generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
9c9d6743d4
vpx/vp8/encoder/onyx_if.c

4612 lines
149 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
*/
Fix building with --disable-postproc Change-Id: I7e6bc28e7974a376da747300744e0dd5dc1d21e9
2011-08-01 22:42:14 +02:00
#include "vpx_config.h"
Fix relative include paths Allow compiling without adding vp8/{common,encoder,decoder} to the include paths. Change-Id: Ifeb5dac351cdfadcd659736f5158b315a0030b6c
2011-02-10 20:41:38 +01:00
#include "vp8/common/onyxc_int.h"
Initial WebM release
2010-05-18 17:58:33 +02:00
#include "onyx_int.h"
Fix relative include paths Allow compiling without adding vp8/{common,encoder,decoder} to the include paths. Change-Id: Ifeb5dac351cdfadcd659736f5158b315a0030b6c
2011-02-10 20:41:38 +01:00
#include "vp8/common/systemdependent.h"
Initial WebM release
2010-05-18 17:58:33 +02:00
#include "quantize.h"
Fix relative include paths Allow compiling without adding vp8/{common,encoder,decoder} to the include paths. Change-Id: Ifeb5dac351cdfadcd659736f5158b315a0030b6c
2011-02-10 20:41:38 +01:00
#include "vp8/common/alloccommon.h"
Initial WebM release
2010-05-18 17:58:33 +02:00
#include "mcomp.h"
#include "firstpass.h"
#include "psnr.h"
#include "vpx_scale/vpxscale.h"
Fix relative include paths Allow compiling without adding vp8/{common,encoder,decoder} to the include paths. Change-Id: Ifeb5dac351cdfadcd659736f5158b315a0030b6c
2011-02-10 20:41:38 +01:00
#include "vp8/common/extend.h"
Initial WebM release
2010-05-18 17:58:33 +02:00
#include "ratectrl.h"
Fix relative include paths Allow compiling without adding vp8/{common,encoder,decoder} to the include paths. Change-Id: Ifeb5dac351cdfadcd659736f5158b315a0030b6c
2011-02-10 20:41:38 +01:00
#include "vp8/common/quant_common.h"
move segmentation_common to encoder vp8_update_gf_useage_maps() is only used by the encoder. This patch fixes the ability to build in decode-only or encode-only configurations. Change-Id: I3a5211428e539886ba998e09e8abd747ac55c9aa
2010-08-13 20:50:51 +02:00
#include "segmentation.h"
Fix relative include paths Allow compiling without adding vp8/{common,encoder,decoder} to the include paths. Change-Id: Ifeb5dac351cdfadcd659736f5158b315a0030b6c
2011-02-10 20:41:38 +01:00
#include "vp8/common/g_common.h"
Initial WebM release
2010-05-18 17:58:33 +02:00
#include "vpx_scale/yv12extend.h"
Fix building with --disable-postproc Change-Id: I7e6bc28e7974a376da747300744e0dd5dc1d21e9
2011-08-01 22:42:14 +02:00
#if CONFIG_POSTPROC
Fix relative include paths Allow compiling without adding vp8/{common,encoder,decoder} to the include paths. Change-Id: Ifeb5dac351cdfadcd659736f5158b315a0030b6c
2011-02-10 20:41:38 +01:00
#include "vp8/common/postproc.h"
Fix building with --disable-postproc Change-Id: I7e6bc28e7974a376da747300744e0dd5dc1d21e9
2011-08-01 22:42:14 +02:00
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
#include "vpx_mem/vpx_mem.h"
Fix relative include paths Allow compiling without adding vp8/{common,encoder,decoder} to the include paths. Change-Id: Ifeb5dac351cdfadcd659736f5158b315a0030b6c
2011-02-10 20:41:38 +01:00
#include "vp8/common/swapyv12buffer.h"
Initial WebM release
2010-05-18 17:58:33 +02:00
#include "vpx_ports/vpx_timer.h"
isolate new temporal filtering code onyx_if is getting pretty big. split out the temporal code to make it easier to look at. Change-Id: I207c3a94c90e91b32e3ea5e1836a53b7a990fabd
2010-10-20 22:27:33 +02:00
#include "temporal_filter.h"
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
#include "vp8/common/seg_common.h"
#include "mbgraph.h"
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
#include "vp8/common/pred_common.h"
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
#include "vp8/encoder/rdopt.h"
Fix a bunch of warnings. Change-Id: I883b4f63559c1cf66a25016e946b1b1ae98a20da
2012-06-15 03:14:43 +02:00
#include "bitstream.h"
#include "ratectrl.h"
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
Add runtime CPU detection support for ARM. The primary goal is to allow a binary to be built which supports NEON, but can fall back to non-NEON routines, since some Android devices do not have NEON, even if they are otherwise ARMv7 (e.g., Tegra). The configure-generated flags HAVE_ARMV7, etc., are used to decide which versions of each function to build, and when CONFIG_RUNTIME_CPU_DETECT is enabled, the correct version is chosen at run time. In order for this to work, the CFLAGS must be set to something appropriate (e.g., without -mfpu=neon for ARMv7, and with appropriate -march and -mcpu for even earlier configurations), or the native C code will not be able to run. The ASFLAGS must remain set for the most advanced instruction set required at build time, since the ARM assembler will refuse to emit them otherwise. I have not attempted to make any changes to configure to do this automatically. Doing so will probably require the addition of new configure options. Many of the hooks for RTCD on ARM were already there, but a lot of the code had bit-rotted, and a good deal of the ARM-specific code is not integrated into the RTCD structs at all. I did not try to resolve the latter, merely to add the minimal amount of protection around them to allow RTCD to work. Those functions that were called based on an ifdef at the calling site were expanded to check the RTCD flags at that site, but they should be added to an RTCD struct somewhere in the future. The functions invoked with global function pointers still are, but these should be moved into an RTCD struct for thread safety (I believe every platform currently supported has atomic pointer stores, but this is not guaranteed). The encoder's boolhuff functions did not even have _c and armv7 suffixes, and the correct version was resolved at link time. The token packing functions did have appropriate suffixes, but the version was selected with a define, with no associated RTCD struct. However, for both of these, the only armv7 instruction they actually used was rbit, and this was completely superfluous, so I reworked them to avoid it. The only non-ARMv4 instruction remaining in them is clz, which is ARMv5 (not even ARMv5TE is required). Considering that there are no ARM-specific configs which are not at least ARMv5TE, I did not try to detect these at runtime, and simply enable them for ARMv5 and above. Finally, the NEON register saving code was completely non-reentrant, since it saved the registers to a global, static variable. I moved the storage for this onto the stack. A single binary built with this code was tested on an ARM11 (ARMv6) and a Cortex A8 (ARMv7 w/NEON), for both the encoder and decoder, and produced identical output, while using the correct accelerated functions on each. I did not test on any earlier processors. Change-Id: I45cbd63a614f4554c3b325c45d46c0806f009eaa
2010-10-21 00:39:11 +02:00
#if ARCH_ARM
#include "vpx_ports/arm.h"
#endif
Reduced the size of MB_MODE_INFO Moved partition_bmi and partition_count out of MB_MODE_INFO and placed into MACROBLOCK. Also reduced the size of other members of the MB_MODE_INFO struct. For 1080p, the memory was reduced by 1,209,516 bytes. The decoder performance appeared to improve by 3% for the clip used. Note: The main goal for this change is to improve the decoder performance. The encoder will be revisited at a later date for further structure cleanup. Change-Id: I4733621292ee9cc3fffa4046cb3fd4d99bd14613
2010-09-02 22:17:52 +02:00
Initial WebM release
2010-05-18 17:58:33 +02:00
#include <math.h>
#include <stdio.h>
#include <limits.h>
#if CONFIG_RUNTIME_CPU_DETECT
#define IF_RTCD(x) (x)
#define RTCD(x) &cpi->common.rtcd.x
#else
#define IF_RTCD(x) NULL
#define RTCD(x) NULL
#endif
extern void vp8cx_pick_filter_level_fast(YV12_BUFFER_CONFIG *sd, VP8_COMP *cpi);
extern void vp8cx_set_alt_lf_level(VP8_COMP *cpi, int filt_val);
extern void vp8cx_pick_filter_level(YV12_BUFFER_CONFIG *sd, VP8_COMP *cpi);
extern void vp8_dmachine_specific_config(VP8_COMP *cpi);
extern void vp8_cmachine_specific_config(VP8_COMP *cpi);
extern void vp8_deblock_frame(YV12_BUFFER_CONFIG *source, YV12_BUFFER_CONFIG *post, int filt_lvl, int low_var_thresh, int flag);
extern void print_parms(VP8_CONFIG *ocf, char *filenam);
extern unsigned int vp8_get_processor_freq();
extern void print_tree_update_probs();
extern void vp8cx_create_encoder_threads(VP8_COMP *cpi);
extern void vp8cx_remove_encoder_threads(VP8_COMP *cpi);
#if HAVE_ARMV7
extern void vp8_yv12_copy_frame_func_neon(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc);
extern void vp8_yv12_copy_src_frame_func_neon(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc);
#endif
int vp8_calc_ss_err(YV12_BUFFER_CONFIG *source, YV12_BUFFER_CONFIG *dest, const vp8_variance_rtcd_vtable_t *rtcd);
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
extern void vp8_temporal_filter_prepare_c(VP8_COMP *cpi, int distance);
Initial WebM release
2010-05-18 17:58:33 +02:00
Rename mode_ref_lf_test_function This function graduated from being a test func to something that's on by default. Rename it and remove some spurious comments that confuse its status. Change-Id: I689695a3ad29c35e9a72a43ec93766733ac6c20b
2010-09-29 19:53:08 +02:00
static void set_default_lf_deltas(VP8_COMP *cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
extern const int vp8_gf_interval_table[101];
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
#if CONFIG_ENHANCED_INTERP
Turning off filter search for now to improve encode speed. Change-Id: I87291fb40c745f34c36b067f47abdf69774a812f
2012-04-30 21:24:22 +02:00
#define SEARCH_BEST_FILTER 0 /* to search exhaustively for best filter */
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
#define RESET_FOREACH_FILTER 0 /* whether to reset the encoder state
before trying each new filter */
#endif
Initial refactoring of high_precision mv code. This is the first patch for refactoring of the code related to high-precision mv, so that 1/4 and 1/8 pel motion vectors can co-exist in the same bit-stream by use of a frame level flag. The current patch works fine for only use of 1/4th and only use of 1/8th pel mv, but there are some issues with the mode switching in between. Subsequent patches on this change Id will fix the remaining issues. Patch 2: Adds fixes to make sure that multiple mv precisions can co-exist in the bit-stream. Frame level switching has been tested to work correctly. Patch 3: Fixes lines exceeding 80 char Patch 4: http://www.corp.google.com/~debargha/vp8_results/enhinterp.html Results on derf after ssse3 bugfix, compared to everything enabled but the 8-tap, 1/8-subpel and 1/16-subpel uv. Overall the gains are about 3% now. Hopefully there are no more bugs lingering. Apparently the sse3 bug affected the quartel subpel results more than the eighth pel ones (which is understandabale because one bad predictor due to the bug, matters less if there are a lot more subpel options available as in the 1/8 subpel case). The results in the 4th column correspond to the current settings. The first two columns correspond to two settings of adaptive switching of the 1/4 or 1/8 subpel mode based on initial Q estimate. These do not work as good as just using 1/8 all the time yet. Change-Id: I3ef392ad338329f4d68a85257a49f2b14f3af472
2012-02-27 19:22:38 +01:00
#if CONFIG_HIGH_PRECISION_MV
#define ALTREF_HIGH_PRECISION_MV 1 /* whether to use high precision mv for altref computation */
#define HIGH_PRECISION_MV_QTHRESH 200 /* Q threshold for use of high precision mv */
/* Choose a very high value for now so
* that HIGH_PRECISION is always chosen
*/
#endif
changed configure option name to reduce confusion Renamed configure option "enable-psnr" to "enable-internal-stats" to better reflect the purpose of the option and eliminate the confusion reported in http://code.google.com/p/webm/issues/detail?id=35 Change-Id: If72df6fdb9f1e33dab1329240ba4d8911d2f1f7a
2011-04-29 18:37:59 +02:00
#if CONFIG_INTERNAL_STATS
Initial WebM release
2010-05-18 17:58:33 +02:00
#include "math.h"
extern double vp8_calc_ssim
(
YV12_BUFFER_CONFIG *source,
YV12_BUFFER_CONFIG *dest,
int lumamask,
vp8cx- alternate ssim function with optimizations Change-Id: I91921b0a90dbaddc7010380b038955be347964b3
2011-03-08 15:05:18 +01:00
double *weight,
const vp8_variance_rtcd_vtable_t *rtcd
Initial WebM release
2010-05-18 17:58:33 +02:00
);
vp8cx- alternate ssim function with optimizations Change-Id: I91921b0a90dbaddc7010380b038955be347964b3
2011-03-08 15:05:18 +01:00
Initial WebM release
2010-05-18 17:58:33 +02:00
extern double vp8_calc_ssimg
(
YV12_BUFFER_CONFIG *source,
YV12_BUFFER_CONFIG *dest,
double *ssim_y,
double *ssim_u,
change to use fast ssim code for internal ssim calculations The commit also removed the slow ssim calculation that uses a 7x7 kernel, and revised the comments to better describe how sample ssim values are computed and averaged Change-Id: I1d874073cddca00f3c997f4b9a9a3db0aa212276
2011-05-03 00:27:14 +02:00
double *ssim_v,
const vp8_variance_rtcd_vtable_t *rtcd
Initial WebM release
2010-05-18 17:58:33 +02:00
);
#endif
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
//#define OUTPUT_YUV_REC
Initial WebM release
2010-05-18 17:58:33 +02:00
#ifdef OUTPUT_YUV_SRC
FILE *yuv_file;
#endif
Add 8x8 transform to experimental branch Please refer to previous commit messages for detailed info: https://on2-git.corp.google.com/g/#change,5940 https://on2-git.corp.google.com/g/#change,6045 Change-Id: I8b16992f2f69c5a808ad40a3e32ef589cce7c59d
2011-02-14 23:18:18 +01:00
#ifdef OUTPUT_YUV_REC
FILE *yuv_rec_file;
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
#if 0
FILE *framepsnr;
fix more merge issues With this fix, the experimental branch now builds and encodes correctly with the following two configure options respectively: --enable-experimental --enable-t8x8 --enable-experimental Change-Id: I3147c33c503fe713a85fd371e4f1a974805778bf
2011-07-20 23:21:24 +02:00
FILE *kf_list;
Initial WebM release
2010-05-18 17:58:33 +02:00
FILE *keyfile;
#endif
#if 0
extern int skip_true_count;
extern int skip_false_count;
#endif
#ifdef ENTROPY_STATS
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
extern int intra_mode_stats[VP8_BINTRAMODES][VP8_BINTRAMODES][VP8_BINTRAMODES];
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
#ifdef SPEEDSTATS
unsigned int frames_at_speed[16] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
unsigned int tot_pm = 0;
unsigned int cnt_pm = 0;
unsigned int tot_ef = 0;
unsigned int cnt_ef = 0;
#endif
Add 8x8 transform to experimental branch Please refer to previous commit messages for detailed info: https://on2-git.corp.google.com/g/#change,5940 https://on2-git.corp.google.com/g/#change,6045 Change-Id: I8b16992f2f69c5a808ad40a3e32ef589cce7c59d
2011-02-14 23:18:18 +01:00
#if defined(SECTIONBITS_OUTPUT)
extern unsigned __int64 Sectionbits[500];
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
#ifdef MODE_STATS
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
extern INT64 Sectionbits[500];
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
extern unsigned int y_modes[VP8_YMODES] ;
extern unsigned int i8x8_modes[VP8_I8X8_MODES];
extern unsigned int uv_modes[VP8_UV_MODES] ;
extern unsigned int uv_modes_y[VP8_YMODES][VP8_UV_MODES];
extern unsigned int b_modes[B_MODE_COUNT];
extern unsigned int inter_y_modes[MB_MODE_COUNT] ;
extern unsigned int inter_uv_modes[VP8_UV_MODES] ;
fixed a crash when MODE_STATS is enabled The MODE_STATS macro was used to #ifdef around code for mode entropy stats collection, this commit fixed a crash when MODE_STATS is on. The commit also changed a number of array definitions to use defined macros instead of hard-coded numbers. Change-Id: I114592f53a1e44e31e455f5725f036ae6168735a
2011-12-06 16:37:39 +01:00
extern unsigned int inter_b_modes[B_MODE_COUNT];
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
extern void (*vp8_short_fdct4x4)(short *input, short *output, int pitch);
extern void (*vp8_short_fdct8x4)(short *input, short *output, int pitch);
extern void vp8cx_init_quantizer(VP8_COMP *cpi);
Further QIndex realted Fixes: Added code to support 256 index steps instead of 128 but disabled for now. Replace hard wired table vp8cx_base_skip_false_prob[128] Observed Qindex problem with setting minimum loop filter value. (Experiment code using real Q in place but for now just returning 0. This has a big beneficial effect on some clips, particularly waterfall which shows 5% ssim gain) Change-Id: I2f7117de8adc1797164c106aa13effc900a1467e
2011-12-12 19:27:25 +01:00
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#if CONFIG_NEWENTROPY
int vp8cx_base_skip_false_prob[QINDEX_RANGE][3];
#else
Further QIndex realted Fixes: Added code to support 256 index steps instead of 128 but disabled for now. Replace hard wired table vp8cx_base_skip_false_prob[128] Observed Qindex problem with setting minimum loop filter value. (Experiment code using real Q in place but for now just returning 0. This has a big beneficial effect on some clips, particularly waterfall which shows 5% ssim gain) Change-Id: I2f7117de8adc1797164c106aa13effc900a1467e
2011-12-12 19:27:25 +01:00
int vp8cx_base_skip_false_prob[QINDEX_RANGE];
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#endif
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
Control of active min quantizer for two pass. Create look up tables for controlling the active quantizer range. Some initial tuning to improve quality circa 0.5% on test set. Clean up of some stats output code Change-Id: Ia698a8525f8b8129a503cadace3ee73fe888f543
2010-09-29 13:03:19 +02:00
// Tables relating active max Q to active min 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
static int kf_low_motion_minq[QINDEX_RANGE];
static int kf_high_motion_minq[QINDEX_RANGE];
static int gf_low_motion_minq[QINDEX_RANGE];
static int gf_high_motion_minq[QINDEX_RANGE];
static int inter_minq[QINDEX_RANGE];
// Functions to compute the active minq lookup table entries based on a
// formulaic approach to facilitate easier adjustment of the Q tables.
// The formulae were derived from computing a 3rd order polynomial best
// fit to the original data (after plotting real maxq vs minq (not q index))
int calculate_minq_index( double maxq,
double x3, double x2, double x, double c )
Control of active min quantizer for two pass. Create look up tables for controlling the active quantizer range. Some initial tuning to improve quality circa 0.5% on test set. Clean up of some stats output code Change-Id: Ia698a8525f8b8129a503cadace3ee73fe888f543
2010-09-29 13:03:19 +02:00
{
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
int i;
double minqtarget;
double thisq;
minqtarget = ( (x3 * maxq * maxq * maxq) +
(x2 * maxq * maxq) +
(x * maxq) +
c );
if ( minqtarget > maxq )
Correctly capped minqtarget to maxq This line of code incorrectly set maxq = maxq rather than capping minqtarget. Change-Id: Ifbc86df8b0ff2779e7b2a5f7349724d04a18bd62
2012-01-31 21:58:34 +01:00
minqtarget = maxq;
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
for ( i = 0; i < QINDEX_RANGE; i++ )
{
thisq = vp8_convert_qindex_to_q(i);
if ( minqtarget <= vp8_convert_qindex_to_q(i) )
return i;
}
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
return QINDEX_RANGE-1;
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
}
void init_minq_luts()
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
{
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
int i;
double maxq;
for ( i = 0; i < QINDEX_RANGE; i++ )
{
maxq = vp8_convert_qindex_to_q(i);
MinQ equations. Slight tweaks to the new minq equations to bring results more into line with original lookup tables. Change-Id: I969fc87d95912df549b6775e83ee2345e84d4da0
2011-11-28 12:48:07 +01:00
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
kf_low_motion_minq[i] = calculate_minq_index( maxq,
MinQ equations. Slight tweaks to the new minq equations to bring results more into line with original lookup tables. Change-Id: I969fc87d95912df549b6775e83ee2345e84d4da0
2011-11-28 12:48:07 +01:00
0.0000003,
-0.000015,
0.074,
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
0.0 );
kf_high_motion_minq[i] = calculate_minq_index( maxq,
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
0.0000004,
MinQ equations. Slight tweaks to the new minq equations to bring results more into line with original lookup tables. Change-Id: I969fc87d95912df549b6775e83ee2345e84d4da0
2011-11-28 12:48:07 +01:00
-0.000125,
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
0.14,
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
0.0 );
gf_low_motion_minq[i] = calculate_minq_index( maxq,
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
0.0000015,
-0.0009,
0.33,
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
0.0 );
gf_high_motion_minq[i] = calculate_minq_index( maxq,
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
0.0000021,
-0.00125,
0.45,
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
0.0 );
inter_minq[i] = calculate_minq_index( maxq,
0.00000271,
-0.00113,
0.697,
0.0 );
}
}
Further QIndex realted Fixes: Added code to support 256 index steps instead of 128 but disabled for now. Replace hard wired table vp8cx_base_skip_false_prob[128] Observed Qindex problem with setting minimum loop filter value. (Experiment code using real Q in place but for now just returning 0. This has a big beneficial effect on some clips, particularly waterfall which shows 5% ssim gain) Change-Id: I2f7117de8adc1797164c106aa13effc900a1467e
2011-12-12 19:27:25 +01:00
void init_base_skip_probs()
{
int i;
double q;
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
int skip_prob, t;
Further QIndex realted Fixes: Added code to support 256 index steps instead of 128 but disabled for now. Replace hard wired table vp8cx_base_skip_false_prob[128] Observed Qindex problem with setting minimum loop filter value. (Experiment code using real Q in place but for now just returning 0. This has a big beneficial effect on some clips, particularly waterfall which shows 5% ssim gain) Change-Id: I2f7117de8adc1797164c106aa13effc900a1467e
2011-12-12 19:27:25 +01:00
for ( i = 0; i < QINDEX_RANGE; i++ )
{
q = vp8_convert_qindex_to_q(i);
// Exponential decay caluclation of baseline skip prob with clamping
// Based on crude best fit of old table.
