular blocks. Change-Id: Iece79048fb4e84741ee1ada982da129a7bf00470
2013-04-15 19:00:34 +02:00
#if CONFIG_SBSEGMENT
case BLOCK_32X16: return BLOCK_16X8;
case BLOCK_16X32: return BLOCK_8X16;
#endif
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
case BLOCK_16X16: return BLOCK_8X8;
default:
assert(0);
return -1;
}
}
static enum BlockSize y_bsizet_to_block_size(BLOCK_SIZE_TYPE bs) {
switch (bs) {
case BLOCK_SIZE_SB64X64: return BLOCK_64X64;
Add rectangular block size variance/sad functions. With this, the RD loop properly supports rectangular blocks. Change-Id: Iece79048fb4e84741ee1ada982da129a7bf00470
2013-04-15 19:00:34 +02:00
#if CONFIG_SBSEGMENT
case BLOCK_SIZE_SB64X32: return BLOCK_64X32;
case BLOCK_SIZE_SB32X64: return BLOCK_32X64;
#endif
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
case BLOCK_SIZE_SB32X32: return BLOCK_32X32;
Add rectangular block size variance/sad functions. With this, the RD loop properly supports rectangular blocks. Change-Id: Iece79048fb4e84741ee1ada982da129a7bf00470
2013-04-15 19:00:34 +02:00
#if CONFIG_SBSEGMENT
case BLOCK_SIZE_SB32X16: return BLOCK_32X16;
case BLOCK_SIZE_SB16X32: return BLOCK_16X32;
#endif
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
case BLOCK_SIZE_MB16X16: return BLOCK_16X16;
default:
assert(0);
return -1;
}
}
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
static int64_t handle_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
BLOCK_SIZE_TYPE bsize,
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
int mdcounts[4], int64_t txfm_cache[],
int *rate2, int *distortion, int *skippable,
int *compmode_cost,
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
int *compmode_interintra_cost,
#endif
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
int *rate_y, int *distortion_y,
int *rate_uv, int *distortion_uv,
int *mode_excluded, int *disable_skip,
Pass macroblock index to pick inter functions Pass the current mb row and column around rather than the recon_yoffset and recon_uvoffset, since those offsets will change from predictor to predictor, based on the reference frame selection. Change-Id: If3f9df059e00f5048ca729d3d083ff428e1859c1
2013-02-07 19:09:05 +01:00
int mode_index,
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
INTERPOLATIONFILTERTYPE *best_filter,
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
int_mv frame_mv[MB_MODE_COUNT]
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
[MAX_REF_FRAMES],
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
YV12_BUFFER_CONFIG *scaled_ref_frame,
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
int mb_row, int mb_col) {
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
const int bw = 1 << mb_width_log2(bsize), bh = 1 << mb_height_log2(bsize);
const enum BlockSize block_size = y_bsizet_to_block_size(bsize);
const enum BlockSize uv_block_size = y_to_uv_block_size(block_size);
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
VP9_COMMON *cm = &cpi->common;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
MACROBLOCKD *xd = &x->e_mbd;
MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
BLOCK *b = &x->block[0];
BLOCKD *d = &xd->block[0];
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
const int is_comp_pred = (mbmi->second_ref_frame > 0);
#if CONFIG_COMP_INTERINTRA_PRED
const int is_comp_interintra_pred = (mbmi->second_ref_frame == INTRA_FRAME);
#endif
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
const int num_refs = is_comp_pred ? 2 : 1;
const int this_mode = mbmi->mode;
int i;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
int refs[2] = { mbmi->ref_frame,
(mbmi->second_ref_frame < 0 ? 0 : mbmi->second_ref_frame) };
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
int_mv cur_mv[2];
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
int_mv ref_mv[2];
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
int64_t this_rd = 0;
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
unsigned char tmp_ybuf[64 * 64];
unsigned char tmp_ubuf[32 * 32];
unsigned char tmp_vbuf[32 * 32];
int pred_exists = 0;
int interpolating_intpel_seen = 0;
int intpel_mv;
int64_t rd, best_rd = INT64_MAX;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
switch (this_mode) {
case NEWMV:
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
ref_mv[0] = mbmi->ref_mvs[refs[0]][0];
ref_mv[1] = mbmi->ref_mvs[refs[1]][0];
Added update-able mv-ref probabilities. Part of NEW_MVREF experiment. Added update-able probabilities. Change-Id: I5a4fcf4aaed1d0d1dac980f69d535639a3d59401
2012-12-10 13:38:48 +01:00
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
if (is_comp_pred) {
if (frame_mv[NEWMV][refs[0]].as_int == INVALID_MV ||
frame_mv[NEWMV][refs[1]].as_int == INVALID_MV)
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
return INT64_MAX;
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
*rate2 += vp9_mv_bit_cost(&frame_mv[NEWMV][refs[0]],
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
&ref_mv[0],
remove macros obfuscating mv costing cleanup Change-Id: I565eee40d900e0441ad211b65ac829fc5b93d94a
2012-11-09 00:44:39 +01:00
x->nmvjointcost, x->mvcost, 96,
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
x->e_mbd.allow_high_precision_mv);
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
*rate2 += vp9_mv_bit_cost(&frame_mv[NEWMV][refs[1]],
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
&ref_mv[1],
remove macros obfuscating mv costing cleanup Change-Id: I565eee40d900e0441ad211b65ac829fc5b93d94a
2012-11-09 00:44:39 +01:00
x->nmvjointcost, x->mvcost, 96,
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
x->e_mbd.allow_high_precision_mv);
} else {
Move pre, second_pre to per-plane MACROBLOCKD data Continue moving framebuffers to per-plane data. Change-Id: I237e5a998b364c4ec20316e7249206c0bff8631a
2013-04-20 04:16:14 +02:00
struct buf_2d backup_yv12[MAX_MB_PLANE] = {{0}};
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
int bestsme = INT_MAX;
int further_steps, step_param = cpi->sf.first_step;
int sadpb = x->sadperbit16;
int_mv mvp_full, tmp_mv;
int sr = 0;
int tmp_col_min = x->mv_col_min;
int tmp_col_max = x->mv_col_max;
int tmp_row_min = x->mv_row_min;
int tmp_row_max = x->mv_row_max;
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
if (scaled_ref_frame) {
Move pre, second_pre to per-plane MACROBLOCKD data Continue moving framebuffers to per-plane data. Change-Id: I237e5a998b364c4ec20316e7249206c0bff8631a
2013-04-20 04:16:14 +02:00
int i;
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
// Swap out the reference frame for a version that's been scaled to
// match the resolution of the current frame, allowing the existing
// motion search code to be used without additional modifications.
Move pre, second_pre to per-plane MACROBLOCKD data Continue moving framebuffers to per-plane data. Change-Id: I237e5a998b364c4ec20316e7249206c0bff8631a
2013-04-20 04:16:14 +02:00
for (i = 0; i < MAX_MB_PLANE; i++)
backup_yv12[i] = xd->plane[i].pre[0];
setup_pre_planes(xd, scaled_ref_frame, NULL, mb_row, mb_col,
NULL, NULL);
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
}
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
vp9_clamp_mv_min_max(x, &ref_mv[0]);
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
Code cleanup: lower case variable names. Renaming Width to width, Height to height and Version to version in several structs and function signatures. Change-Id: I084c3f7e747cb2ce3345aff27a3dff9b13a87543
2013-03-21 00:41:30 +01:00
sr = vp9_init_search_range(cpi->common.width, cpi->common.height);
Extend diff MV limit from +/-256 to +/-1024 Increase the motion search range by 4x. Change MV_CLASS tree of the entropy coding to allow two additional mv classes to cover the extended motion vector limit. The codec determines the effective motion search range conditioned on the actual frame dimension. It provides coding gains: stdhd 0.39% yt 0.56% hd 0.47% Major coding performance gains are packed in several sequences with intense motion activities, e.g., ped_1080p gains 7% at high bit-rates, and on average 3%. TODO: Need to further tune the rate control and motion search units. Change-Id: Ib842540a6796fbee5a797809433ef6a477c6d78d
2013-03-06 21:02:15 +01:00
Added update-able mv-ref probabilities. Part of NEW_MVREF experiment. Added update-able probabilities. Change-Id: I5a4fcf4aaed1d0d1dac980f69d535639a3d59401
2012-12-10 13:38:48 +01:00
// mvp_full.as_int = ref_mv[0].as_int;
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
mvp_full.as_int =
Added update-able mv-ref probabilities. Part of NEW_MVREF experiment. Added update-able probabilities. Change-Id: I5a4fcf4aaed1d0d1dac980f69d535639a3d59401
2012-12-10 13:38:48 +01:00
mbmi->ref_mvs[refs[0]][x->mv_best_ref_index[refs[0]]].as_int;
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
mvp_full.as_mv.col >>= 3;
mvp_full.as_mv.row >>= 3;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
// adjust search range according to sr from mv prediction
step_param = MAX(step_param, sr);
// Further step/diamond searches as necessary
further_steps = (cpi->sf.max_step_search_steps - 1) - step_param;
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
bestsme = vp9_full_pixel_diamond(cpi, x, b, d, &mvp_full, step_param,
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
sadpb, further_steps, 1,
&cpi->fn_ptr[block_size],
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
&ref_mv[0], &tmp_mv);
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
x->mv_col_min = tmp_col_min;
x->mv_col_max = tmp_col_max;
x->mv_row_min = tmp_row_min;
x->mv_row_max = tmp_row_max;
if (bestsme < INT_MAX) {
int dis; /* TODO: use dis in distortion calculation later. */
unsigned int sse;
cpi->find_fractional_mv_step(x, b, d, &tmp_mv,
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
&ref_mv[0],
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
x->errorperbit,
&cpi->fn_ptr[block_size],
remove macros obfuscating mv costing cleanup Change-Id: I565eee40d900e0441ad211b65ac829fc5b93d94a
2012-11-09 00:44:39 +01:00
x->nmvjointcost, x->mvcost,
&dis, &sse);
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
}
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
d->bmi.as_mv[0].as_int = tmp_mv.as_int;
frame_mv[NEWMV][refs[0]].as_int = d->bmi.as_mv[0].as_int;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
// Add the new motion vector cost to our rolling cost variable
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
*rate2 += vp9_mv_bit_cost(&tmp_mv, &ref_mv[0],
remove macros obfuscating mv costing cleanup Change-Id: I565eee40d900e0441ad211b65ac829fc5b93d94a
2012-11-09 00:44:39 +01:00
x->nmvjointcost, x->mvcost,
96, xd->allow_high_precision_mv);
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
// restore the predictor, if required
if (scaled_ref_frame) {
Move pre, second_pre to per-plane MACROBLOCKD data Continue moving framebuffers to per-plane data. Change-Id: I237e5a998b364c4ec20316e7249206c0bff8631a
2013-04-20 04:16:14 +02:00
int i;
for (i = 0; i < MAX_MB_PLANE; i++)
xd->plane[i].pre[0] = backup_yv12[i];
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
}
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
}
break;
case NEARMV:
Change definition of NearestMV. This commit makes the NearestMV match the chosen best reference MV. It can be a 0,0 or non zero vector which means the the compound nearest mv mode can combine a 0,0 and a non zero vector. Change-Id: I2213d09996ae2916e53e6458d7d110350dcffd7a
2013-02-05 11:28:26 +01:00
case NEARESTMV:
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
case ZEROMV:
default:
break;
}
for (i = 0; i < num_refs; ++i) {
cur_mv[i] = frame_mv[this_mode][refs[i]];
// Clip "next_nearest" so that it does not extend to far out of image
Clamp inferred motion vectors only Clamp only the motion vectors inferred from neighboring reference macroblocks. The motion vectors obtained through motion search in NEWMV mode are constrained during the search process, which allows a relatively larger referencing region than the inferred mvs. Hence further clamping the best mv provided by the motion search may affect the efficacy of NEWMV mode. Synchronized the decoding process. The decoded mvs in NEWMV modes should be guaranteed to fit in the effective range. Put a mv range clamping function there for security purpose. This improves the coding performance of high motion sequences, e.g., derf set: foreman 0.233% husky 0.175% icd 0.135% mother_daughter 0.337% pamphlet 0.561% stdhd set: blue_sky 0.408% city 0.455% also saw sunflower goes down by -0.469%. Change-Id: I3fcbba669e56dab779857a8126a91b926e899cb5
2013-04-06 19:00:53 +02:00
if (this_mode == NEWMV)
assert(!clamp_mv2(&cur_mv[i], xd));
else
clamp_mv2(&cur_mv[i], xd);
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
if (mv_check_bounds(x, &cur_mv[i]))
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
return INT64_MAX;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
mbmi->mv[i].as_int = cur_mv[i].as_int;
}
/* We don't include the cost of the second reference here, because there
* are only three options: Last/Golden, ARF/Last or Golden/ARF, or in other
* words if you present them in that order, the second one is always known
* if the first is known */
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
*compmode_cost = vp9_cost_bit(vp9_get_pred_prob(cm, xd, PRED_COMP),
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
is_comp_pred);
New inter mode context This change is a fix / extension of the newbestrefmv experiment. As such it is presented without IFDEF. The change creates a new context for coding inter modes in vp9_find_mv_refs(). This replaces the context that was previously calculated in vp9_find_near_mvs(). The new context is unoptimized and not necessarily any better at this stage (results pending), but eliminates the need for a legacy call to vp9_find_near_mvs(). Based on numbers from Scott, this could help decode speed by several %. In a later patch I will add support for forward update of context (assuming this helps) and refine the context as necessary. Change-Id: I1cd991b82c8df86cc02237a34185e6d67510698a
2012-11-12 16:09:25 +01:00
*rate2 += vp9_cost_mv_ref(cpi, this_mode,
mbmi->mb_mode_context[mbmi->ref_frame]);
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
if (!is_comp_pred) {
*compmode_interintra_cost = vp9_cost_bit(cm->fc.interintra_prob,
is_comp_interintra_pred);
if (is_comp_interintra_pred) {
*compmode_interintra_cost +=
x->mbmode_cost[xd->frame_type][mbmi->interintra_mode];
#if SEPARATE_INTERINTRA_UV
*compmode_interintra_cost +=
x->intra_uv_mode_cost[xd->frame_type][mbmi->interintra_uv_mode];
#endif
}
}
#endif
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
pred_exists = 0;
interpolating_intpel_seen = 0;
// Are all MVs integer pel for Y and UV
intpel_mv = (mbmi->mv[0].as_mv.row & 15) == 0 &&
(mbmi->mv[0].as_mv.col & 15) == 0;
if (is_comp_pred)
intpel_mv &= (mbmi->mv[1].as_mv.row & 15) == 0 &&
(mbmi->mv[1].as_mv.col & 15) == 0;
// Search for best switchable filter by checking the variance of
// pred error irrespective of whether the filter will be used
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
if (bsize != BLOCK_SIZE_MB16X16) {
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
int switchable_filter_index, newbest;
int tmp_rate_y_i = 0, tmp_rate_u_i = 0, tmp_rate_v_i = 0;
int tmp_dist_y_i = 0, tmp_dist_u_i = 0, tmp_dist_v_i = 0;
for (switchable_filter_index = 0;
switchable_filter_index < VP9_SWITCHABLE_FILTERS;
++switchable_filter_index) {
int rs = 0;
mbmi->interp_filter = vp9_switchable_interp[switchable_filter_index];
vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
if (cpi->common.mcomp_filter_type == SWITCHABLE) {
const int c = vp9_get_pred_context(cm, xd, PRED_SWITCHABLE_INTERP);
const int m = vp9_switchable_interp_map[mbmi->interp_filter];
rs = SWITCHABLE_INTERP_RATE_FACTOR * x->switchable_interp_costs[c][m];
}
if (interpolating_intpel_seen && intpel_mv &&
vp9_is_interpolating_filter[mbmi->interp_filter]) {
rd = RDCOST(x->rdmult, x->rddiv,
rs + tmp_rate_y_i + tmp_rate_u_i + tmp_rate_v_i,
tmp_dist_y_i + tmp_dist_u_i + tmp_dist_v_i);
} else {
unsigned int sse, var;
int tmp_rate_y, tmp_rate_u, tmp_rate_v;
int tmp_dist_y, tmp_dist_u, tmp_dist_v;
Enable inter predictor for rectangular block size Combine superblock inter predictors into a unified function that allows configurable block width and height. The inter predictions of block sizes smaller than 16x16 are handled differently. To be continued on merging them later. Change-Id: I14075959dd5e221f00c205c99ca35c1c31ef728e
2013-04-12 20:25:53 +02:00
vp9_build_inter_predictors_sb(xd, mb_row, mb_col, bsize);
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
var = cpi->fn_ptr[block_size].vf(*(b->base_src), b->src_stride,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[0].dst.buf,
xd->plane[0].dst.stride,
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
&sse);
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
// Note our transform coeffs are 8 times an orthogonal transform.
// Hence quantizer step is also 8 times. To get effective quantizer
// we need to divide by 8 before sending to modeling function.
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
model_rd_from_var_lapndz(var, 16 * bw * 16 * bh,
xd->block[0].dequant[1] >> 3,
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
&tmp_rate_y, &tmp_dist_y);
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
var = cpi->fn_ptr[uv_block_size].vf(x->src.u_buffer, x->src.uv_stride,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[1].dst.buf,
xd->plane[1].dst.stride,
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
&sse);
model_rd_from_var_lapndz(var, 8 * bw * 8 * bh,
xd->block[16].dequant[1] >> 3,
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
&tmp_rate_u, &tmp_dist_u);
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
var = cpi->fn_ptr[uv_block_size].vf(x->src.v_buffer, x->src.uv_stride,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[2].dst.buf,
xd->plane[1].dst.stride,
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
&sse);
model_rd_from_var_lapndz(var, 8 * bw * 8 * bh,
xd->block[20].dequant[1] >> 3,
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
&tmp_rate_v, &tmp_dist_v);
rd = RDCOST(x->rdmult, x->rddiv,
rs + tmp_rate_y + tmp_rate_u + tmp_rate_v,
tmp_dist_y + tmp_dist_u + tmp_dist_v);
if (!interpolating_intpel_seen && intpel_mv &&
vp9_is_interpolating_filter[mbmi->interp_filter]) {
tmp_rate_y_i = tmp_rate_y;
tmp_rate_u_i = tmp_rate_u;
tmp_rate_v_i = tmp_rate_v;
tmp_dist_y_i = tmp_dist_y;
tmp_dist_u_i = tmp_dist_u;
tmp_dist_v_i = tmp_dist_v;
}
}
newbest = (switchable_filter_index == 0 || rd < best_rd);
if (newbest) {
best_rd = rd;
*best_filter = mbmi->interp_filter;
}
if ((cm->mcomp_filter_type == SWITCHABLE && newbest) ||
(cm->mcomp_filter_type != SWITCHABLE &&
cm->mcomp_filter_type == mbmi->interp_filter)) {
int i;
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
for (i = 0; i < 16 * bh; ++i)
vpx_memcpy(tmp_ybuf + i * 16 * bw,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[0].dst.buf + i * xd->plane[0].dst.stride,
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
sizeof(unsigned char) * 16 * bw);
for (i = 0; i < 8 * bh; ++i)
vpx_memcpy(tmp_ubuf + i * 8 * bw,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[1].dst.buf + i * xd->plane[1].dst.stride,
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
sizeof(unsigned char) * 8 * bw);
for (i = 0; i < 8 * bh; ++i)
vpx_memcpy(tmp_vbuf + i * 8 * bw,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[2].dst.buf + i * xd->plane[1].dst.stride,
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
sizeof(unsigned char) * 8 * bw);
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
pred_exists = 1;
}
interpolating_intpel_seen |=
intpel_mv && vp9_is_interpolating_filter[mbmi->interp_filter];
}
Merge superblocks (32x32) experiment. Change-Id: I0df99742029834a85c4933652b0587cf5b6b2587
2013-01-08 19:29:22 +01:00
} else {
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
int switchable_filter_index, newbest;
int tmp_rate_y_i = 0, tmp_rate_u_i = 0, tmp_rate_v_i = 0;
int tmp_dist_y_i = 0, tmp_dist_u_i = 0, tmp_dist_v_i = 0;
for (switchable_filter_index = 0;
switchable_filter_index < VP9_SWITCHABLE_FILTERS;
++switchable_filter_index) {
int rs = 0;
mbmi->interp_filter = vp9_switchable_interp[switchable_filter_index];
vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
if (cpi->common.mcomp_filter_type == SWITCHABLE) {
const int c = vp9_get_pred_context(cm, xd, PRED_SWITCHABLE_INTERP);
const int m = vp9_switchable_interp_map[mbmi->interp_filter];
rs = SWITCHABLE_INTERP_RATE_FACTOR * x->switchable_interp_costs[c][m];
}
if (interpolating_intpel_seen && intpel_mv &&
vp9_is_interpolating_filter[mbmi->interp_filter]) {
rd = RDCOST(x->rdmult, x->rddiv,
rs + tmp_rate_y_i + tmp_rate_u_i + tmp_rate_v_i,
tmp_dist_y_i + tmp_dist_u_i + tmp_dist_v_i);
} else {
unsigned int sse, var;
int tmp_rate_y, tmp_rate_u, tmp_rate_v;
int tmp_dist_y, tmp_dist_u, tmp_dist_v;
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
vp9_build_inter_predictors_sb(xd, mb_row, mb_col, BLOCK_SIZE_MB16X16);
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
var = vp9_variance16x16(*(b->base_src), b->src_stride,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[0].dst.buf, xd->plane[0].dst.stride,
&sse);
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
// Note our transform coeffs are 8 times an orthogonal transform.