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
t = (int)( 564.25 * pow( 2.71828, (-0.012*q) ) );
skip_prob = t;
if ( skip_prob < 1 )
skip_prob = 1;
else if ( skip_prob > 255 )
skip_prob = 255;
#if CONFIG_NEWENTROPY
vp8cx_base_skip_false_prob[i][1] = skip_prob;
skip_prob = t * 0.75;
Further QIndex realted Fixes: Added code to support 256 index steps instead of 128 but disabled for now. Replace hard wired table vp8cx_base_skip_false_prob[128] Observed Qindex problem with setting minimum loop filter value. (Experiment code using real Q in place but for now just returning 0. This has a big beneficial effect on some clips, particularly waterfall which shows 5% ssim gain) Change-Id: I2f7117de8adc1797164c106aa13effc900a1467e
2011-12-12 19:27:25 +01:00
if ( skip_prob < 1 )
skip_prob = 1;
else if ( skip_prob > 255 )
skip_prob = 255;
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
vp8cx_base_skip_false_prob[i][2] = skip_prob;
Further QIndex realted Fixes: Added code to support 256 index steps instead of 128 but disabled for now. Replace hard wired table vp8cx_base_skip_false_prob[128] Observed Qindex problem with setting minimum loop filter value. (Experiment code using real Q in place but for now just returning 0. This has a big beneficial effect on some clips, particularly waterfall which shows 5% ssim gain) Change-Id: I2f7117de8adc1797164c106aa13effc900a1467e
2011-12-12 19:27:25 +01:00
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
skip_prob = t * 1.25;
if ( skip_prob < 1 )
skip_prob = 1;
else if ( skip_prob > 255 )
skip_prob = 255;
vp8cx_base_skip_false_prob[i][0] = skip_prob;
#else
Further QIndex realted Fixes: Added code to support 256 index steps instead of 128 but disabled for now. Replace hard wired table vp8cx_base_skip_false_prob[128] Observed Qindex problem with setting minimum loop filter value. (Experiment code using real Q in place but for now just returning 0. This has a big beneficial effect on some clips, particularly waterfall which shows 5% ssim gain) Change-Id: I2f7117de8adc1797164c106aa13effc900a1467e
2011-12-12 19:27:25 +01:00
vp8cx_base_skip_false_prob[i] = skip_prob;
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#endif
Further QIndex realted Fixes: Added code to support 256 index steps instead of 128 but disabled for now. Replace hard wired table vp8cx_base_skip_false_prob[128] Observed Qindex problem with setting minimum loop filter value. (Experiment code using real Q in place but for now just returning 0. This has a big beneficial effect on some clips, particularly waterfall which shows 5% ssim gain) Change-Id: I2f7117de8adc1797164c106aa13effc900a1467e
2011-12-12 19:27:25 +01:00
}
}
Changes to costing of skip. Update the costing of skip in the recode loop and rd code. Change-Id: I2e5ebbd7ddf201212b32441321e12626cd0423e9
2012-04-11 15:37:48 +02:00
void update_base_skip_probs(VP8_COMP *cpi)
{
VP8_COMMON *cm = &cpi->common;
if (cm->frame_type != KEY_FRAME)
{
update_skip_probs(cpi);
if (cm->refresh_alt_ref_frame)
{
#if CONFIG_NEWENTROPY
int k;
for (k=0; k<MBSKIP_CONTEXTS; ++k)
cpi->last_skip_false_probs[2][k] = cm->mbskip_pred_probs[k];
#else
cpi->last_skip_false_probs[2] = cpi->prob_skip_false;
#endif
cpi->last_skip_probs_q[2] = cm->base_qindex;
}
else if (cpi->common.refresh_golden_frame)
{
#if CONFIG_NEWENTROPY
int k;
for (k=0; k<MBSKIP_CONTEXTS; ++k)
cpi->last_skip_false_probs[1][k] = cm->mbskip_pred_probs[k];
#else
cpi->last_skip_false_probs[1] = cpi->prob_skip_false;
#endif
cpi->last_skip_probs_q[1] = cm->base_qindex;
}
else
{
#if CONFIG_NEWENTROPY
int k;
for (k=0; k<MBSKIP_CONTEXTS; ++k)
cpi->last_skip_false_probs[0][k] = cm->mbskip_pred_probs[k];
#else
cpi->last_skip_false_probs[0] = cpi->prob_skip_false;
#endif
cpi->last_skip_probs_q[0] = cm->base_qindex;
// update the baseline table for the current q
#if CONFIG_NEWENTROPY
for (k=0; k<MBSKIP_CONTEXTS; ++k)
cpi->base_skip_false_prob[cm->base_qindex][k] =
cm->mbskip_pred_probs[k];
#else
cpi->base_skip_false_prob[cm->base_qindex] = cpi->prob_skip_false;
#endif
}
}
}
Further QIndex realted Fixes: Added code to support 256 index steps instead of 128 but disabled for now. Replace hard wired table vp8cx_base_skip_false_prob[128] Observed Qindex problem with setting minimum loop filter value. (Experiment code using real Q in place but for now just returning 0. This has a big beneficial effect on some clips, particularly waterfall which shows 5% ssim gain) Change-Id: I2f7117de8adc1797164c106aa13effc900a1467e
2011-12-12 19:27:25 +01:00
Initial WebM release
2010-05-18 17:58:33 +02:00
void vp8_initialize()
{
static int init_done = 0;
if (!init_done)
{
vp8_scale_machine_specific_config();
vp8_initialize_common();
//vp8_dmachine_specific_config();
vp8_tokenize_initialize();
Further work on extended Q range. Fixed some further QIndex related issues and replaced some tables (eg zbin and rounding) Also Added function (currently disabled by default) to populate the main AC and DC quantizer tables. Using the original AC range the resulting computed DC values give behavior broadly comparable on the DERF set. That is not to say that the equations will hold good over a more extended range. The purpose of this code is to make it easier to experiment with further alterations to the Q range and distribution of Q values plus the relative weights given to AC and DC. The function find_fp_qindex() ensures that changes to the Q tables are reflected in the value passed in to the first pass code. Slight experimental adjustment to static segment Q offset. Change-Id: I36186267d55dfc2a3d565d0cff7218ef300d1cd5
2011-12-02 15:57:21 +01:00
vp8_init_quant_tables();
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
vp8_init_me_luts();
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
init_minq_luts();
Further QIndex realted Fixes: Added code to support 256 index steps instead of 128 but disabled for now. Replace hard wired table vp8cx_base_skip_false_prob[128] Observed Qindex problem with setting minimum loop filter value. (Experiment code using real Q in place but for now just returning 0. This has a big beneficial effect on some clips, particularly waterfall which shows 5% ssim gain) Change-Id: I2f7117de8adc1797164c106aa13effc900a1467e
2011-12-12 19:27:25 +01:00
init_base_skip_probs();
Initial WebM release
2010-05-18 17:58:33 +02:00
init_done = 1;
}
}
#ifdef PACKET_TESTING
extern FILE *vpxlogc;
#endif
static void setup_features(VP8_COMP *cpi)
{
Segment Feature Signaling Plumbing for tuning new segment features on and off. Change-Id: If86cd6f103296b73030e8af7cf85c5b9bbffdbaf
2011-09-14 19:20:25 +02:00
MACROBLOCKD *xd = &cpi->mb.e_mbd;
Initial WebM release
2010-05-18 17:58:33 +02:00
// Set up default state for MB feature flags
Merging and testing of SEGMENTATION experiment. Removed code in #if CONFIG_SEGMENTATION that enables segmentation and creates a test segmentation map, to avoid conflicts with the other segmentation test code, Change-Id: I7a21a44ed188b814cd80b30dd628c62474eba730
2011-11-08 16:40:32 +01:00
xd->segmentation_enabled = 0; // Default segmentation disabled
Segment Feature Signaling Plumbing for tuning new segment features on and off. Change-Id: If86cd6f103296b73030e8af7cf85c5b9bbffdbaf
2011-09-14 19:20:25 +02:00
xd->update_mb_segmentation_map = 0;
xd->update_mb_segmentation_data = 0;
vpx_memset(xd->mb_segment_tree_probs, 255, sizeof(xd->mb_segment_tree_probs));
Segment Feature Data Access No change to functionality or output. Updates to the segment feature data structure now all done through functions such as set_segdata() and get_segdata() in seg_common.c. The reason for this is to make changing the structures (if needed) and debug easier. In addition it provides a single location for subsequent addition of range and validity checks. For example valid combination of mode and reference frame. Change-Id: I2e866505562db4e4cb6f17a472b25b4465f01add
2011-11-03 17:58:26 +01:00
clearall_segfeatures( xd );
Segment Feature Signaling Plumbing for tuning new segment features on and off. Change-Id: If86cd6f103296b73030e8af7cf85c5b9bbffdbaf
2011-09-14 19:20:25 +02:00
xd->mode_ref_lf_delta_enabled = 0;
xd->mode_ref_lf_delta_update = 0;
vpx_memset(xd->ref_lf_deltas, 0, sizeof(xd->ref_lf_deltas));
vpx_memset(xd->mode_lf_deltas, 0, sizeof(xd->mode_lf_deltas));
vpx_memset(xd->last_ref_lf_deltas, 0, sizeof(xd->ref_lf_deltas));
vpx_memset(xd->last_mode_lf_deltas, 0, sizeof(xd->mode_lf_deltas));
Initial WebM release
2010-05-18 17:58:33 +02:00
Rename mode_ref_lf_test_function This function graduated from being a test func to something that's on by default. Rename it and remove some spurious comments that confuse its status. Change-Id: I689695a3ad29c35e9a72a43ec93766733ac6c20b
2010-09-29 19:53:08 +02:00
set_default_lf_deltas(cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Increase static linkage, remove unused functions A large number of functions were defined with external linkage, even though they were only used from within one file. This patch changes their linkage to static and removes the vp8_ prefix from their names, which should make it more obvious to the reader that the function is contained within the current translation unit. Functions that were not referenced were removed. These symbols were identified by: $ nm -A libvpx.a | sort -k3 | uniq -c -f2 | grep ' [A-Z] ' \ | sort | grep '^ *1 ' Change-Id: I59609f58ab65312012c047036ae1e0634f795779
2011-03-17 22:07:59 +01:00
static void dealloc_compressor_data(VP8_COMP *cpi)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Remove temporal alt ref from realtime only build It is not used in realtime mode. Reduces memory footprint. Change-Id: I7f163225762368df5457cfd413050161d3704a3f
2011-02-22 09:29:23 +01:00
vpx_free(cpi->tplist);
Put more code under #if CONFIG_MULTITHREAD. Change-Id: Icf4b692099d7d249fe3553852b1022b027b28e4b
2011-02-05 06:29:25 +01:00
cpi->tplist = NULL;
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
// Delete last frame MV storage buffers
Remove temporal alt ref from realtime only build It is not used in realtime mode. Reduces memory footprint. Change-Id: I7f163225762368df5457cfd413050161d3704a3f
2011-02-22 09:29:23 +01:00
vpx_free(cpi->lfmv);
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
cpi->lfmv = 0;
Remove temporal alt ref from realtime only build It is not used in realtime mode. Reduces memory footprint. Change-Id: I7f163225762368df5457cfd413050161d3704a3f
2011-02-22 09:29:23 +01:00
vpx_free(cpi->lf_ref_frame_sign_bias);
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
cpi->lf_ref_frame_sign_bias = 0;
Remove temporal alt ref from realtime only build It is not used in realtime mode. Reduces memory footprint. Change-Id: I7f163225762368df5457cfd413050161d3704a3f
2011-02-22 09:29:23 +01:00
vpx_free(cpi->lf_ref_frame);
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
cpi->lf_ref_frame = 0;
Initial WebM release
2010-05-18 17:58:33 +02:00
// Delete sementation map
Remove temporal alt ref from realtime only build It is not used in realtime mode. Reduces memory footprint. Change-Id: I7f163225762368df5457cfd413050161d3704a3f
2011-02-22 09:29:23 +01:00
vpx_free(cpi->segmentation_map);
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->segmentation_map = 0;
Moved some segmentation data structures. Moved some segmentation data structures into VP8_COMMON Change-Id: I59c6e2edf7a0176e35319936eea450027aeb3b39
2012-01-27 19:29:07 +01:00
vpx_free(cpi->common.last_frame_seg_map);
cpi->common.last_frame_seg_map = 0;
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
vpx_free(cpi->coding_context.last_frame_seg_map_copy);
cpi->coding_context.last_frame_seg_map_copy = 0;
Initial WebM release
2010-05-18 17:58:33 +02:00
Remove temporal alt ref from realtime only build It is not used in realtime mode. Reduces memory footprint. Change-Id: I7f163225762368df5457cfd413050161d3704a3f
2011-02-22 09:29:23 +01:00
vpx_free(cpi->active_map);
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->active_map = 0;
vp8_de_alloc_frame_buffers(&cpi->common);
vp8_yv12_de_alloc_frame_buffer(&cpi->last_frame_uf);
vp8_yv12_de_alloc_frame_buffer(&cpi->scaled_source);
#if VP8_TEMPORAL_ALT_REF
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
vp8_yv12_de_alloc_frame_buffer(&cpi->alt_ref_buffer);
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
vp8_lookahead_destroy(cpi->lookahead);
Initial WebM release
2010-05-18 17:58:33 +02:00
vpx_free(cpi->tok);
cpi->tok = 0;
Fix encoder real-time only configuration. Remove allocation/deallocation of stats storage. Remove full search functions in machine specific encoder inits. Remove last pass validation in validate_config. Change-Id: I7f29be69273981a4fef6e80ecdb6217c68cbad4e
2011-01-10 10:14:10 +01:00
// Structure used to monitor GF usage
Remove temporal alt ref from realtime only build It is not used in realtime mode. Reduces memory footprint. Change-Id: I7f163225762368df5457cfd413050161d3704a3f
2011-02-22 09:29:23 +01:00
vpx_free(cpi->gf_active_flags);
Moved gf_active code to encoder only The gf_active code is only used by the encoder, so it was moved from common and decoder. Change-Id: Iada15acd5b2b33ff70c34668ca87d4cfd0d05025
2010-08-11 17:02:31 +02:00
cpi->gf_active_flags = 0;
Restructure of activity masking code. This commit restructures the mb activity masking code to better facilitate experimentation using different metrics etc. and also allows for adjustment of the zero bin either for encode only or both the encode and mode selection stages It also uses information from the current frame rather than the previous frame and the default strength has been reduced. Change-Id: Id39b19eace37574dc429f25aae810c203709629b
2011-05-12 18:01:55 +02:00
// Activity mask based per mb zbin adjustments
vpx_free(cpi->mb_activity_map);
cpi->mb_activity_map = 0;
vpx_free(cpi->mb_norm_activity_map);
cpi->mb_norm_activity_map = 0;
Remove temporal alt ref from realtime only build It is not used in realtime mode. Reduces memory footprint. Change-Id: I7f163225762368df5457cfd413050161d3704a3f
2011-02-22 09:29:23 +01:00
vpx_free(cpi->mb.pip);
Reduced the size of MB_MODE_INFO Moved partition_bmi and partition_count out of MB_MODE_INFO and placed into MACROBLOCK. Also reduced the size of other members of the MB_MODE_INFO struct. For 1080p, the memory was reduced by 1,209,516 bytes. The decoder performance appeared to improve by 3% for the clip used. Note: The main goal for this change is to improve the decoder performance. The encoder will be revisited at a later date for further structure cleanup. Change-Id: I4733621292ee9cc3fffa4046cb3fd4d99bd14613
2010-09-02 22:17:52 +02:00
cpi->mb.pip = 0;
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
vpx_free(cpi->twopass.total_stats);
cpi->twopass.total_stats = 0;
Fix a memory leak problem in encoder Deallocating the buffers before re-allocating them. The fix passed James Berry's test program for memory leak check. Change-Id: I18c3cf665412c0e313a523e3d435106c03ca438d
2010-12-06 23:21:37 +01:00
Two pass rate control code changes. This comitt brings accross changes from the public branch commit number Icf74d13af77437c08602571dc7a97e747cce5066. The main puurpose of this comit relates to CQ mode but it also includes some refactoring of the two pass code which I hope will make tuning the experimental branch for the new quantizer range a little less painfull. Change-Id: I278e989436a928fc1fe7761068960048f9d7a376
2011-11-23 18:18:31 +01:00
vpx_free(cpi->twopass.total_left_stats);
cpi->twopass.total_left_stats = 0;
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
vpx_free(cpi->twopass.this_frame_stats);
cpi->twopass.this_frame_stats = 0;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Some further QIndex issues with extended Q Resolved or factored out some further issues with Q index. Put in a 3rd order polynomial instead of less accurate power function as the best fit on gf and kf boost adjustment. Added avg_q value to use instead of ni_av_qi. Compute segment delta Q values based on avg_q. Fixed bug in adjust_maxq_qrange(). The extended range Q on the derf set, using standard data rates (which do not extend high enough to get big benefits) still show a shortfall of between 0.5 and 1% though so there would appear to be further issues that need to be tracked down. Change-Id: Icfd49b9f401906ba487ef1bef7d397048295d959
2011-12-06 15:48:52 +01:00
// Computes a q delta (in "q index" terms) to get from a starting q value
// to a target value
// target q value
static int compute_qdelta( VP8_COMP *cpi, double qstart, double qtarget )
{
int i;
int start_index = cpi->worst_quality;
int target_index = cpi->worst_quality;
// Convert the average q value to an index.
for ( i = cpi->best_quality; i < cpi->worst_quality; i++ )
{
start_index = i;
if ( vp8_convert_qindex_to_q(i) >= qstart )
break;
}
// Convert the q target to an index
for ( i = cpi->best_quality; i < cpi->worst_quality; i++ )
{
target_index = i;
if ( vp8_convert_qindex_to_q(i) >= qtarget )
break;
}
return target_index - start_index;
}
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
static void init_seg_features(VP8_COMP *cpi)
{
VP8_COMMON *cm = &cpi->common;
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
MACROBLOCKD *xd = &cpi->mb.e_mbd;
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
Reference frame prediction: Extended prediction and coding of reference frame where a subset of options are flagged as available at the segment level. Updated copyright notices. Switch to SAD in mbgraph code as SATD problematic for the foreground and background separation as it can ignore large DC shifts. Change-Id: I661dbbb2f94f3ec0f96bb928c1655e5e415a7de1
2012-01-31 13:45:30 +01:00
int high_q = (int)(cpi->avg_q > 48.0);
Some further QIndex issues with extended Q Resolved or factored out some further issues with Q index. Put in a 3rd order polynomial instead of less accurate power function as the best fit on gf and kf boost adjustment. Added avg_q value to use instead of ni_av_qi. Compute segment delta Q values based on avg_q. Fixed bug in adjust_maxq_qrange(). The extended range Q on the derf set, using standard data rates (which do not extend high enough to get big benefits) still show a shortfall of between 0.5 and 1% though so there would appear to be further issues that need to be tracked down. Change-Id: Icfd49b9f401906ba487ef1bef7d397048295d959
2011-12-06 15:48:52 +01:00
int qi_delta;
Segmentation Entropy and tweaks. Some correction for entropy impact of segment signaled (EOB and ref frame) Other slight tweaks. Derf VBR average gain now over 1% (best over 7%) One YT test clip has gains of circa 30% (VBR) There is still an issue with noisy clips where making the background static and coded with 0,0 can have a negative effect, especially at low Q. This is probably because of the loss of smoothing by fractional pixel filters. Change-Id: I7a225613c98067b96f8fc7a7e36f95d465b2b834
2011-10-28 16:27:23 +02:00
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
// Disable and clear down for KF
if ( cm->frame_type == KEY_FRAME )
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
{
// Clear down the global segmentation map
vpx_memset( cpi->segmentation_map, 0, (cm->mb_rows * cm->mb_cols));
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
xd->update_mb_segmentation_map = 0;
xd->update_mb_segmentation_data = 0;
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
cpi->static_mb_pct = 0;
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
// Disable segmentation
vp8_disable_segmentation((VP8_PTR)cpi);
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
// Clear down the segment features.
clearall_segfeatures(xd);
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
}
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
// If this is an alt ref frame
else if ( cm->refresh_alt_ref_frame )
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
{
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
// Clear down the global segmentation map
vpx_memset( cpi->segmentation_map, 0, (cm->mb_rows * cm->mb_cols));
xd->update_mb_segmentation_map = 0;
xd->update_mb_segmentation_data = 0;
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
cpi->static_mb_pct = 0;
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
// Disable segmentation and individual segment features by default
vp8_disable_segmentation((VP8_PTR)cpi);
clearall_segfeatures(xd);
// Scan frames from current to arf frame.
// This function re-enables segmentation if appropriate.
vp8_update_mbgraph_stats(cpi);
// If segmentation was enabled set those features needed for the
// arf itself.
if ( xd->segmentation_enabled )
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
{
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
xd->update_mb_segmentation_map = 1;
xd->update_mb_segmentation_data = 1;
Some further QIndex issues with extended Q Resolved or factored out some further issues with Q index. Put in a 3rd order polynomial instead of less accurate power function as the best fit on gf and kf boost adjustment. Added avg_q value to use instead of ni_av_qi. Compute segment delta Q values based on avg_q. Fixed bug in adjust_maxq_qrange(). The extended range Q on the derf set, using standard data rates (which do not extend high enough to get big benefits) still show a shortfall of between 0.5 and 1% though so there would appear to be further issues that need to be tracked down. Change-Id: Icfd49b9f401906ba487ef1bef7d397048295d959
2011-12-06 15:48:52 +01:00
qi_delta = compute_qdelta( cpi, cpi->avg_q, (cpi->avg_q * 0.875) );
set_segdata( xd, 1, SEG_LVL_ALT_Q, (qi_delta - 2) );
Segment Feature Data Access No change to functionality or output. Updates to the segment feature data structure now all done through functions such as set_segdata() and get_segdata() in seg_common.c. The reason for this is to make changing the structures (if needed) and debug easier. In addition it provides a single location for subsequent addition of range and validity checks. For example valid combination of mode and reference frame. Change-Id: I2e866505562db4e4cb6f17a472b25b4465f01add
2011-11-03 17:58:26 +01:00
set_segdata( xd, 1, SEG_LVL_ALT_LF, -2 );
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
enable_segfeature(xd, 1, SEG_LVL_ALT_Q);
enable_segfeature(xd, 1, SEG_LVL_ALT_LF);
// Where relevant assume segment data is delta data
Corrected spelling Apparently the correct spelling of segement is segment ! Change-Id: I88593ee0523f251b3a96794c6166ef8c7898a029
2012-02-27 22:42:36 +01:00
xd->mb_segment_abs_delta = SEGMENT_DELTADATA;
T8x8 experiment merge. For ease of testing and merging experiments I have removed in line code in encode_frame() that assigns MBs to be t8x8 or t4x4 coded segments and have moved the decision point and segment setup to the init_seg_features0 test function. Keeping everything in one place helps make sure for now that experiments using segmentation are not fighting each other. Also made sure mode selection code can't choose 4x4 modes if t8x8 is selected. Patch2: In init_seg_features() add checks for SEG_LVL_TRANSFORM active. Change-Id: Ia1767edd99b78510011d4251539f9bc325842e3a
2011-11-09 15:05:28 +01:00
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
}
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
}
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
// All other frames if segmentation has been enabled
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
else if ( xd->segmentation_enabled )
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
{
Comment out segref segmentation filter changes. Commented out changes from earlier checking: "Change Iab7f1eff: vpnext use segref segmentation filter" Which in its current state breaks the decoder. Change-Id: I9185098aeda8ce65310f338c4c9375f4a39005d3
2012-01-31 14:09:49 +01:00
/*
vpnext use segref segmentation filter Goes through set of ref frames used by each macroblock and sets seg_lvl_ref_frame flags accordingly.. http://www.corp.google.com/~jimbankoski/no_crawl/segref.html Change-Id: Iab7f1effd75a839b34eb310d7168692c8f105411
2012-01-26 19:04:34 +01:00
int i;
// clears prior frame seg lev refs
for (i = 0; i < MAX_MB_SEGMENTS; i++)
{
// only do it if the force drop the background stuff is off
if(!segfeature_active(xd, i, SEG_LVL_MODE))
{
disable_segfeature(xd,i,SEG_LVL_REF_FRAME);
set_segdata( xd,i, SEG_LVL_REF_FRAME, 0xffffff);
}
}
Comment out segref segmentation filter changes. Commented out changes from earlier checking: "Change Iab7f1eff: vpnext use segref segmentation filter" Which in its current state breaks the decoder. Change-Id: I9185098aeda8ce65310f338c4c9375f4a39005d3
2012-01-31 14:09:49 +01:00
*/
vpnext use segref segmentation filter Goes through set of ref frames used by each macroblock and sets seg_lvl_ref_frame flags accordingly.. http://www.corp.google.com/~jimbankoski/no_crawl/segref.html Change-Id: Iab7f1effd75a839b34eb310d7168692c8f105411
2012-01-26 19:04:34 +01:00
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
// First normal frame in a valid gf or alt ref group
if ( cpi->common.frames_since_golden == 0 )
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
{
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
// Set up segment features for normal frames in an af group
if ( cpi->source_alt_ref_active )
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
{
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
xd->update_mb_segmentation_map = 0;
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
xd->update_mb_segmentation_data = 1;
Corrected spelling Apparently the correct spelling of segement is segment ! Change-Id: I88593ee0523f251b3a96794c6166ef8c7898a029
2012-02-27 22:42:36 +01:00
xd->mb_segment_abs_delta = SEGMENT_DELTADATA;
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
Some further QIndex issues with extended Q Resolved or factored out some further issues with Q index. Put in a 3rd order polynomial instead of less accurate power function as the best fit on gf and kf boost adjustment. Added avg_q value to use instead of ni_av_qi. Compute segment delta Q values based on avg_q. Fixed bug in adjust_maxq_qrange(). The extended range Q on the derf set, using standard data rates (which do not extend high enough to get big benefits) still show a shortfall of between 0.5 and 1% though so there would appear to be further issues that need to be tracked down. Change-Id: Icfd49b9f401906ba487ef1bef7d397048295d959
2011-12-06 15:48:52 +01:00
qi_delta = compute_qdelta( cpi, cpi->avg_q,
(cpi->avg_q * 1.125) );
set_segdata( xd, 1, SEG_LVL_ALT_Q, (qi_delta + 2) );
set_segdata( xd, 1, SEG_LVL_ALT_Q, 0 );
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
enable_segfeature(xd, 1, SEG_LVL_ALT_Q);
Segment Feature Data Access No change to functionality or output. Updates to the segment feature data structure now all done through functions such as set_segdata() and get_segdata() in seg_common.c. The reason for this is to make changing the structures (if needed) and debug easier. In addition it provides a single location for subsequent addition of range and validity checks. For example valid combination of mode and reference frame. Change-Id: I2e866505562db4e4cb6f17a472b25b4465f01add
2011-11-03 17:58:26 +01:00
set_segdata( xd, 1, SEG_LVL_ALT_LF, -2 );
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
enable_segfeature(xd, 1, SEG_LVL_ALT_LF);
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
// Segment coding disabled for compred testing
if ( high_q || (cpi->static_mb_pct == 100) )
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
{
Reference frame prediction: Extended prediction and coding of reference frame where a subset of options are flagged as available at the segment level. Updated copyright notices. Switch to SAD in mbgraph code as SATD problematic for the foreground and background separation as it can ignore large DC shifts. Change-Id: I661dbbb2f94f3ec0f96bb928c1655e5e415a7de1
2012-01-31 13:45:30 +01:00
//set_segref(xd, 1, LAST_FRAME);
Segmentation: Reference frames Modify reference frame segmentation so that ONE or MORE reference frames may be marked as a available for a given segment. Fixed bugs relating to segment coding of INTRA and some INTER modes at the segment level. Modified Q boost for static areas based on ambient average Q. Strong results now on clips with significant static areas. (some data points in derf set as high as 9% and some static & slide show type content in YT set > 20%) Change-Id: Ia79f912efa84b977f35a23683ae3643251e24f0c
2011-11-02 14:30:10 +01:00
set_segref(xd, 1, ALTREF_FRAME);
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
enable_segfeature(xd, 1, SEG_LVL_REF_FRAME);
Segment Feature Data Access No change to functionality or output. Updates to the segment feature data structure now all done through functions such as set_segdata() and get_segdata() in seg_common.c. The reason for this is to make changing the structures (if needed) and debug easier. In addition it provides a single location for subsequent addition of range and validity checks. For example valid combination of mode and reference frame. Change-Id: I2e866505562db4e4cb6f17a472b25b4465f01add
2011-11-03 17:58:26 +01:00
set_segdata( xd, 1, SEG_LVL_MODE, ZEROMV );
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
enable_segfeature(xd, 1, SEG_LVL_MODE);
Segment Feature Data Access No change to functionality or output. Updates to the segment feature data structure now all done through functions such as set_segdata() and get_segdata() in seg_common.c. The reason for this is to make changing the structures (if needed) and debug easier. In addition it provides a single location for subsequent addition of range and validity checks. For example valid combination of mode and reference frame. Change-Id: I2e866505562db4e4cb6f17a472b25b4465f01add
2011-11-03 17:58:26 +01:00
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
// EOB segment coding not fixed for 8x8 yet
set_segdata( xd, 1, SEG_LVL_EOB, 0 );
enable_segfeature(xd, 1, SEG_LVL_EOB);
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
}
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
}
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
// Disable segmentation and clear down features if alt ref
// is not active for this group
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
else
{
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
vp8_disable_segmentation((VP8_PTR)cpi);
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
vpx_memset( cpi->segmentation_map, 0,
(cm->mb_rows * cm->mb_cols));
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
xd->update_mb_segmentation_map = 0;
xd->update_mb_segmentation_data = 0;
clearall_segfeatures(xd);
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
}
}
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
// Special case where we are coding over the top of a previous
// alt ref frame
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
// Segment coding disabled for compred testing
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
else if ( cpi->is_src_frame_alt_ref )
{
// Enable mode and ref frame features for segment 0 as well
enable_segfeature(xd, 0, SEG_LVL_REF_FRAME);
enable_segfeature(xd, 0, SEG_LVL_MODE);
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
enable_segfeature(xd, 1, SEG_LVL_REF_FRAME);
enable_segfeature(xd, 1, SEG_LVL_MODE);
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
Segmentation: Reference frames Modify reference frame segmentation so that ONE or MORE reference frames may be marked as a available for a given segment. Fixed bugs relating to segment coding of INTRA and some INTER modes at the segment level. Modified Q boost for static areas based on ambient average Q. Strong results now on clips with significant static areas. (some data points in derf set as high as 9% and some static & slide show type content in YT set > 20%) Change-Id: Ia79f912efa84b977f35a23683ae3643251e24f0c
2011-11-02 14:30:10 +01:00
// All mbs should use ALTREF_FRAME, ZEROMV exclusively
clear_segref(xd, 0);
set_segref(xd, 0, ALTREF_FRAME);
clear_segref(xd, 1);
set_segref(xd, 1, ALTREF_FRAME);
Segment Feature Data Access No change to functionality or output. Updates to the segment feature data structure now all done through functions such as set_segdata() and get_segdata() in seg_common.c. The reason for this is to make changing the structures (if needed) and debug easier. In addition it provides a single location for subsequent addition of range and validity checks. For example valid combination of mode and reference frame. Change-Id: I2e866505562db4e4cb6f17a472b25b4465f01add
2011-11-03 17:58:26 +01:00
set_segdata( xd, 0, SEG_LVL_MODE, ZEROMV );
set_segdata( xd, 1, SEG_LVL_MODE, ZEROMV );
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
Segmentation Entropy and tweaks. Some correction for entropy impact of segment signaled (EOB and ref frame) Other slight tweaks. Derf VBR average gain now over 1% (best over 7%) One YT test clip has gains of circa 30% (VBR) There is still an issue with noisy clips where making the background static and coded with 0,0 can have a negative effect, especially at low Q. This is probably because of the loss of smoothing by fractional pixel filters. Change-Id: I7a225613c98067b96f8fc7a7e36f95d465b2b834
2011-10-28 16:27:23 +02:00
// Skip all MBs if high Q
if ( high_q )
{
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
enable_segfeature(xd, 0, SEG_LVL_EOB);
set_segdata( xd, 0, SEG_LVL_EOB, 0 );
enable_segfeature(xd, 1, SEG_LVL_EOB);
set_segdata( xd, 1, SEG_LVL_EOB, 0 );
Segmentation Entropy and tweaks. Some correction for entropy impact of segment signaled (EOB and ref frame) Other slight tweaks. Derf VBR average gain now over 1% (best over 7%) One YT test clip has gains of circa 30% (VBR) There is still an issue with noisy clips where making the background static and coded with 0,0 can have a negative effect, especially at low Q. This is probably because of the loss of smoothing by fractional pixel filters. Change-Id: I7a225613c98067b96f8fc7a7e36f95d465b2b834
2011-10-28 16:27:23 +02:00
}
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
// Enable data udpate
xd->update_mb_segmentation_data = 1;
}
// All other frames.
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
else
{
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
// No updates.. leave things as they are.
Segmentation Features; Only encode sign bit for feature data that can have a sign. Tweaks to the test segmentation rules so that it now actually gives a net benefit on the derf set of about 0.4% though much higher on some clips at the low end. Change-Id: I8e61f1aebf41c9037db7e67e2f8975aa18a0c986
2011-10-07 17:58:28 +02:00
xd->update_mb_segmentation_map = 0;
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
xd->update_mb_segmentation_data = 0;
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
}
}
}
// DEBUG: Print out the segment id of each MB in the current frame.
static void print_seg_map(VP8_COMP *cpi)
{
VP8_COMMON *cm = & cpi->common;
int row,col;
int map_index = 0;
FILE *statsfile;
statsfile = fopen("segmap.stt", "a");
fprintf(statsfile, "%10d\n",
cm->current_video_frame );
for ( row = 0; row < cpi->common.mb_rows; row++ )
{
for ( col = 0; col < cpi->common.mb_cols; col++ )
{
fprintf(statsfile, "%10d",
cpi->segmentation_map[map_index]);
map_index++;
}
fprintf(statsfile, "\n");
}
fprintf(statsfile, "\n");
fclose(statsfile);
}
Rename mode_ref_lf_test_function This function graduated from being a test func to something that's on by default. Rename it and remove some spurious comments that confuse its status. Change-Id: I689695a3ad29c35e9a72a43ec93766733ac6c20b
2010-09-29 19:53:08 +02:00
static void set_default_lf_deltas(VP8_COMP *cpi)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
cpi->mb.e_mbd.mode_ref_lf_delta_enabled = 1;
cpi->mb.e_mbd.mode_ref_lf_delta_update = 1;
vpx_memset(cpi->mb.e_mbd.ref_lf_deltas, 0, sizeof(cpi->mb.e_mbd.ref_lf_deltas));
vpx_memset(cpi->mb.e_mbd.mode_lf_deltas, 0, sizeof(cpi->mb.e_mbd.mode_lf_deltas));
// Test of ref frame deltas
cpi->mb.e_mbd.ref_lf_deltas[INTRA_FRAME] = 2;
cpi->mb.e_mbd.ref_lf_deltas[LAST_FRAME] = 0;
cpi->mb.e_mbd.ref_lf_deltas[GOLDEN_FRAME] = -2;
cpi->mb.e_mbd.ref_lf_deltas[ALTREF_FRAME] = -2;
cpi->mb.e_mbd.mode_lf_deltas[0] = 4; // BPRED
cpi->mb.e_mbd.mode_lf_deltas[1] = -2; // Zero
cpi->mb.e_mbd.mode_lf_deltas[2] = 2; // New mv
cpi->mb.e_mbd.mode_lf_deltas[3] = 4; // Split mv
}
void vp8_set_speed_features(VP8_COMP *cpi)
{
SPEED_FEATURES *sf = &cpi->sf;
int Mode = cpi->compressor_speed;
int Speed = cpi->Speed;
int i;
VP8_COMMON *cm = &cpi->common;
Code simplification Removal of the pickinter.c and .h files and calls to this code. Removal of some code relating to real time and one pass settings though there is more to be done in this regard. However, vp8_set_speed_features() now only supports modes 0 and 1 and speeds up to 3 so rd should always be set. Change-Id: I62c0c1b6154ab499785baef310536080e87bc4d8
2012-02-16 18:02:17 +01:00
// Only modes 0 and 1 supported for now in experimental code basae
if ( Mode > 1 )
Mode = 1;
Initial WebM release
2010-05-18 17:58:33 +02:00
// Initialise default mode frequency sampling variables
for (i = 0; i < MAX_MODES; i ++)
{
cpi->mode_check_freq[i] = 0;
cpi->mode_test_hit_counts[i] = 0;
cpi->mode_chosen_counts[i] = 0;
}
Disable some features for first pass. The following features don't make sense for the first pass in its current form and have a significant impact on its speed (up to 50%). Slow quantizer, slow dct and trellis optimization. Change-Id: Id9943f6765ffbd71fc0084ec7dfbc9d376fd6fcd
2011-01-06 18:10:07 +01:00
// best quality defaults
Initial WebM release
2010-05-18 17:58:33 +02:00
sf->RD = 1;
sf->search_method = NSTEP;
sf->improved_dct = 1;
sf->auto_filter = 1;
sf->recode_loop = 1;
sf->quarter_pixel_search = 1;
sf->half_pixel_search = 1;
sf->iterative_sub_pixel = 1;
sf->optimize_coefficients = 1;
Add no_skip_block4x4_search flag in SPLITMV mode Add a flag to always enable block4x4 search for speed=0 (good quality) to guarantee no quality loss for speed0. Change-Id: Ie04bbc25f7e6a33a7bfa30e05775d33148731c81
2011-01-10 23:21:55 +01:00
sf->no_skip_block4x4_search = 1;
Initial WebM release
2010-05-18 17:58:33 +02:00
sf->first_step = 0;
sf->max_step_search_steps = MAX_MVSEARCH_STEPS;
Add improved_mv_pred flag in real-time mode As mentioned in check-in "Improve motion search in real-time mode", MV prediction calculation causes speed loss for speed 7 and above. This change added a flag to turn off this calculation for speed>6 in real-time mode. Change-Id: I9f4ae5a8bf449222d1784b54e7d315fc8347b2d1
2011-02-11 15:43:37 +01:00
sf->improved_mv_pred = 1;
Initial WebM release
2010-05-18 17:58:33 +02:00
// default thresholds to 0
for (i = 0; i < MAX_MODES; i++)
sf->thresh_mult[i] = 0;
switch (Mode)
{
case 0: // best quality mode
sf->thresh_mult[THR_ZEROMV ] = 0;
sf->thresh_mult[THR_ZEROG ] = 0;