// Hence quantizer step is also 8 times. To get effective quantizer
// we need to divide by 8 before sending to modeling function.
model_rd_from_var_lapndz(var, 16 * 16, xd->block[0].dequant[1] >> 3,
&tmp_rate_y, &tmp_dist_y);
var = vp9_variance8x8(x->src.u_buffer, x->src.uv_stride,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[1].dst.buf, xd->plane[1].dst.stride,
&sse);
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
model_rd_from_var_lapndz(var, 8 * 8, xd->block[16].dequant[1] >> 3,
&tmp_rate_u, &tmp_dist_u);
var = vp9_variance8x8(x->src.v_buffer, x->src.uv_stride,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[2].dst.buf, xd->plane[1].dst.stride,
&sse);
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
model_rd_from_var_lapndz(var, 8 * 8, xd->block[20].dequant[1] >> 3,
&tmp_rate_v, &tmp_dist_v);
rd = RDCOST(x->rdmult, x->rddiv,
rs + tmp_rate_y + tmp_rate_u + tmp_rate_v,
tmp_dist_y + tmp_dist_u + tmp_dist_v);
if (!interpolating_intpel_seen && intpel_mv &&
vp9_is_interpolating_filter[mbmi->interp_filter]) {
tmp_rate_y_i = tmp_rate_y;
tmp_rate_u_i = tmp_rate_u;
tmp_rate_v_i = tmp_rate_v;
tmp_dist_y_i = tmp_dist_y;
tmp_dist_u_i = tmp_dist_u;
tmp_dist_v_i = tmp_dist_v;
}
}
newbest = (switchable_filter_index == 0 || rd < best_rd);
if (newbest) {
best_rd = rd;
*best_filter = mbmi->interp_filter;
}
if ((cm->mcomp_filter_type == SWITCHABLE && newbest) ||
(cm->mcomp_filter_type != SWITCHABLE &&
cm->mcomp_filter_type == mbmi->interp_filter)) {
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
int i;
for (i = 0; i < 16 * bh; ++i)
vpx_memcpy(tmp_ybuf + i * 16 * bw,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[0].dst.buf + i * xd->plane[0].dst.stride,
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
sizeof(unsigned char) * 16 * bw);
for (i = 0; i < 8 * bh; ++i)
vpx_memcpy(tmp_ubuf + i * 8 * bw,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[1].dst.buf + i * xd->plane[1].dst.stride,
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
sizeof(unsigned char) * 8 * bw);
for (i = 0; i < 8 * bh; ++i)
vpx_memcpy(tmp_vbuf + i * 8 * bw,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[2].dst.buf + i * xd->plane[1].dst.stride,
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
sizeof(unsigned char) * 8 * bw);
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
pred_exists = 1;
}
interpolating_intpel_seen |=
intpel_mv && vp9_is_interpolating_filter[mbmi->interp_filter];
}
}
// Set the appripriate filter
if (cm->mcomp_filter_type != SWITCHABLE)
mbmi->interp_filter = cm->mcomp_filter_type;
else
mbmi->interp_filter = *best_filter;
vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
if (pred_exists) {
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
// FIXME(rbultje): mb code still predicts into xd->predictor
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
for (i = 0; i < bh * 16; ++i)
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
vpx_memcpy(xd->plane[0].dst.buf + i * xd->plane[0].dst.stride,
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
tmp_ybuf + i * bw * 16, sizeof(unsigned char) * bw * 16);
for (i = 0; i < bh * 8; ++i)
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
vpx_memcpy(xd->plane[1].dst.buf + i * xd->plane[1].dst.stride,
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
tmp_ubuf + i * bw * 8, sizeof(unsigned char) * bw * 8);
for (i = 0; i < bh * 8; ++i)
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
vpx_memcpy(xd->plane[2].dst.buf + i * xd->plane[1].dst.stride,
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
tmp_vbuf + i * bw * 8, sizeof(unsigned char) * bw * 8);
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
} else {
// Handles the special case when a filter that is not in the
// switchable list (ex. bilinear, 6-tap) is indicated at the frame level
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
vp9_build_inter_predictors_sb(xd, mb_row, mb_col, bsize);
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
}
if (cpi->common.mcomp_filter_type == SWITCHABLE) {
const int c = vp9_get_pred_context(cm, xd, PRED_SWITCHABLE_INTERP);
const int m = vp9_switchable_interp_map[mbmi->interp_filter];
*rate2 += SWITCHABLE_INTERP_RATE_FACTOR * x->switchable_interp_costs[c][m];
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
}
if (cpi->active_map_enabled && x->active_ptr[0] == 0)
x->skip = 1;
else if (x->encode_breakout) {
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
unsigned int var, sse;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
int threshold = (xd->block[0].dequant[1]
* xd->block[0].dequant[1] >> 4);
if (threshold < x->encode_breakout)
threshold = x->encode_breakout;
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
if (bsize != BLOCK_SIZE_MB16X16) {
var = cpi->fn_ptr[block_size].vf(*(b->base_src), b->src_stride,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[0].dst.buf,
xd->plane[0].dst.stride,
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
&sse);
Merge superblocks (32x32) experiment. Change-Id: I0df99742029834a85c4933652b0587cf5b6b2587
2013-01-08 19:29:22 +01:00
} else {
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
var = vp9_variance16x16(*(b->base_src), b->src_stride,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[0].dst.buf, xd->plane[0].dst.stride,
&sse);
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
}
silent a lot of MSVC compiler warnings there are still a couple type of warning left, which are related to double constants assigned to float type. As those would be addressed by the conversion of transforms into integer version. This commit has left those un-dealt with. Change-Id: I48fd9b489c0c27ad6b543f4177423419f929f2bb
2012-11-05 23:22:59 +01:00
if ((int)sse < threshold) {
Remove some Y2-related code. Change-Id: I4f46d142c2a8d1e8a880cfac63702dcbfb999b78
2013-02-15 19:15:42 +01:00
unsigned int q2dc = xd->block[0].dequant[0];
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
/* If there is no codeable 2nd order dc
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
or a very small uniform pixel change change */
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
if ((sse - var < q2dc * q2dc >> 4) ||
(sse / 2 > var && sse - var < 64)) {
// Check u and v to make sure skip is ok
int sse2;
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
if (bsize != BLOCK_SIZE_MB16X16) {
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
unsigned int sse2u, sse2v;
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
// FIXME(rbultje): mb predictors predict into xd->predictor
var = cpi->fn_ptr[uv_block_size].vf(x->src.u_buffer, x->src.uv_stride,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[1].dst.buf,
xd->plane[1].dst.stride, &sse2u);
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
var = cpi->fn_ptr[uv_block_size].vf(x->src.v_buffer, x->src.uv_stride,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[2].dst.buf,
xd->plane[1].dst.stride, &sse2v);
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
sse2 = sse2u + sse2v;
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
} else {
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
unsigned int sse2u, sse2v;
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
var = vp9_variance8x8(x->src.u_buffer, x->src.uv_stride,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[1].dst.buf, xd->plane[1].dst.stride,
&sse2u);
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
var = vp9_variance8x8(x->src.v_buffer, x->src.uv_stride,
Move dst to per-plane MACROBLOCKD data First in a series of commits moving the framebuffers pointers to per-plane data, so that they can be indexed numerically rather than by name. Change-Id: I6e0d60fd4d51e6375c384eb7321776564df21775
2013-04-20 00:52:17 +02:00
xd->plane[2].dst.buf, xd->plane[1].dst.stride,
&sse2v);
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
sse2 = sse2u + sse2v;
}
if (sse2 * 2 < threshold) {
x->skip = 1;
*distortion = sse + sse2;
*rate2 = 500;
/* for best_yrd calculation */
*rate_uv = 0;
*distortion_uv = sse2;
*disable_skip = 1;
this_rd = RDCOST(x->rdmult, x->rddiv, *rate2, *distortion);
}
}
}
}
if (!x->skip) {
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
int skippable_y, skippable_uv;
// Y cost and distortion
super_block_yrd(cpi, x, rate_y, distortion_y, &skippable_y,
bsize, txfm_cache);
*rate2 += *rate_y;
*distortion += *distortion_y;
super_block_uvrd(cm, x, rate_uv, distortion_uv,
&skippable_uv, bsize);
*rate2 += *rate_uv;
*distortion += *distortion_uv;
*skippable = skippable_y && skippable_uv;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
}
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
if (!(*mode_excluded)) {
if (is_comp_pred) {
*mode_excluded = (cpi->common.comp_pred_mode == SINGLE_PREDICTION_ONLY);
} else {
*mode_excluded = (cpi->common.comp_pred_mode == COMP_PREDICTION_ONLY);
}
#if CONFIG_COMP_INTERINTRA_PRED
if (is_comp_interintra_pred && !cm->use_interintra) *mode_excluded = 1;
#endif
}
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
return this_rd; // if 0, this will be re-calculated by caller
}
Place non-static function prototypes in a header file. Change-Id: I7cd21b9f1e69f4e0b3338bfe27b3c67e4b47de58
2012-11-02 19:22:57 +01:00
static void rd_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
Pass macroblock index to pick inter functions Pass the current mb row and column around rather than the recon_yoffset and recon_uvoffset, since those offsets will change from predictor to predictor, based on the reference frame selection. Change-Id: If3f9df059e00f5048ca729d3d083ff428e1859c1
2013-02-07 19:09:05 +01:00
int mb_row, int mb_col,
Place non-static function prototypes in a header file. Change-Id: I7cd21b9f1e69f4e0b3338bfe27b3c67e4b47de58
2012-11-02 19:22:57 +01:00
int *returnrate, int *returndistortion,
int64_t *returnintra) {
Initial support for resolution changes on P-frames Allows inter-frames to change resolution. Currently these are almost equivalent to keyframes, as only intra prediction modes are allowed, but without the other context resets that occur on keyframes. Change-Id: Icd1a2a5af0d9462cc792588427b0a1f5b12e40d3
2013-02-07 00:54:52 +01:00
static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG,
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
VP9_ALT_FLAG };
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
VP9_COMMON *cm = &cpi->common;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
MACROBLOCKD *xd = &x->e_mbd;
union b_mode_info best_bmodes[16];
MB_MODE_INFO best_mbmode;
PARTITION_INFO best_partition;
int_mv best_ref_mv, second_best_ref_mv;
MB_PREDICTION_MODE this_mode;
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
MB_PREDICTION_MODE best_mode = DC_PRED;
Inconsistent reference to MacroblockD data structure. Cleaned up some inconsistent references using both xd-> and x->e_mbd. to access the same data structure in the same function. Change-Id: Ieb496fa22bf1feec6aa7bc70b941ea4f16e0f8b5
2012-10-29 11:36:11 +01:00
MB_MODE_INFO * mbmi = &xd->mode_info_context->mbmi;
Some refactoring of mcomp functions. Change-Id: Ic7a7cb1199b085e98ede0e634619b3077c348d57
2012-08-08 01:44:26 +02:00
int i, best_mode_index = 0;
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
int mode8x8[4];
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
unsigned char segment_id = mbmi->segment_id;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int mode_index;
int mdcounts[4];
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
int rate, distortion;
int rate2, distortion2;
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
int64_t best_txfm_rd[NB_TXFM_MODES];
int64_t best_txfm_diff[NB_TXFM_MODES];
Refactor to remove some arguments from vp8_rd_pick_inter_mode Change-Id: I8e72279cc68f34d269705f06cdaf8f3d06eed635
2012-08-10 03:25:29 +02:00
int64_t best_pred_diff[NB_PREDICTION_TYPES];
int64_t best_pred_rd[NB_PREDICTION_TYPES];
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t best_rd = INT64_MAX, best_intra_rd = INT64_MAX;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
int is_best_interintra = 0;
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t best_intra16_rd = INT64_MAX;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
int best_intra16_mode = DC_PRED;
#if SEPARATE_INTERINTRA_UV
int best_intra16_uv_mode = DC_PRED;
#endif
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
#endif
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t best_overall_rd = INT64_MAX;
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
INTERPOLATIONFILTERTYPE best_filter = SWITCHABLE;
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
INTERPOLATIONFILTERTYPE tmp_best_filter = SWITCHABLE;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
int uv_intra_rate[2], uv_intra_distortion[2], uv_intra_rate_tokenonly[2];
int uv_intra_skippable[2];
MB_PREDICTION_MODE uv_intra_mode[2];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int rate_y, UNINITIALIZED_IS_SAFE(rate_uv);
Fix compiler warnings Change-Id: I887cdcf90b87736f2a3805420aa41eed93324bee
2012-08-22 00:08:54 +02:00
int distortion_uv = INT_MAX;
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t best_yrd = INT64_MAX;
Moved ref frame calculations Moved ref frame calculations to outside of the mode_index loop. Change-Id: I06103fc7e8af88b54b84443acf6691d29b1272ac
2011-01-11 21:00:00 +01:00
Refactor RD to take same codepath for single and comp pred Change-Id: Id38baf1b89648ef534e28be72f583137871f920c
2012-08-10 01:07:41 +02:00
int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int frame_mdcounts[4][4];
Pass macroblock index to pick inter functions Pass the current mb row and column around rather than the recon_yoffset and recon_uvoffset, since those offsets will change from predictor to predictor, based on the reference frame selection. Change-Id: If3f9df059e00f5048ca729d3d083ff428e1859c1
2013-02-07 19:09:05 +01:00
YV12_BUFFER_CONFIG yv12_mb[4];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
unsigned int ref_costs[MAX_REF_FRAMES];
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
int_mv seg_mvs[NB_PARTITIONINGS][16 /* n_blocks */][MAX_REF_FRAMES - 1];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Problem of over smoothing with intra modes. In some cases intra modes in inter frames give an over smoothed appearance. Especially with noisy but flat content. Also in some cases there were problems with key frame sizing again with very flat but noisy content. These are temporary changes to help alleviate the visual problems but will almost certainly hurt metric results especially at the very low data rate end. Change-Id: I11549179a19277ffc283d9788bc70168f2a8bdc9
2012-12-14 18:49:46 +01:00
int intra_cost_penalty = 20 * vp9_dc_quant(cpi->common.base_qindex,
Renaming y1dc_delta_q, uvdc_delta_q, uvac_delta_q fields from VP9Common. New names are y_dc_delta_q, uv_dc_delta_q, uv_ac_delta_q. Change-Id: I4acae1fc23a4697ce2c5a5becb8dc28ef0a4b552
2013-04-17 00:05:52 +02:00
cpi->common.y_dc_delta_q);
Problem of over smoothing with intra modes. In some cases intra modes in inter frames give an over smoothed appearance. Especially with noisy but flat content. Also in some cases there were problems with key frame sizing again with very flat but noisy content. These are temporary changes to help alleviate the visual problems but will almost certainly hurt metric results especially at the very low data rate end. Change-Id: I11549179a19277ffc283d9788bc70168f2a8bdc9
2012-12-14 18:49:46 +01:00
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
struct scale_factors scale_factor[4];
Shut up some compiler warnings on gcc Change-Id: I8548aebd3c39797b253eef472bfe1c40f2f7ec59
2012-08-14 01:20:20 +02:00
vpx_memset(mode8x8, 0, sizeof(mode8x8));
Refactor RD to take same codepath for single and comp pred Change-Id: Id38baf1b89648ef534e28be72f583137871f920c
2012-08-10 01:07:41 +02:00
vpx_memset(&frame_mv, 0, sizeof(frame_mv));
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
vpx_memset(&best_mbmode, 0, sizeof(best_mbmode));
vpx_memset(&best_bmodes, 0, sizeof(best_bmodes));
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
vpx_memset(&x->mb_context[xd->sb_index][xd->mb_index], 0,
sizeof(PICK_MODE_CONTEXT));
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Refactor RD to take same codepath for single and comp pred Change-Id: Id38baf1b89648ef534e28be72f583137871f920c
2012-08-10 01:07:41 +02:00
for (i = 0; i < MAX_REF_FRAMES; i++)
frame_mv[NEWMV][i].as_int = INVALID_MV;
Refactor to remove some arguments from vp8_rd_pick_inter_mode Change-Id: I8e72279cc68f34d269705f06cdaf8f3d06eed635
2012-08-10 03:25:29 +02:00
for (i = 0; i < NB_PREDICTION_TYPES; ++i)
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
best_pred_rd[i] = INT64_MAX;
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
for (i = 0; i < NB_TXFM_MODES; i++)
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
best_txfm_rd[i] = INT64_MAX;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Use SPLITMV_PARTITIONING instead of a plain integer type. This can be used to distinguish between 16x8, 8x16, 8x8 and 4x4 partitioning modes when choosing splitmv as a MB mode. Change-Id: Idc8b59772e1a80ccc4ad44d63c5c2ec3fc061a3c
2012-10-22 20:25:48 +02:00
for (i = 0; i < NB_PARTITIONINGS; i++) {
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
int j, k;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
for (j = 0; j < 16; j++)
for (k = 0; k < MAX_REF_FRAMES - 1; k++)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
seg_mvs[i][j][k].as_int = INVALID_MV;
}
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
if (cpi->ref_frame_flags & VP9_LAST_FLAG) {
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
setup_buffer_inter(cpi, x, cpi->lst_fb_idx,
Pass macroblock index to pick inter functions Pass the current mb row and column around rather than the recon_yoffset and recon_uvoffset, since those offsets will change from predictor to predictor, based on the reference frame selection. Change-Id: If3f9df059e00f5048ca729d3d083ff428e1859c1
2013-02-07 19:09:05 +01:00
LAST_FRAME, BLOCK_16X16, mb_row, mb_col,
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
frame_mv[NEARESTMV], frame_mv[NEARMV],
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
frame_mdcounts, yv12_mb, scale_factor);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
if (cpi->ref_frame_flags & VP9_GOLD_FLAG) {
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
setup_buffer_inter(cpi, x, cpi->gld_fb_idx,
Pass macroblock index to pick inter functions Pass the current mb row and column around rather than the recon_yoffset and recon_uvoffset, since those offsets will change from predictor to predictor, based on the reference frame selection. Change-Id: If3f9df059e00f5048ca729d3d083ff428e1859c1
2013-02-07 19:09:05 +01:00
GOLDEN_FRAME, BLOCK_16X16, mb_row, mb_col,
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
frame_mv[NEARESTMV], frame_mv[NEARMV],
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
frame_mdcounts, yv12_mb, scale_factor);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
if (cpi->ref_frame_flags & VP9_ALT_FLAG) {
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
setup_buffer_inter(cpi, x, cpi->alt_fb_idx,
Pass macroblock index to pick inter functions Pass the current mb row and column around rather than the recon_yoffset and recon_uvoffset, since those offsets will change from predictor to predictor, based on the reference frame selection. Change-Id: If3f9df059e00f5048ca729d3d083ff428e1859c1
2013-02-07 19:09:05 +01:00
ALTREF_FRAME, BLOCK_16X16, mb_row, mb_col,
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
frame_mv[NEARESTMV], frame_mv[NEARMV],
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
frame_mdcounts, yv12_mb, scale_factor);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
*returnintra = INT64_MAX;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
mbmi->ref_frame = INTRA_FRAME;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/* Initialize zbin mode boost for uv costing */
cpi->zbin_mode_boost = 0;
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
vp9_update_zbin_extra(cpi, x);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
xd->mode_info_context->mbmi.mode = DC_PRED;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
for (i = 0; i <= TX_8X8; i++) {
mbmi->txfm_size = i;
rd_pick_intra_sbuv_mode(cpi, x, &uv_intra_rate[i],
&uv_intra_rate_tokenonly[i],
&uv_intra_distortion[i],
&uv_intra_skippable[i],
BLOCK_SIZE_MB16X16);
uv_intra_mode[i] = mbmi->uv_mode;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
// Get estimates of reference frame costs for each reference frame
// that depend on the current prediction etc.