sf->thresh_mult[THR_ZEROA ] = 0;
sf->thresh_mult[THR_NEARESTMV] = 0;
sf->thresh_mult[THR_NEARESTG ] = 0;
sf->thresh_mult[THR_NEARESTA ] = 0;
sf->thresh_mult[THR_NEARMV ] = 0;
sf->thresh_mult[THR_NEARG ] = 0;
sf->thresh_mult[THR_NEARA ] = 0;
sf->thresh_mult[THR_DC ] = 0;
sf->thresh_mult[THR_V_PRED ] = 1000;
sf->thresh_mult[THR_H_PRED ] = 1000;
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
#if CONFIG_NEWINTRAMODES
sf->thresh_mult[THR_D45_PRED ] = 1000;
sf->thresh_mult[THR_D135_PRED] = 1000;
sf->thresh_mult[THR_D117_PRED] = 1000;
sf->thresh_mult[THR_D153_PRED] = 1000;
sf->thresh_mult[THR_D27_PRED ] = 1000;
sf->thresh_mult[THR_D63_PRED ] = 1000;
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
sf->thresh_mult[THR_B_PRED ] = 2000;
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
sf->thresh_mult[THR_I8X8_PRED] = 2000;
Initial WebM release
2010-05-18 17:58:33 +02:00
sf->thresh_mult[THR_TM ] = 1000;
sf->thresh_mult[THR_NEWMV ] = 1000;
sf->thresh_mult[THR_NEWG ] = 1000;
sf->thresh_mult[THR_NEWA ] = 1000;
sf->thresh_mult[THR_SPLITMV ] = 2500;
sf->thresh_mult[THR_SPLITG ] = 5000;
sf->thresh_mult[THR_SPLITA ] = 5000;
Rename "dual" prediction to "compound" prediction. Change-Id: Ibcd2b9b247ff9f83331dac47f91ec285e8955ff1
2012-02-29 02:25:45 +01:00
sf->thresh_mult[THR_COMP_ZEROLG ] = 0;
sf->thresh_mult[THR_COMP_NEARESTLG] = 0;
sf->thresh_mult[THR_COMP_NEARLG ] = 0;
sf->thresh_mult[THR_COMP_ZEROLA ] = 0;
sf->thresh_mult[THR_COMP_NEARESTLA] = 0;
sf->thresh_mult[THR_COMP_NEARLA ] = 0;
sf->thresh_mult[THR_COMP_ZEROGA ] = 0;
sf->thresh_mult[THR_COMP_NEARESTGA] = 0;
sf->thresh_mult[THR_COMP_NEARGA ] = 0;
sf->thresh_mult[THR_COMP_NEWLG ] = 1000;
sf->thresh_mult[THR_COMP_NEWLA ] = 1000;
sf->thresh_mult[THR_COMP_NEWGA ] = 1000;
Initial WebM release
2010-05-18 17:58:33 +02:00
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
sf->thresh_mult[THR_COMP_SPLITLA ] = 2500;
sf->thresh_mult[THR_COMP_SPLITGA ] = 5000;
sf->thresh_mult[THR_COMP_SPLITLG ] = 5000;
Initial WebM release
2010-05-18 17:58:33 +02:00
sf->first_step = 0;
sf->max_step_search_steps = MAX_MVSEARCH_STEPS;
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
#if CONFIG_ENHANCED_INTERP
sf->search_best_filter = SEARCH_BEST_FILTER;
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
break;
case 1:
sf->thresh_mult[THR_NEARESTMV] = 0;
sf->thresh_mult[THR_ZEROMV ] = 0;
sf->thresh_mult[THR_DC ] = 0;
sf->thresh_mult[THR_NEARMV ] = 0;
sf->thresh_mult[THR_V_PRED ] = 1000;
sf->thresh_mult[THR_H_PRED ] = 1000;
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
#if CONFIG_NEWINTRAMODES
sf->thresh_mult[THR_D45_PRED ] = 1000;
sf->thresh_mult[THR_D135_PRED] = 1000;
sf->thresh_mult[THR_D117_PRED] = 1000;
sf->thresh_mult[THR_D153_PRED] = 1000;
sf->thresh_mult[THR_D27_PRED ] = 1000;
sf->thresh_mult[THR_D63_PRED ] = 1000;
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
sf->thresh_mult[THR_B_PRED ] = 2500;
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
sf->thresh_mult[THR_I8X8_PRED] = 2500;
Initial WebM release
2010-05-18 17:58:33 +02:00
sf->thresh_mult[THR_TM ] = 1000;
sf->thresh_mult[THR_NEARESTG ] = 1000;
sf->thresh_mult[THR_NEARESTA ] = 1000;
sf->thresh_mult[THR_ZEROG ] = 1000;
sf->thresh_mult[THR_ZEROA ] = 1000;
sf->thresh_mult[THR_NEARG ] = 1000;
sf->thresh_mult[THR_NEARA ] = 1000;
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
sf->thresh_mult[THR_ZEROMV ] = 0;
sf->thresh_mult[THR_ZEROG ] = 0;
sf->thresh_mult[THR_ZEROA ] = 0;
sf->thresh_mult[THR_NEARESTMV] = 0;
sf->thresh_mult[THR_NEARESTG ] = 0;
sf->thresh_mult[THR_NEARESTA ] = 0;
sf->thresh_mult[THR_NEARMV ] = 0;
sf->thresh_mult[THR_NEARG ] = 0;
sf->thresh_mult[THR_NEARA ] = 0;
sf->thresh_mult[THR_NEWMV ] = 1000;
sf->thresh_mult[THR_NEWG ] = 1000;
sf->thresh_mult[THR_NEWA ] = 1000;
sf->thresh_mult[THR_SPLITMV ] = 1700;
sf->thresh_mult[THR_SPLITG ] = 4500;
sf->thresh_mult[THR_SPLITA ] = 4500;
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 "dual" prediction to "compound" prediction. Change-Id: Ibcd2b9b247ff9f83331dac47f91ec285e8955ff1
2012-02-29 02:25:45 +01:00
sf->thresh_mult[THR_COMP_ZEROLG ] = 0;
sf->thresh_mult[THR_COMP_NEARESTLG] = 0;
sf->thresh_mult[THR_COMP_NEARLG ] = 0;
sf->thresh_mult[THR_COMP_ZEROLA ] = 0;
sf->thresh_mult[THR_COMP_NEARESTLA] = 0;
sf->thresh_mult[THR_COMP_NEARLA ] = 0;
sf->thresh_mult[THR_COMP_ZEROGA ] = 0;
sf->thresh_mult[THR_COMP_NEARESTGA] = 0;
sf->thresh_mult[THR_COMP_NEARGA ] = 0;
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 "dual" prediction to "compound" prediction. Change-Id: Ibcd2b9b247ff9f83331dac47f91ec285e8955ff1
2012-02-29 02:25:45 +01:00
sf->thresh_mult[THR_COMP_NEWLG ] = 1000;
sf->thresh_mult[THR_COMP_NEWLA ] = 1000;
sf->thresh_mult[THR_COMP_NEWGA ] = 1000;
Initial WebM release
2010-05-18 17:58:33 +02:00
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
sf->thresh_mult[THR_COMP_SPLITLA ] = 1700;
sf->thresh_mult[THR_COMP_SPLITGA ] = 4500;
sf->thresh_mult[THR_COMP_SPLITLG ] = 4500;
Redefining good quality speed settings Created a new speed 1 which is in the middle of the old speed 0 and speed 1. (for both quality and performance) Change-Id: I4802133cdb43f359ca787646c090899679dd5d84
2011-02-09 23:18:28 +01:00
if (Speed > 0)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Redefining good quality speed settings Created a new speed 1 which is in the middle of the old speed 0 and speed 1. (for both quality and performance) Change-Id: I4802133cdb43f359ca787646c090899679dd5d84
2011-02-09 23:18:28 +01:00
/* Disable coefficient optimization above speed 0 */
sf->optimize_coefficients = 0;
sf->no_skip_block4x4_search = 0;
Fix breakout thresh computation for golden & AltRef frames 1. Unavailability of each reference frame type should be tested independently, 2. Also, only the VP8_GOLD_FLAG needs to be tested before setting golden frame specific thresholds, and only VP8_ALT_FLAG needs testing before setting thresholds relevant to the AltRef frame. (Raised by gbvalor, in response to Issue 47) Change-Id: I6a06fc2a6592841d85422bc1661e33349bb6c3b8
2010-06-21 14:44:42 +02:00
Redefining good quality speed settings Created a new speed 1 which is in the middle of the old speed 0 and speed 1. (for both quality and performance) Change-Id: I4802133cdb43f359ca787646c090899679dd5d84
2011-02-09 23:18:28 +01:00
sf->first_step = 1;
Fix breakout thresh computation for golden & AltRef frames 1. Unavailability of each reference frame type should be tested independently, 2. Also, only the VP8_GOLD_FLAG needs to be tested before setting golden frame specific thresholds, and only VP8_ALT_FLAG needs testing before setting thresholds relevant to the AltRef frame. (Raised by gbvalor, in response to Issue 47) Change-Id: I6a06fc2a6592841d85422bc1661e33349bb6c3b8
2010-06-21 14:44:42 +02:00
Redefining good quality speed settings Created a new speed 1 which is in the middle of the old speed 0 and speed 1. (for both quality and performance) Change-Id: I4802133cdb43f359ca787646c090899679dd5d84
2011-02-09 23:18:28 +01:00
cpi->mode_check_freq[THR_SPLITG] = 2;
cpi->mode_check_freq[THR_SPLITA] = 2;
cpi->mode_check_freq[THR_SPLITMV] = 0;
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
cpi->mode_check_freq[THR_COMP_SPLITGA] = 2;
cpi->mode_check_freq[THR_COMP_SPLITLG] = 2;
cpi->mode_check_freq[THR_COMP_SPLITLA] = 0;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Redefining good quality speed settings Created a new speed 1 which is in the middle of the old speed 0 and speed 1. (for both quality and performance) Change-Id: I4802133cdb43f359ca787646c090899679dd5d84
2011-02-09 23:18:28 +01:00
if (Speed > 1)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
cpi->mode_check_freq[THR_SPLITG] = 4;
cpi->mode_check_freq[THR_SPLITA] = 4;
Use the fast quantizer for inter mode selection Use the fast quantizer for inter mode selection and the regular quantizer for the rest of the encode for good quality, speed 1. Both performance and quality were improved. The quality gains will make up for the quality loss mentioned in I9dc089007ca08129fb6c11fe7692777ebb8647b0. Change-Id: Ia90bc9cf326a7c65d60d31fa32f6465ab6984d21
2010-12-28 20:51:46 +01:00
cpi->mode_check_freq[THR_SPLITMV] = 2;
Initial WebM release
2010-05-18 17:58:33 +02:00
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
cpi->mode_check_freq[THR_COMP_SPLITGA] = 4;
cpi->mode_check_freq[THR_COMP_SPLITLG] = 4;
cpi->mode_check_freq[THR_COMP_SPLITLA] = 2;
Initial WebM release
2010-05-18 17:58:33 +02:00
sf->thresh_mult[THR_TM ] = 1500;
sf->thresh_mult[THR_V_PRED ] = 1500;
sf->thresh_mult[THR_H_PRED ] = 1500;
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
#if CONFIG_NEWINTRAMODES
sf->thresh_mult[THR_D45_PRED ] = 1500;
sf->thresh_mult[THR_D135_PRED] = 1500;
sf->thresh_mult[THR_D117_PRED] = 1500;
sf->thresh_mult[THR_D153_PRED] = 1500;
sf->thresh_mult[THR_D27_PRED ] = 1500;
sf->thresh_mult[THR_D63_PRED ] = 1500;
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
sf->thresh_mult[THR_B_PRED ] = 5000;
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
sf->thresh_mult[THR_I8X8_PRED] = 5000;
Removed #if CONFIG_I8X8 This commit removed the macro CONFIG_I8X8, which was used to indicate the 8x8 intra prediction experiment, made the change fully merged in. Change-Id: Iafa4443781ce6e83f5591c12ba615a0e92ce0ea0
2011-12-07 22:03:57 +01:00
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->ref_frame_flags & VP8_LAST_FLAG)
{
sf->thresh_mult[THR_NEWMV ] = 2000;
sf->thresh_mult[THR_SPLITMV ] = 10000;
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
sf->thresh_mult[THR_COMP_SPLITLG ] = 20000;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
if (cpi->ref_frame_flags & VP8_GOLD_FLAG)
{
sf->thresh_mult[THR_NEARESTG ] = 1500;
sf->thresh_mult[THR_ZEROG ] = 1500;
sf->thresh_mult[THR_NEARG ] = 1500;
sf->thresh_mult[THR_NEWG ] = 2000;
sf->thresh_mult[THR_SPLITG ] = 20000;
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
sf->thresh_mult[THR_COMP_SPLITGA ] = 20000;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
if (cpi->ref_frame_flags & VP8_ALT_FLAG)
{
sf->thresh_mult[THR_NEARESTA ] = 1500;
sf->thresh_mult[THR_ZEROA ] = 1500;
sf->thresh_mult[THR_NEARA ] = 1500;
sf->thresh_mult[THR_NEWA ] = 2000;
sf->thresh_mult[THR_SPLITA ] = 20000;
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
sf->thresh_mult[THR_COMP_SPLITLA ] = 10000;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
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 "dual" prediction to "compound" prediction. Change-Id: Ibcd2b9b247ff9f83331dac47f91ec285e8955ff1
2012-02-29 02:25:45 +01:00
sf->thresh_mult[THR_COMP_ZEROLG ] = 1500;
sf->thresh_mult[THR_COMP_NEARESTLG] = 1500;
sf->thresh_mult[THR_COMP_NEARLG ] = 1500;
sf->thresh_mult[THR_COMP_ZEROLA ] = 1500;
sf->thresh_mult[THR_COMP_NEARESTLA] = 1500;
sf->thresh_mult[THR_COMP_NEARLA ] = 1500;
sf->thresh_mult[THR_COMP_ZEROGA ] = 1500;
sf->thresh_mult[THR_COMP_NEARESTGA] = 1500;
sf->thresh_mult[THR_COMP_NEARGA ] = 1500;
sf->thresh_mult[THR_COMP_NEWLG ] = 2000;
sf->thresh_mult[THR_COMP_NEWLA ] = 2000;
sf->thresh_mult[THR_COMP_NEWGA ] = 2000;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Redefining good quality speed settings Created a new speed 1 which is in the middle of the old speed 0 and speed 1. (for both quality and performance) Change-Id: I4802133cdb43f359ca787646c090899679dd5d84
2011-02-09 23:18:28 +01:00
if (Speed > 2)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
cpi->mode_check_freq[THR_SPLITG] = 15;
cpi->mode_check_freq[THR_SPLITA] = 15;
cpi->mode_check_freq[THR_SPLITMV] = 7;
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
cpi->mode_check_freq[THR_COMP_SPLITGA] = 15;
cpi->mode_check_freq[THR_COMP_SPLITLG] = 15;
cpi->mode_check_freq[THR_COMP_SPLITLA] = 7;
Initial WebM release
2010-05-18 17:58:33 +02:00
sf->thresh_mult[THR_TM ] = 2000;
sf->thresh_mult[THR_V_PRED ] = 2000;
sf->thresh_mult[THR_H_PRED ] = 2000;
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
#if CONFIG_NEWINTRAMODES
sf->thresh_mult[THR_D45_PRED ] = 2000;
sf->thresh_mult[THR_D135_PRED] = 2000;
sf->thresh_mult[THR_D117_PRED] = 2000;
sf->thresh_mult[THR_D153_PRED] = 2000;
sf->thresh_mult[THR_D27_PRED ] = 2000;
sf->thresh_mult[THR_D63_PRED ] = 2000;
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
sf->thresh_mult[THR_B_PRED ] = 7500;
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
sf->thresh_mult[THR_I8X8_PRED] = 7500;
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->ref_frame_flags & VP8_LAST_FLAG)
{
sf->thresh_mult[THR_NEWMV ] = 2000;
sf->thresh_mult[THR_SPLITMV ] = 25000;
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
sf->thresh_mult[THR_COMP_SPLITLG ] = 50000;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
if (cpi->ref_frame_flags & VP8_GOLD_FLAG)
{
sf->thresh_mult[THR_NEARESTG ] = 2000;
sf->thresh_mult[THR_ZEROG ] = 2000;
sf->thresh_mult[THR_NEARG ] = 2000;
sf->thresh_mult[THR_NEWG ] = 2500;
sf->thresh_mult[THR_SPLITG ] = 50000;
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
sf->thresh_mult[THR_COMP_SPLITGA ] = 50000;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
if (cpi->ref_frame_flags & VP8_ALT_FLAG)
{
sf->thresh_mult[THR_NEARESTA ] = 2000;
sf->thresh_mult[THR_ZEROA ] = 2000;
sf->thresh_mult[THR_NEARA ] = 2000;
sf->thresh_mult[THR_NEWA ] = 2500;
sf->thresh_mult[THR_SPLITA ] = 50000;
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
sf->thresh_mult[THR_COMP_SPLITLA ] = 25000;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Rename "dual" prediction to "compound" prediction. Change-Id: Ibcd2b9b247ff9f83331dac47f91ec285e8955ff1
2012-02-29 02:25:45 +01:00
sf->thresh_mult[THR_COMP_ZEROLG ] = 2000;
sf->thresh_mult[THR_COMP_NEARESTLG] = 2000;
sf->thresh_mult[THR_COMP_NEARLG ] = 2000;
sf->thresh_mult[THR_COMP_ZEROLA ] = 2000;
sf->thresh_mult[THR_COMP_NEARESTLA] = 2000;
sf->thresh_mult[THR_COMP_NEARLA ] = 2000;
sf->thresh_mult[THR_COMP_ZEROGA ] = 2000;
sf->thresh_mult[THR_COMP_NEARESTGA] = 2000;
sf->thresh_mult[THR_COMP_NEARGA ] = 2000;
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 "dual" prediction to "compound" prediction. Change-Id: Ibcd2b9b247ff9f83331dac47f91ec285e8955ff1
2012-02-29 02:25:45 +01:00
sf->thresh_mult[THR_COMP_NEWLG ] = 2500;
sf->thresh_mult[THR_COMP_NEWLA ] = 2500;
sf->thresh_mult[THR_COMP_NEWGA ] = 2500;
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
Use the fast quantizer for inter mode selection Use the fast quantizer for inter mode selection and the regular quantizer for the rest of the encode for good quality, speed 1. Both performance and quality were improved. The quality gains will make up for the quality loss mentioned in I9dc089007ca08129fb6c11fe7692777ebb8647b0. Change-Id: Ia90bc9cf326a7c65d60d31fa32f6465ab6984d21
2010-12-28 20:51:46 +01:00
sf->improved_dct = 0;
Replaced recode loop test with a function call Replaced existing code to decide if a frame recode is required with a function call. This is to simplify addition of extra clauses that may be needed for the planned constrained quality mode. Also fixed a bug where by alt ref not considered in the test. Change-Id: I3d40bb21abe3e19f8456761e6849deb171738b60
2010-11-17 16:12:04 +01:00
// Only do recode loop on key frames, golden frames and
// alt ref frames
Initial WebM release
2010-05-18 17:58:33 +02:00
sf->recode_loop = 2;
}
break;
Redefining good quality speed settings Created a new speed 1 which is in the middle of the old speed 0 and speed 1. (for both quality and performance) Change-Id: I4802133cdb43f359ca787646c090899679dd5d84
2011-02-09 23:18:28 +01:00
}; /* switch */
/* disable frame modes if flags not set */
if (!(cpi->ref_frame_flags & VP8_LAST_FLAG))
{
sf->thresh_mult[THR_NEWMV ] = INT_MAX;
sf->thresh_mult[THR_NEARESTMV] = INT_MAX;
sf->thresh_mult[THR_ZEROMV ] = INT_MAX;
sf->thresh_mult[THR_NEARMV ] = INT_MAX;
sf->thresh_mult[THR_SPLITMV ] = INT_MAX;
}
if (!(cpi->ref_frame_flags & VP8_GOLD_FLAG))
{
sf->thresh_mult[THR_NEARESTG ] = INT_MAX;
sf->thresh_mult[THR_ZEROG ] = INT_MAX;
sf->thresh_mult[THR_NEARG ] = INT_MAX;
sf->thresh_mult[THR_NEWG ] = INT_MAX;
sf->thresh_mult[THR_SPLITG ] = INT_MAX;
}
if (!(cpi->ref_frame_flags & VP8_ALT_FLAG))
{
sf->thresh_mult[THR_NEARESTA ] = INT_MAX;
sf->thresh_mult[THR_ZEROA ] = INT_MAX;
sf->thresh_mult[THR_NEARA ] = INT_MAX;
sf->thresh_mult[THR_NEWA ] = INT_MAX;
sf->thresh_mult[THR_SPLITA ] = INT_MAX;
}
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
if ((cpi->ref_frame_flags & (VP8_LAST_FLAG | VP8_GOLD_FLAG)) != (VP8_LAST_FLAG | VP8_GOLD_FLAG))
{
Rename "dual" prediction to "compound" prediction. Change-Id: Ibcd2b9b247ff9f83331dac47f91ec285e8955ff1
2012-02-29 02:25:45 +01:00
sf->thresh_mult[THR_COMP_ZEROLG ] = INT_MAX;
sf->thresh_mult[THR_COMP_NEARESTLG] = INT_MAX;
sf->thresh_mult[THR_COMP_NEARLG ] = INT_MAX;
sf->thresh_mult[THR_COMP_NEWLG ] = INT_MAX;
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
sf->thresh_mult[THR_COMP_SPLITLG ] = INT_MAX;
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
}
if ((cpi->ref_frame_flags & (VP8_LAST_FLAG | VP8_ALT_FLAG)) != (VP8_LAST_FLAG | VP8_ALT_FLAG))
{
Rename "dual" prediction to "compound" prediction. Change-Id: Ibcd2b9b247ff9f83331dac47f91ec285e8955ff1
2012-02-29 02:25:45 +01:00
sf->thresh_mult[THR_COMP_ZEROLA ] = INT_MAX;
sf->thresh_mult[THR_COMP_NEARESTLA] = INT_MAX;
sf->thresh_mult[THR_COMP_NEARLA ] = INT_MAX;
sf->thresh_mult[THR_COMP_NEWLA ] = INT_MAX;
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
sf->thresh_mult[THR_COMP_SPLITLA ] = INT_MAX;
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
}
if ((cpi->ref_frame_flags & (VP8_GOLD_FLAG | VP8_ALT_FLAG)) != (VP8_GOLD_FLAG | VP8_ALT_FLAG))
{
Rename "dual" prediction to "compound" prediction. Change-Id: Ibcd2b9b247ff9f83331dac47f91ec285e8955ff1
2012-02-29 02:25:45 +01:00
sf->thresh_mult[THR_COMP_ZEROGA ] = INT_MAX;
sf->thresh_mult[THR_COMP_NEARESTGA] = INT_MAX;
sf->thresh_mult[THR_COMP_NEARGA ] = INT_MAX;
sf->thresh_mult[THR_COMP_NEWGA ] = INT_MAX;
Compound prediction for splitmv macroblocks. Change-Id: I0af3395500b1cb0ed629249eb6636a0c9322cb18
2012-04-18 22:51:58 +02:00
sf->thresh_mult[THR_COMP_SPLITGA ] = INT_MAX;
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
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Disable some features for first pass. The following features don't make sense for the first pass in its current form and have a significant impact on its speed (up to 50%). Slow quantizer, slow dct and trellis optimization. Change-Id: Id9943f6765ffbd71fc0084ec7dfbc9d376fd6fcd
2011-01-06 18:10:07 +01:00
// Slow quant, dct and trellis not worthwhile for first pass
// so make sure they are always turned off.
if ( cpi->pass == 1 )
{
sf->optimize_coefficients = 0;
sf->improved_dct = 0;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->sf.search_method == NSTEP)
{
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
vp8_init3smotion_compensation(&cpi->mb, cm->yv12_fb[cm->lst_fb_idx].y_stride);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
else if (cpi->sf.search_method == DIAMOND)
{
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
vp8_init_dsmotion_compensation(&cpi->mb, cm->yv12_fb[cm->lst_fb_idx].y_stride);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
if (cpi->sf.improved_dct)
{
Add 8x8 transform to experimental branch Please refer to previous commit messages for detailed info: https://on2-git.corp.google.com/g/#change,5940 https://on2-git.corp.google.com/g/#change,6045 Change-Id: I8b16992f2f69c5a808ad40a3e32ef589cce7c59d
2011-02-14 23:18:18 +01:00
cpi->mb.vp8_short_fdct8x8 = FDCT_INVOKE(&cpi->rtcd.fdct, short8x8);
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->mb.vp8_short_fdct8x4 = FDCT_INVOKE(&cpi->rtcd.fdct, short8x4);
cpi->mb.vp8_short_fdct4x4 = FDCT_INVOKE(&cpi->rtcd.fdct, short4x4);
}
else
{
Add 8x8 transform to experimental branch Please refer to previous commit messages for detailed info: https://on2-git.corp.google.com/g/#change,5940 https://on2-git.corp.google.com/g/#change,6045 Change-Id: I8b16992f2f69c5a808ad40a3e32ef589cce7c59d
2011-02-14 23:18:18 +01:00
cpi->mb.vp8_short_fdct8x8 = FDCT_INVOKE(&cpi->rtcd.fdct, short8x8);
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->mb.vp8_short_fdct8x4 = FDCT_INVOKE(&cpi->rtcd.fdct, fast8x4);
cpi->mb.vp8_short_fdct4x4 = FDCT_INVOKE(&cpi->rtcd.fdct, fast4x4);
}
cpi->mb.short_walsh4x4 = FDCT_INVOKE(&cpi->rtcd.fdct, walsh_short4x4);
Add 8x8 transform to experimental branch Please refer to previous commit messages for detailed info: https://on2-git.corp.google.com/g/#change,5940 https://on2-git.corp.google.com/g/#change,6045 Change-Id: I8b16992f2f69c5a808ad40a3e32ef589cce7c59d
2011-02-14 23:18:18 +01:00
cpi->mb.short_fhaar2x2 = FDCT_INVOKE(&cpi->rtcd.fdct, haar_short2x2);
Initial WebM release
2010-05-18 17:58:33 +02:00
Only support improved quant Deprecate fast quant and strict_quant code. Small effect on quality as fast was used in first pass but the effect is basically neutral across the derf set. The rationale here is to reduce the number of code paths for now to make experimentation easier. Optimized and fast code options can be re-introduced later along with other encode speed options. Change-Id: Ia30c5daf3dbc52e72c83b277a1d281e3c934cdad
2012-03-21 19:22:33 +01:00
cpi->mb.quantize_b = vp8_regular_quantize_b;
cpi->mb.quantize_b_pair = vp8_regular_quantize_b_pair;
cpi->mb.quantize_b_8x8 = vp8_regular_quantize_b_8x8;
cpi->mb.quantize_b_2x2 = vp8_regular_quantize_b_2x2;
vp8cx_init_quantizer(cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
#if CONFIG_RUNTIME_CPU_DETECT
cpi->mb.e_mbd.rtcd = &cpi->common.rtcd;
#endif
if (cpi->sf.iterative_sub_pixel == 1)
{
cpi->find_fractional_mv_step = vp8_find_best_sub_pixel_step_iteratively;
}
else if (cpi->sf.quarter_pixel_search)
{
cpi->find_fractional_mv_step = vp8_find_best_sub_pixel_step;
}
else if (cpi->sf.half_pixel_search)
{
cpi->find_fractional_mv_step = vp8_find_best_half_pixel_step;
}
disable trellis optimization for first pass also remove 2 #defines and 1 function declaration that are not in use. Change-Id: I8f743d0e3dd9ebf1de24a8b0c30ff09f29b00c53
2011-05-20 01:00:28 +02:00
if (cpi->sf.optimize_coefficients == 1 && cpi->pass!=1)
remove a variable no longer in use The variable is introduced in commit 2e53e9e53 to make more use of trellis quantization, but this is no longer necessary after RDMULT was made adaptive in a number of later commits. Change-Id: I7420522ec7723f38cf77033466c25afb405d52ae
2011-05-11 04:57:51 +02:00
cpi->mb.optimize = 1;
Initial WebM release
2010-05-18 17:58:33 +02:00
else
cpi->mb.optimize = 0;
#ifdef SPEEDSTATS
frames_at_speed[cpi->Speed]++;
#endif
}
static void alloc_raw_frame_buffers(VP8_COMP *cpi)
{
Allocate source buffers to be multiples of 16 Currently, when the video frame width is not multiples of 16, the source buffer has a stride of non-multiples of 16, which forces an unaligned load in SAD function and hurts the performance. To avoid that, this change allocates source buffers to be multiples of 16. Change-Id: Ib7506e3eb2cea06657d56be5a899f38dfe3eeb39
2011-02-16 18:00:25 +01:00
int width = (cpi->oxcf.Width + 15) & ~15;
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
int height = (cpi->oxcf.Height + 15) & ~15;
Initial WebM release
2010-05-18 17:58:33 +02:00
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
cpi->lookahead = vp8_lookahead_init(cpi->oxcf.Width, cpi->oxcf.Height,
cpi->oxcf.lag_in_frames);
if(!cpi->lookahead)
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate lag buffers");
Initial WebM release
2010-05-18 17:58:33 +02:00
#if VP8_TEMPORAL_ALT_REF
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
if (vp8_yv12_alloc_frame_buffer(&cpi->alt_ref_buffer,
Increase chrow row alignment to 16 bytes. This is done by expanding luma row to 32-byte alignment, since there is currently a bunch of code that assumes that uv_stride == y_stride/2 (see, for example, vp8/common/postproc.c, common/reconinter.c, common/arm/neon/recon16x16mb_neon.asm, encoder/temporal_filter.c, and possibly others; I haven't done a full audit). It also uses replaces the hardcoded border of 16 in a number of encoder buffers with VP8BORDERINPIXELS (currently 32), as the chroma rows start at an offset of border/2. Together, these two changes have the nice advantage that simply dumping the frame memory as a contiguous blob produces a valid, if padded, image. Change-Id: Iaf5ea722ae5c82d5daa50f6e2dade9de753f1003
2011-07-20 19:20:31 +02:00
width, height, VP8BORDERINPIXELS))
Initial WebM release
2010-05-18 17:58:33 +02:00
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate altref buffer");
#endif
}
Reduced the size of MB_MODE_INFO Moved partition_bmi and partition_count out of MB_MODE_INFO and placed into MACROBLOCK. Also reduced the size of other members of the MB_MODE_INFO struct. For 1080p, the memory was reduced by 1,209,516 bytes. The decoder performance appeared to improve by 3% for the clip used. Note: The main goal for this change is to improve the decoder performance. The encoder will be revisited at a later date for further structure cleanup. Change-Id: I4733621292ee9cc3fffa4046cb3fd4d99bd14613
2010-09-02 22:17:52 +02:00
static int vp8_alloc_partition_data(VP8_COMP *cpi)
{
Fix a memory leak problem in encoder Deallocating the buffers before re-allocating them. The fix passed James Berry's test program for memory leak check. Change-Id: I18c3cf665412c0e313a523e3d435106c03ca438d
2010-12-06 23:21:37 +01:00
vpx_free(cpi->mb.pip);
Reduced the size of MB_MODE_INFO Moved partition_bmi and partition_count out of MB_MODE_INFO and placed into MACROBLOCK. Also reduced the size of other members of the MB_MODE_INFO struct. For 1080p, the memory was reduced by 1,209,516 bytes. The decoder performance appeared to improve by 3% for the clip used. Note: The main goal for this change is to improve the decoder performance. The encoder will be revisited at a later date for further structure cleanup. Change-Id: I4733621292ee9cc3fffa4046cb3fd4d99bd14613
2010-09-02 22:17:52 +02:00
cpi->mb.pip = vpx_calloc((cpi->common.mb_cols + 1) *
(cpi->common.mb_rows + 1),
sizeof(PARTITION_INFO));
if(!cpi->mb.pip)
clean up unused files Removed a number of files that were unused or little-used. Change-Id: If9ae5e5b11390077581a9a879e8a0defe709f5da
2011-02-17 12:59:48 +01:00
return 1;
Reduced the size of MB_MODE_INFO Moved partition_bmi and partition_count out of MB_MODE_INFO and placed into MACROBLOCK. Also reduced the size of other members of the MB_MODE_INFO struct. For 1080p, the memory was reduced by 1,209,516 bytes. The decoder performance appeared to improve by 3% for the clip used. Note: The main goal for this change is to improve the decoder performance. The encoder will be revisited at a later date for further structure cleanup. Change-Id: I4733621292ee9cc3fffa4046cb3fd4d99bd14613
2010-09-02 22:17:52 +02:00
cpi->mb.pi = cpi->mb.pip + cpi->common.mode_info_stride + 1;
return 0;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
void vp8_alloc_compressor_data(VP8_COMP *cpi)
{
VP8_COMMON *cm = & cpi->common;
int width = cm->Width;
int height = cm->Height;
if (vp8_alloc_frame_buffers(cm, width, height))
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate frame buffers");
Reduced the size of MB_MODE_INFO Moved partition_bmi and partition_count out of MB_MODE_INFO and placed into MACROBLOCK. Also reduced the size of other members of the MB_MODE_INFO struct. For 1080p, the memory was reduced by 1,209,516 bytes. The decoder performance appeared to improve by 3% for the clip used. Note: The main goal for this change is to improve the decoder performance. The encoder will be revisited at a later date for further structure cleanup. Change-Id: I4733621292ee9cc3fffa4046cb3fd4d99bd14613
2010-09-02 22:17:52 +02:00
if (vp8_alloc_partition_data(cpi))
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate partition data");
Initial WebM release
2010-05-18 17:58:33 +02:00
if ((width & 0xf) != 0)
width += 16 - (width & 0xf);
if ((height & 0xf) != 0)
height += 16 - (height & 0xf);
if (vp8_yv12_alloc_frame_buffer(&cpi->last_frame_uf,
width, height, VP8BORDERINPIXELS))
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate last frame buffer");
Increase chrow row alignment to 16 bytes. This is done by expanding luma row to 32-byte alignment, since there is currently a bunch of code that assumes that uv_stride == y_stride/2 (see, for example, vp8/common/postproc.c, common/reconinter.c, common/arm/neon/recon16x16mb_neon.asm, encoder/temporal_filter.c, and possibly others; I haven't done a full audit). It also uses replaces the hardcoded border of 16 in a number of encoder buffers with VP8BORDERINPIXELS (currently 32), as the chroma rows start at an offset of border/2. Together, these two changes have the nice advantage that simply dumping the frame memory as a contiguous blob produces a valid, if padded, image. Change-Id: Iaf5ea722ae5c82d5daa50f6e2dade9de753f1003
2011-07-20 19:20:31 +02:00
if (vp8_yv12_alloc_frame_buffer(&cpi->scaled_source,
width, height, VP8BORDERINPIXELS))
Initial WebM release
2010-05-18 17:58:33 +02:00
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate scaled source buffer");
vpx_free(cpi->tok);
{
unsigned int tokens = cm->mb_rows * cm->mb_cols * 24 * 16;
CHECK_MEM_ERROR(cpi->tok, vpx_calloc(tokens, sizeof(*cpi->tok)));
}
// Data used for real time vc mode to see if gf needs refreshing
cpi->inter_zz_count = 0;
cpi->gf_bad_count = 0;
cpi->gf_update_recommended = 0;
Moved gf_active code to encoder only The gf_active code is only used by the encoder, so it was moved from common and decoder. Change-Id: Iada15acd5b2b33ff70c34668ca87d4cfd0d05025
2010-08-11 17:02:31 +02:00
// Structures used to minitor GF usage
Restructure of activity masking code. This commit restructures the mb activity masking code to better facilitate experimentation using different metrics etc. and also allows for adjustment of the zero bin either for encode only or both the encode and mode selection stages It also uses information from the current frame rather than the previous frame and the default strength has been reduced. Change-Id: Id39b19eace37574dc429f25aae810c203709629b
2011-05-12 18:01:55 +02:00
vpx_free(cpi->gf_active_flags);
CHECK_MEM_ERROR(cpi->gf_active_flags,
vpx_calloc(1, cm->mb_rows * cm->mb_cols));
cpi->gf_active_count = cm->mb_rows * cm->mb_cols;
Moved gf_active code to encoder only The gf_active code is only used by the encoder, so it was moved from common and decoder. Change-Id: Iada15acd5b2b33ff70c34668ca87d4cfd0d05025
2010-08-11 17:02:31 +02:00
Restructure of activity masking code. This commit restructures the mb activity masking code to better facilitate experimentation using different metrics etc. and also allows for adjustment of the zero bin either for encode only or both the encode and mode selection stages It also uses information from the current frame rather than the previous frame and the default strength has been reduced. Change-Id: Id39b19eace37574dc429f25aae810c203709629b
2011-05-12 18:01:55 +02:00
vpx_free(cpi->mb_activity_map);
CHECK_MEM_ERROR(cpi->mb_activity_map,
vpx_calloc(sizeof(unsigned int),
cm->mb_rows * cm->mb_cols));
Moved gf_active code to encoder only The gf_active code is only used by the encoder, so it was moved from common and decoder. Change-Id: Iada15acd5b2b33ff70c34668ca87d4cfd0d05025
2010-08-11 17:02:31 +02:00
Restructure of activity masking code. This commit restructures the mb activity masking code to better facilitate experimentation using different metrics etc. and also allows for adjustment of the zero bin either for encode only or both the encode and mode selection stages It also uses information from the current frame rather than the previous frame and the default strength has been reduced. Change-Id: Id39b19eace37574dc429f25aae810c203709629b
2011-05-12 18:01:55 +02:00
vpx_free(cpi->mb_norm_activity_map);
CHECK_MEM_ERROR(cpi->mb_norm_activity_map,
vpx_calloc(sizeof(unsigned int),
cm->mb_rows * cm->mb_cols));
Move firstpass motion map to stats packet The first implementation of the firstpass motion map for motion compensated temporal filtering created a file, fpmotionmap.stt, in the current working directory. This was not safe for multiple encoder instances. This patch merges this data into the first pass stats packet interface, so that it is handled like the other (numerical) firstpass stats. The new stats packet is defined as follows: Numerical Stats (16 doubles) -- 128 bytes Motion Map -- 1 byte / Macroblock Padding -- to align packet to 8 bytes The fpmotionmap.stt file can still be generated for debugging purposes in the same way that the textual version of the stats are available (defining OUTPUT_FPF in firstpass.c) Change-Id: I083ffbfd95e7d6a42bb4039ba0e81f678c8183ca
2010-10-14 22:40:12 +02:00
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
vpx_free(cpi->twopass.total_stats);
Fix a memory leak problem in encoder Deallocating the buffers before re-allocating them. The fix passed James Berry's test program for memory leak check. Change-Id: I18c3cf665412c0e313a523e3d435106c03ca438d
2010-12-06 23:21:37 +01:00
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
cpi->twopass.total_stats = vpx_calloc(1, sizeof(FIRSTPASS_STATS));
Fix a memory leak problem in encoder Deallocating the buffers before re-allocating them. The fix passed James Berry's test program for memory leak check. Change-Id: I18c3cf665412c0e313a523e3d435106c03ca438d
2010-12-06 23:21:37 +01:00
Two pass rate control code changes. This comitt brings accross changes from the public branch commit number Icf74d13af77437c08602571dc7a97e747cce5066. The main puurpose of this comit relates to CQ mode but it also includes some refactoring of the two pass code which I hope will make tuning the experimental branch for the new quantizer range a little less painfull. Change-Id: I278e989436a928fc1fe7761068960048f9d7a376
2011-11-23 18:18:31 +01:00
vpx_free(cpi->twopass.total_left_stats);
cpi->twopass.total_left_stats = vpx_calloc(1, sizeof(FIRSTPASS_STATS));
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
vpx_free(cpi->twopass.this_frame_stats);
Fix a memory leak problem in encoder Deallocating the buffers before re-allocating them. The fix passed James Berry's test program for memory leak check. Change-Id: I18c3cf665412c0e313a523e3d435106c03ca438d
2010-12-06 23:21:37 +01:00
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
cpi->twopass.this_frame_stats = vpx_calloc(1, sizeof(FIRSTPASS_STATS));
Fix a memory leak problem in encoder Deallocating the buffers before re-allocating them. The fix passed James Berry's test program for memory leak check. Change-Id: I18c3cf665412c0e313a523e3d435106c03ca438d
2010-12-06 23:21:37 +01:00
Two pass rate control code changes. This comitt brings accross changes from the public branch commit number Icf74d13af77437c08602571dc7a97e747cce5066. The main puurpose of this comit relates to CQ mode but it also includes some refactoring of the two pass code which I hope will make tuning the experimental branch for the new quantizer range a little less painfull. Change-Id: I278e989436a928fc1fe7761068960048f9d7a376
2011-11-23 18:18:31 +01:00
if( !cpi->twopass.total_stats ||
!cpi->twopass.total_left_stats ||
!cpi->twopass.this_frame_stats)
Move firstpass motion map to stats packet The first implementation of the firstpass motion map for motion compensated temporal filtering created a file, fpmotionmap.stt, in the current working directory. This was not safe for multiple encoder instances. This patch merges this data into the first pass stats packet interface, so that it is handled like the other (numerical) firstpass stats. The new stats packet is defined as follows: Numerical Stats (16 doubles) -- 128 bytes Motion Map -- 1 byte / Macroblock Padding -- to align packet to 8 bytes The fpmotionmap.stt file can still be generated for debugging purposes in the same way that the textual version of the stats are available (defining OUTPUT_FPF in firstpass.c) Change-Id: I083ffbfd95e7d6a42bb4039ba0e81f678c8183ca
2010-10-14 22:40:12 +02:00
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate firstpass stats");
Improved encoder threading Reduce the number of sync points by letting each thread continue imediatly with a new MB row. Better multicore scaling, improves performance by 5-20% on ARM multicore. Change-Id: Ic97e4d1c4886a842c85dd3539a93cb217188ed1b
2011-01-26 09:29:46 +01:00
mem leak fix for cpi->tplist checks added to make sure that cpi->tplist is freed correctly in vp8_dealloc_compressor_data and vp8_alloc_compressor_data. Change-Id: I66149dbbd25c958800ad94f4379d723191d9680d
2011-02-14 20:02:52 +01:00
vpx_free(cpi->tplist);
Put more code under #if CONFIG_MULTITHREAD. Change-Id: Icf4b692099d7d249fe3553852b1022b027b28e4b
2011-02-05 06:29:25 +01:00
CHECK_MEM_ERROR(cpi->tplist, vpx_malloc(sizeof(TOKENLIST) * cpi->common.mb_rows));
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Further QIndex realted Fixes: Added code to support 256 index steps instead of 128 but disabled for now. Replace hard wired table vp8cx_base_skip_false_prob[128] Observed Qindex problem with setting minimum loop filter value. (Experiment code using real Q in place but for now just returning 0. This has a big beneficial effect on some clips, particularly waterfall which shows 5% ssim gain) Change-Id: I2f7117de8adc1797164c106aa13effc900a1467e
2011-12-12 19:27:25 +01:00
// TODO perhaps change number of steps expose to outside world when setting
// max and min limits. Also this will likely want refining for the extended Q
// range.
//
// Table that converts 0-63 Q range values passed in outside to the Qindex
// range used internally.
static const int q_trans[] =
{
0, 4, 8, 12, 16, 20, 24, 28,
32, 36, 40, 44, 48, 52, 56, 60,
64, 68, 72, 76, 80, 84, 88, 92,
96, 100, 104, 108, 112, 116, 120, 124,
128, 132, 136, 140, 144, 148, 152, 156,
160, 164, 168, 172, 176, 180, 184, 188,
192, 196, 200, 204, 208, 212, 216, 220,
224, 228, 232, 236, 240, 244, 249, 255,
};
Initial WebM release
2010-05-18 17:58:33 +02:00
int vp8_reverse_trans(int x)
{
int i;
for (i = 0; i < 64; i++)
if (q_trans[i] >= x)
return i;
return 63;
};
void vp8_new_frame_rate(VP8_COMP *cpi, double framerate)
{
safety check to avoid divide by 0s
2010-10-14 22:19:06 +02:00
if(framerate < .1)
framerate = 30;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->oxcf.frame_rate = framerate;
cpi->output_frame_rate = cpi->oxcf.frame_rate;
cpi->per_frame_bandwidth = (int)(cpi->oxcf.target_bandwidth / cpi->output_frame_rate);
cpi->av_per_frame_bandwidth = (int)(cpi->oxcf.target_bandwidth / cpi->output_frame_rate);
cpi->min_frame_bandwidth = (int)(cpi->av_per_frame_bandwidth * cpi->oxcf.two_pass_vbrmin_section / 100);
Allow for frame overheads in min frame bandwidth. Change-Id: I6ade229ff400fe492709010ac5bada37f8afa73e
2012-03-02 02:24:30 +01:00
if (cpi->min_frame_bandwidth < FRAME_OVERHEAD_BITS )
cpi->min_frame_bandwidth = FRAME_OVERHEAD_BITS;
Further CQ, Key frame and ARF changes This code fixes a bug in the calculation of the minimum Q for alt ref frames. It also allows an extended gf/arf interval for sections of clips that completely static (or nearly so). Change-Id: I1a21aaa16d4f0578e5f99b13bebd78d59403c73b
2011-01-18 16:19:05 +01:00
// Set Maximum gf/arf interval
vp8e -removed undefined max call Change-Id: I42a86b0488f44115f09551fc5ad6d711fd470f0d
2011-01-18 17:14:17 +01:00
cpi->max_gf_interval = ((int)(cpi->output_frame_rate / 2.0) + 2);
Initial WebM release
2010-05-18 17:58:33 +02:00
vp8e -removed undefined max call Change-Id: I42a86b0488f44115f09551fc5ad6d711fd470f0d
2011-01-18 17:14:17 +01:00
if(cpi->max_gf_interval < 12)
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->max_gf_interval = 12;
Further CQ, Key frame and ARF changes This code fixes a bug in the calculation of the minimum Q for alt ref frames. It also allows an extended gf/arf interval for sections of clips that completely static (or nearly so). Change-Id: I1a21aaa16d4f0578e5f99b13bebd78d59403c73b
2011-01-18 16:19:05 +01:00
// Extended interval for genuinely static scenes
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
cpi->twopass.static_scene_max_gf_interval = cpi->key_frame_frequency >> 1;
Initial WebM release
2010-05-18 17:58:33 +02:00
Further CQ, Key frame and ARF changes This code fixes a bug in the calculation of the minimum Q for alt ref frames. It also allows an extended gf/arf interval for sections of clips that completely static (or nearly so). Change-Id: I1a21aaa16d4f0578e5f99b13bebd78d59403c73b
2011-01-18 16:19:05 +01:00
// Special conditions when altr ref frame enabled in lagged compress mode
Fix GF interval for non-lagged ARFs When ARFs are enabled in non-lagged compress modes, the GF interval was being reset to zero. Non-lagged ARF updates were enabled in commit 63ccfbd, but this incorrect GF interval caused a quality regression. Change-Id: I615c3b493f4ce2127044f4e68d0bcb07d6b730c3
2010-09-09 18:57:23 +02:00
if (cpi->oxcf.play_alternate && cpi->oxcf.lag_in_frames)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
if (cpi->max_gf_interval > cpi->oxcf.lag_in_frames - 1)
cpi->max_gf_interval = cpi->oxcf.lag_in_frames - 1;
Further CQ, Key frame and ARF changes This code fixes a bug in the calculation of the minimum Q for alt ref frames. It also allows an extended gf/arf interval for sections of clips that completely static (or nearly so). Change-Id: I1a21aaa16d4f0578e5f99b13bebd78d59403c73b
2011-01-18 16:19:05 +01:00
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
if (cpi->twopass.static_scene_max_gf_interval > cpi->oxcf.lag_in_frames - 1)
cpi->twopass.static_scene_max_gf_interval = cpi->oxcf.lag_in_frames - 1;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Further CQ, Key frame and ARF changes This code fixes a bug in the calculation of the minimum Q for alt ref frames. It also allows an extended gf/arf interval for sections of clips that completely static (or nearly so). Change-Id: I1a21aaa16d4f0578e5f99b13bebd78d59403c73b
2011-01-18 16:19:05 +01:00
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
if ( cpi->max_gf_interval > cpi->twopass.static_scene_max_gf_interval )
cpi->max_gf_interval = cpi->twopass.static_scene_max_gf_interval;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
increase rate control buffer level precision The external API exposes the RC initial/optimal/full buffer level in milliseconds, but this value was truncated internally to seconds. This patch allows the use of the full precision during the conversion from time to bits. Change-Id: If8dd2a87614c05747f81432cbe75dd9e6ed2f04e
2010-08-20 17:04:10 +02:00
static int
rescale(int val, int num, int denom)
{
int64_t llnum = num;
int64_t llden = denom;
int64_t llval = val;
return llval * llnum / llden;
}
Increase static linkage, remove unused functions A large number of functions were defined with external linkage, even though they were only used from within one file. This patch changes their linkage to static and removes the vp8_ prefix from their names, which should make it more obvious to the reader that the function is contained within the current translation unit. Functions that were not referenced were removed. These symbols were identified by: $ nm -A libvpx.a | sort -k3 | uniq -c -f2 | grep ' [A-Z] ' \ | sort | grep '^ *1 ' Change-Id: I59609f58ab65312012c047036ae1e0634f795779
2011-03-17 22:07:59 +01:00
static void init_config(VP8_PTR ptr, VP8_CONFIG *oxcf)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
VP8_COMP *cpi = (VP8_COMP *)(ptr);
VP8_COMMON *cm = &cpi->common;
Clean up of vp8_init_config() Clean up vp8_init_config() a bit and remove null pointer case, as this code can't be called any more and is not an adequate trap anyway, as a null pointer would cause exceptions before hitting the test. Change-Id: I937c00167cc039b3aa3f645f29c319d58ae8d3ee
2011-03-11 15:51:40 +01:00
cpi->oxcf = *oxcf;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->goldfreq = 7;
cm->version = oxcf->Version;
vp8_setup_version(cm);
1 Pass CQ and VBR bug fixes Issue 291 highlighted the fact that CQ mode was not working as expected in 1 pass mode, This commit fixes that specific problem but in so doing I also uncovered an overflow issue in the VBR code for 1 pass and some data values not being correctly initialized. For some clips (particularly short clips), the resulting improvement is dramatic. Change-Id: Ieefd6c6e4776eb8f1b0550dbfdfb72f86b33c960
2011-03-10 17:11:39 +01:00
// change includes all joint functionality
vp8_change_config(ptr, oxcf);
Initial WebM release
2010-05-18 17:58:33 +02:00
1 Pass CQ and VBR bug fixes Issue 291 highlighted the fact that CQ mode was not working as expected in 1 pass mode, This commit fixes that specific problem but in so doing I also uncovered an overflow issue in the VBR code for 1 pass and some data values not being correctly initialized. For some clips (particularly short clips), the resulting improvement is dramatic. Change-Id: Ieefd6c6e4776eb8f1b0550dbfdfb72f86b33c960
2011-03-10 17:11:39 +01:00
// Initialize active best and worst q and average q values.