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
estimate_ref_frame_costs(cpi, segment_id, ref_costs);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
for (mode_index = 0; mode_index < MAX_MODES; ++mode_index) {
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t this_rd = INT64_MAX;
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
int disable_skip = 0, skippable = 0;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int other_cost = 0;
int compmode_cost = 0;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
int compmode_interintra_cost = 0;
#endif
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int mode_excluded = 0;
Use transform-size selection for i8x8_pred also. Change-Id: Iecb282fc89f9b5145ef31c5eda294ad42bc32a5d
2012-10-11 02:18:22 +02:00
int64_t txfm_cache[NB_TXFM_MODES] = { 0 };
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
YV12_BUFFER_CONFIG *scaled_ref_frame;
refactored code that checks if a macroblock is skippable Change-Id: I4ea6d819bbbde312792c4f813ab63ea50cf0cd1d
2012-02-22 03:10:18 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// These variables hold are rolling total cost and distortion for this mode
rate2 = 0;
distortion2 = 0;
rate_y = 0;
rate_uv = 0;
Reset x->skip for each iteration in the RD loop. This prevents ill-defined behaviour, such as setting x->skip for a mode that is excluded because of frame-level flags (e.g. filter selection, compound prediction selection), then not breaking out of the RD loop because the mode is not allowed, but keeping the flag on. Whatever mode is iterated through next in the RD loop will then carry this flag, and all sort of bad stuff happens, such as x->skip being set on intra pred modes. Change-Id: I5bec46b36e38292174acb1c564b3caf00a9b4b9a
2013-01-14 21:43:12 +01:00
x->skip = 0;
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
this_mode = vp9_mode_order[mode_index].mode;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
mbmi->mode = this_mode;
mbmi->uv_mode = DC_PRED;
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
mbmi->ref_frame = vp9_mode_order[mode_index].ref_frame;
mbmi->second_ref_frame = vp9_mode_order[mode_index].second_ref_frame;
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
mbmi->interp_filter = cm->mcomp_filter_type;
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
set_scale_factors(xd, mbmi->ref_frame, mbmi->second_ref_frame,
scale_factor);
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
Adds support for switchable interpolation filters. Allows for swtiching/setting interpolation filters at the MB level. A frame level flag indicates whether to use a specifc filter for the entire frame or to signal the interpolation filter for each MB. When switchable filters are used, the encoder chooses between 8-tap and 8-tap sharp filters. The code currently has options to explore other variations as well, which will be cleaned up subsequently. One issue with the framework is that encoding is slow. I tried to do some tricks to speed things up but it is still slow. Decoding speed should not be affected since the number of filter taps remain unchanged. With the current version, we are up 0.5% on derf on average but some videos city/mobile improve by close to 4 and 2% respectively. If we did a full-search by turning the SEARCH_BEST_FILTER flag on, the results are somewhat better. The framework can be combined with filtered prediction, and I seek feedback regarding that. Rebased. Change-Id: I8f632cb2c111e76284140a2bd480945d6d42b77a
2012-07-18 22:43:01 +02:00
// Test best rd so far against threshold for trying this mode.
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
if (best_rd <= cpi->rd_threshes[mode_index])
Adds support for switchable interpolation filters. Allows for swtiching/setting interpolation filters at the MB level. A frame level flag indicates whether to use a specifc filter for the entire frame or to signal the interpolation filter for each MB. When switchable filters are used, the encoder chooses between 8-tap and 8-tap sharp filters. The code currently has options to explore other variations as well, which will be cleaned up subsequently. One issue with the framework is that encoding is slow. I tried to do some tricks to speed things up but it is still slow. Decoding speed should not be affected since the number of filter taps remain unchanged. With the current version, we are up 0.5% on derf on average but some videos city/mobile improve by close to 4 and 2% respectively. If we did a full-search by turning the SEARCH_BEST_FILTER flag on, the results are somewhat better. The framework can be combined with filtered prediction, and I seek feedback regarding that. Rebased. Change-Id: I8f632cb2c111e76284140a2bd480945d6d42b77a
2012-07-18 22:43:01 +02:00
continue;
Superblock encoding order This is the first patch to add superblock (32x32) coding order capabilities. It does not yet do any mode selection at the SB level, that will follow in a further patch. This patch encodes rows of SBs rather than MBs, each SB contains 2x2 MBs. Two intra prediction modes have been disabled since they require reconstructed data for the above-right MB which may not have been encoded yet (e.g. for the bottom right MB in each SB). Results on the one test clip I have tried (720p GIPS clip) suggest that it is somewhere around 0.2dB worse than the baseline version, so there may be bugs. It has been tested with no experiments enabled and with the following 3 experiments enabled: --enable-enhanced_interp --enable-high_precision_mv --enable-sixteenth_subpel_uv in each case the decode buffer matches the recon buffer (using "cmp" to compare the dumped/decoded frames). Note: Testing these experiments individually created errors. Some problems were found with other experiments but it is unclear what state these experiments are in: --enable-comp_intra_pred --enable-newentropy --enable-uvintra This code has not been extensively tested yet, so there is every likelihood that further bugs remain. I also intend to do some code cleanup & refactoring in tandem with the next patch that adds the 32x32 modes. Change-Id: I1eba7f740a70b3510df58db53464535ef881b4d9
2012-04-07 01:38:34 +02:00
Initial support for resolution changes on P-frames Allows inter-frames to change resolution. Currently these are almost equivalent to keyframes, as only intra prediction modes are allowed, but without the other context resets that occur on keyframes. Change-Id: Icd1a2a5af0d9462cc792588427b0a1f5b12e40d3
2013-02-07 00:54:52 +01:00
// Ensure that the references used by this mode are available.
if (mbmi->ref_frame &&
!(cpi->ref_frame_flags & flag_list[mbmi->ref_frame]))
continue;
if (mbmi->second_ref_frame > 0 &&
!(cpi->ref_frame_flags & flag_list[mbmi->second_ref_frame]))
continue;
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
// only scale on zeromv.
if (mbmi->ref_frame > 0 &&
(yv12_mb[mbmi->ref_frame].y_width != cm->mb_cols * 16 ||
yv12_mb[mbmi->ref_frame].y_height != cm->mb_rows * 16) &&
this_mode != ZEROMV)
continue;
set up a new speed 1 slightly worse results for faster encodes Change-Id: I25ea82a18ce20635dbcd328808c1d05ac1f58fd7
2013-04-19 17:04:57 +02:00
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
if (mbmi->second_ref_frame > 0 &&
(yv12_mb[mbmi->second_ref_frame].y_width != cm->mb_cols * 16 ||
yv12_mb[mbmi->second_ref_frame].y_height != cm->mb_rows * 16) &&
this_mode != ZEROMV)
continue;
Adds hybrid transform Adds ADST/DCT hybrid transform coding for Intra4x4 mode. The ADST is applied to directions in which the boundary pixels are used for prediction, while DCT applied to directions without corresponding boundary prediction. Adds enum TX_TYPE in b_mode_infor to indicate the transform type used. Make coding style consistent with google style. Fixed the commented issues. Experimental results in terms of bit-rate reduction: derf: 0.731% yt: 0.982% std-hd: 0.459% hd: 0.725% Will be looking at 8x8 transforms next. Change-Id: I46dbd7b80dbb3e8856e9c34fbc58cb3764a12fcf
2012-06-25 21:26:09 +02:00
// current coding mode under rate-distortion optimization test loop
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
mbmi->interintra_mode = (MB_PREDICTION_MODE)(DC_PRED - 1);
mbmi->interintra_uv_mode = (MB_PREDICTION_MODE)(DC_PRED - 1);
#endif
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// If the segment reference frame feature is enabled....
// then do nothing if the current ref frame is not allowed..
seg_common: give all exported symbols a vp9_ prefix. Change-Id: Ie8ba934a757acec1c80ac37ab9307c9a2783254e
2012-10-30 06:15:27 +01:00
if (vp9_segfeature_active(xd, segment_id, SEG_LVL_REF_FRAME) &&
!vp9_check_segref(xd, segment_id, mbmi->ref_frame)) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
continue;
Segment Skip Flag First step in simplifying the segment mode and segment EOB flags into a simpler segment skip flag that implies 0,0 mv and EOB at position 0. Change-Id: Ib750cac31a7a02dc21082580498efd9f7d8d72a5
2013-01-28 16:22:53 +01:00
// If the segment skip feature is enabled....
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// then do nothing if the current mode is not allowed..
Segment Skip Flag First step in simplifying the segment mode and segment EOB flags into a simpler segment skip flag that implies 0,0 mv and EOB at position 0. Change-Id: Ib750cac31a7a02dc21082580498efd9f7d8d72a5
2013-01-28 16:22:53 +01:00
} else if (vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP) &&
(this_mode != ZEROMV)) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
continue;
Segment Skip Flag First step in simplifying the segment mode and segment EOB flags into a simpler segment skip flag that implies 0,0 mv and EOB at position 0. Change-Id: Ib750cac31a7a02dc21082580498efd9f7d8d72a5
2013-01-28 16:22:53 +01:00
// Disable this drop out case if the ref frame segment
// level feature is enabled for this segment. This is to
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// prevent the possibility that the we end up unable to pick any mode.
Segment Skip Flag First step in simplifying the segment mode and segment EOB flags into a simpler segment skip flag that implies 0,0 mv and EOB at position 0. Change-Id: Ib750cac31a7a02dc21082580498efd9f7d8d72a5
2013-01-28 16:22:53 +01:00
} else if (!vp9_segfeature_active(xd, segment_id, SEG_LVL_REF_FRAME)) {
// Only consider ZEROMV/ALTREF_FRAME for alt ref frame overlay,
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// unless ARNR filtering is enabled in which case we want
// an unfiltered alternative
if (cpi->is_src_frame_alt_ref && (cpi->oxcf.arnr_max_frames == 0)) {
if (this_mode != ZEROMV ||
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
mbmi->ref_frame != ALTREF_FRAME) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
continue;
}
}
}
/* everything but intra */
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
scaled_ref_frame = NULL;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
if (mbmi->ref_frame) {
int ref = mbmi->ref_frame;
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
int fb;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
best_ref_mv = mbmi->ref_mvs[ref][0];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
vpx_memcpy(mdcounts, frame_mdcounts[ref], sizeof(mdcounts));
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
if (mbmi->ref_frame == LAST_FRAME) {
fb = cpi->lst_fb_idx;
} else if (mbmi->ref_frame == GOLDEN_FRAME) {
fb = cpi->gld_fb_idx;
} else {
fb = cpi->alt_fb_idx;
}
Use ref_frame_map vice active_ref_idx on the encoder This patch makes the encoder's use of ref_frame_map and active_ref_idx consistent with the decoder. ref_frame_map[] maps a reference buffer index to its actual location in the yv12_fb array, since many references may share an underlying buffer. active_ref_idx[] mirrors cpi->{lst,gld,alt}_fb_idx, holding the active references in each slot. This also fixes a bug in setup_buffer_inter() where the incorrect reference was used to populate the scaling factors. Change-Id: Id3728f6d77cffcd27c248903bf51f9c3e594287e
2013-02-22 20:22:03 +01:00
if (cpi->scaled_ref_idx[fb] != cm->ref_frame_map[fb])
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
scaled_ref_frame = &cm->yv12_fb[cpi->scaled_ref_idx[fb]];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0) {
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
int ref = mbmi->second_ref_frame;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Further changes to mv reference code. Some further changes and refactoring of mv reference code and selection of center point for searches. Mainly relates to not passing so many different local copies of things around. Some place holder comments. Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
2012-12-05 17:23:38 +01:00
second_best_ref_mv = mbmi->ref_mvs[ref][0];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Move pre, second_pre to per-plane MACROBLOCKD data Continue moving framebuffers to per-plane data. Change-Id: I237e5a998b364c4ec20316e7249206c0bff8631a
2013-04-20 04:16:14 +02:00
// TODO(jkoleszar) scaling/translation handled during creation of yv12_mb
// currently.
setup_pre_planes(xd, &yv12_mb[mbmi->ref_frame],
mbmi->second_ref_frame > 0 ? &yv12_mb[mbmi->second_ref_frame] : NULL,
0, 0, NULL, NULL);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// Experimental code. Special case for gf and arf zeromv modes.
// Increase zbin size to suppress noise
if (cpi->zbin_mode_boost_enabled) {
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
if (vp9_mode_order[mode_index].ref_frame == INTRA_FRAME)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
cpi->zbin_mode_boost = 0;
else {
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
if (vp9_mode_order[mode_index].mode == ZEROMV) {
if (vp9_mode_order[mode_index].ref_frame != LAST_FRAME)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
cpi->zbin_mode_boost = GF_ZEROMV_ZBIN_BOOST;
else
cpi->zbin_mode_boost = LF_ZEROMV_ZBIN_BOOST;
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
} else if (vp9_mode_order[mode_index].mode == SPLITMV)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
cpi->zbin_mode_boost = 0;
else
cpi->zbin_mode_boost = MV_ZBIN_BOOST;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Change encoder vp8_ and vp8cx_ public symbol prefixes to vp9_. Change-Id: Ie2e3652591b010ded10c216501ce24fd95d0aec5
2012-10-30 20:58:42 +01:00
vp9_update_zbin_extra(cpi, x);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Begin refactoring to help merge parts of rd inter. Also refactor some RD code, remove some comments, and cosmetic changes. Change-Id: I43806ae573ab5a5036602ac4425d750bbfa1e812
2012-08-08 20:13:53 +02:00
// Intra
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
if (!mbmi->ref_frame) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
switch (this_mode) {
Begin refactoring to help merge parts of rd inter. Also refactor some RD code, remove some comments, and cosmetic changes. Change-Id: I43806ae573ab5a5036602ac4425d750bbfa1e812
2012-08-08 20:13:53 +02:00
default:
Reorganiziation of the RD loop to make more sense. Change-Id: Ia1e3b0ff9b70747b3725391de59f230edab767d7
2012-08-07 23:33:15 +02:00
case V_PRED:
case H_PRED:
case D45_PRED:
case D135_PRED:
case D117_PRED:
case D153_PRED:
case D27_PRED:
case D63_PRED:
Problem of over smoothing with intra modes. In some cases intra modes in inter frames give an over smoothed appearance. Especially with noisy but flat content. Also in some cases there were problems with key frame sizing again with very flat but noisy content. These are temporary changes to help alleviate the visual problems but will almost certainly hurt metric results especially at the very low data rate end. Change-Id: I11549179a19277ffc283d9788bc70168f2a8bdc9
2012-12-14 18:49:46 +01:00
rate2 += intra_cost_penalty;
case DC_PRED:
case TM_PRED:
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
mbmi->ref_frame = INTRA_FRAME;
Reorganiziation of the RD loop to make more sense. Change-Id: Ia1e3b0ff9b70747b3725391de59f230edab767d7
2012-08-07 23:33:15 +02:00
// FIXME compound intra prediction
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
vp9_build_intra_predictors_sby_s(&x->e_mbd, BLOCK_SIZE_MB16X16);
// vp9_build_intra_predictors_mby(&x->e_mbd);
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
super_block_yrd(cpi, x, &rate_y, &distortion, &skippable,
BLOCK_SIZE_MB16X16, txfm_cache);
Reorganiziation of the RD loop to make more sense. Change-Id: Ia1e3b0ff9b70747b3725391de59f230edab767d7
2012-08-07 23:33:15 +02:00
rate2 += rate_y;
distortion2 += distortion;
Inconsistent reference to MacroblockD data structure. Cleaned up some inconsistent references using both xd-> and x->e_mbd. to access the same data structure in the same function. Change-Id: Ieb496fa22bf1feec6aa7bc70b941ea4f16e0f8b5
2012-10-29 11:36:11 +01:00
rate2 += x->mbmode_cost[xd->frame_type][mbmi->mode];
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
rate2 += uv_intra_rate[mbmi->txfm_size != TX_4X4];
rate_uv = uv_intra_rate_tokenonly[mbmi->txfm_size != TX_4X4];
distortion2 += uv_intra_distortion[mbmi->txfm_size != TX_4X4];
distortion_uv = uv_intra_distortion[mbmi->txfm_size != TX_4X4];
skippable = skippable &&
uv_intra_skippable[mbmi->txfm_size != TX_4X4];
Reorganiziation of the RD loop to make more sense. Change-Id: Ia1e3b0ff9b70747b3725391de59f230edab767d7
2012-08-07 23:33:15 +02:00
break;
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
case I4X4_PRED: {
Make RDCOST() results 64-bit. This prevents various integer overflows that several of us have been seeing in the past few days. Change-Id: If0020613a608e13d311b7838796c7ad53cb97255
2012-08-02 19:07:33 +02:00
int64_t tmp_rd;
Added early breakout for vp8_rd_pick_intra4x4mby_modes Improved performance of good quality, speed 0 (3% average) with no average quality loss. Change-Id: Ica34473f99bd74260eaebde6b132185e09e3c09d
2011-02-08 22:50:43 +01:00
Begin refactoring to help merge parts of rd inter. Also refactor some RD code, remove some comments, and cosmetic changes. Change-Id: I43806ae573ab5a5036602ac4425d750bbfa1e812
2012-08-08 20:13:53 +02:00
// Note the rate value returned here includes the cost of coding
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
// the I4X4_PRED mode : x->mbmode_cost[xd->frame_type][I4X4_PRED];
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
mbmi->txfm_size = TX_4X4;
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
tmp_rd = rd_pick_intra4x4mby_modes(cpi, x, &rate, &rate_y,
Remove "update_context" variable from VP9_COMP context. The variable is always zero. Change-Id: Id5cdbecad543bca465a5b1d471badaec7e112c8d
2013-01-25 01:28:53 +01:00
&distortion, best_yrd);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
rate2 += rate;
Problem of over smoothing with intra modes. In some cases intra modes in inter frames give an over smoothed appearance. Especially with noisy but flat content. Also in some cases there were problems with key frame sizing again with very flat but noisy content. These are temporary changes to help alleviate the visual problems but will almost certainly hurt metric results especially at the very low data rate end. Change-Id: I11549179a19277ffc283d9788bc70168f2a8bdc9
2012-12-14 18:49:46 +01:00
rate2 += intra_cost_penalty;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
distortion2 += distortion;
if (tmp_rd < best_yrd) {
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
rate2 += uv_intra_rate[TX_4X4];
rate_uv = uv_intra_rate_tokenonly[TX_4X4];
distortion2 += uv_intra_distortion[TX_4X4];
distortion_uv = uv_intra_distortion[TX_4X4];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
} else {
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
this_rd = INT64_MAX;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
disable_skip = 1;
}
Added early breakout for vp8_rd_pick_intra4x4mby_modes Improved performance of good quality, speed 0 (3% average) with no average quality loss. Change-Id: Ica34473f99bd74260eaebde6b132185e09e3c09d
2011-02-08 22:50:43 +01:00
}
break;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
case I8X8_PRED: {
Add support for tx_select in i8x8 encoding in keyframes. Also enable tx_select for keyframes. Change-Id: Iadb1231d9fa7af0c8dce3d9b41830b93a302479e
2013-03-07 19:45:25 +01:00
int64_t tmp_rd;
Use transform-size selection for i8x8_pred also. Change-Id: Iecb282fc89f9b5145ef31c5eda294ad42bc32a5d
2012-10-11 02:18:22 +02:00
Add support for tx_select in i8x8 encoding in keyframes. Also enable tx_select for keyframes. Change-Id: Iadb1231d9fa7af0c8dce3d9b41830b93a302479e
2013-03-07 19:45:25 +01:00
tmp_rd = rd_pick_intra8x8mby_modes_and_txsz(cpi, x, &rate, &rate_y,
&distortion, mode8x8,
best_yrd, txfm_cache);
Use transform-size selection for i8x8_pred also. Change-Id: Iecb282fc89f9b5145ef31c5eda294ad42bc32a5d
2012-10-11 02:18:22 +02:00
rate2 += rate;
Problem of over smoothing with intra modes. In some cases intra modes in inter frames give an over smoothed appearance. Especially with noisy but flat content. Also in some cases there were problems with key frame sizing again with very flat but noisy content. These are temporary changes to help alleviate the visual problems but will almost certainly hurt metric results especially at the very low data rate end. Change-Id: I11549179a19277ffc283d9788bc70168f2a8bdc9
2012-12-14 18:49:46 +01:00
rate2 += intra_cost_penalty;
Use transform-size selection for i8x8_pred also. Change-Id: Iecb282fc89f9b5145ef31c5eda294ad42bc32a5d
2012-10-11 02:18:22 +02:00
distortion2 += distortion;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/* TODO: uv rate maybe over-estimated here since there is UV intra
mode coded in I8X8_PRED prediction */
if (tmp_rd < best_yrd) {
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
rate2 += uv_intra_rate[TX_4X4];
rate_uv = uv_intra_rate_tokenonly[TX_4X4];
distortion2 += uv_intra_distortion[TX_4X4];
distortion_uv = uv_intra_distortion[TX_4X4];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
} else {
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
this_rd = INT64_MAX;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
disable_skip = 1;
}
enabled 8x8 intra prediction modes on inter frames This commit enabled the usage of 8x8 intra prediction modes on inter frames. There are a few TODO items related to this: 1)baseline entropy need be calibrated; 2)cost of UV need to be done more properly rather than using decision only relying on Y; 3)Threshold for allowing picking 8x8 intra prediction should be lowered to lower than the B_PRED. Even with all the TODOs, tests showed consistent gain on derf set ~0.1% (PSNR:0.08% and SSIM:0.14%). It is assumed that 8x8 intra prediction will help more on large resolution clips, especially with above TODOs addressed. Change-Id: I398ada49dfc32575cfab962a569c2885111ae3ba
2011-12-01 01:25:00 +01:00
}
break;
Begin refactoring to help merge parts of rd inter. Also refactor some RD code, remove some comments, and cosmetic changes. Change-Id: I43806ae573ab5a5036602ac4425d750bbfa1e812
2012-08-08 20:13:53 +02:00
}
}
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
// Split MV. The code is very different from the other inter modes so
// special case it.
else if (this_mode == SPLITMV) {
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
const int is_comp_pred = mbmi->second_ref_frame > 0;
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
int64_t this_rd_thresh;
int64_t tmp_rd, tmp_best_rd = INT64_MAX, tmp_best_rdu = INT64_MAX;
int tmp_best_rate = INT_MAX, tmp_best_ratey = INT_MAX;
int tmp_best_distortion = INT_MAX, tmp_best_skippable = 0;
int switchable_filter_index;
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
int_mv *second_ref = is_comp_pred ? &second_best_ref_mv : NULL;
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
union b_mode_info tmp_best_bmodes[16];
MB_MODE_INFO tmp_best_mbmode;
PARTITION_INFO tmp_best_partition;
int pred_exists = 0;
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
this_rd_thresh =
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
(mbmi->ref_frame == LAST_FRAME) ?
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
cpi->rd_threshes[THR_NEWMV] : cpi->rd_threshes[THR_NEWA];
this_rd_thresh =
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
(mbmi->ref_frame == GOLDEN_FRAME) ?
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
cpi->rd_threshes[THR_NEWG] : this_rd_thresh;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
xd->mode_info_context->mbmi.txfm_size = TX_4X4;
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
for (switchable_filter_index = 0;
switchable_filter_index < VP9_SWITCHABLE_FILTERS;
++switchable_filter_index) {
int newbest;
mbmi->interp_filter =
vp9_switchable_interp[switchable_filter_index];
vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
tmp_rd = rd_pick_best_mbsegmentation(cpi, x, &best_ref_mv,
second_ref, best_yrd, mdcounts,
&rate, &rate_y, &distortion,
&skippable,
(int)this_rd_thresh, seg_mvs,
txfm_cache);
if (cpi->common.mcomp_filter_type == SWITCHABLE) {
int rs = SWITCHABLE_INTERP_RATE_FACTOR * x->switchable_interp_costs
[vp9_get_pred_context(&cpi->common, xd,
PRED_SWITCHABLE_INTERP)]
[vp9_switchable_interp_map[mbmi->interp_filter]];
tmp_rd += RDCOST(x->rdmult, x->rddiv, rs, 0);
}
newbest = (tmp_rd < tmp_best_rd);
if (newbest) {
tmp_best_filter = mbmi->interp_filter;
tmp_best_rd = tmp_rd;
}
if ((newbest && cm->mcomp_filter_type == SWITCHABLE) ||
(mbmi->interp_filter == cm->mcomp_filter_type &&
cm->mcomp_filter_type != SWITCHABLE)) {
tmp_best_rdu = tmp_rd;
tmp_best_rate = rate;
tmp_best_ratey = rate_y;
tmp_best_distortion = distortion;
tmp_best_skippable = skippable;
vpx_memcpy(&tmp_best_mbmode, mbmi, sizeof(MB_MODE_INFO));
vpx_memcpy(&tmp_best_partition, x->partition_info,
sizeof(PARTITION_INFO));
for (i = 0; i < 16; i++) {
tmp_best_bmodes[i] = xd->block[i].bmi;
}
pred_exists = 1;
}
} // switchable_filter_index loop
mbmi->interp_filter = (cm->mcomp_filter_type == SWITCHABLE ?
tmp_best_filter : cm->mcomp_filter_type);
vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
if (!pred_exists) {
// Handles the special case when a filter that is not in the
// switchable list (bilinear, 6-tap) is indicated at the frame level
tmp_rd = rd_pick_best_mbsegmentation(cpi, x, &best_ref_mv,
second_ref, best_yrd, mdcounts,
&rate, &rate_y, &distortion,
&skippable,
(int)this_rd_thresh, seg_mvs,
txfm_cache);
} else {
if (cpi->common.mcomp_filter_type == SWITCHABLE) {
int rs = SWITCHABLE_INTERP_RATE_FACTOR * x->switchable_interp_costs
[vp9_get_pred_context(&cpi->common, xd,
PRED_SWITCHABLE_INTERP)]
[vp9_switchable_interp_map[mbmi->interp_filter]];
tmp_best_rdu -= RDCOST(x->rdmult, x->rddiv, rs, 0);
}
tmp_rd = tmp_best_rdu;
rate = tmp_best_rate;
rate_y = tmp_best_ratey;
distortion = tmp_best_distortion;
skippable = tmp_best_skippable;
vpx_memcpy(mbmi, &tmp_best_mbmode, sizeof(MB_MODE_INFO));
vpx_memcpy(x->partition_info, &tmp_best_partition,
sizeof(PARTITION_INFO));
for (i = 0; i < 16; i++) {
Implicit weighted prediction experiment Adds an experiment to use a weighted prediction of two INTER predictors, where the weight is one of (1/4, 3/4), (3/8, 5/8), (1/2, 1/2), (5/8, 3/8) or (3/4, 1/4), and is chosen implicitly based on consistency of the predictors to the already reconstructed pixels to the top and left of the current macroblock or superblock. Currently the weighting is not applied to SPLITMV modes, which default to the usual (1/2, 1/2) weighting. However the code is in place controlled by a macro. The same weighting is used for Y and UV components, where the weight is derived from analyzing the Y component only. Results (over compound inter-intra experiment) derf: +0.18% yt: +0.34% hd: +0.49% stdhd: +0.23% The experiment suggests bigger benefit for explicitly signaled weights. Change-Id: I5438539ff4485c5752874cd1eb078ff14bf5235a
2013-03-12 22:21:08 +01:00
xd->block[i].bmi = xd->mode_info_context->bmi[i] = tmp_best_bmodes[i];
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
}
}
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
rate2 += rate;
distortion2 += distortion;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
if (cpi->common.mcomp_filter_type == SWITCHABLE)
rate2 += SWITCHABLE_INTERP_RATE_FACTOR * x->switchable_interp_costs
Name space clean up. Preparation for project restructuring. Added vp9_ prefix on some function names that have global scope. Added static declaration on some that dont. Change-Id: If072f78b4300e8c17cfeed82c5d17b59946dcc5e
2012-10-29 14:44:18 +01:00
[vp9_get_pred_context(&cpi->common, xd, PRED_SWITCHABLE_INTERP)]
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
[vp9_switchable_interp_map[mbmi->interp_filter]];
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
// If even the 'Y' rd value of split is higher than best so far
// then dont bother looking at UV
if (tmp_rd < best_yrd) {
Detect skippable blocks in splitmv RD loop case. Change-Id: Id8a7265dc721e5cdcaa144e0041beeb32a98f51c
2012-10-17 23:32:17 +02:00
int uv_skippable;
reconinter: remove unnecessary functions, params Removes the redundant dst pointers from vp9_build_inter_predictors_sb{y,uv} and the remaining mb specific functions. Change-Id: I7b6bf439d9394b85ea79b4fe61a3ffc1025720da
2013-04-19 19:45:50 +02:00
vp9_build_inter_predictors_sbuv(&x->e_mbd, mb_row, mb_col,
BLOCK_SIZE_MB16X16);
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
Move src_diff to per-plane MACROBLOCK data First in a series of commits making certain MACROBLOCK members addressable per-plane. This commit also refactors the block subtraction functions vp9_subtract_b, vp9_subtract_sby_c, etc to be loops-over-planes and variable subsampling aware. Change-Id: I371d092b914ae0a495dfd852ea1a3d2467be6ec3
2013-04-23 17:26:10 +02:00
vp9_subtract_sbuv(x, BLOCK_SIZE_MB16X16);
Make the use of pred buffers consistent in MB/SB Use in-place buffers (dst of MACROBLOCKD) for macroblock prediction. This makes the macroblock buffer handling consistent with those of superblock. Remove predictor buffer MACROBLOCKD. Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df
2013-04-15 18:31:27 +02:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
super_block_uvrd_4x4(cm, x, &rate_uv, &distortion_uv,
&uv_skippable, BLOCK_SIZE_MB16X16);
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
rate2 += rate_uv;
distortion2 += distortion_uv;
Detect skippable blocks in splitmv RD loop case. Change-Id: Id8a7265dc721e5cdcaa144e0041beeb32a98f51c
2012-10-17 23:32:17 +02:00
skippable = skippable && uv_skippable;
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
} else {
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
this_rd = INT64_MAX;
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
disable_skip = 1;
}
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
if (!mode_excluded) {
if (is_comp_pred)
mode_excluded = cpi->common.comp_pred_mode == SINGLE_PREDICTION_ONLY;
else
mode_excluded = cpi->common.comp_pred_mode == COMP_PREDICTION_ONLY;
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
compmode_cost =
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
vp9_cost_bit(vp9_get_pred_prob(cm, xd, PRED_COMP), is_comp_pred);
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
mbmi->mode = this_mode;
Refactor SPLITMV in RD to a special case. SPLITMV shares very little code with the other inter modes. Change-Id: I0beee95de1fbedeea1f71650951ffc7db3a118eb
2012-08-08 20:49:56 +02:00
}
Refactor RD to take same codepath for single and comp pred Change-Id: Id38baf1b89648ef534e28be72f583137871f920c
2012-08-10 01:07:41 +02:00
else {
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
if (mbmi->second_ref_frame == INTRA_FRAME) {
if (best_intra16_mode == DC_PRED - 1) continue;
mbmi->interintra_mode = best_intra16_mode;
#if SEPARATE_INTERINTRA_UV
mbmi->interintra_uv_mode = best_intra16_uv_mode;
#else
mbmi->interintra_uv_mode = best_intra16_mode;
#endif
}
#endif
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
this_rd = handle_inter_mode(cpi, x, BLOCK_SIZE_MB16X16,
Remove unused parameters in handle_inter_mode Removed 2 unused parameters. Change-Id: Ic2862569313c404047072b268c3d2be3f635492c
2013-04-18 20:55:46 +02:00
mdcounts, txfm_cache,
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
&rate2, &distortion2, &skippable,
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
&compmode_cost,
#if CONFIG_COMP_INTERINTRA_PRED
&compmode_interintra_cost,
#endif
&rate_y, &distortion,
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
&rate_uv, &distortion_uv,
Pass macroblock index to pick inter functions Pass the current mb row and column around rather than the recon_yoffset and recon_uvoffset, since those offsets will change from predictor to predictor, based on the reference frame selection. Change-Id: If3f9df059e00f5048ca729d3d083ff428e1859c1
2013-02-07 19:09:05 +01:00
&mode_excluded, &disable_skip,
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
mode_index, &tmp_best_filter, frame_mv,
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
scaled_ref_frame, mb_row, mb_col);
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
if (this_rd == INT64_MAX)
Refactor RD to take same codepath for single and comp pred Change-Id: Id38baf1b89648ef534e28be72f583137871f920c
2012-08-10 01:07:41 +02:00
continue;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
if (cpi->common.use_interintra)
rate2 += compmode_interintra_cost;
#endif
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
if (cpi->common.comp_pred_mode == HYBRID_PREDICTION)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
rate2 += compmode_cost;
// Estimate the reference frame signaling cost and add it
// to the rolling cost variable.
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
rate2 += ref_costs[mbmi->ref_frame];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (!disable_skip) {
// Test for the condition where skip block will be activated
// because there are no non zero coefficients and make any
// necessary adjustment for rate. Ignore if skip is coded at
// segment level as the cost wont have been added in.
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
int mb_skip_allowed;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
// Is Mb level skip allowed (i.e. not coded at segment level).
mb_skip_allowed = !vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
if (skippable) {
mbmi->mb_skip_coeff = 1;
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
// Back out the coefficient coding costs
rate2 -= (rate_y + rate_uv);
// for best_yrd calculation
rate_uv = 0;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
if (mb_skip_allowed) {
int prob_skip_cost;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
// Cost the skip mb case
vp9_prob skip_prob =
vp9_get_pred_prob(cm, &x->e_mbd, PRED_MBSKIP);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
if (skip_prob) {
prob_skip_cost = vp9_cost_bit(skip_prob, 1);
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
rate2 += prob_skip_cost;
other_cost += prob_skip_cost;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
} else {
// Add in the cost of the no skip flag.
mbmi->mb_skip_coeff = 0;
if (mb_skip_allowed) {
int prob_skip_cost = vp9_cost_bit(
vp9_get_pred_prob(cm, &x->e_mbd, PRED_MBSKIP), 0);
rate2 += prob_skip_cost;
other_cost += prob_skip_cost;
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// Calculate the final RD estimate for this mode.
this_rd = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
}
Dual 16x16 inter prediction. This patch introduces the concept of dual inter16x16 prediction. A 16x16 inter-predicted macroblock can use 2 references instead of 1, where both references use the same mvmode (new, near/est, zero). In the case of newmv, this means that two MVs are coded instead of one. The frame can be encoded in 3 ways: all MBs single-prediction, all MBs dual prediction, or per-MB single/dual prediction selection ("hybrid"), in which case a single bit is coded per-MB to indicate whether the MB uses single or dual inter prediction. In the future, we can (maybe?) get further gains by mixing this with Adrian's 32x32 work, per-segment dual prediction settings, or adding support for dual splitmv/8x8mv inter prediction. Gain (on derf-set, CQ mode) is ~2.8% (SSIM) or ~3.6% (glb PSNR). Most gain is at medium/high bitrates, but there's minor gains at low bitrates also. Output was confirmed to match between encoder and decoder. Note for optimization people: this patch introduces a 2nd version of 16x16/8x8 sixtap/bilin functions, which does an avg instead of a store. They may want to look and make sure this is implemented to their satisfaction so we can optimize it best in the future. Change-ID: I59dc84b07cbb3ccf073ac0f756d03d294cb19281
2011-12-06 20:53:02 +01:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// Keep record of best intra distortion
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
if ((mbmi->ref_frame == INTRA_FRAME) &&
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
(this_rd < best_intra_rd)) {
best_intra_rd = this_rd;
*returnintra = distortion2;
}
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
if ((mbmi->ref_frame == INTRA_FRAME) &&
warnings in various experiments Change-Id: Ib5106d4772450f8026f823dd743f162ab833b1d6
2012-11-30 16:29:43 +01:00
(this_mode <= TM_PRED) &&
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
(this_rd < best_intra16_rd)) {
best_intra16_rd = this_rd;
best_intra16_mode = this_mode;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
#if SEPARATE_INTERINTRA_UV
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
best_intra16_uv_mode = uv_intra_mode[mbmi->txfm_size != TX_4X4];
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
#endif
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
}
#endif
Refactor to remove some arguments from vp8_rd_pick_inter_mode Change-Id: I8e72279cc68f34d269705f06cdaf8f3d06eed635
2012-08-10 03:25:29 +02:00
if (!disable_skip && mbmi->ref_frame == INTRA_FRAME)
for (i = 0; i < NB_PREDICTION_TYPES; ++i)
best_pred_rd[i] = MIN(best_pred_rd[i], this_rd);
Initial WebM release
2010-05-18 17:58:33 +02:00
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (this_rd < best_overall_rd) {
best_overall_rd = this_rd;
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
best_filter = tmp_best_filter;
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
best_mode = this_mode;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
is_best_interintra = (mbmi->second_ref_frame == INTRA_FRAME);
#endif
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
// Did this mode help.. i.e. is it the new best mode
if (this_rd < best_rd || x->skip) {
if (!mode_excluded) {
/*
if (mbmi->second_ref_frame == INTRA_FRAME) {
printf("rd %d best %d bestintra16 %d\n", this_rd, best_rd, best_intra16_rd);
}
*/
// Note index of best mode so far
best_mode_index = mode_index;
Added Prediction Filter to Mode Selection Added the ability to optionally filter the prediction data when inter modes are selected (excludes SPLITMV, for now). The mode selection loop considers both the filtered and non-filtered prediction data when choosing mode. The filter can be turned on/off at the frame-level, or signaled for each MB. Change-Id: I1b783c71d95a361ab36c761b07e8a6b06bc36822
2012-06-26 01:23:58 +02:00
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
if (this_mode <= I4X4_PRED) {
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
if (mbmi->txfm_size != TX_4X4
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
&& this_mode != I4X4_PRED
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
&& this_mode != I8X8_PRED)
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
mbmi->uv_mode = uv_intra_mode[TX_8X8];
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
else
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
mbmi->uv_mode = uv_intra_mode[TX_4X4];
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
/* required for left and above block mv */
mbmi->mv[0].as_int = 0;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
other_cost += ref_costs[mbmi->ref_frame];
Initial WebM release
2010-05-18 17:58:33 +02:00
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
/* Calculate the final y RD estimate for this mode */
best_yrd = RDCOST(x->rdmult, x->rddiv, (rate2 - rate_uv - other_cost),
(distortion2 - distortion_uv));
Initial WebM release
2010-05-18 17:58:33 +02:00
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
*returnrate = rate2;
*returndistortion = distortion2;
best_rd = this_rd;
vpx_memcpy(&best_mbmode, mbmi, sizeof(MB_MODE_INFO));
vpx_memcpy(&best_partition, x->partition_info, sizeof(PARTITION_INFO));
Initial WebM release
2010-05-18 17:58:33 +02:00
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
if ((this_mode == I4X4_PRED)
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
|| (this_mode == I8X8_PRED)
|| (this_mode == SPLITMV))
for (i = 0; i < 16; i++) {
best_bmodes[i] = xd->block[i].bmi;
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
// Testing this mode gave rise to an improvement in best error score.
// Lower threshold a bit for next time
cpi->rd_thresh_mult[mode_index] =
(cpi->rd_thresh_mult[mode_index] >= (MIN_THRESHMULT + 2)) ?
cpi->rd_thresh_mult[mode_index] - 2 : MIN_THRESHMULT;
cpi->rd_threshes[mode_index] =
(cpi->rd_baseline_thresh[mode_index] >> 7) *
cpi->rd_thresh_mult[mode_index];
} else {
Adds support for switchable interpolation filters. Allows for swtiching/setting interpolation filters at the MB level. A frame level flag indicates whether to use a specifc filter for the entire frame or to signal the interpolation filter for each MB. When switchable filters are used, the encoder chooses between 8-tap and 8-tap sharp filters. The code currently has options to explore other variations as well, which will be cleaned up subsequently. One issue with the framework is that encoding is slow. I tried to do some tricks to speed things up but it is still slow. Decoding speed should not be affected since the number of filter taps remain unchanged. With the current version, we are up 0.5% on derf on average but some videos city/mobile improve by close to 4 and 2% respectively. If we did a full-search by turning the SEARCH_BEST_FILTER flag on, the results are somewhat better. The framework can be combined with filtered prediction, and I seek feedback regarding that. Rebased. Change-Id: I8f632cb2c111e76284140a2bd480945d6d42b77a
2012-07-18 22:43:01 +02:00
// If the mode did not help improve the best error case then raise the
// threshold for testing that mode next time around.