cpi->active_worst_quality = cpi->oxcf.worst_allowed_q;
cpi->active_best_quality = cpi->oxcf.best_allowed_q;
cpi->avg_frame_qindex = cpi->oxcf.worst_allowed_q;
Initial WebM release
2010-05-18 17:58:33 +02:00
1 Pass CQ and VBR bug fixes Issue 291 highlighted the fact that CQ mode was not working as expected in 1 pass mode, This commit fixes that specific problem but in so doing I also uncovered an overflow issue in the VBR code for 1 pass and some data values not being correctly initialized. For some clips (particularly short clips), the resulting improvement is dramatic. Change-Id: Ieefd6c6e4776eb8f1b0550dbfdfb72f86b33c960
2011-03-10 17:11:39 +01:00
// Initialise the starting buffer levels
Configuration updates:Making a clear distinction between Init and Change Change-Id: I7b2fb326e1aabc08b032177a7b914a5b8bb7376f
2011-03-02 00:07:47 +01:00
cpi->buffer_level = cpi->oxcf.starting_buffer_level;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->bits_off_target = cpi->oxcf.starting_buffer_level;
Badly placed initialization of rolling rate monitors. This affects control of the active quantizer range. Change-Id: I30511fc81ac9f75ff20d9f1372382423d56739da
2010-09-24 18:52:55 +02:00
cpi->rolling_target_bits = cpi->av_per_frame_bandwidth;
cpi->rolling_actual_bits = cpi->av_per_frame_bandwidth;
Configuration updates:Making a clear distinction between Init and Change Change-Id: I7b2fb326e1aabc08b032177a7b914a5b8bb7376f
2011-03-02 00:07:47 +01:00
cpi->long_rolling_target_bits = cpi->av_per_frame_bandwidth;
cpi->long_rolling_actual_bits = cpi->av_per_frame_bandwidth;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->total_actual_bits = 0;
Configuration updates:Making a clear distinction between Init and Change Change-Id: I7b2fb326e1aabc08b032177a7b914a5b8bb7376f
2011-03-02 00:07:47 +01:00
cpi->total_target_vs_actual = 0;
Initial WebM release
2010-05-18 17:58:33 +02:00
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
cpi->static_mb_pct = 0;
Initial WebM release
2010-05-18 17:58:33 +02:00
#if VP8_TEMPORAL_ALT_REF
{
int i;
cpi->fixed_divide[0] = 0;
Made AltRef filter adaptive & added motion compensation Modified AltRef temporal filter to adapt filter length based on macroblock coding modes selected during first-pass encode. Also added sub-pixel motion compensation to the AltRef filter.
2010-09-28 16:23:41 +02:00
for (i = 1; i < 512; i++)
cpi->fixed_divide[i] = 0x80000 / i;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
#endif
}
void vp8_change_config(VP8_PTR ptr, VP8_CONFIG *oxcf)
{
VP8_COMP *cpi = (VP8_COMP *)(ptr);
VP8_COMMON *cm = &cpi->common;
if (!cpi)
return;
if (!oxcf)
return;
if (cm->version != oxcf->Version)
{
cm->version = oxcf->Version;
vp8_setup_version(cm);
}
cpi->oxcf = *oxcf;
switch (cpi->oxcf.Mode)
{
Code simplification Removal of the pickinter.c and .h files and calls to this code. Removal of some code relating to real time and one pass settings though there is more to be done in this regard. However, vp8_set_speed_features() now only supports modes 0 and 1 and speeds up to 3 so rd should always be set. Change-Id: I62c0c1b6154ab499785baef310536080e87bc4d8
2012-02-16 18:02:17 +01:00
// Real time and one pass deprecated in test code base
Initial WebM release
2010-05-18 17:58:33 +02:00
case MODE_FIRSTPASS:
cpi->pass = 1;
cpi->compressor_speed = 1;
break;
Code simplification Removal of the pickinter.c and .h files and calls to this code. Removal of some code relating to real time and one pass settings though there is more to be done in this regard. However, vp8_set_speed_features() now only supports modes 0 and 1 and speeds up to 3 so rd should always be set. Change-Id: I62c0c1b6154ab499785baef310536080e87bc4d8
2012-02-16 18:02:17 +01:00
Initial WebM release
2010-05-18 17:58:33 +02:00
case MODE_SECONDPASS:
cpi->pass = 2;
cpi->compressor_speed = 1;
if (cpi->oxcf.cpu_used < -5)
{
cpi->oxcf.cpu_used = -5;
}
if (cpi->oxcf.cpu_used > 5)
cpi->oxcf.cpu_used = 5;
break;
Code simplification Removal of the pickinter.c and .h files and calls to this code. Removal of some code relating to real time and one pass settings though there is more to be done in this regard. However, vp8_set_speed_features() now only supports modes 0 and 1 and speeds up to 3 so rd should always be set. Change-Id: I62c0c1b6154ab499785baef310536080e87bc4d8
2012-02-16 18:02:17 +01:00
Initial WebM release
2010-05-18 17:58:33 +02:00
case MODE_SECONDPASS_BEST:
cpi->pass = 2;
cpi->compressor_speed = 0;
break;
}
cpi->oxcf.worst_allowed_q = q_trans[oxcf->worst_allowed_q];
cpi->oxcf.best_allowed_q = q_trans[oxcf->best_allowed_q];
Fix CQ range and experimental KF sizing changes. The CQ level was not using the q_trans[] array to convert to a 0-127 range as per min and maxq Experimental change to try and match the reconstruction error for forced key frames approximately to that of the previous frame by means of the recode loop. Though this may cause extra recodes and the recode behavior has not been optimized, it can only happen on forced key frames. Change-Id: I1f7e42d526f1b1cb556dd461eff1a692bd1b5b2f
2011-01-17 18:23:11 +01:00
cpi->oxcf.cq_level = q_trans[cpi->oxcf.cq_level];
Initial WebM release
2010-05-18 17:58:33 +02:00
Further code simplification and clean up. Change-Id: Ifdb17b56090a317b2aa82cf125d57934902c5298
2012-02-25 00:27:59 +01:00
cpi->baseline_gf_interval = DEFAULT_GF_INTERVAL;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->ref_frame_flags = VP8_ALT_FLAG | VP8_GOLD_FLAG | VP8_LAST_FLAG;
//cpi->use_golden_frame_only = 0;
//cpi->use_last_frame_only = 0;
cm->refresh_golden_frame = 0;
cm->refresh_last_frame = 1;
cm->refresh_entropy_probs = 1;
setup_features(cpi);
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
#if CONFIG_HIGH_PRECISION_MV
Initial refactoring of high_precision mv code. This is the first patch for refactoring of the code related to high-precision mv, so that 1/4 and 1/8 pel motion vectors can co-exist in the same bit-stream by use of a frame level flag. The current patch works fine for only use of 1/4th and only use of 1/8th pel mv, but there are some issues with the mode switching in between. Subsequent patches on this change Id will fix the remaining issues. Patch 2: Adds fixes to make sure that multiple mv precisions can co-exist in the bit-stream. Frame level switching has been tested to work correctly. Patch 3: Fixes lines exceeding 80 char Patch 4: http://www.corp.google.com/~debargha/vp8_results/enhinterp.html Results on derf after ssse3 bugfix, compared to everything enabled but the 8-tap, 1/8-subpel and 1/16-subpel uv. Overall the gains are about 3% now. Hopefully there are no more bugs lingering. Apparently the sse3 bug affected the quartel subpel results more than the eighth pel ones (which is understandabale because one bad predictor due to the bug, matters less if there are a lot more subpel options available as in the 1/8 subpel case). The results in the 4th column correspond to the current settings. The first two columns correspond to two settings of adaptive switching of the 1/4 or 1/8 subpel mode based on initial Q estimate. These do not work as good as just using 1/8 all the time yet. Change-Id: I3ef392ad338329f4d68a85257a49f2b14f3af472
2012-02-27 19:22:38 +01:00
cpi->mb.e_mbd.allow_high_precision_mv = 0; // Default mv precision adaptation
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
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
{
int i;
for (i = 0; i < MAX_MB_SEGMENTS; i++)
cpi->segment_encode_breakout[i] = cpi->oxcf.encode_breakout;
}
// At the moment the first order values may not be > MAXQ
if (cpi->oxcf.fixed_q > MAXQ)
cpi->oxcf.fixed_q = MAXQ;
// local file playback mode == really big buffer
if (cpi->oxcf.end_usage == USAGE_LOCAL_FILE_PLAYBACK)
{
increase rate control buffer level precision The external API exposes the RC initial/optimal/full buffer level in milliseconds, but this value was truncated internally to seconds. This patch allows the use of the full precision during the conversion from time to bits. Change-Id: If8dd2a87614c05747f81432cbe75dd9e6ed2f04e
2010-08-20 17:04:10 +02:00
cpi->oxcf.starting_buffer_level = 60000;
cpi->oxcf.optimal_buffer_level = 60000;
cpi->oxcf.maximum_buffer_size = 240000;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
// Convert target bandwidth from Kbit/s to Bit/s
cpi->oxcf.target_bandwidth *= 1000;
Limit size of initial keyframe in one-pass. Rather than using a default size of 1/2 or 3/2 seconds for the first frame, use a fraction of the initial buffer level to give the application some control. This will likely undergo further refinement as size limits on key frames are currently under discussion on codec-devel@, but this gives much better behavior for small buffer sizes as a starting point. Change-Id: Ieba55b86517b81e51e6f0a9fe27aabba295acab0
2011-04-19 20:05:27 +02:00
cpi->oxcf.starting_buffer_level =
rescale(cpi->oxcf.starting_buffer_level,
cpi->oxcf.target_bandwidth, 1000);
1 Pass CQ and VBR bug fixes Issue 291 highlighted the fact that CQ mode was not working as expected in 1 pass mode, This commit fixes that specific problem but in so doing I also uncovered an overflow issue in the VBR code for 1 pass and some data values not being correctly initialized. For some clips (particularly short clips), the resulting improvement is dramatic. Change-Id: Ieefd6c6e4776eb8f1b0550dbfdfb72f86b33c960
2011-03-10 17:11:39 +01:00
// Set or reset optimal and maximum buffer levels.
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->oxcf.optimal_buffer_level == 0)
cpi->oxcf.optimal_buffer_level = cpi->oxcf.target_bandwidth / 8;
else
increase rate control buffer level precision The external API exposes the RC initial/optimal/full buffer level in milliseconds, but this value was truncated internally to seconds. This patch allows the use of the full precision during the conversion from time to bits. Change-Id: If8dd2a87614c05747f81432cbe75dd9e6ed2f04e
2010-08-20 17:04:10 +02:00
cpi->oxcf.optimal_buffer_level =
rescale(cpi->oxcf.optimal_buffer_level,
cpi->oxcf.target_bandwidth, 1000);
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->oxcf.maximum_buffer_size == 0)
cpi->oxcf.maximum_buffer_size = cpi->oxcf.target_bandwidth / 8;
else
increase rate control buffer level precision The external API exposes the RC initial/optimal/full buffer level in milliseconds, but this value was truncated internally to seconds. This patch allows the use of the full precision during the conversion from time to bits. Change-Id: If8dd2a87614c05747f81432cbe75dd9e6ed2f04e
2010-08-20 17:04:10 +02:00
cpi->oxcf.maximum_buffer_size =
rescale(cpi->oxcf.maximum_buffer_size,
cpi->oxcf.target_bandwidth, 1000);
Initial WebM release
2010-05-18 17:58:33 +02:00
1 Pass CQ and VBR bug fixes Issue 291 highlighted the fact that CQ mode was not working as expected in 1 pass mode, This commit fixes that specific problem but in so doing I also uncovered an overflow issue in the VBR code for 1 pass and some data values not being correctly initialized. For some clips (particularly short clips), the resulting improvement is dramatic. Change-Id: Ieefd6c6e4776eb8f1b0550dbfdfb72f86b33c960
2011-03-10 17:11:39 +01:00
// Set up frame rate and related parameters rate control values.
Initial WebM release
2010-05-18 17:58:33 +02:00
vp8_new_frame_rate(cpi, cpi->oxcf.frame_rate);
1 Pass CQ and VBR bug fixes Issue 291 highlighted the fact that CQ mode was not working as expected in 1 pass mode, This commit fixes that specific problem but in so doing I also uncovered an overflow issue in the VBR code for 1 pass and some data values not being correctly initialized. For some clips (particularly short clips), the resulting improvement is dramatic. Change-Id: Ieefd6c6e4776eb8f1b0550dbfdfb72f86b33c960
2011-03-10 17:11:39 +01:00
// Set absolute upper and lower quality limits
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->worst_quality = cpi->oxcf.worst_allowed_q;
cpi->best_quality = cpi->oxcf.best_allowed_q;
Configuration updates:Making a clear distinction between Init and Change Change-Id: I7b2fb326e1aabc08b032177a7b914a5b8bb7376f
2011-03-02 00:07:47 +01:00
// active values should only be modified if out of new range
if (cpi->active_worst_quality > cpi->oxcf.worst_allowed_q)
{
cpi->active_worst_quality = cpi->oxcf.worst_allowed_q;
}
// less likely
else if (cpi->active_worst_quality < cpi->oxcf.best_allowed_q)
{
cpi->active_worst_quality = cpi->oxcf.best_allowed_q;
}
if (cpi->active_best_quality < cpi->oxcf.best_allowed_q)
{
cpi->active_best_quality = cpi->oxcf.best_allowed_q;
}
// less likely
else if (cpi->active_best_quality > cpi->oxcf.worst_allowed_q)
{
cpi->active_best_quality = cpi->oxcf.worst_allowed_q;
}
CQ Mode The merge includes hooks to for CQ mode and other code changes merged from the test branch. CQ mode attempts to maintain a more stable quantizer within a clip whilst also trying to adhere to a guidline maximum bitrate. The existing target data rate parameter is used to specify the guideline maximum bitrate. A new parameter allows the user to specify a target CQ level. For normal (non kf/gf/arf) frames, the quantizer will not drop BELOW the user specified value (0-63). However, in some cases the encoder may choose to impose a target CQ that is above that specified by the user, if it estimates that consistent use of the target value is not compatible with guideline maximum bitrate. Change-Id: I2221f9eecae8cc3c431d36caf83503941b25e4c1
2011-01-07 19:29:37 +01:00
Configuration updates:Making a clear distinction between Init and Change Change-Id: I7b2fb326e1aabc08b032177a7b914a5b8bb7376f
2011-03-02 00:07:47 +01:00
cpi->buffered_mode = (cpi->oxcf.optimal_buffer_level > 0) ? TRUE : FALSE;
Initial WebM release
2010-05-18 17:58:33 +02:00
Configuration updates:Making a clear distinction between Init and Change Change-Id: I7b2fb326e1aabc08b032177a7b914a5b8bb7376f
2011-03-02 00:07:47 +01:00
cpi->cq_target_quality = cpi->oxcf.cq_level;
Initial WebM release
2010-05-18 17:58:33 +02:00
if (!cm->use_bilinear_mc_filter)
Making subpel filters switchable at frame level Various refactoring to make the subpel motion compensation filters switchable by a frame level field. Two types of 8-tap filters are supported in addition to the existing bilinar and sixtap filters. One is the default 8-tap and the other has a sharper cut-off for use with frames with substantial edge content. Patch 2: Added a preliminary strategy for filter selection based on edginess detecton. Also includes some filter changes. Change-Id: I866085bda5ae143cfdf2ec88157feaabdf7bd63a
2012-03-19 15:53:05 +01:00
#if CONFIG_ENHANCED_INTERP
cm->mcomp_filter_type = EIGHTTAP;
#else
Initial WebM release
2010-05-18 17:58:33 +02:00
cm->mcomp_filter_type = SIXTAP;
Making subpel filters switchable at frame level Various refactoring to make the subpel motion compensation filters switchable by a frame level field. Two types of 8-tap filters are supported in addition to the existing bilinar and sixtap filters. One is the default 8-tap and the other has a sharper cut-off for use with frames with substantial edge content. Patch 2: Added a preliminary strategy for filter selection based on edginess detecton. Also includes some filter changes. Change-Id: I866085bda5ae143cfdf2ec88157feaabdf7bd63a
2012-03-19 15:53:05 +01:00
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
else
cm->mcomp_filter_type = BILINEAR;
cpi->target_bandwidth = cpi->oxcf.target_bandwidth;
cm->Width = cpi->oxcf.Width ;
cm->Height = cpi->oxcf.Height ;
cm->horiz_scale = cpi->horiz_scale;
cm->vert_scale = cpi->vert_scale ;
// VP8 sharpness level mapping 0-7 (vs 0-10 in general VPx dialogs)
if (cpi->oxcf.Sharpness > 7)
cpi->oxcf.Sharpness = 7;
cm->sharpness_level = cpi->oxcf.Sharpness;
if (cm->horiz_scale != NORMAL || cm->vert_scale != NORMAL)
{
int UNINITIALIZED_IS_SAFE(hr), UNINITIALIZED_IS_SAFE(hs);
int UNINITIALIZED_IS_SAFE(vr), UNINITIALIZED_IS_SAFE(vs);
Scale2Ratio(cm->horiz_scale, &hr, &hs);
Scale2Ratio(cm->vert_scale, &vr, &vs);
// always go to the next whole number
cm->Width = (hs - 1 + cpi->oxcf.Width * hr) / hs;
cm->Height = (vs - 1 + cpi->oxcf.Height * vr) / vs;
}