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
cpi->rd_thresh_mult[mode_index] += 4;
Initial WebM release
2010-05-18 17:58:33 +02:00
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
if (cpi->rd_thresh_mult[mode_index] > MAX_THRESHMULT)
cpi->rd_thresh_mult[mode_index] = MAX_THRESHMULT;
Initial WebM release
2010-05-18 17:58:33 +02:00
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
cpi->rd_threshes[mode_index] = (cpi->rd_baseline_thresh[mode_index] >> 7)
* cpi->rd_thresh_mult[mode_index];
}
Initial WebM release
2010-05-18 17:58:33 +02:00
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
/* keep record of best compound/single-only prediction */
if (!disable_skip && mbmi->ref_frame != INTRA_FRAME) {
int64_t single_rd, hybrid_rd;
int single_rate, hybrid_rate;
Initial WebM release
2010-05-18 17:58:33 +02:00
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
if (cpi->common.comp_pred_mode == HYBRID_PREDICTION) {
single_rate = rate2 - compmode_cost;
hybrid_rate = rate2;
} else {
single_rate = rate2;
hybrid_rate = rate2 + compmode_cost;
}
enabled 8x8 intra prediction modes on inter frames This commit enabled the usage of 8x8 intra prediction modes on inter frames. There are a few TODO items related to this: 1)baseline entropy need be calibrated; 2)cost of UV need to be done more properly rather than using decision only relying on Y; 3)Threshold for allowing picking 8x8 intra prediction should be lowered to lower than the B_PRED. Even with all the TODOs, tests showed consistent gain on derf set ~0.1% (PSNR:0.08% and SSIM:0.14%). It is assumed that 8x8 intra prediction will help more on large resolution clips, especially with above TODOs addressed. Change-Id: I398ada49dfc32575cfab962a569c2885111ae3ba
2011-12-01 01:25:00 +01:00
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
single_rd = RDCOST(x->rdmult, x->rddiv, single_rate, distortion2);
hybrid_rd = RDCOST(x->rdmult, x->rddiv, hybrid_rate, distortion2);
Initial WebM release
2010-05-18 17:58:33 +02:00
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
if (mbmi->second_ref_frame <= INTRA_FRAME &&
single_rd < best_pred_rd[SINGLE_PREDICTION_ONLY]) {
best_pred_rd[SINGLE_PREDICTION_ONLY] = single_rd;
} else if (mbmi->second_ref_frame > INTRA_FRAME &&
single_rd < best_pred_rd[COMP_PREDICTION_ONLY]) {
best_pred_rd[COMP_PREDICTION_ONLY] = single_rd;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
if (hybrid_rd < best_pred_rd[HYBRID_PREDICTION])
best_pred_rd[HYBRID_PREDICTION] = hybrid_rd;
}
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
/* keep record of best txfm size */
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
if (!mode_excluded && this_rd != INT64_MAX) {
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
for (i = 0; i < NB_TXFM_MODES; i++) {
int64_t adj_rd;
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
if (this_mode != I4X4_PRED) {
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
const int64_t txfm_mode_diff =
txfm_cache[i] - txfm_cache[cm->txfm_mode];
adj_rd = this_rd + txfm_mode_diff;
} else {
adj_rd = this_rd;
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
}
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
if (adj_rd < best_txfm_rd[i])
best_txfm_rd[i] = adj_rd;
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
}
Using partition_info instead of blockd info for splitmv The partition_info struct contains info just for SPLITMV, so it should be used instead of BLOCKD. Eventually, I want to reduce the size of B_MODE_INFO struct found in BLOCKD, so this is the first step toward that goal. Also, since SPLITMV is not supported in vp8_pick_inter_mode(), the unnecessary mem copies and checks were removed. For rt encodes, this gave a slight performance improvement. Change-Id: I5585c98fa9d5acbde1c7e0f452a01d9ecc080574
2011-05-19 21:03:36 +02:00
}
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
if (x->skip && !mode_excluded)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
break;
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
}
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
assert((cm->mcomp_filter_type == SWITCHABLE) ||
(cm->mcomp_filter_type == best_mbmode.interp_filter) ||
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
(best_mbmode.mode <= I4X4_PRED));
Superblock encoding order This is the first patch to add superblock (32x32) coding order capabilities. It does not yet do any mode selection at the SB level, that will follow in a further patch. This patch encodes rows of SBs rather than MBs, each SB contains 2x2 MBs. Two intra prediction modes have been disabled since they require reconstructed data for the above-right MB which may not have been encoded yet (e.g. for the bottom right MB in each SB). Results on the one test clip I have tried (720p GIPS clip) suggest that it is somewhere around 0.2dB worse than the baseline version, so there may be bugs. It has been tested with no experiments enabled and with the following 3 experiments enabled: --enable-enhanced_interp --enable-high_precision_mv --enable-sixteenth_subpel_uv in each case the decode buffer matches the recon buffer (using "cmp" to compare the dumped/decoded frames). Note: Testing these experiments individually created errors. Some problems were found with other experiments but it is unclear what state these experiments are in: --enable-comp_intra_pred --enable-newentropy --enable-uvintra This code has not been extensively tested yet, so there is every likelihood that further bugs remain. I also intend to do some code cleanup & refactoring in tandem with the next patch that adds the 32x32 modes. Change-Id: I1eba7f740a70b3510df58db53464535ef881b4d9
2012-04-07 01:38:34 +02:00
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
++cpi->interintra_select_count[is_best_interintra];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
#endif
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
// Accumulate filter usage stats
// TODO(agrange): Use RD criteria to select interpolation filter mode.
if ((best_mode >= NEARESTMV) && (best_mode <= SPLITMV))
++cpi->best_switchable_interp_count[vp9_switchable_interp_map[best_filter]];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// Reduce the activation RD thresholds for the best choice mode
Adds support for switchable interpolation filters. Allows for swtiching/setting interpolation filters at the MB level. A frame level flag indicates whether to use a specifc filter for the entire frame or to signal the interpolation filter for each MB. When switchable filters are used, the encoder chooses between 8-tap and 8-tap sharp filters. The code currently has options to explore other variations as well, which will be cleaned up subsequently. One issue with the framework is that encoding is slow. I tried to do some tricks to speed things up but it is still slow. Decoding speed should not be affected since the number of filter taps remain unchanged. With the current version, we are up 0.5% on derf on average but some videos city/mobile improve by close to 4 and 2% respectively. If we did a full-search by turning the SEARCH_BEST_FILTER flag on, the results are somewhat better. The framework can be combined with filtered prediction, and I seek feedback regarding that. Rebased. Change-Id: I8f632cb2c111e76284140a2bd480945d6d42b77a
2012-07-18 22:43:01 +02:00
if ((cpi->rd_baseline_thresh[best_mode_index] > 0) &&
(cpi->rd_baseline_thresh[best_mode_index] < (INT_MAX >> 2))) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
int best_adjustment = (cpi->rd_thresh_mult[best_mode_index] >> 2);
Adds support for switchable interpolation filters. Allows for swtiching/setting interpolation filters at the MB level. A frame level flag indicates whether to use a specifc filter for the entire frame or to signal the interpolation filter for each MB. When switchable filters are used, the encoder chooses between 8-tap and 8-tap sharp filters. The code currently has options to explore other variations as well, which will be cleaned up subsequently. One issue with the framework is that encoding is slow. I tried to do some tricks to speed things up but it is still slow. Decoding speed should not be affected since the number of filter taps remain unchanged. With the current version, we are up 0.5% on derf on average but some videos city/mobile improve by close to 4 and 2% respectively. If we did a full-search by turning the SEARCH_BEST_FILTER flag on, the results are somewhat better. The framework can be combined with filtered prediction, and I seek feedback regarding that. Rebased. Change-Id: I8f632cb2c111e76284140a2bd480945d6d42b77a
2012-07-18 22:43:01 +02:00
cpi->rd_thresh_mult[best_mode_index] =
(cpi->rd_thresh_mult[best_mode_index] >=
(MIN_THRESHMULT + best_adjustment)) ?
cpi->rd_thresh_mult[best_mode_index] - best_adjustment : MIN_THRESHMULT;
cpi->rd_threshes[best_mode_index] =
(cpi->rd_baseline_thresh[best_mode_index] >> 7) *
cpi->rd_thresh_mult[best_mode_index];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Segment Skip Flag First step in simplifying the segment mode and segment EOB flags into a simpler segment skip flag that implies 0,0 mv and EOB at position 0. Change-Id: Ib750cac31a7a02dc21082580498efd9f7d8d72a5
2013-01-28 16:22:53 +01:00
// This code forces Altref,0,0 and skip for the frame that overlays a
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// an alrtef unless Altref is filtered. However, this is unsafe if
Segment Skip Flag First step in simplifying the segment mode and segment EOB flags into a simpler segment skip flag that implies 0,0 mv and EOB at position 0. Change-Id: Ib750cac31a7a02dc21082580498efd9f7d8d72a5
2013-01-28 16:22:53 +01:00
// segment level coding of ref frame is enabled for this
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// segment.
seg_common: give all exported symbols a vp9_ prefix. Change-Id: Ie8ba934a757acec1c80ac37ab9307c9a2783254e
2012-10-30 06:15:27 +01:00
if (!vp9_segfeature_active(xd, segment_id, SEG_LVL_REF_FRAME) &&
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
cpi->is_src_frame_alt_ref &&
(cpi->oxcf.arnr_max_frames == 0) &&
(best_mbmode.mode != ZEROMV || best_mbmode.ref_frame != ALTREF_FRAME)) {
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
mbmi->mode = ZEROMV;
Don't use tx32x32 for macroblocks. Change-Id: Ib674e0153ca360867ab7a20ba291ac9171a01250
2013-01-04 21:05:40 +01:00
if (cm->txfm_mode <= ALLOW_8X8)
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
mbmi->txfm_size = cm->txfm_mode;
else
mbmi->txfm_size = TX_16X16;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
mbmi->ref_frame = ALTREF_FRAME;
Refactor RD to take same codepath for single and comp pred Change-Id: Id38baf1b89648ef534e28be72f583137871f920c
2012-08-10 01:07:41 +02:00
mbmi->mv[0].as_int = 0;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
mbmi->uv_mode = DC_PRED;
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
mbmi->mb_skip_coeff = 1;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
mbmi->partitioning = 0;
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
set_scale_factors(xd, mbmi->ref_frame, mbmi->second_ref_frame,
scale_factor);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Refactor to remove some arguments from vp8_rd_pick_inter_mode Change-Id: I8e72279cc68f34d269705f06cdaf8f3d06eed635
2012-08-10 03:25:29 +02:00
vpx_memset(best_pred_diff, 0, sizeof(best_pred_diff));
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
vpx_memset(best_txfm_diff, 0, sizeof(best_txfm_diff));
Refactor to remove some arguments from vp8_rd_pick_inter_mode Change-Id: I8e72279cc68f34d269705f06cdaf8f3d06eed635
2012-08-10 03:25:29 +02:00
goto end;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
// macroblock modes
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
vpx_memcpy(mbmi, &best_mbmode, sizeof(MB_MODE_INFO));
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
if (best_mbmode.mode == I4X4_PRED) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
for (i = 0; i < 16; i++) {
xd->mode_info_context->bmi[i].as_mode = best_bmodes[i].as_mode;
xd->block[i].bmi.as_mode = xd->mode_info_context->bmi[i].as_mode;
}
}
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
if (best_mbmode.mode == I8X8_PRED)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
set_i8x8_block_modes(x, mode8x8);
if (best_mbmode.mode == SPLITMV) {
for (i = 0; i < 16; i++)
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
xd->mode_info_context->bmi[i].as_mv[0].as_int =
best_bmodes[i].as_mv[0].as_int;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame > 0)
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
for (i = 0; i < 16; i++)
Replace as_mv struct with array Replace as_mv.{first, second} with a two element array, so that they can easily be processed with an index variable. Change-Id: I1e429155544d2a94a5b72a5b467c53d8b8728190
2013-02-09 04:46:36 +01:00
xd->mode_info_context->bmi[i].as_mv[1].as_int =
best_bmodes[i].as_mv[1].as_int;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
vpx_memcpy(x->partition_info, &best_partition, sizeof(PARTITION_INFO));
Refactor RD to take same codepath for single and comp pred Change-Id: Id38baf1b89648ef534e28be72f583137871f920c
2012-08-10 01:07:41 +02:00
mbmi->mv[0].as_int = x->partition_info->bmi[15].mv.as_int;
mbmi->mv[1].as_int = x->partition_info->bmi[15].second_mv.as_int;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Refactor to remove some arguments from vp8_rd_pick_inter_mode Change-Id: I8e72279cc68f34d269705f06cdaf8f3d06eed635
2012-08-10 03:25:29 +02:00
for (i = 0; i < NB_PREDICTION_TYPES; ++i) {
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
if (best_pred_rd[i] == INT64_MAX)
Refactor to remove some arguments from vp8_rd_pick_inter_mode Change-Id: I8e72279cc68f34d269705f06cdaf8f3d06eed635
2012-08-10 03:25:29 +02:00
best_pred_diff[i] = INT_MIN;
else
best_pred_diff[i] = best_rd - best_pred_rd[i];
}
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
if (!x->skip) {
for (i = 0; i < NB_TXFM_MODES; i++) {
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
if (best_txfm_rd[i] == INT64_MAX)
Prevent filling transform size cache with uninitialized values. The 32x32 value in case of splitmv was uninitialized. this leads to all kind of erratic behaviour down the line. Also fill in dummy values for superblocks in keyframes (the values are currently unused, but we run into integer overflows anyway, which makes detecting bad cases harder). Lastly, in case we did not find any RD value at all, don't set tx_diff to INT_MIN, but instead set it to zero (since if we couldn't find a mode, it's unlikely that any particular transform would have made that worse or better; rather, it's likely equally bad for all tx_sizes). Change-Id: If236fd3aa2037e5b398d03f3b1978fbbc5ce740e
2013-02-16 00:55:31 +01:00
best_txfm_diff[i] = 0;
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
else
best_txfm_diff[i] = best_rd - best_txfm_rd[i];
}
} else {
vpx_memset(best_txfm_diff, 0, sizeof(best_txfm_diff));
}
Refactor to remove some arguments from vp8_rd_pick_inter_mode Change-Id: I8e72279cc68f34d269705f06cdaf8f3d06eed635
2012-08-10 03:25:29 +02:00
end:
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
set_scale_factors(xd, mbmi->ref_frame, mbmi->second_ref_frame,
scale_factor);
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
store_coding_context(x, &x->mb_context[xd->sb_index][xd->mb_index],
best_mode_index, &best_partition,
&mbmi->ref_mvs[mbmi->ref_frame][0],
&mbmi->ref_mvs[mbmi->second_ref_frame < 0 ? 0 :
mbmi->second_ref_frame][0],
best_pred_diff, best_txfm_diff);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
void vp9_rd_pick_intra_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
int *returnrate, int *returndist,
Merge pick_sb_modes and pick_sb64_modes. Change-Id: Iad69e7a3b7e470acf6094f6a52e7da69066fd552
2013-04-11 18:33:49 +02:00
BLOCK_SIZE_TYPE bsize,
PICK_MODE_CONTEXT *ctx) {
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
VP9_COMMON *cm = &cpi->common;
Add proper skip support to intra frames. Change-Id: I7524883fb29f42303fb46a5bc6772fbcf8781d1d
2012-08-29 19:43:20 +02:00
MACROBLOCKD *xd = &x->e_mbd;
fix a number issues that cause failures During master jenkins verification proces Change-Id: I3722b8753eaf39f99b45979ce407a8ea0bea0b89
2013-01-15 03:03:34 +01:00
int rate_y = 0, rate_uv;
int rate_y_tokenonly = 0, rate_uv_tokenonly;
int dist_y = 0, dist_uv;
int y_skip = 0, uv_skip;
Prevent filling transform size cache with uninitialized values. The 32x32 value in case of splitmv was uninitialized. this leads to all kind of erratic behaviour down the line. Also fill in dummy values for superblocks in keyframes (the values are currently unused, but we run into integer overflows anyway, which makes detecting bad cases harder). Lastly, in case we did not find any RD value at all, don't set tx_diff to INT_MIN, but instead set it to zero (since if we couldn't find a mode, it's unlikely that any particular transform would have made that worse or better; rather, it's likely equally bad for all tx_sizes). Change-Id: If236fd3aa2037e5b398d03f3b1978fbbc5ce740e
2013-02-16 00:55:31 +01:00
int64_t txfm_cache[NB_TXFM_MODES], err;
int i;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Fix lingering x->skip settings if static_threshold is used. Keyframes don't set this variable, so it would use the last set values from inter frames. Change-Id: Ie1ef45ece2c44b21b5d55f6cea9f7d6e7a445692
2013-04-15 21:50:32 +02:00
ctx->skip = 0;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
xd->mode_info_context->mbmi.mode = DC_PRED;
Prevent filling transform size cache with uninitialized values. The 32x32 value in case of splitmv was uninitialized. this leads to all kind of erratic behaviour down the line. Also fill in dummy values for superblocks in keyframes (the values are currently unused, but we run into integer overflows anyway, which makes detecting bad cases harder). Lastly, in case we did not find any RD value at all, don't set tx_diff to INT_MIN, but instead set it to zero (since if we couldn't find a mode, it's unlikely that any particular transform would have made that worse or better; rather, it's likely equally bad for all tx_sizes). Change-Id: If236fd3aa2037e5b398d03f3b1978fbbc5ce740e
2013-02-16 00:55:31 +01:00
err = rd_pick_intra_sby_mode(cpi, x, &rate_y, &rate_y_tokenonly,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
&dist_y, &y_skip, bsize, txfm_cache);
Fix compiler warnings The warnings caused verify failure with gerrit for several commits Change-Id: I030df8638bd69b8783a3ac58e720ff9f0bfd546c
2013-01-14 22:56:52 +01:00
rd_pick_intra_sbuv_mode(cpi, x, &rate_uv, &rate_uv_tokenonly,
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
&dist_uv, &uv_skip, bsize);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
if (y_skip && uv_skip) {
Add proper skip support to intra frames. Change-Id: I7524883fb29f42303fb46a5bc6772fbcf8781d1d
2012-08-29 19:43:20 +02:00
*returnrate = rate_y + rate_uv - rate_y_tokenonly - rate_uv_tokenonly +
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
vp9_cost_bit(vp9_get_pred_prob(cm, xd, PRED_MBSKIP), 1);
Add proper skip support to intra frames. Change-Id: I7524883fb29f42303fb46a5bc6772fbcf8781d1d
2012-08-29 19:43:20 +02:00
*returndist = dist_y + (dist_uv >> 2);
Merge pick_sb_modes and pick_sb64_modes. Change-Id: Iad69e7a3b7e470acf6094f6a52e7da69066fd552
2013-04-11 18:33:49 +02:00
memset(ctx->txfm_rd_diff, 0,
sizeof(x->sb32_context[xd->sb_index].txfm_rd_diff));
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
} else {
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
*returnrate = rate_y + rate_uv +
vp9_cost_bit(vp9_get_pred_prob(cm, xd, PRED_MBSKIP), 0);
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
*returndist = dist_y + (dist_uv >> 2);
Prevent filling transform size cache with uninitialized values. The 32x32 value in case of splitmv was uninitialized. this leads to all kind of erratic behaviour down the line. Also fill in dummy values for superblocks in keyframes (the values are currently unused, but we run into integer overflows anyway, which makes detecting bad cases harder). Lastly, in case we did not find any RD value at all, don't set tx_diff to INT_MIN, but instead set it to zero (since if we couldn't find a mode, it's unlikely that any particular transform would have made that worse or better; rather, it's likely equally bad for all tx_sizes). Change-Id: If236fd3aa2037e5b398d03f3b1978fbbc5ce740e
2013-02-16 00:55:31 +01:00
for (i = 0; i < NB_TXFM_MODES; i++) {
Merge pick_sb_modes and pick_sb64_modes. Change-Id: Iad69e7a3b7e470acf6094f6a52e7da69066fd552
2013-04-11 18:33:49 +02:00
ctx->txfm_rd_diff[i] = err - txfm_cache[i];
Prevent filling transform size cache with uninitialized values. The 32x32 value in case of splitmv was uninitialized. this leads to all kind of erratic behaviour down the line. Also fill in dummy values for superblocks in keyframes (the values are currently unused, but we run into integer overflows anyway, which makes detecting bad cases harder). Lastly, in case we did not find any RD value at all, don't set tx_diff to INT_MIN, but instead set it to zero (since if we couldn't find a mode, it's unlikely that any particular transform would have made that worse or better; rather, it's likely equally bad for all tx_sizes). Change-Id: If236fd3aa2037e5b398d03f3b1978fbbc5ce740e
2013-02-16 00:55:31 +01:00
}
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
}
Merge pick_sb_modes and pick_sb64_modes. Change-Id: Iad69e7a3b7e470acf6094f6a52e7da69066fd552
2013-04-11 18:33:49 +02:00
vpx_memcpy(&ctx->mic, xd->mode_info_context, sizeof(MODE_INFO));
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
}
Move RD intra block mode selection to rdopt.c This change is analogous to I0b67dae1f8a74902378da7bdf565e39ab832dda7, which made the move for the non-RD path. Change-Id: If63fc1b0cd1eb7f932e710f83ff24d91454f8ed1
2011-06-08 18:05:05 +02:00
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
void vp9_rd_pick_intra_mode(VP9_COMP *cpi, MACROBLOCK *x,
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
int *returnrate, int *returndist) {
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
VP9_COMMON *cm = &cpi->common;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
MACROBLOCKD *xd = &x->e_mbd;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
MB_MODE_INFO * mbmi = &x->e_mbd.mode_info_context->mbmi;
Make RDCOST() results 64-bit. This prevents various integer overflows that several of us have been seeing in the past few days. Change-Id: If0020613a608e13d311b7838796c7ad53cb97255
2012-08-02 19:07:33 +02:00
int64_t error4x4, error16x16;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
int rate4x4, rate16x16 = 0, rateuv[2];
int dist4x4 = 0, dist16x16 = 0, distuv[2];
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
int rate;
int rate4x4_tokenonly = 0;
int rate16x16_tokenonly = 0;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
int rateuv_tokenonly[2];
Make RDCOST() results 64-bit. This prevents various integer overflows that several of us have been seeing in the past few days. Change-Id: If0020613a608e13d311b7838796c7ad53cb97255
2012-08-02 19:07:33 +02:00
int64_t error8x8;
int rate8x8_tokenonly=0;
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
int rate8x8, dist8x8;
int mode16x16;
Remove compound intra-intra experiment. This experiment gives little gains and adds relatively much code complexity (and it hinders other experiments), so let's get rid of it. Change-Id: Id25e79a137a1b8a01138aa27a1fa0ba4a2df274a
2013-01-14 23:37:53 +01:00
int mode8x8[4];
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
int dist;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
int modeuv[2], uv_intra_skippable[2];
warnings in various experiments Change-Id: Ib5106d4772450f8026f823dd743f162ab833b1d6
2012-11-30 16:29:43 +01:00
int y_intra16x16_skippable = 0;
Add support for tx_select in i8x8 encoding in keyframes. Also enable tx_select for keyframes. Change-Id: Iadb1231d9fa7af0c8dce3d9b41830b93a302479e
2013-03-07 19:45:25 +01:00
int64_t txfm_cache[2][NB_TXFM_MODES];
TX_SIZE txfm_size_16x16, txfm_size_8x8;
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
int i;
Adds hybrid transform Adds ADST/DCT hybrid transform coding for Intra4x4 mode. The ADST is applied to directions in which the boundary pixels are used for prediction, while DCT applied to directions without corresponding boundary prediction. Adds enum TX_TYPE in b_mode_infor to indicate the transform type used. Make coding style consistent with google style. Fixed the commented issues. Experimental results in terms of bit-rate reduction: derf: 0.731% yt: 0.982% std-hd: 0.459% hd: 0.725% Will be looking at 8x8 transforms next. Change-Id: I46dbd7b80dbb3e8856e9c34fbc58cb3764a12fcf
2012-06-25 21:26:09 +02:00
Fix lingering x->skip settings if static_threshold is used. Keyframes don't set this variable, so it would use the last set values from inter frames. Change-Id: Ie1ef45ece2c44b21b5d55f6cea9f7d6e7a445692
2013-04-15 21:50:32 +02:00
x->mb_context[xd->sb_index][xd->mb_index].skip = 0;
Fix some tabbing in rdopt.c Change-Id: Ie7e7727512cc1333bd16e7688cd1d8764371dde3
2012-08-10 19:00:18 +02:00
mbmi->ref_frame = INTRA_FRAME;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
mbmi->mode = DC_PRED;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
for (i = 0; i <= TX_8X8; i++) {
mbmi->txfm_size = i;
rd_pick_intra_sbuv_mode(cpi, x, &rateuv[i], &rateuv_tokenonly[i],
&distuv[i], &uv_intra_skippable[i],
BLOCK_SIZE_MB16X16);
modeuv[i] = mbmi->uv_mode;
Add proper skip support to intra frames. Change-Id: I7524883fb29f42303fb46a5bc6772fbcf8781d1d
2012-08-29 19:43:20 +02:00
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Fix some tabbing in rdopt.c Change-Id: Ie7e7727512cc1333bd16e7688cd1d8764371dde3
2012-08-10 19:00:18 +02:00
// current macroblock under rate-distortion optimization test loop
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
error16x16 = rd_pick_intra_sby_mode(cpi, x, &rate16x16,
&rate16x16_tokenonly, &dist16x16,
&y_intra16x16_skippable,
BLOCK_SIZE_MB16X16, txfm_cache[1]);
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
mode16x16 = mbmi->mode;
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
txfm_size_16x16 = mbmi->txfm_size;
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
if (y_intra16x16_skippable &&
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
((cm->txfm_mode == ONLY_4X4 && uv_intra_skippable[TX_4X4]) ||
(cm->txfm_mode != ONLY_4X4 && uv_intra_skippable[TX_8X8]))) {
Add support for tx_select in i8x8 encoding in keyframes. Also enable tx_select for keyframes. Change-Id: Iadb1231d9fa7af0c8dce3d9b41830b93a302479e
2013-03-07 19:45:25 +01:00
error16x16 -= RDCOST(x->rdmult, x->rddiv, rate16x16_tokenonly, 0);
rate16x16 -= rate16x16_tokenonly;
}
for (i = 0; i < NB_TXFM_MODES; i++) {
txfm_cache[0][i] = error16x16 - txfm_cache[1][cm->txfm_mode] +
txfm_cache[1][i];
}
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Add support for tx_select in i8x8 encoding in keyframes. Also enable tx_select for keyframes. Change-Id: Iadb1231d9fa7af0c8dce3d9b41830b93a302479e
2013-03-07 19:45:25 +01:00
error8x8 = rd_pick_intra8x8mby_modes_and_txsz(cpi, x, &rate8x8,
&rate8x8_tokenonly,
&dist8x8, mode8x8,
error16x16, txfm_cache[1]);
txfm_size_8x8 = mbmi->txfm_size;
for (i = 0; i < NB_TXFM_MODES; i++) {
int64_t tmp_rd = error8x8 - txfm_cache[1][cm->txfm_mode] + txfm_cache[1][i];
if (tmp_rd < txfm_cache[0][i])
txfm_cache[0][i] = tmp_rd;
}
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
Fixed a subtle issue that breaks encoding process This issue breaks the encoding process of the codebase. The effect emerges only in particular test sequence at certain bit-rates and frame limits. Change-Id: I02e080f2a49624eef9a21c424053dc2a1d902452
2013-02-15 23:45:20 +01:00
mbmi->txfm_size = TX_4X4;
Minor cosmetic changes to rdopt.c. Change-Id: Ib7b972486ccbbcb82fa4c05a5f6d7f8aaec32155
2012-08-07 01:21:23 +02:00
error4x4 = rd_pick_intra4x4mby_modes(cpi, x,
&rate4x4, &rate4x4_tokenonly,
Remove "update_context" variable from VP9_COMP context. The variable is always zero. Change-Id: Id5cdbecad543bca465a5b1d471badaec7e112c8d
2013-01-25 01:28:53 +01:00
&dist4x4, error16x16);
Add support for tx_select in i8x8 encoding in keyframes. Also enable tx_select for keyframes. Change-Id: Iadb1231d9fa7af0c8dce3d9b41830b93a302479e
2013-03-07 19:45:25 +01:00
for (i = 0; i < NB_TXFM_MODES; i++) {
if (error4x4 < txfm_cache[0][i])
txfm_cache[0][i] = error4x4;
}
Move RD intra block mode selection to rdopt.c This change is analogous to I0b67dae1f8a74902378da7bdf565e39ab832dda7, which made the move for the non-RD path. Change-Id: If63fc1b0cd1eb7f932e710f83ff24d91454f8ed1
2011-06-08 18:05:05 +02:00
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
mbmi->mb_skip_coeff = 0;
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
if (y_intra16x16_skippable &&
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
((cm->txfm_mode == ONLY_4X4 && uv_intra_skippable[TX_4X4]) ||
(cm->txfm_mode != ONLY_4X4 && uv_intra_skippable[TX_8X8]))) {
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
mbmi->mb_skip_coeff = 1;
Fix a mismatch if a keyframe contains skip-blocks. Change-Id: I0681d3183f51627be8c2bb76f343b7270f9116d8
2012-10-16 21:41:56 +02:00
mbmi->mode = mode16x16;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
mbmi->uv_mode = modeuv[cm->txfm_mode != ONLY_4X4];
Add support for tx_select in i8x8 encoding in keyframes. Also enable tx_select for keyframes. Change-Id: Iadb1231d9fa7af0c8dce3d9b41830b93a302479e
2013-03-07 19:45:25 +01:00
rate = rate16x16 + vp9_cost_bit(vp9_get_pred_prob(cm, xd, PRED_MBSKIP), 1);
dist = dist16x16;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
rate += rateuv[cm->txfm_mode != ONLY_4X4] -
rateuv_tokenonly[cm->txfm_mode != ONLY_4X4];
dist += (distuv[cm->txfm_mode != ONLY_4X4] >> 2);
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
mbmi->txfm_size = txfm_size_16x16;
Add proper skip support to intra frames. Change-Id: I7524883fb29f42303fb46a5bc6772fbcf8781d1d
2012-08-29 19:43:20 +02:00
} else if (error8x8 > error16x16) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (error4x4 < error16x16) {
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
rate = rateuv[TX_4X4] + rate4x4;
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
mbmi->mode = I4X4_PRED;
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
mbmi->txfm_size = TX_4X4;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
dist = dist4x4 + (distuv[TX_4X4] >> 2);
mbmi->uv_mode = modeuv[TX_4X4];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
} else {
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
mbmi->txfm_size = txfm_size_16x16;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
mbmi->mode = mode16x16;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
rate = rate16x16 + rateuv[mbmi->txfm_size != TX_4X4];
dist = dist16x16 + (distuv[mbmi->txfm_size != TX_4X4] >> 2);
mbmi->uv_mode = modeuv[mbmi->txfm_size != TX_4X4];
add 8x8 intra prediction modes Patch 1 to Patch 3 is an initial implementation of 8x8 intra prediction modes, here are with the following assumptions: a. 8x8 has 4 prediction modes DC, H, V and TM b. UV 4x4 block use the same mode as corresponding 8x8 area c. i8x8 modes are enabled for key frame only for now Patch 4: d. removed debug code from previous patches Patch 5: e. added stats code to collect entropy stats and further cleaned up Patch 6: f. changed mode stats code to collect finer stats of modes Patch 7: g. normalized i8x8 modes distribution to total at 256 (8bits). Patch 8: h. fixed a bug in decoder and removed debug printf output. Patch 9: i. more cleanups to address paul's comment Patch 10: j. messy rebase/merges to bring the commit up to date. Tests on HD clips encoded with all key frame showing consistent gain on all clips and all metrics:~0.5%(psnr) and 0.6%(ssim): http://www.corp.google.com/~yaowu/no_crawl/i8x8hd_allkey_fixedq.html To build and test, configure with: --enable-experimental --enable-i8x8 Change-Id: I9813fe07ae48cab5fdb5d904bca022514ad01e7f
2011-08-05 01:30:27 +02:00
}
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
rate += vp9_cost_bit(vp9_get_pred_prob(cm, xd, PRED_MBSKIP), 0);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
} else {
if (error4x4 < error8x8) {
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
rate = rateuv[TX_4X4] + rate4x4;
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
mbmi->mode = I4X4_PRED;
Selectable transform size try 2. Results: derf (vanilla or +hybridtx) +0.2% and (+hybrid16x16 or +tx16x16) +0.7%-0.8%; HD (vanilla or +hybridtx) +0.1-0.2% and (+hybrid16x16 or +tx16x16) +1.4%, STD/HD (vanilla or +hybridtx) about even, and (+hybrid16x16 or +tx16x16) +0.8-1.0%. Change-Id: I03899e2f7a64e725a863f32e55366035ba77aa62
2012-10-09 18:18:21 +02:00
mbmi->txfm_size = TX_4X4;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
dist = dist4x4 + (distuv[TX_4X4] >> 2);
mbmi->uv_mode = modeuv[TX_4X4];
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
} else {
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
mbmi->mode = I8X8_PRED;
Add support for tx_select in i8x8 encoding in keyframes. Also enable tx_select for keyframes. Change-Id: Iadb1231d9fa7af0c8dce3d9b41830b93a302479e
2013-03-07 19:45:25 +01:00
mbmi->txfm_size = txfm_size_8x8;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
set_i8x8_block_modes(x, mode8x8);
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
rate = rate8x8 + rateuv[TX_4X4];
dist = dist8x8 + (distuv[TX_4X4] >> 2);
add 8x8 intra prediction modes Patch 1 to Patch 3 is an initial implementation of 8x8 intra prediction modes, here are with the following assumptions: a. 8x8 has 4 prediction modes DC, H, V and TM b. UV 4x4 block use the same mode as corresponding 8x8 area c. i8x8 modes are enabled for key frame only for now Patch 4: d. removed debug code from previous patches Patch 5: e. added stats code to collect entropy stats and further cleaned up Patch 6: f. changed mode stats code to collect finer stats of modes Patch 7: g. normalized i8x8 modes distribution to total at 256 (8bits). Patch 8: h. fixed a bug in decoder and removed debug printf output. Patch 9: i. more cleanups to address paul's comment Patch 10: j. messy rebase/merges to bring the commit up to date. Tests on HD clips encoded with all key frame showing consistent gain on all clips and all metrics:~0.5%(psnr) and 0.6%(ssim): http://www.corp.google.com/~yaowu/no_crawl/i8x8hd_allkey_fixedq.html To build and test, configure with: --enable-experimental --enable-i8x8 Change-Id: I9813fe07ae48cab5fdb5d904bca022514ad01e7f
2011-08-05 01:30:27 +02:00
}
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
rate += vp9_cost_bit(vp9_get_pred_prob(cm, xd, PRED_MBSKIP), 0);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
}
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Add support for tx_select in i8x8 encoding in keyframes. Also enable tx_select for keyframes. Change-Id: Iadb1231d9fa7af0c8dce3d9b41830b93a302479e
2013-03-07 19:45:25 +01:00
for (i = 0; i < NB_TXFM_MODES; i++) {
x->mb_context[xd->sb_index][xd->mb_index].txfm_rd_diff[i] =
txfm_cache[0][cm->txfm_mode] - txfm_cache[0][i];
}
Add proper skip support to intra frames. Change-Id: I7524883fb29f42303fb46a5bc6772fbcf8781d1d
2012-08-29 19:43:20 +02:00
*returnrate = rate;
*returndist = dist;
Superblock encoding order This is the first patch to add superblock (32x32) coding order capabilities. It does not yet do any mode selection at the SB level, that will follow in a further patch. This patch encodes rows of SBs rather than MBs, each SB contains 2x2 MBs. Two intra prediction modes have been disabled since they require reconstructed data for the above-right MB which may not have been encoded yet (e.g. for the bottom right MB in each SB). Results on the one test clip I have tried (720p GIPS clip) suggest that it is somewhere around 0.2dB worse than the baseline version, so there may be bugs. It has been tested with no experiments enabled and with the following 3 experiments enabled: --enable-enhanced_interp --enable-high_precision_mv --enable-sixteenth_subpel_uv in each case the decode buffer matches the recon buffer (using "cmp" to compare the dumped/decoded frames). Note: Testing these experiments individually created errors. Some problems were found with other experiments but it is unclear what state these experiments are in: --enable-comp_intra_pred --enable-newentropy --enable-uvintra This code has not been extensively tested yet, so there is every likelihood that further bugs remain. I also intend to do some code cleanup & refactoring in tandem with the next patch that adds the 32x32 modes. Change-Id: I1eba7f740a70b3510df58db53464535ef881b4d9
2012-04-07 01:38:34 +02:00
}
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
int64_t vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
int mb_row, int mb_col,
int *returnrate,
int *returndistortion,
Merge pick_sb_modes and pick_sb64_modes. Change-Id: Iad69e7a3b7e470acf6094f6a52e7da69066fd552
2013-04-11 18:33:49 +02:00
BLOCK_SIZE_TYPE bsize,
PICK_MODE_CONTEXT *ctx) {
Make filter RD code and encode breakout variance size-independent. Static threshold results slightly up (+0.1% on derf), probably b/c we now take the filter (sharp/lowpass) into account for the breakout decision. Change-Id: I9f597601da434205142afd05f32690e7ba8fd690
2013-04-13 19:01:46 +02:00
const enum BlockSize block_size = y_bsizet_to_block_size(bsize);
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
VP9_COMMON *cm = &cpi->common;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
MACROBLOCKD *xd = &x->e_mbd;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
MB_PREDICTION_MODE this_mode;
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
MB_PREDICTION_MODE best_mode = DC_PRED;
Move pre, second_pre to per-plane MACROBLOCKD data Continue moving framebuffers to per-plane data. Change-Id: I237e5a998b364c4ec20316e7249206c0bff8631a
2013-04-20 04:16:14 +02:00
MV_REFERENCE_FRAME ref_frame, second_ref;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
unsigned char segment_id = xd->mode_info_context->mbmi.segment_id;
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
int comp_pred, i;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES];
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
int frame_mdcounts[4][4];
Pass macroblock index to pick inter functions Pass the current mb row and column around rather than the recon_yoffset and recon_uvoffset, since those offsets will change from predictor to predictor, based on the reference frame selection. Change-Id: If3f9df059e00f5048ca729d3d083ff428e1859c1
2013-02-07 19:09:05 +01:00
YV12_BUFFER_CONFIG yv12_mb[4];
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG,
VP9_ALT_FLAG };
Start to anonymize reference frames Remove lst_fb_idx, gld_fb_idx, alt_fb_idx, refresh_last_frame, refresh_golden_frame, refresh_alt_ref_frame from common. Gold/Alt are encode side conventions. From the decoder's perspective, we want to be dealing with numbered references. Updates to active_ref 2 signal mode context switches, vestigial from refresh_alt_ref_frame. This needs some clean up to make sense with increased numbers of reference frames, as well as reimplementing the swapping of alt/golden which was previously done using the buffer-to-buffer copy mechanism removed in an earlier commit. Change-Id: I7334445158b7666f9295d2a2dd22aa03f4485f58
2013-01-15 22:49:44 +01:00
int idx_list[4] = {0,
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
cpi->lst_fb_idx,
cpi->gld_fb_idx,
cpi->alt_fb_idx};
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
int mdcounts[4];
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t best_rd = INT64_MAX;
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
int64_t best_txfm_rd[NB_TXFM_MODES];
int64_t best_txfm_diff[NB_TXFM_MODES];
int64_t best_pred_diff[NB_PREDICTION_TYPES];
int64_t best_pred_rd[NB_PREDICTION_TYPES];
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
MB_MODE_INFO best_mbmode;
set up a new speed 1 slightly worse results for faster encodes Change-Id: I25ea82a18ce20635dbcd328808c1d05ac1f58fd7
2013-04-19 17:04:57 +02:00
int j;
warnings in various experiments Change-Id: Ib5106d4772450f8026f823dd743f162ab833b1d6
2012-11-30 16:29:43 +01:00
int mode_index, best_mode_index = 0;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
unsigned int ref_costs[MAX_REF_FRAMES];
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
int is_best_interintra = 0;
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t best_intra16_rd = INT64_MAX;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
int best_intra16_mode = DC_PRED;
#if SEPARATE_INTERINTRA_UV
int best_intra16_uv_mode = DC_PRED;
#endif
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#endif
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t best_overall_rd = INT64_MAX;
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
INTERPOLATIONFILTERTYPE best_filter = SWITCHABLE;
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
INTERPOLATIONFILTERTYPE tmp_best_filter = SWITCHABLE;
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
int rate_uv_intra[TX_SIZE_MAX_SB], rate_uv_tokenonly[TX_SIZE_MAX_SB];
int dist_uv[TX_SIZE_MAX_SB], skip_uv[TX_SIZE_MAX_SB];
MB_PREDICTION_MODE mode_uv[TX_SIZE_MAX_SB];
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
struct scale_factors scale_factor[4];
set up a new speed 1 slightly worse results for faster encodes Change-Id: I25ea82a18ce20635dbcd328808c1d05ac1f58fd7
2013-04-19 17:04:57 +02:00
unsigned int ref_frame_mask = 0;
unsigned int mode_mask = 0;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
xd->mode_info_context->mbmi.segment_id = segment_id;
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
estimate_ref_frame_costs(cpi, segment_id, ref_costs);
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
vpx_memset(&best_mbmode, 0, sizeof(best_mbmode));
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
for (i = 0; i < NB_PREDICTION_TYPES; ++i)
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
best_pred_rd[i] = INT64_MAX;
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
for (i = 0; i < NB_TXFM_MODES; i++)
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
best_txfm_rd[i] = INT64_MAX;
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
set up a new speed 1 slightly worse results for faster encodes Change-Id: I25ea82a18ce20635dbcd328808c1d05ac1f58fd7
2013-04-19 17:04:57 +02:00
// Create a mask set to 1 for each frame used by a smaller resolution.p
if (cpi->Speed > 0) {
switch (block_size) {
case BLOCK_64X64:
for (i = 0; i < 4; i++) {
for (j = 0; j < 4; j++) {
ref_frame_mask |= (1 << x->mb_context[i][j].mic.mbmi.ref_frame);
mode_mask |= (1 << x->mb_context[i][j].mic.mbmi.mode);
}
}
for (i = 0; i < 4; i++) {
ref_frame_mask |= (1 << x->sb32_context[i].mic.mbmi.ref_frame);
mode_mask |= (1 << x->sb32_context[i].mic.mbmi.mode);
}
break;
case BLOCK_32X32:
for (i = 0; i < 4; i++) {
ref_frame_mask |= (1
<< x->mb_context[xd->sb_index][i].mic.mbmi.ref_frame);
mode_mask |= (1 << x->mb_context[xd->sb_index][i].mic.mbmi.mode);
}
break;
default:
catch all for new block sizes Just make sure we don't stop them from testing in speed 1. Change-Id: Iec9b3dba0a32616ff7a451207e0f54b81bb72575
2013-04-19 18:48:56 +02:00
// Until we handle all block sizes set it to present;
ref_frame_mask = 0xff;
mode_mask = 0xff;
set up a new speed 1 slightly worse results for faster encodes Change-Id: I25ea82a18ce20635dbcd328808c1d05ac1f58fd7
2013-04-19 17:04:57 +02:00
break;
}
}
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ref_frame++) {
if (cpi->ref_frame_flags & flag_list[ref_frame]) {
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
setup_buffer_inter(cpi, x, idx_list[ref_frame], ref_frame, block_size,
set up a new speed 1 slightly worse results for faster encodes Change-Id: I25ea82a18ce20635dbcd328808c1d05ac1f58fd7
2013-04-19 17:04:57 +02:00
mb_row, mb_col, frame_mv[NEARESTMV], frame_mv[NEARMV],
frame_mdcounts, yv12_mb, scale_factor);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
frame_mv[NEWMV][ref_frame].as_int = INVALID_MV;
frame_mv[ZEROMV][ref_frame].as_int = 0;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
set up a new speed 1 slightly worse results for faster encodes Change-Id: I25ea82a18ce20635dbcd328808c1d05ac1f58fd7
2013-04-19 17:04:57 +02:00
if (cpi->Speed == 0
|| (cpi->Speed > 0 && (ref_frame_mask & (1 << INTRA_FRAME)))) {
mbmi->mode = DC_PRED;
for (i = 0; i <= ((bsize < BLOCK_SIZE_SB64X64) ? TX_16X16 : TX_32X32);
i++) {
mbmi->txfm_size = i;
rd_pick_intra_sbuv_mode(cpi, x, &rate_uv_intra[i], &rate_uv_tokenonly[i],
&dist_uv[i], &skip_uv[i], bsize);
mode_uv[i] = mbmi->uv_mode;
}
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
}
Support 32x32 intra modes in non-keyframe superblocks. Change-Id: Icf8ad313c543462e523bff89690e5daa8d49bcc0
2012-11-16 00:50:07 +01:00
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
for (mode_index = 0; mode_index < MAX_MODES; ++mode_index) {
Support 32x32 intra modes in non-keyframe superblocks. Change-Id: Icf8ad313c543462e523bff89690e5daa8d49bcc0
2012-11-16 00:50:07 +01:00
int mode_excluded = 0;
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
int64_t this_rd = INT64_MAX;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
int disable_skip = 0;
int other_cost = 0;
int compmode_cost = 0;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
int rate2 = 0, rate_y = 0, rate_uv = 0;
int distortion2 = 0, distortion_y = 0, distortion_uv = 0;
int skippable;
int64_t txfm_cache[NB_TXFM_MODES];
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
int compmode_interintra_cost = 0;
#endif
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
// Test best rd so far against threshold for trying this mode.