1 Pass CQ and VBR bug fixes Issue 291 highlighted the fact that CQ mode was not working as expected in 1 pass mode, This commit fixes that specific problem but in so doing I also uncovered an overflow issue in the VBR code for 1 pass and some data values not being correctly initialized. For some clips (particularly short clips), the resulting improvement is dramatic. Change-Id: Ieefd6c6e4776eb8f1b0550dbfdfb72f86b33c960
2011-03-10 17:11:39 +01:00
if (((cm->Width + 15) & 0xfffffff0) !=
cm->yv12_fb[cm->lst_fb_idx].y_width ||
((cm->Height + 15) & 0xfffffff0) !=
cm->yv12_fb[cm->lst_fb_idx].y_height ||
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
cm->yv12_fb[cm->lst_fb_idx].y_width == 0)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
alloc_raw_frame_buffers(cpi);
vp8_alloc_compressor_data(cpi);
}
if (cpi->oxcf.fixed_q >= 0)
{
cpi->last_q[0] = cpi->oxcf.fixed_q;
cpi->last_q[1] = cpi->oxcf.fixed_q;
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
cpi->last_boosted_qindex = cpi->oxcf.fixed_q;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
cpi->Speed = cpi->oxcf.cpu_used;
// force to allowlag to 0 if lag_in_frames is 0;
if (cpi->oxcf.lag_in_frames == 0)
{
cpi->oxcf.allow_lag = 0;
}
// Limit on lag buffers as these are not currently dynamically allocated
else if (cpi->oxcf.lag_in_frames > MAX_LAG_BUFFERS)
cpi->oxcf.lag_in_frames = MAX_LAG_BUFFERS;
// YX Temp
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
cpi->alt_ref_source = NULL;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->is_src_frame_alt_ref = 0;
remove a variable no longer in use The variable is introduced in commit 2e53e9e53 to make more use of trellis quantization, but this is no longer necessary after RDMULT was made adaptive in a number of later commits. Change-Id: I7420522ec7723f38cf77033466c25afb405d52ae
2011-05-11 04:57:51 +02:00
Initial WebM release
2010-05-18 17:58:33 +02:00
#if 0
// Experimental RD Code
cpi->frame_distortion = 0;
cpi->last_frame_distortion = 0;
#endif
}
#define M_LOG2_E 0.693147180559945309417
#define log2f(x) (log (x) / (float) M_LOG2_E)
static void cal_mvsadcosts(int *mvsadcost[2])
{
int i = 1;
mvsadcost [0] [0] = 300;
mvsadcost [1] [0] = 300;
do
{
Use full-pixel MV in mvsadcost calculation MV sad cost error is only used in full-pixel motion search, which only need full-pixel resolution instead of quarter-pixel resolution. This change reduced mvsadcost table size, and removed unneccessary pamameter passing since this table is constant once it is generated. Change-Id: I9f931e55f6abc3c99011321f1dfb2f3562e6f6b0
2011-04-01 22:41:58 +02:00
double z = 256 * (2 * (log2f(8 * i) + .6));
Initial WebM release
2010-05-18 17:58:33 +02:00
mvsadcost [0][i] = (int) z;
mvsadcost [1][i] = (int) z;
mvsadcost [0][-i] = (int) z;
mvsadcost [1][-i] = (int) z;
}
Use full-pixel MV in mvsadcost calculation MV sad cost error is only used in full-pixel motion search, which only need full-pixel resolution instead of quarter-pixel resolution. This change reduced mvsadcost table size, and removed unneccessary pamameter passing since this table is constant once it is generated. Change-Id: I9f931e55f6abc3c99011321f1dfb2f3562e6f6b0
2011-04-01 22:41:58 +02:00
while (++i <= mvfp_max);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Initial refactoring of high_precision mv code. This is the first patch for refactoring of the code related to high-precision mv, so that 1/4 and 1/8 pel motion vectors can co-exist in the same bit-stream by use of a frame level flag. The current patch works fine for only use of 1/4th and only use of 1/8th pel mv, but there are some issues with the mode switching in between. Subsequent patches on this change Id will fix the remaining issues. Patch 2: Adds fixes to make sure that multiple mv precisions can co-exist in the bit-stream. Frame level switching has been tested to work correctly. Patch 3: Fixes lines exceeding 80 char Patch 4: http://www.corp.google.com/~debargha/vp8_results/enhinterp.html Results on derf after ssse3 bugfix, compared to everything enabled but the 8-tap, 1/8-subpel and 1/16-subpel uv. Overall the gains are about 3% now. Hopefully there are no more bugs lingering. Apparently the sse3 bug affected the quartel subpel results more than the eighth pel ones (which is understandabale because one bad predictor due to the bug, matters less if there are a lot more subpel options available as in the 1/8 subpel case). The results in the 4th column correspond to the current settings. The first two columns correspond to two settings of adaptive switching of the 1/4 or 1/8 subpel mode based on initial Q estimate. These do not work as good as just using 1/8 all the time yet. Change-Id: I3ef392ad338329f4d68a85257a49f2b14f3af472
2012-02-27 19:22:38 +01:00
#if CONFIG_HIGH_PRECISION_MV
static void cal_mvsadcosts_hp(int *mvsadcost[2])
{
int i = 1;
mvsadcost [0] [0] = 300;
mvsadcost [1] [0] = 300;
do
{
double z = 256 * (2 * (log2f(8 * i) + .6));
mvsadcost [0][i] = (int) z;
mvsadcost [1][i] = (int) z;
mvsadcost [0][-i] = (int) z;
mvsadcost [1][-i] = (int) z;
}
while (++i <= mvfp_max_hp);
}
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
VP8_PTR vp8_create_compressor(VP8_CONFIG *oxcf)
{
int i;
volatile union
{
VP8_COMP *cpi;
VP8_PTR ptr;
} ctx;
Delta updates to segmentation map using left and above contexts. -Updates by making use of spatial correlation. -Checks if the segment_id is same as above or left context and encodes only the update to the map instead of updating individual segment_ids. Change-Id: Ib861df97e8aa2b37516219eeddcdbaf552b6a249
2010-09-01 02:43:14 +02:00
Initial WebM release
2010-05-18 17:58:33 +02:00
VP8_COMP *cpi;
VP8_COMMON *cm;
cpi = ctx.cpi = vpx_memalign(32, sizeof(VP8_COMP));
// Check that the CPI instance is valid
if (!cpi)
return 0;
cm = &cpi->common;
vpx_memset(cpi, 0, sizeof(VP8_COMP));
if (setjmp(cm->error.jmp))
{
VP8_PTR ptr = ctx.ptr;
ctx.cpi->common.error.setjmp = 0;
vp8_remove_compressor(&ptr);
return 0;
}
cpi->common.error.setjmp = 1;
CHECK_MEM_ERROR(cpi->mb.ss, vpx_calloc(sizeof(search_site), (MAX_MVSEARCH_STEPS * 8) + 1));
vp8_create_common(&cpi->common);
Add runtime CPU detection support for ARM. The primary goal is to allow a binary to be built which supports NEON, but can fall back to non-NEON routines, since some Android devices do not have NEON, even if they are otherwise ARMv7 (e.g., Tegra). The configure-generated flags HAVE_ARMV7, etc., are used to decide which versions of each function to build, and when CONFIG_RUNTIME_CPU_DETECT is enabled, the correct version is chosen at run time. In order for this to work, the CFLAGS must be set to something appropriate (e.g., without -mfpu=neon for ARMv7, and with appropriate -march and -mcpu for even earlier configurations), or the native C code will not be able to run. The ASFLAGS must remain set for the most advanced instruction set required at build time, since the ARM assembler will refuse to emit them otherwise. I have not attempted to make any changes to configure to do this automatically. Doing so will probably require the addition of new configure options. Many of the hooks for RTCD on ARM were already there, but a lot of the code had bit-rotted, and a good deal of the ARM-specific code is not integrated into the RTCD structs at all. I did not try to resolve the latter, merely to add the minimal amount of protection around them to allow RTCD to work. Those functions that were called based on an ifdef at the calling site were expanded to check the RTCD flags at that site, but they should be added to an RTCD struct somewhere in the future. The functions invoked with global function pointers still are, but these should be moved into an RTCD struct for thread safety (I believe every platform currently supported has atomic pointer stores, but this is not guaranteed). The encoder's boolhuff functions did not even have _c and armv7 suffixes, and the correct version was resolved at link time. The token packing functions did have appropriate suffixes, but the version was selected with a define, with no associated RTCD struct. However, for both of these, the only armv7 instruction they actually used was rbit, and this was completely superfluous, so I reworked them to avoid it. The only non-ARMv4 instruction remaining in them is clz, which is ARMv5 (not even ARMv5TE is required). Considering that there are no ARM-specific configs which are not at least ARMv5TE, I did not try to detect these at runtime, and simply enable them for ARMv5 and above. Finally, the NEON register saving code was completely non-reentrant, since it saved the registers to a global, static variable. I moved the storage for this onto the stack. A single binary built with this code was tested on an ARM11 (ARMv6) and a Cortex A8 (ARMv7 w/NEON), for both the encoder and decoder, and produced identical output, while using the correct accelerated functions on each. I did not test on any earlier processors. Change-Id: I45cbd63a614f4554c3b325c45d46c0806f009eaa
2010-10-21 00:39:11 +02:00
vp8_cmachine_specific_config(cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
Increase static linkage, remove unused functions A large number of functions were defined with external linkage, even though they were only used from within one file. This patch changes their linkage to static and removes the vp8_ prefix from their names, which should make it more obvious to the reader that the function is contained within the current translation unit. Functions that were not referenced were removed. These symbols were identified by: $ nm -A libvpx.a | sort -k3 | uniq -c -f2 | grep ' [A-Z] ' \ | sort | grep '^ *1 ' Change-Id: I59609f58ab65312012c047036ae1e0634f795779
2011-03-17 22:07:59 +01:00
init_config((VP8_PTR)cpi, oxcf);
Initial WebM release
2010-05-18 17:58:33 +02:00
memcpy(cpi->base_skip_false_prob, vp8cx_base_skip_false_prob, sizeof(vp8cx_base_skip_false_prob));
cpi->common.current_video_frame = 0;
cpi->kf_overspend_bits = 0;
cpi->kf_bitrate_adjustment = 0;
cpi->frames_till_gf_update_due = 0;
cpi->gf_overspend_bits = 0;
cpi->non_gf_bitrate_adjustment = 0;
Moved some reference frame data structures into common. Encoder side changes Change-Id: I8921800e4fccec5e5a9e4755b80cbd472623107b
2012-01-28 09:50:33 +01:00
cm->prob_last_coded = 128;
cm->prob_gf_coded = 128;
cm->prob_intra_coded = 63;
Rename "dual" prediction to "compound" prediction. Change-Id: Ibcd2b9b247ff9f83331dac47f91ec285e8955ff1
2012-02-29 02:25:45 +01:00
for ( i = 0; i < COMP_PRED_CONTEXTS; i++ )
cm->prob_comppred[i] = 128;
Initial WebM release
2010-05-18 17:58:33 +02:00
// Prime the recent reference frame useage counters.
// Hereafter they will be maintained as a sort of moving average
cpi->recent_ref_frame_usage[INTRA_FRAME] = 1;
cpi->recent_ref_frame_usage[LAST_FRAME] = 1;
cpi->recent_ref_frame_usage[GOLDEN_FRAME] = 1;
cpi->recent_ref_frame_usage[ALTREF_FRAME] = 1;
// Set reference frame sign bias for ALTREF frame to 1 (for now)
cpi->common.ref_frame_sign_bias[ALTREF_FRAME] = 1;
cpi->baseline_gf_interval = DEFAULT_GF_INTERVAL;
cpi->gold_is_last = 0 ;
cpi->alt_is_last = 0 ;
cpi->gold_is_alt = 0 ;
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
// allocate memory for storing last frame's MVs for MV prediction.
Refine motion vector prediction for NEWMV mode Adjust checking points in motion vector prediction to better cover possible movements, and get a better prediction. Tests on test clips showed a 0.1% improvement in SSIM, and no change in PSNR and performance. Change-Id: Ifdab05d35e10faea1445c61bb73debf888c9d2f8
2011-01-25 21:54:34 +01:00
CHECK_MEM_ERROR(cpi->lfmv, vpx_calloc((cpi->common.mb_rows+2) * (cpi->common.mb_cols+2), sizeof(int_mv)));
CHECK_MEM_ERROR(cpi->lf_ref_frame_sign_bias, vpx_calloc((cpi->common.mb_rows+2) * (cpi->common.mb_cols+2), sizeof(int)));
CHECK_MEM_ERROR(cpi->lf_ref_frame, vpx_calloc((cpi->common.mb_rows+2) * (cpi->common.mb_cols+2), sizeof(int)));
Initial WebM release
2010-05-18 17:58:33 +02:00
// Create the encoder segmentation map and set all entries to 0
Delta updates to segmentation map using left and above contexts. -Updates by making use of spatial correlation. -Checks if the segment_id is same as above or left context and encodes only the update to the map instead of updating individual segment_ids. Change-Id: Ib861df97e8aa2b37516219eeddcdbaf552b6a249
2010-09-01 02:43:14 +02:00
CHECK_MEM_ERROR(cpi->segmentation_map, vpx_calloc((cpi->common.mb_rows * cpi->common.mb_cols), 1));
Segmentation experiment: Added last_segmentation_map[] structure to keep track of what we had before when doing temporal prediction. With this change the existing code does once again appear to be giving a decodable bitstream for both temporal and standard prediction modes. However, it is still somewhat messy and confused and there is no option to take advantage of spatial prediction so it could do with further work. Some housekeeping / clean out. Change-Id: I368258243f82127b81d8dffa7ada615208513b47
2011-11-11 19:08:06 +01:00
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
// And a copy in common for temporal coding
Moved some segmentation data structures. Moved some segmentation data structures into VP8_COMMON Change-Id: I59c6e2edf7a0176e35319936eea450027aeb3b39
2012-01-27 19:29:07 +01:00
CHECK_MEM_ERROR(cm->last_frame_seg_map,
Segmentation experiment: Added last_segmentation_map[] structure to keep track of what we had before when doing temporal prediction. With this change the existing code does once again appear to be giving a decodable bitstream for both temporal and standard prediction modes. However, it is still somewhat messy and confused and there is no option to take advantage of spatial prediction so it could do with further work. Some housekeeping / clean out. Change-Id: I368258243f82127b81d8dffa7ada615208513b47
2011-11-11 19:08:06 +01:00
vpx_calloc((cpi->common.mb_rows * cpi->common.mb_cols), 1));
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
// And a place holder structure is the coding context
// for use if we want to save and restore it
CHECK_MEM_ERROR(cpi->coding_context.last_frame_seg_map_copy,
vpx_calloc((cpi->common.mb_rows * cpi->common.mb_cols), 1));
Initial WebM release
2010-05-18 17:58:33 +02:00
CHECK_MEM_ERROR(cpi->active_map, vpx_calloc(cpi->common.mb_rows * cpi->common.mb_cols, 1));
vpx_memset(cpi->active_map , 1, (cpi->common.mb_rows * cpi->common.mb_cols));
cpi->active_map_enabled = 0;
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
for (i = 0; i < ( sizeof(cpi->mbgraph_stats) /
sizeof(cpi->mbgraph_stats[0]) ); i++)
{
CHECK_MEM_ERROR(cpi->mbgraph_stats[i].mb_stats,
vpx_calloc(cpi->common.mb_rows * cpi->common.mb_cols *
sizeof(*cpi->mbgraph_stats[i].mb_stats),
1));
}
Initial WebM release
2010-05-18 17:58:33 +02:00
#ifdef ENTROPY_STATS
Added the ability to accumulate coef stats across encodings This commit added the ability to accumulate the coef stats across different encodings using an intermediate binary stats files. The accumulation happens only the binary stats file exists in current directory. The encoder needs to be built with "ENTROPY_STATS" to allow the output. The commit also fixed a few formating issues in output stats file. Change-Id: Ib1a41180aa554845cf51e4421a230b128a3a82b4
2012-05-08 21:38:39 +02:00
if(cpi->pass != 1)
init_context_counters();
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
#ifdef MODE_STATS
vp8_zero(y_modes);
vp8_zero(i8x8_modes);
vp8_zero(uv_modes);
vp8_zero(uv_modes_y);
vp8_zero(b_modes);
vp8_zero(inter_y_modes);
vp8_zero(inter_uv_modes);
vp8_zero(inter_b_modes);
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
Add simple version of activity masking. This uses MB variance to change the RDO weight for mode decision and quantization. Activity is normalized against the average for the frame, which is currently tracked using feed-forward statistics. This could also be used to adjust the quantizer for the entire frame, but that requires more extensive rate control changes. This does not yet attempt to adapt the quantizer within the frame, but the signaling cost means that will likely only be useful at very high rates. Change-Id: I26cd7c755cac3ff33cfe0688b1da50b2b87b9c93
2010-10-11 23:37:27 +02:00
/*Initialize the feed-forward activity masking.*/
cpi->activity_avg = 90<<12;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->frames_since_key = 8; // Give a sensible default for the first frame.