Fix crash in pick_inter_mode_sb(). It didn't handle rd_thresh == INT_MAX, which means the reference is unavailable. Change-Id: Ie6fa8b2577437411db81a8c24e8dcdfd856a0e8d
2012-11-24 17:19:04 +01:00
if (best_rd <= cpi->rd_threshes[mode_index] ||
cpi->rd_threshes[mode_index] == INT_MAX) {
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
continue;
}
Reset skip flag in superblock RD loop. This is the superblock equivalent of commit 290b83a. Change-Id: Ib3945dd9e992fa9ec1fdea5a11e17a3cc0e37637
2013-01-30 06:42:46 +01:00
x->skip = 0;
Change non-function symbol vp8_ prefixes to vp9_. For local symbols, make them static instead. Change-Id: I13d60947a46f711bc8991e16100cea2a13e3a22e
2012-10-31 01:12:12 +01:00
this_mode = vp9_mode_order[mode_index].mode;
ref_frame = vp9_mode_order[mode_index].ref_frame;
set up a new speed 1 slightly worse results for faster encodes Change-Id: I25ea82a18ce20635dbcd328808c1d05ac1f58fd7
2013-04-19 17:04:57 +02:00
if (!(ref_frame == INTRA_FRAME
|| (cpi->ref_frame_flags & flag_list[ref_frame]))) {
Fixing 8x8/4x4 ADST for intra modes with tx select This patch allows use of 8x8 and 4x4 ADST correctly for Intra 16x16 modes and Intra 8x8 modes when the block size selected is smaller than the prediction mode. Also includes some cleanups and refactoring. Rebase. Change-Id: Ie3257bdf07bdb9c6e9476915e3a80183c8fa005a
2012-11-16 00:14:38 +01:00
continue;
}
set up a new speed 1 slightly worse results for faster encodes Change-Id: I25ea82a18ce20635dbcd328808c1d05ac1f58fd7
2013-04-19 17:04:57 +02:00
if (cpi->Speed > 0) {
if (!(ref_frame_mask & (1 << ref_frame))) {
continue;
}
if (vp9_mode_order[mode_index].second_ref_frame != NONE
&& !(ref_frame_mask
& (1 << vp9_mode_order[mode_index].second_ref_frame))) {
continue;
}
}
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
mbmi->ref_frame = ref_frame;
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
mbmi->second_ref_frame = vp9_mode_order[mode_index].second_ref_frame;
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
set_scale_factors(xd, mbmi->ref_frame, mbmi->second_ref_frame,
scale_factor);
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
comp_pred = mbmi->second_ref_frame > INTRA_FRAME;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
mbmi->mode = this_mode;
mbmi->uv_mode = DC_PRED;
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
mbmi->interintra_mode = (MB_PREDICTION_MODE)(DC_PRED - 1);
mbmi->interintra_uv_mode = (MB_PREDICTION_MODE)(DC_PRED - 1);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
#endif
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
// Evaluate all sub-pel filters irrespective of whether we can use
// them for this frame.
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
mbmi->interp_filter = cm->mcomp_filter_type;
vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
// if (!(cpi->ref_frame_flags & flag_list[ref_frame]))
// continue;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
if (this_mode == I8X8_PRED ||
this_mode == I4X4_PRED ||
this_mode == SPLITMV)
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
continue;
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
// if (vp9_mode_order[mode_index].second_ref_frame == INTRA_FRAME)
// continue;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
if (comp_pred) {
if (ref_frame == ALTREF_FRAME) {
second_ref = LAST_FRAME;
} else {
second_ref = ref_frame + 1;
}
if (!(cpi->ref_frame_flags & flag_list[second_ref]))
continue;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
mbmi->second_ref_frame = second_ref;
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
set_scale_factors(xd, mbmi->ref_frame, mbmi->second_ref_frame,
scale_factor);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
mode_excluded =
mode_excluded ?
mode_excluded : cm->comp_pred_mode == SINGLE_PREDICTION_ONLY;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
} else {
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
// mbmi->second_ref_frame = vp9_mode_order[mode_index].second_ref_frame;
if (ref_frame != INTRA_FRAME) {
if (mbmi->second_ref_frame != INTRA_FRAME)
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
mode_excluded =
mode_excluded ?
mode_excluded : cm->comp_pred_mode == COMP_PREDICTION_ONLY;
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
else
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
mode_excluded = mode_excluded ? mode_excluded : !cm->use_interintra;
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
#endif
}
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
Move pre, second_pre to per-plane MACROBLOCKD data Continue moving framebuffers to per-plane data. Change-Id: I237e5a998b364c4ec20316e7249206c0bff8631a
2013-04-20 04:16:14 +02:00
setup_pre_planes(xd, &yv12_mb[ref_frame],
comp_pred ? &yv12_mb[second_ref] : NULL, 0, 0, NULL, NULL);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
vpx_memcpy(mdcounts, frame_mdcounts[ref_frame], sizeof(mdcounts));
// If the segment reference frame feature is enabled....
// then do nothing if the current ref frame is not allowed..
seg_common: give all exported symbols a vp9_ prefix. Change-Id: Ie8ba934a757acec1c80ac37ab9307c9a2783254e
2012-10-30 06:15:27 +01:00
if (vp9_segfeature_active(xd, segment_id, SEG_LVL_REF_FRAME) &&
!vp9_check_segref(xd, segment_id, ref_frame)) {
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
continue;
Segment Skip Flag First step in simplifying the segment mode and segment EOB flags into a simpler segment skip flag that implies 0,0 mv and EOB at position 0. Change-Id: Ib750cac31a7a02dc21082580498efd9f7d8d72a5
2013-01-28 16:22:53 +01:00
// If the segment skip feature is enabled....
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
// then do nothing if the current mode is not allowed..
Segment Skip Flag First step in simplifying the segment mode and segment EOB flags into a simpler segment skip flag that implies 0,0 mv and EOB at position 0. Change-Id: Ib750cac31a7a02dc21082580498efd9f7d8d72a5
2013-01-28 16:22:53 +01:00
} else if (vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP) &&
(this_mode != ZEROMV)) {
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
continue;
Segment Skip Flag First step in simplifying the segment mode and segment EOB flags into a simpler segment skip flag that implies 0,0 mv and EOB at position 0. Change-Id: Ib750cac31a7a02dc21082580498efd9f7d8d72a5
2013-01-28 16:22:53 +01:00
// Disable this drop out case if the ref frame
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
// segment level feature is enabled for this segment. This is to
// prevent the possibility that we end up unable to pick any mode.
Segment Skip Flag First step in simplifying the segment mode and segment EOB flags into a simpler segment skip flag that implies 0,0 mv and EOB at position 0. Change-Id: Ib750cac31a7a02dc21082580498efd9f7d8d72a5
2013-01-28 16:22:53 +01:00
} else if (!vp9_segfeature_active(xd, segment_id, SEG_LVL_REF_FRAME)) {
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
// Only consider ZEROMV/ALTREF_FRAME for alt ref frame,
// unless ARNR filtering is enabled in which case we want
// an unfiltered alternative
if (cpi->is_src_frame_alt_ref && (cpi->oxcf.arnr_max_frames == 0)) {
if (this_mode != ZEROMV || ref_frame != ALTREF_FRAME) {
continue;
}
}
}
Support 32x32 intra modes in non-keyframe superblocks. Change-Id: Icf8ad313c543462e523bff89690e5daa8d49bcc0
2012-11-16 00:50:07 +01:00
if (ref_frame == INTRA_FRAME) {
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
TX_SIZE uv_tx;
Make intra predictor support rectangular blocks The intra predictor supports configurable block sizes. It can handle intra prediction down to 4x4 sizes, when enabled in BLOCK_SIZE_TYPE. Change-Id: I7399ec2512393aa98aadda9813ca0c83e19af854
2013-04-11 21:12:11 +02:00
vp9_build_intra_predictors_sby_s(xd, bsize);
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
super_block_yrd(cpi, x, &rate_y, &distortion_y, &skippable,
bsize, txfm_cache);
uv_tx = mbmi->txfm_size;
if (bsize < BLOCK_SIZE_SB32X32 && uv_tx == TX_16X16)
uv_tx = TX_8X8;
else if (bsize < BLOCK_SIZE_SB64X64 && uv_tx == TX_32X32)
uv_tx = TX_16X16;
rate_uv = rate_uv_intra[uv_tx];
distortion_uv = dist_uv[uv_tx];
skippable = skippable && skip_uv[uv_tx];
mbmi->uv_mode = mode_uv[uv_tx];
Support 32x32 intra modes in non-keyframe superblocks. Change-Id: Icf8ad313c543462e523bff89690e5daa8d49bcc0
2012-11-16 00:50:07 +01:00
rate2 = rate_y + x->mbmode_cost[cm->frame_type][mbmi->mode] + rate_uv;
distortion2 = distortion_y + distortion_uv;
} else {
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
YV12_BUFFER_CONFIG *scaled_ref_frame = NULL;
int fb;
if (mbmi->ref_frame == LAST_FRAME) {
fb = cpi->lst_fb_idx;
} else if (mbmi->ref_frame == GOLDEN_FRAME) {
fb = cpi->gld_fb_idx;
} else {
fb = cpi->alt_fb_idx;
}
Use ref_frame_map vice active_ref_idx on the encoder This patch makes the encoder's use of ref_frame_map and active_ref_idx consistent with the decoder. ref_frame_map[] maps a reference buffer index to its actual location in the yv12_fb array, since many references may share an underlying buffer. active_ref_idx[] mirrors cpi->{lst,gld,alt}_fb_idx, holding the active references in each slot. This also fixes a bug in setup_buffer_inter() where the incorrect reference was used to populate the scaling factors. Change-Id: Id3728f6d77cffcd27c248903bf51f9c3e594287e
2013-02-22 20:22:03 +01:00
if (cpi->scaled_ref_idx[fb] != cm->ref_frame_map[fb])
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
scaled_ref_frame = &cm->yv12_fb[cpi->scaled_ref_idx[fb]];
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
if (mbmi->second_ref_frame == INTRA_FRAME) {
if (best_intra16_mode == DC_PRED - 1) continue;
mbmi->interintra_mode = best_intra16_mode;
#if SEPARATE_INTERINTRA_UV
mbmi->interintra_uv_mode = best_intra16_uv_mode;
#else
mbmi->interintra_uv_mode = best_intra16_mode;
#endif
}
#endif
Make RD superblock mode search size-agnostic. Merge various super_block_yrd and super_block_uvrd versions into one common function that works for all sizes. Make transform size selection size-agnostic also. This fixes a slight bug in the intra UV superblock code where it used the wrong transform size for txsz > 8x8, and stores the txsz selection for superblocks properly (instead of forgetting it). Lastly, it removes the trellis search that was done for 16x16 intra predictors, since trellis is relatively expensive and should thus only be done after RD mode selection. Gives basically identical results on derf (+0.009%). Change-Id: If4485c6f0a0fe4038b3172f7a238477c35a6f8d3
2013-04-11 00:55:59 +02:00
this_rd = handle_inter_mode(cpi, x, bsize,
Remove unused parameters in handle_inter_mode Removed 2 unused parameters. Change-Id: Ic2862569313c404047072b268c3d2be3f635492c
2013-04-18 20:55:46 +02:00
mdcounts, txfm_cache,
Support 32x32 intra modes in non-keyframe superblocks. Change-Id: Icf8ad313c543462e523bff89690e5daa8d49bcc0
2012-11-16 00:50:07 +01:00
&rate2, &distortion2, &skippable,
&compmode_cost,
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
Support 32x32 intra modes in non-keyframe superblocks. Change-Id: Icf8ad313c543462e523bff89690e5daa8d49bcc0
2012-11-16 00:50:07 +01:00
&compmode_interintra_cost,
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
#endif
Support 32x32 intra modes in non-keyframe superblocks. Change-Id: Icf8ad313c543462e523bff89690e5daa8d49bcc0
2012-11-16 00:50:07 +01:00
&rate_y, &distortion_y,
&rate_uv, &distortion_uv,
Pass macroblock index to pick inter functions Pass the current mb row and column around rather than the recon_yoffset and recon_uvoffset, since those offsets will change from predictor to predictor, based on the reference frame selection. Change-Id: If3f9df059e00f5048ca729d3d083ff428e1859c1
2013-02-07 19:09:05 +01:00
&mode_excluded, &disable_skip,
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
mode_index, &tmp_best_filter, frame_mv,
Combined motion compensation with scaled predictors This patch extends the previous support for using references of a different resolution in ZEROMV mode to all inter prediction modes. Subpixel based best-mv scoring is disabled when the reference frame differs in resolution from the current frame. Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac
2013-02-20 21:34:31 +01:00
scaled_ref_frame, mb_row, mb_col);
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
if (this_rd == INT64_MAX)
Support 32x32 intra modes in non-keyframe superblocks. Change-Id: Icf8ad313c543462e523bff89690e5daa8d49bcc0
2012-11-16 00:50:07 +01:00
continue;
}
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
if (cpi->common.use_interintra) {
rate2 += compmode_interintra_cost;
}
#endif
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
if (cpi->common.comp_pred_mode == HYBRID_PREDICTION) {
rate2 += compmode_cost;
}
// Estimate the reference frame signaling cost and add it
// to the rolling cost variable.
rate2 += ref_costs[xd->mode_info_context->mbmi.ref_frame];
if (!disable_skip) {
// Test for the condition where skip block will be activated
// because there are no non zero coefficients and make any
// necessary adjustment for rate. Ignore if skip is coded at
// segment level as the cost wont have been added in.
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
int mb_skip_allowed;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
// Is Mb level skip allowed (i.e. not coded at segment level).
mb_skip_allowed = !vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
if (skippable) {
// Back out the coefficient coding costs
rate2 -= (rate_y + rate_uv);
// for best_yrd calculation
rate_uv = 0;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
if (mb_skip_allowed) {
int prob_skip_cost;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
// Cost the skip mb case
vp9_prob skip_prob =
vp9_get_pred_prob(cm, xd, PRED_MBSKIP);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
if (skip_prob) {
prob_skip_cost = vp9_cost_bit(skip_prob, 1);
rate2 += prob_skip_cost;
other_cost += prob_skip_cost;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
}
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
} else if (mb_skip_allowed) {
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
// Add in the cost of the no skip flag.
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
int prob_skip_cost = vp9_cost_bit(vp9_get_pred_prob(cm, xd,
PRED_MBSKIP), 0);
rate2 += prob_skip_cost;
other_cost += prob_skip_cost;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
// Calculate the final RD estimate for this mode.
this_rd = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
}
#if 0
// Keep record of best intra distortion
if ((xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) &&
(this_rd < best_intra_rd)) {
best_intra_rd = this_rd;
*returnintra = distortion2;
}
#endif
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
if ((mbmi->ref_frame == INTRA_FRAME) &&
(this_mode <= TM_PRED) &&
(this_rd < best_intra16_rd)) {
best_intra16_rd = this_rd;
best_intra16_mode = this_mode;
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
#if SEPARATE_INTERINTRA_UV
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
best_intra16_uv_mode = (mbmi->txfm_size != TX_4X4 ?
mode_uv_8x8 : mode_uv_4x4);
Coding con-zero count rather than EOB for coeffs This patch revamps the entropy coding of coefficients to code first a non-zero count per coded block and correspondingly remove the EOB token from the token set. STATUS: Main encode/decode code achieving encode/decode sync - done. Forward and backward probability updates to the nzcs - done. Rd costing updates for nzcs - done. Note: The dynamic progrmaming apporach used in trellis quantization is not exactly compatible with nzcs. A suboptimal approach has been used instead where branch costs are updated to account for changes in the nzcs. TODO: Training the default probs/counts for nzcs Change-Id: I951bc1e22f47885077a7453a09b0493daa77883d
2013-02-20 19:16:24 +01:00
#endif
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
}
#endif
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
if (!disable_skip && mbmi->ref_frame == INTRA_FRAME)
for (i = 0; i < NB_PREDICTION_TYPES; ++i)
best_pred_rd[i] = MIN(best_pred_rd[i], this_rd);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
if (this_rd < best_overall_rd) {
best_overall_rd = this_rd;
Refactoring of switchable filter search for speed Refactors the switchable filter search in the rd loop to improve encode speed. Uses a piecewise approximation to a closed form expression to estimate rd cost for a Laplacian source with a given variance and quantization step-size. About 40% encode time reduction is achieved. Results (on a feb 12 baseline) show a slight drop: derf: -0.019% yt: +0.010% std-hd: -0.162% hd: -0.050% Change-Id: Ie861badf5bba1e3b1052e29a0ef1b7e256edbcd0
2013-02-12 02:08:52 +01:00
best_filter = tmp_best_filter;
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
best_mode = this_mode;
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
is_best_interintra = (mbmi->second_ref_frame == INTRA_FRAME);
#endif
}
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
// Did this mode help.. i.e. is it the new best mode
if (this_rd < best_rd || x->skip) {
if (!mode_excluded) {
// Note index of best mode so far
best_mode_index = mode_index;
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
if (this_mode <= I4X4_PRED) {
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
/* required for left and above block mv */
Support 32x32 intra modes in non-keyframe superblocks. Change-Id: Icf8ad313c543462e523bff89690e5daa8d49bcc0
2012-11-16 00:50:07 +01:00
mbmi->mv[0].as_int = 0;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
other_cost += ref_costs[xd->mode_info_context->mbmi.ref_frame];
*returnrate = rate2;
*returndistortion = distortion2;
best_rd = this_rd;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
vpx_memcpy(&best_mbmode, mbmi, sizeof(MB_MODE_INFO));
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
#if 0
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
// Testing this mode gave rise to an improvement in best error score.
// Lower threshold a bit for next time
cpi->rd_thresh_mult[mode_index] =
(cpi->rd_thresh_mult[mode_index] >= (MIN_THRESHMULT + 2)) ?
cpi->rd_thresh_mult[mode_index] - 2 : MIN_THRESHMULT;
cpi->rd_threshes[mode_index] =
(cpi->rd_baseline_thresh[mode_index] >> 7)
* cpi->rd_thresh_mult[mode_index];
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
#endif
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
} else {
// If the mode did not help improve the best error case then
// raise the threshold for testing that mode next time around.
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
#if 0
cpi->rd_thresh_mult[mode_index] += 4;
if (cpi->rd_thresh_mult[mode_index] > MAX_THRESHMULT)
cpi->rd_thresh_mult[mode_index] = MAX_THRESHMULT;
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
cpi->rd_threshes[mode_index] =
(cpi->rd_baseline_thresh[mode_index] >> 7)
* cpi->rd_thresh_mult[mode_index];
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
#endif
}
/* keep record of best compound/single-only prediction */
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
if (!disable_skip && mbmi->ref_frame != INTRA_FRAME) {
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
int single_rd, hybrid_rd, single_rate, hybrid_rate;
if (cpi->common.comp_pred_mode == HYBRID_PREDICTION) {
single_rate = rate2 - compmode_cost;
hybrid_rate = rate2;
} else {
single_rate = rate2;
hybrid_rate = rate2 + compmode_cost;
}
single_rd = RDCOST(x->rdmult, x->rddiv, single_rate, distortion2);
hybrid_rd = RDCOST(x->rdmult, x->rddiv, hybrid_rate, distortion2);
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
if (mbmi->second_ref_frame <= INTRA_FRAME &&
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
single_rd < best_pred_rd[SINGLE_PREDICTION_ONLY]) {
best_pred_rd[SINGLE_PREDICTION_ONLY] = single_rd;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
} else if (mbmi->second_ref_frame > INTRA_FRAME &&
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
single_rd < best_pred_rd[COMP_PREDICTION_ONLY]) {
best_pred_rd[COMP_PREDICTION_ONLY] = single_rd;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
if (hybrid_rd < best_pred_rd[HYBRID_PREDICTION])
best_pred_rd[HYBRID_PREDICTION] = hybrid_rd;
}
/* keep record of best txfm size */
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
if (!mode_excluded && this_rd != INT64_MAX) {
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
for (i = 0; i < NB_TXFM_MODES; i++) {
int64_t adj_rd;
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
if (this_mode != I4X4_PRED) {
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
adj_rd = this_rd + txfm_cache[i] - txfm_cache[cm->txfm_mode];
} else {
adj_rd = this_rd;
}
if (adj_rd < best_txfm_rd[i])
best_txfm_rd[i] = adj_rd;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
}
if (x->skip && !mode_excluded)
break;
}
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
assert((cm->mcomp_filter_type == SWITCHABLE) ||
(cm->mcomp_filter_type == best_mbmode.interp_filter) ||
Rename B_PRED to I4X4_PRED So it is consistent with I8x8_PRED. Change-Id: Iefa65124b2419690d83e526c611129c0ede29d11
2013-04-11 21:16:35 +02:00
(best_mbmode.mode <= I4X4_PRED));
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
Supports inter-intra prediction with superblocks Adds support for compound inter-intra prediction with superblocks. Also, fixes a bug that disabled intra modes for superblocks. Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
2012-11-30 20:46:20 +01:00
#if CONFIG_COMP_INTERINTRA_PRED
++cpi->interintra_select_count[is_best_interintra];
// if (is_best_interintra) printf("best_interintra\n");
#endif
New interpolation filter selection algorithm Old Scheme: When SWITCHABLE filter selection is enabled the encoder evaluates the use of each interpolation filter type and selects the best one to use at the MB level. A frame- level flag can be set to force the use of a particular filter type for all MBs in a frame if it is more efficient to encode that way. The logic here involved a Q dependent threshold that assumed that the second 8-tap filter was a high-pass filter. However, this requires a trip around the recode loop. If the frame-level flag indicates use of a particular filter, the other filters are not evaluated in the pick_mode loop. New Scheme: Each filter type is evaluated at the MB level and a record of the best filter is kept, irrespective of what filter is signaled at the frame-level. Once all MBs have been encoded, a decision is made as to what frame-level mode to set for the *next* frame. If one filter is used by 80% or more of the MBs, then this filter is forced since it is assumed that this will be more efficient if the next frame has similar characteristics. i.e. there is a one-frame lag between measuring the filter selection and setting the frame-level mode to use. Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
2012-12-20 23:56:19 +01:00
// Accumulate filter usage stats
// TODO(agrange): Use RD criteria to select interpolation filter mode.
if ((best_mode >= NEARESTMV) && (best_mode <= SPLITMV))
++cpi->best_switchable_interp_count[vp9_switchable_interp_map[best_filter]];
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
// TODO(rbultje) integrate with RD thresholding
#if 0
// Reduce the activation RD thresholds for the best choice mode
if ((cpi->rd_baseline_thresh[best_mode_index] > 0) &&
(cpi->rd_baseline_thresh[best_mode_index] < (INT_MAX >> 2))) {
int best_adjustment = (cpi->rd_thresh_mult[best_mode_index] >> 2);
cpi->rd_thresh_mult[best_mode_index] =
(cpi->rd_thresh_mult[best_mode_index] >= (MIN_THRESHMULT + best_adjustment)) ?
cpi->rd_thresh_mult[best_mode_index] - best_adjustment : MIN_THRESHMULT;
cpi->rd_threshes[best_mode_index] =
(cpi->rd_baseline_thresh[best_mode_index] >> 7) * cpi->rd_thresh_mult[best_mode_index];
}
#endif
// This code forces Altref,0,0 and skip for the frame that overlays a
// an alrtef unless Altref is filtered. However, this is unsafe if
Segment Skip Flag First step in simplifying the segment mode and segment EOB flags into a simpler segment skip flag that implies 0,0 mv and EOB at position 0. Change-Id: Ib750cac31a7a02dc21082580498efd9f7d8d72a5
2013-01-28 16:22:53 +01:00
// segment level coding of ref frame is enabled for this segment.
seg_common: give all exported symbols a vp9_ prefix. Change-Id: Ie8ba934a757acec1c80ac37ab9307c9a2783254e
2012-10-30 06:15:27 +01:00
if (!vp9_segfeature_active(xd, segment_id, SEG_LVL_REF_FRAME) &&
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
cpi->is_src_frame_alt_ref &&
(cpi->oxcf.arnr_max_frames == 0) &&
(best_mbmode.mode != ZEROMV || best_mbmode.ref_frame != ALTREF_FRAME)) {
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
mbmi->mode = ZEROMV;
mbmi->ref_frame = ALTREF_FRAME;
Compound inter-intra experiment A patch on compound inter-intra prediction. In compound inter-intra prediction, a new predictor for 16x16 inter coded MBs are obtained by combining a single inter predictor with a 16x16 intra predictor, in a manner that the weight varies with distance from the top/left boundary. The current search strategy is to combine the best inter mode with the best intra mode obtained independently. Results so far: derf +0.31% yt +0.32% std-hd +0.35% hd +0.42% It is conceivable that the results would improve somewhat with a more thorough search strategy where all intra modes are searched given the best mv, or even a joint search for the best mv and the best intra mode. Change-Id: I7951f1ed0d6eb31ca32ac24d120f1585bcd8d79b
2012-11-07 15:50:25 +01:00
mbmi->second_ref_frame = INTRA_FRAME;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
mbmi->mv[0].as_int = 0;
mbmi->uv_mode = DC_PRED;
Remove the mb_no_coeff_skip flag This flag was added to VP8 to allow a mode where MB-level skipping was not allowed, saving a bit per mb. It was never used in practice, and hasn't been tested in VP9, so remove it. Change-Id: Id450ec6904c6d06c1919508e7efc52d05cde5631
2013-04-09 19:54:19 +02:00
mbmi->mb_skip_coeff = 1;
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
mbmi->partitioning = 0;
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
mbmi->txfm_size = cm->txfm_mode == TX_MODE_SELECT ?
Default superblock skip flag to 32x32 for skip-blocks. This is identical to the later decisions made in encode_superblock(). This commit doesn't actually change anything, but makes the mbmi state more consistent between the RD loop and the final encode result. Change-Id: I9e735afb7c5a52e5b61728cb88c67ef9b9bf59be
2013-01-30 06:46:31 +01:00
TX_32X32 : cm->txfm_mode;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
vpx_memset(best_txfm_diff, 0, sizeof(best_txfm_diff));
vpx_memset(best_pred_diff, 0, sizeof(best_pred_diff));
goto end;
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
}
// macroblock modes
Fix superblock experiment. Also merge some duplicate code related to the superblock experiment in the RD loop. Change-Id: Ic93f1d4d1ed81220fd7ecf6e65da2821a215b2de
2012-10-30 01:58:18 +01:00
vpx_memcpy(mbmi, &best_mbmode, sizeof(MB_MODE_INFO));
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
for (i = 0; i < NB_PREDICTION_TYPES; ++i) {
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
if (best_pred_rd[i] == INT64_MAX)
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
best_pred_diff[i] = INT_MIN;
else
best_pred_diff[i] = best_rd - best_pred_rd[i];
}
if (!x->skip) {
for (i = 0; i < NB_TXFM_MODES; i++) {
Use INT64_MAX instead of LLONG_MAX These variables have the type int64_t, not long long. long long could be a larger type than 64 bits. Emulate INT64_MAX for older versions of MSVC, and remove the unreferenced vpx_ports/vpxtypes.h Change-Id: Ideaca71838fcd3849d816d5ab17aa347c97d03b0
2013-01-14 20:49:30 +01:00
if (best_txfm_rd[i] == INT64_MAX)
Prevent filling transform size cache with uninitialized values. The 32x32 value in case of splitmv was uninitialized. this leads to all kind of erratic behaviour down the line. Also fill in dummy values for superblocks in keyframes (the values are currently unused, but we run into integer overflows anyway, which makes detecting bad cases harder). Lastly, in case we did not find any RD value at all, don't set tx_diff to INT_MIN, but instead set it to zero (since if we couldn't find a mode, it's unlikely that any particular transform would have made that worse or better; rather, it's likely equally bad for all tx_sizes). Change-Id: If236fd3aa2037e5b398d03f3b1978fbbc5ce740e
2013-02-16 00:55:31 +01:00
best_txfm_diff[i] = 0;
Implement tx_select for superblock encoding. Also split superblock handling code out of decode_macroblock() into a new function decode_superblock(), for easier readability. Derf +0.05%, HD +0.2%, STDHD +0.1%. We can likely get further gains by allowing to select mb_skip_coeff for a subset of the complete SB or something along those lines, because although this change allows coding smaller transforms for bigger predictors, it increases the overhead of coding EOBs to skip the parts where the residual is near-zero, and thus the overall gain is not as high as we'd expect. Change-Id: I552ce1286487267f504e3090b683e15515791efa
2012-11-08 20:03:00 +01:00
else
best_txfm_diff[i] = best_rd - best_txfm_rd[i];
}
} else {
vpx_memset(best_txfm_diff, 0, sizeof(best_txfm_diff));
}
end:
Spatial resamping of ZEROMV predictors This patch allows coding frames using references of different resolution, in ZEROMV mode. For compound prediction, either reference may be scaled. To test, I use the resize_test and enable WRITE_RECON_BUFFER in vp9_onyxd_if.c. It's also useful to apply this patch to test/i420_video_source.h: --- a/test/i420_video_source.h +++ b/test/i420_video_source.h @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource { virtual void FillFrame() { // Read a frame from input_file. + if (frame_ != 3) if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) { limit_ = frame_; } This forces the frame that the resolution changes on to be coded with no motion, only scaling, and improves the quality of the result. Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
set_scale_factors(xd, mbmi->ref_frame, mbmi->second_ref_frame,
scale_factor);
Merge pick_sb_modes and pick_sb64_modes. Change-Id: Iad69e7a3b7e470acf6094f6a52e7da69066fd552
2013-04-11 18:33:49 +02:00
store_coding_context(x, ctx, best_mode_index, NULL,
&mbmi->ref_mvs[mbmi->ref_frame][0],
&mbmi->ref_mvs[mbmi->second_ref_frame < 0 ? 0 :
mbmi->second_ref_frame][0],
best_pred_diff, best_txfm_diff);
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
return best_rd;
}
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
Change name of common top-level structures from VP8 to VP9. This change encompasses VP8_PTR, VP8_COMP, VP8D_COMP, VP8_COMMON, VP8Decompressor and VP8Common. Change-Id: I514ef4ad4e682370f36d656af1c09ee20da216ad
2012-10-31 01:53:32 +01:00
void vp9_pick_mode_inter_macroblock(VP9_COMP *cpi, MACROBLOCK *x,
Pass macroblock index to pick inter functions Pass the current mb row and column around rather than the recon_yoffset and recon_uvoffset, since those offsets will change from predictor to predictor, based on the reference frame selection. Change-Id: If3f9df059e00f5048ca729d3d083ff428e1859c1
2013-02-07 19:09:05 +01:00
int mb_row, int mb_col,
Adjust style to match Google Coding Style a little more closely. Most of these were picked up in the previous commit (prefix change from vp8_ to vp9_), but I'm pushing this separately so that it's easier to review. Change-Id: I91e959895778b8632d7d33375523df8a7568a490
2012-10-30 22:25:33 +01:00
int *totalrate, int *totaldist) {
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
MACROBLOCKD *const xd = &x->e_mbd;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
MB_MODE_INFO * mbmi = &x->e_mbd.mode_info_context->mbmi;
Refactor to remove some arguments from vp8_rd_pick_inter_mode Change-Id: I8e72279cc68f34d269705f06cdaf8f3d06eed635
2012-08-10 03:25:29 +02:00
int rate, distortion;
A few more conversions from 32-bit to 64-bit RD calculations. Change-Id: Ida01bb8f4b35fde24b0f096329d326833f1fdc7e
2012-08-02 20:01:48 +02:00
int64_t intra_error = 0;
Code cleanup. Add local variable in several places to reference the MB mode info structure. Currently this is usually accessed in the code as x->e_mbd.mode_info_context->mbmi.* or in some places xd->mode_info_context->mbmi.* Resolved some uses of x-> for the MACROBLOCKD structure. Rebased without dependency on motion reference experiment. Change-Id: If6718276ee4f2ef131825d1524dfdb02a3793aed
2012-08-10 15:12:43 +02:00
unsigned char *segment_id = &mbmi->segment_id;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
if (xd->segmentation_enabled)
x->encode_breakout = cpi->segment_encode_breakout[*segment_id];
else
x->encode_breakout = cpi->oxcf.encode_breakout;
// if (cpi->sf.RD)
// For now this codebase is limited to a single rd encode path
{
int zbin_mode_boost_enabled = cpi->zbin_mode_boost_enabled;
Pass macroblock index to pick inter functions Pass the current mb row and column around rather than the recon_yoffset and recon_uvoffset, since those offsets will change from predictor to predictor, based on the reference frame selection. Change-Id: If3f9df059e00f5048ca729d3d083ff428e1859c1
2013-02-07 19:09:05 +01:00
rd_pick_inter_mode(cpi, x, mb_row, mb_col, &rate,
Place non-static function prototypes in a header file. Change-Id: I7cd21b9f1e69f4e0b3338bfe27b3c67e4b47de58
2012-11-02 19:22:57 +01:00
&distortion, &intra_error);
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
/* restore cpi->zbin_mode_boost_enabled */
cpi->zbin_mode_boost_enabled = zbin_mode_boost_enabled;
}
// else
// The non rd encode path has been deleted from this code base
// to simplify development
Remove vp8 in local symbols. For non-static functions, change the prefix to vp9_. For static functions, remove the prefix. Also fix some comments, remove unused code or unused function prototypes. Change-Id: I1f8be05362f66060fe421c3d4c9a906fdf835de5
2012-10-31 22:40:53 +01:00
// vp9_pick_inter_mode
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
// Store metrics so they can be added in to totals if this mode is picked
64x64 blocksize support. 3.2% gains on std/hd, 1.0% gains on hd. Change-Id: I481d5df23d8a4fc650a5bcba956554490b2bd200
2013-01-06 03:20:25 +01:00
x->mb_context[xd->sb_index][xd->mb_index].distortion = distortion;
x->mb_context[xd->sb_index][xd->mb_index].intra_error = intra_error;
Restyle code Approximate the Google style guide[1] so that that there's a written document to follow and tools to check compliance[2]. [1]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml [2]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py Change-Id: Idf40e3d8dddcc72150f6af127b13e5dab838685f
2012-07-14 00:21:29 +02:00
Superblock coding. This commit adds a pick_sb_mode() function which selects the best 32x32 superblock coding mode. Then it selects the best per-MB modes, compares the two and encodes that in the bitstream. The bitstream coding is rather simplistic right now. At the SB level, we code a bit to indicate whether this block uses SB-coding (32x32 prediction) or MB-coding (anything else), and then we follow with the actual modes. This could and should be modified in the future, but is omitted from this commit because it will likely involve reorganizing much more code rather than just adding SB coding, so it's better to let that be judged on its own merits. Gains on derf: about even, YT/HD: +0.75%, STD/HD: +1.5%. Change-Id: Iae313a7cbd8f75b3c66d04a68b991cb096eaaba6
2012-08-20 23:43:34 +02:00
*totalrate = rate;
*totaldist = distortion;
Move RD intra block mode selection to rdopt.c This change is analogous to I0b67dae1f8a74902378da7bdf565e39ab832dda7, which made the move for the non-RD path. Change-Id: If63fc1b0cd1eb7f932e710f83ff24d91454f8ed1
2011-06-08 18:05:05 +02:00
}
Reference in New Issue Copy Permalink