cpi->key_frame_frequency = cpi->oxcf.key_freq;
Limit key frame quantizer for forced key frames. Where a key frame occurs because of a minimum interval selected by the user, then these forced key frames ideally need to be more closely matched in quality to the surrounding frame. Change-Id: Ia55b1f047e77dc7fbd78379c45869554f25b3df7
2011-01-12 18:08:42 +01:00
cpi->this_key_frame_forced = FALSE;
cpi->next_key_frame_forced = FALSE;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->source_alt_ref_pending = FALSE;
cpi->source_alt_ref_active = FALSE;
cpi->common.refresh_alt_ref_frame = 0;
changed configure option name to reduce confusion Renamed configure option "enable-psnr" to "enable-internal-stats" to better reflect the purpose of the option and eliminate the confusion reported in http://code.google.com/p/webm/issues/detail?id=35 Change-Id: If72df6fdb9f1e33dab1329240ba4d8911d2f1f7a
2011-04-29 18:37:59 +02:00
cpi->b_calculate_psnr = CONFIG_INTERNAL_STATS;
#if CONFIG_INTERNAL_STATS
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->b_calculate_ssimg = 0;
cpi->count = 0;
cpi->bytes = 0;
if (cpi->b_calculate_psnr)
{
cpi->total_sq_error = 0.0;
cpi->total_sq_error2 = 0.0;
cpi->total_y = 0.0;
cpi->total_u = 0.0;
cpi->total_v = 0.0;
cpi->total = 0.0;
cpi->totalp_y = 0.0;
cpi->totalp_u = 0.0;
cpi->totalp_v = 0.0;
cpi->totalp = 0.0;
cpi->tot_recode_hits = 0;
cpi->summed_quality = 0;
cpi->summed_weights = 0;
}
if (cpi->b_calculate_ssimg)
{
cpi->total_ssimg_y = 0;
cpi->total_ssimg_u = 0;
cpi->total_ssimg_v = 0;
cpi->total_ssimg_all = 0;
}
Initialize first_time_stamp_ever Misplaced #endif caused first_time_stamp_ever to only be initialized if CONFIG_INTERNAL_STATS was set. Change-Id: I2296a4ab00f7dfb767583edcc5d59b94f48c0621
2011-05-31 18:37:45 +02:00
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
#ifndef LLONG_MAX
#define LLONG_MAX 9223372036854775807LL
#endif
cpi->first_time_stamp_ever = LLONG_MAX;
cpi->frames_till_gf_update_due = 0;
cpi->key_frame_count = 1;
cpi->ni_av_qi = cpi->oxcf.worst_allowed_q;
cpi->ni_tot_qi = 0;
cpi->ni_frames = 0;
Some further QIndex issues with extended Q Resolved or factored out some further issues with Q index. Put in a 3rd order polynomial instead of less accurate power function as the best fit on gf and kf boost adjustment. Added avg_q value to use instead of ni_av_qi. Compute segment delta Q values based on avg_q. Fixed bug in adjust_maxq_qrange(). The extended range Q on the derf set, using standard data rates (which do not extend high enough to get big benefits) still show a shortfall of between 0.5 and 1% though so there would appear to be further issues that need to be tracked down. Change-Id: Icfd49b9f401906ba487ef1bef7d397048295d959
2011-12-06 15:48:52 +01:00
cpi->tot_q = 0.0;
cpi->avg_q = vp8_convert_qindex_to_q( cpi->oxcf.worst_allowed_q );
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->total_byte_count = 0;
cpi->rate_correction_factor = 1.0;
cpi->key_frame_rate_correction_factor = 1.0;
cpi->gf_rate_correction_factor = 1.0;
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
cpi->twopass.est_max_qcorrection_factor = 1.0;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->mb.mvcost[0] = &cpi->mb.mvcosts[0][mv_max+1];
cpi->mb.mvcost[1] = &cpi->mb.mvcosts[1][mv_max+1];
Use full-pixel MV in mvsadcost calculation MV sad cost error is only used in full-pixel motion search, which only need full-pixel resolution instead of quarter-pixel resolution. This change reduced mvsadcost table size, and removed unneccessary pamameter passing since this table is constant once it is generated. Change-Id: I9f931e55f6abc3c99011321f1dfb2f3562e6f6b0
2011-04-01 22:41:58 +02:00
cpi->mb.mvsadcost[0] = &cpi->mb.mvsadcosts[0][mvfp_max+1];
cpi->mb.mvsadcost[1] = &cpi->mb.mvsadcosts[1][mvfp_max+1];
Initial WebM release
2010-05-18 17:58:33 +02:00
cal_mvsadcosts(cpi->mb.mvsadcost);
Initial refactoring of high_precision mv code. This is the first patch for refactoring of the code related to high-precision mv, so that 1/4 and 1/8 pel motion vectors can co-exist in the same bit-stream by use of a frame level flag. The current patch works fine for only use of 1/4th and only use of 1/8th pel mv, but there are some issues with the mode switching in between. Subsequent patches on this change Id will fix the remaining issues. Patch 2: Adds fixes to make sure that multiple mv precisions can co-exist in the bit-stream. Frame level switching has been tested to work correctly. Patch 3: Fixes lines exceeding 80 char Patch 4: http://www.corp.google.com/~debargha/vp8_results/enhinterp.html Results on derf after ssse3 bugfix, compared to everything enabled but the 8-tap, 1/8-subpel and 1/16-subpel uv. Overall the gains are about 3% now. Hopefully there are no more bugs lingering. Apparently the sse3 bug affected the quartel subpel results more than the eighth pel ones (which is understandabale because one bad predictor due to the bug, matters less if there are a lot more subpel options available as in the 1/8 subpel case). The results in the 4th column correspond to the current settings. The first two columns correspond to two settings of adaptive switching of the 1/4 or 1/8 subpel mode based on initial Q estimate. These do not work as good as just using 1/8 all the time yet. Change-Id: I3ef392ad338329f4d68a85257a49f2b14f3af472
2012-02-27 19:22:38 +01:00
#if CONFIG_HIGH_PRECISION_MV
cpi->mb.mvcost_hp[0] = &cpi->mb.mvcosts_hp[0][mv_max_hp+1];
cpi->mb.mvcost_hp[1] = &cpi->mb.mvcosts_hp[1][mv_max_hp+1];
cpi->mb.mvsadcost_hp[0] = &cpi->mb.mvsadcosts_hp[0][mvfp_max_hp+1];
cpi->mb.mvsadcost_hp[1] = &cpi->mb.mvsadcosts_hp[1][mvfp_max_hp+1];
cal_mvsadcosts_hp(cpi->mb.mvsadcost_hp);
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
for (i = 0; i < KEY_FRAME_CONTEXT; i++)
{
cpi->prior_key_frame_distance[i] = (int)cpi->output_frame_rate;
}
#ifdef OUTPUT_YUV_SRC
yuv_file = fopen("bd.yuv", "ab");
#endif
Add 8x8 transform to experimental branch Please refer to previous commit messages for detailed info: https://on2-git.corp.google.com/g/#change,5940 https://on2-git.corp.google.com/g/#change,6045 Change-Id: I8b16992f2f69c5a808ad40a3e32ef589cce7c59d
2011-02-14 23:18:18 +01:00
#ifdef OUTPUT_YUV_REC
yuv_rec_file = fopen("rec.yuv", "wb");
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
#if 0
framepsnr = fopen("framepsnr.stt", "a");
kf_list = fopen("kf_list.stt", "w");
#endif
cpi->output_pkt_list = oxcf->output_pkt_list;
if (cpi->pass == 1)
{
vp8_init_first_pass(cpi);
}
else if (cpi->pass == 2)
{
Removed firstpass motion map The firstpass motion map consists of an 8-bit flag for each MB indicating how strongly the firstpass code believes it should be filtered during the second pass ARNR filtering. For long or large format material the motion map can become extremely large and hamper the operation of the encoding process. This change removes the motion map altogether, leaving the second pass to rely on the magnitude of the motion compensated error to determine the filter weight to use for the MB during ARNR filtering. Tests on the derf set indicate that the effect of this change is neutral, with some small wins and losses. The motion map has therefore been removed based on a cost/benefit evaluation. Change-Id: I53e07d236f5ce09a6f0c54e7c4ffbb490fb870f6
2011-03-10 20:32:48 +01:00
size_t packet_sz = sizeof(FIRSTPASS_STATS);
Move firstpass motion map to stats packet The first implementation of the firstpass motion map for motion compensated temporal filtering created a file, fpmotionmap.stt, in the current working directory. This was not safe for multiple encoder instances. This patch merges this data into the first pass stats packet interface, so that it is handled like the other (numerical) firstpass stats. The new stats packet is defined as follows: Numerical Stats (16 doubles) -- 128 bytes Motion Map -- 1 byte / Macroblock Padding -- to align packet to 8 bytes The fpmotionmap.stt file can still be generated for debugging purposes in the same way that the textual version of the stats are available (defining OUTPUT_FPF in firstpass.c) Change-Id: I083ffbfd95e7d6a42bb4039ba0e81f678c8183ca
2010-10-14 22:40:12 +02:00
int packets = oxcf->two_pass_stats_in.sz / packet_sz;
Change to arf boost calculation. In this commit I have added an experimental function that tests prediction quality either side of a central position to calculate a suggested boost number for an ARF frame. The function is passed an offset from the current position and a number of frames to search forwards and backwards. It returns a forward, backward and compound boost number. The new code can be deactivated using #define NEW_BOOST 0 In its current default state the code searches forwards and backwards from the proposed position of the next alt ref. The the old code used a boost number calculated by scanning forward from the previous GF up to the proposed alt ref frame position. I have also added some code to try and prevent placement of a gf/arf where there is a brief flash. Change-Id: I98af789a5181148659f10dd5dd2ff2d4250cd51c
2011-06-28 18:29:47 +02:00
cpi->twopass.stats_in_start = oxcf->two_pass_stats_in.buf;
cpi->twopass.stats_in = cpi->twopass.stats_in_start;
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
cpi->twopass.stats_in_end = (void*)((char *)cpi->twopass.stats_in
Move firstpass motion map to stats packet The first implementation of the firstpass motion map for motion compensated temporal filtering created a file, fpmotionmap.stt, in the current working directory. This was not safe for multiple encoder instances. This patch merges this data into the first pass stats packet interface, so that it is handled like the other (numerical) firstpass stats. The new stats packet is defined as follows: Numerical Stats (16 doubles) -- 128 bytes Motion Map -- 1 byte / Macroblock Padding -- to align packet to 8 bytes The fpmotionmap.stt file can still be generated for debugging purposes in the same way that the textual version of the stats are available (defining OUTPUT_FPF in firstpass.c) Change-Id: I083ffbfd95e7d6a42bb4039ba0e81f678c8183ca
2010-10-14 22:40:12 +02:00
+ (packets - 1) * packet_sz);
Initial WebM release
2010-05-18 17:58:33 +02:00
vp8_init_second_pass(cpi);
}
vp8_set_speed_features(cpi);
// Set starting values of RD threshold multipliers (128 = *1)
for (i = 0; i < MAX_MODES; i++)
{
cpi->rd_thresh_mult[i] = 128;
}
#ifdef ENTROPY_STATS
init_mv_ref_counts();
#endif
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
cpi->fn_ptr[BLOCK_16X16].sdf = VARIANCE_INVOKE(&cpi->rtcd.variance, sad16x16);
cpi->fn_ptr[BLOCK_16X16].vf = VARIANCE_INVOKE(&cpi->rtcd.variance, var16x16);
cpi->fn_ptr[BLOCK_16X16].svf = VARIANCE_INVOKE(&cpi->rtcd.variance, subpixvar16x16);
cpi->fn_ptr[BLOCK_16X16].svf_halfpix_h = VARIANCE_INVOKE(&cpi->rtcd.variance, halfpixvar16x16_h);
cpi->fn_ptr[BLOCK_16X16].svf_halfpix_v = VARIANCE_INVOKE(&cpi->rtcd.variance, halfpixvar16x16_v);
cpi->fn_ptr[BLOCK_16X16].svf_halfpix_hv = VARIANCE_INVOKE(&cpi->rtcd.variance, halfpixvar16x16_hv);
cpi->fn_ptr[BLOCK_16X16].sdx3f = VARIANCE_INVOKE(&cpi->rtcd.variance, sad16x16x3);
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
cpi->fn_ptr[BLOCK_16X16].sdx8f = VARIANCE_INVOKE(&cpi->rtcd.variance, sad16x16x8);
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
cpi->fn_ptr[BLOCK_16X16].sdx4df = VARIANCE_INVOKE(&cpi->rtcd.variance, sad16x16x4d);
cpi->fn_ptr[BLOCK_16X8].sdf = VARIANCE_INVOKE(&cpi->rtcd.variance, sad16x8);
cpi->fn_ptr[BLOCK_16X8].vf = VARIANCE_INVOKE(&cpi->rtcd.variance, var16x8);
cpi->fn_ptr[BLOCK_16X8].svf = VARIANCE_INVOKE(&cpi->rtcd.variance, subpixvar16x8);
cpi->fn_ptr[BLOCK_16X8].svf_halfpix_h = NULL;
cpi->fn_ptr[BLOCK_16X8].svf_halfpix_v = NULL;
cpi->fn_ptr[BLOCK_16X8].svf_halfpix_hv = NULL;
cpi->fn_ptr[BLOCK_16X8].sdx3f = VARIANCE_INVOKE(&cpi->rtcd.variance, sad16x8x3);
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
cpi->fn_ptr[BLOCK_16X8].sdx8f = VARIANCE_INVOKE(&cpi->rtcd.variance, sad16x8x8);
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
cpi->fn_ptr[BLOCK_16X8].sdx4df = VARIANCE_INVOKE(&cpi->rtcd.variance, sad16x8x4d);
cpi->fn_ptr[BLOCK_8X16].sdf = VARIANCE_INVOKE(&cpi->rtcd.variance, sad8x16);
cpi->fn_ptr[BLOCK_8X16].vf = VARIANCE_INVOKE(&cpi->rtcd.variance, var8x16);
cpi->fn_ptr[BLOCK_8X16].svf = VARIANCE_INVOKE(&cpi->rtcd.variance, subpixvar8x16);
cpi->fn_ptr[BLOCK_8X16].svf_halfpix_h = NULL;
cpi->fn_ptr[BLOCK_8X16].svf_halfpix_v = NULL;
cpi->fn_ptr[BLOCK_8X16].svf_halfpix_hv = NULL;
cpi->fn_ptr[BLOCK_8X16].sdx3f = VARIANCE_INVOKE(&cpi->rtcd.variance, sad8x16x3);
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
cpi->fn_ptr[BLOCK_8X16].sdx8f = VARIANCE_INVOKE(&cpi->rtcd.variance, sad8x16x8);
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
cpi->fn_ptr[BLOCK_8X16].sdx4df = VARIANCE_INVOKE(&cpi->rtcd.variance, sad8x16x4d);
cpi->fn_ptr[BLOCK_8X8].sdf = VARIANCE_INVOKE(&cpi->rtcd.variance, sad8x8);
cpi->fn_ptr[BLOCK_8X8].vf = VARIANCE_INVOKE(&cpi->rtcd.variance, var8x8);
cpi->fn_ptr[BLOCK_8X8].svf = VARIANCE_INVOKE(&cpi->rtcd.variance, subpixvar8x8);
cpi->fn_ptr[BLOCK_8X8].svf_halfpix_h = NULL;
cpi->fn_ptr[BLOCK_8X8].svf_halfpix_v = NULL;
cpi->fn_ptr[BLOCK_8X8].svf_halfpix_hv = NULL;
cpi->fn_ptr[BLOCK_8X8].sdx3f = VARIANCE_INVOKE(&cpi->rtcd.variance, sad8x8x3);
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
cpi->fn_ptr[BLOCK_8X8].sdx8f = VARIANCE_INVOKE(&cpi->rtcd.variance, sad8x8x8);
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
cpi->fn_ptr[BLOCK_8X8].sdx4df = VARIANCE_INVOKE(&cpi->rtcd.variance, sad8x8x4d);
cpi->fn_ptr[BLOCK_4X4].sdf = VARIANCE_INVOKE(&cpi->rtcd.variance, sad4x4);
cpi->fn_ptr[BLOCK_4X4].vf = VARIANCE_INVOKE(&cpi->rtcd.variance, var4x4);
cpi->fn_ptr[BLOCK_4X4].svf = VARIANCE_INVOKE(&cpi->rtcd.variance, subpixvar4x4);
cpi->fn_ptr[BLOCK_4X4].svf_halfpix_h = NULL;
cpi->fn_ptr[BLOCK_4X4].svf_halfpix_v = NULL;
cpi->fn_ptr[BLOCK_4X4].svf_halfpix_hv = NULL;
cpi->fn_ptr[BLOCK_4X4].sdx3f = VARIANCE_INVOKE(&cpi->rtcd.variance, sad4x4x3);
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
cpi->fn_ptr[BLOCK_4X4].sdx8f = VARIANCE_INVOKE(&cpi->rtcd.variance, sad4x4x8);
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
cpi->fn_ptr[BLOCK_4X4].sdx4df = VARIANCE_INVOKE(&cpi->rtcd.variance, sad4x4x4d);
Initial WebM release
2010-05-18 17:58:33 +02:00
Preload reference area to an intermediate buffer in sub-pixel motion search In sub-pixel motion search, the search range is small(+/- 3 pixels). Preload whole search area from reference buffer into a 32-byte aligned buffer. Then in search, load reference data from this buffer instead. This keeps data in cache, and reduces the crossing cache- line penalty. For tulip clip, tests on Intel Core2 Quad machine(linux) showed encoder speed improvement: 3.4% at --rt --cpu-used =-4 2.8% at --rt --cpu-used =-3 2.3% at --rt --cpu-used =-2 2.2% at --rt --cpu-used =-1 Test on Atom notebook showed only 1.1% speed improvement(speed=-4). Test on Xeon machine also showed less improvement, since unaligned data access latency is greatly reduced in newer cores. Next, I will apply similar idea to other 2 sub-pixel search functions for encoding speed > 4. Make this change exclusively for x86 platforms. Change-Id: Ia7bb9f56169eac0f01009fe2b2f2ab5b61d2eb2f
2011-06-28 15:14:13 +02:00
#if ARCH_X86 || ARCH_X86_64
cpi->fn_ptr[BLOCK_16X16].copymem = VARIANCE_INVOKE(&cpi->rtcd.variance, copy32xn);
cpi->fn_ptr[BLOCK_16X8].copymem = VARIANCE_INVOKE(&cpi->rtcd.variance, copy32xn);
cpi->fn_ptr[BLOCK_8X16].copymem = VARIANCE_INVOKE(&cpi->rtcd.variance, copy32xn);
cpi->fn_ptr[BLOCK_8X8].copymem = VARIANCE_INVOKE(&cpi->rtcd.variance, copy32xn);
cpi->fn_ptr[BLOCK_4X4].copymem = VARIANCE_INVOKE(&cpi->rtcd.variance, copy32xn);
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->full_search_sad = SEARCH_INVOKE(&cpi->rtcd.search, full_search);
cpi->diamond_search_sad = SEARCH_INVOKE(&cpi->rtcd.search, diamond_search);
Use diamond search to replace full search in full-pixel refining search In NEWMV mode, currently, full search is used as the refining search after n-step search. By replacing it with an iterative diamond search of radius 1 largely reduced the computation complexity, but still maintained the same encoding quality since the refining search is done for every macroblock instead of only a small precentage of macroblocks while using full search. Tests on the test set showed a 3.4% encoding speed increase with none psnr & ssim loss. Change-Id: Ife907d7eb9544d15c34f17dc6e4cfd97cb743d41
2011-05-06 18:51:31 +02:00
cpi->refining_search_sad = SEARCH_INVOKE(&cpi->rtcd.search, refining_search);
Initial WebM release
2010-05-18 17:58:33 +02:00
// make sure frame 1 is okay
cpi->error_bins[0] = cpi->common.MBs;
//vp8cx_init_quantizer() is first called here. Add check in vp8cx_frame_init_quantizer() so that vp8cx_init_quantizer is only called later
//when needed. This will avoid unnecessary calls of vp8cx_init_quantizer() for every frame.
vp8cx_init_quantizer(cpi);
fix sharpness bug and clean up sharpness was not recalculated in vp8cx_pick_filter_level_fast remove last_filter_type. all values are calculated, don't need to update the lfi data when it changes. always use cm->sharpness_level. the extra indirection was annoying. don't track last frame_type or sharpness_level manually. frame type only matters for motion search and sharpness_level is taken care of in frame_init move function declarations to their proper header Change-Id: I7ef037bd4bf8cf5e37d2d36bd03b5e22a2ad91db
2011-07-20 21:53:42 +02:00
vp8_loop_filter_init(cm);
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->common.error.setjmp = 0;
Enabled adaptive UV intra coding for inter frames Previously, Y-adaptive UV intra coding only enabled on key frames in UVINTRA experiment. This commit enabled the same coding for inter frames, so the encoding of UV intra modes are consistent cross all frame types. Tests on derf set showed a very small overall gain around .04%: http://www.corp.google.com/~yaowu/no_crawl/interUVintra.html The gain looks to be reasonable given inta coded MBs is only a small portion of MBs in inter frames. Change-Id: Ic6fc261923f2c253f4a0c9f8bccf4797557b9e16
2011-12-08 20:43:09 +01:00
vp8_zero(cpi->y_uv_mode_count)
Initial WebM release
2010-05-18 17:58:33 +02:00
return (VP8_PTR) cpi;
}
void vp8_remove_compressor(VP8_PTR *ptr)
{
VP8_COMP *cpi = (VP8_COMP *)(*ptr);
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
int i;
Initial WebM release
2010-05-18 17:58:33 +02:00
if (!cpi)
return;
if (cpi && (cpi->common.current_video_frame > 0))
{
if (cpi->pass == 2)
{
vp8_end_second_pass(cpi);
}
#ifdef ENTROPY_STATS
Added the ability to accumulate coef stats across encodings This commit added the ability to accumulate the coef stats across different encodings using an intermediate binary stats files. The accumulation happens only the binary stats file exists in current directory. The encoder needs to be built with "ENTROPY_STATS" to allow the output. The commit also fixed a few formating issues in output stats file. Change-Id: Ib1a41180aa554845cf51e4421a230b128a3a82b4
2012-05-08 21:38:39 +02:00
if(cpi->pass != 1)
{
Initial WebM release
2010-05-18 17:58:33 +02:00
print_context_counters();
print_tree_update_probs();
print_mode_context();
Added the ability to accumulate coef stats across encodings This commit added the ability to accumulate the coef stats across different encodings using an intermediate binary stats files. The accumulation happens only the binary stats file exists in current directory. The encoder needs to be built with "ENTROPY_STATS" to allow the output. The commit also fixed a few formating issues in output stats file. Change-Id: Ib1a41180aa554845cf51e4421a230b128a3a82b4
2012-05-08 21:38:39 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
changed configure option name to reduce confusion Renamed configure option "enable-psnr" to "enable-internal-stats" to better reflect the purpose of the option and eliminate the confusion reported in http://code.google.com/p/webm/issues/detail?id=35 Change-Id: If72df6fdb9f1e33dab1329240ba4d8911d2f1f7a
2011-04-29 18:37:59 +02:00
#if CONFIG_INTERNAL_STATS
Initial WebM release
2010-05-18 17:58:33 +02:00
Added an emms to prevent invalid stats output In certain hardware configuration, where mmx code is enabled and other simd (sse2/sse3) disabled, lacking of this emms caused invalid internal stats outputs. Change-Id: I77c61cf6e0448d3f3b8c11781aa9e42f31d231c9
2012-01-05 20:22:06 +01:00
vp8_clear_system_state();
Merge t8x8 experiments Change-Id: I8e9b6b154e1a0d0cb42d596366380d69c00ac15f
2012-02-29 02:11:12 +01:00
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
//printf("\n8x8-4x4:%d-%d\n", cpi->t8x8_count, cpi->t4x4_count);
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->pass != 1)
{
FILE *f = fopen("opsnr.stt", "a");
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
double time_encoded = (cpi->last_end_time_stamp_seen
- cpi->first_time_stamp_ever) / 10000000.000;
Initial WebM release
2010-05-18 17:58:33 +02:00
double total_encode_time = (cpi->time_receive_data + cpi->time_compress_data) / 1000.000;
double dr = (double)cpi->bytes * (double) 8 / (double)1000 / time_encoded;
Merged NEWNEAR experiment This commit merges the NEWNEAR experiment such that it is effectively always on. The fact that there were changes in the threading code again highlights the need to strip out such features during the bitstream development phase as trying to maintain this code (especially as it is not being tested) slows the development cycle. Change-Id: I8b34950a1333231ced9928aa11cd6d6459984b65
2012-02-03 16:10:44 +01:00
#if defined(MODE_STATS)
Revised coding using adaptive mode context to depend on frame type A previous commit 76feb965 made the vp8_mode_context adaptive on a frame frame basis, this commit further made the coding context adaptive to two frame types separately. Tests on derf set showed a further small gain on all metrics: avg psnr 0.10%, glb psnr: 0.11%, ssim: 0.08% http://www.corp.google.com/~yaowu/no_crawl/newNearMode_1209.html Change-Id: I7b3e32ec8729de1903d14a3f1213f1624b78cdee
2011-12-08 20:43:09 +01:00
print_mode_contexts(&cpi->common);
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->b_calculate_psnr)
{
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
YV12_BUFFER_CONFIG *lst_yv12 = &cpi->common.yv12_fb[cpi->common.lst_fb_idx];
double samples = 3.0 / 2 * cpi->count * lst_yv12->y_width * lst_yv12->y_height;
Initial WebM release
2010-05-18 17:58:33 +02:00
double total_psnr = vp8_mse2psnr(samples, 255.0, cpi->total_sq_error);
double total_psnr2 = vp8_mse2psnr(samples, 255.0, cpi->total_sq_error2);
double total_ssim = 100 * pow(cpi->summed_quality / cpi->summed_weights, 8.0);
Fix stats format and correct data size and bit rate output Change-ID: I093abe6094589a0d73f6ca85b825678a19e68285
2010-05-27 19:32:40 +02:00
fprintf(f, "Bitrate\tAVGPsnr\tGLBPsnr\tAVPsnrP\tGLPsnrP\tVPXSSIM\t Time(us)\n");
Use consistent delimiters. opsnr.stt file was using \t for delimiters on everything except between VPXSSIM and Time. Change-Id: I6284c4e40c05ff642bf4b0170dca062c279a42df
2011-04-14 00:06:17 +02:00
fprintf(f, "%7.3f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t%8.0f\n",
Initial WebM release
2010-05-18 17:58:33 +02:00
dr, cpi->total / cpi->count, total_psnr, cpi->totalp / cpi->count, total_psnr2, total_ssim,
total_encode_time);
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
// fprintf(f, "%7.3f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t%8.0f %10ld\n",
// dr, cpi->total / cpi->count, total_psnr, cpi->totalp / cpi->count, total_psnr2, total_ssim,
// total_encode_time, cpi->tot_recode_hits);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
if (cpi->b_calculate_ssimg)
{
fprintf(f, "BitRate\tSSIM_Y\tSSIM_U\tSSIM_V\tSSIM_A\t Time(us)\n");
fprintf(f, "%7.3f\t%6.4f\t%6.4f\t%6.4f\t%6.4f\t%8.0f\n", dr,
cpi->total_ssimg_y / cpi->count, cpi->total_ssimg_u / cpi->count,
cpi->total_ssimg_v / cpi->count, cpi->total_ssimg_all / cpi->count, total_encode_time);
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
// fprintf(f, "%7.3f\t%6.4f\t%6.4f\t%6.4f\t%6.4f\t%8.0f %10ld\n", dr,
// cpi->total_ssimg_y / cpi->count, cpi->total_ssimg_u / cpi->count,
// cpi->total_ssimg_v / cpi->count, cpi->total_ssimg_all / cpi->count, total_encode_time, cpi->tot_recode_hits);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
fclose(f);
}
#endif
#ifdef MODE_STATS
{
extern int count_mb_seg[4];
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
char modes_stats_file[250];
changed find_near_mvs search to include a mb from last frame This is an experiment to include a mv contribution from last frame to nearest and near mv definition. Initial test showed some small though consistent gain. latest patch slightly better result ~.13%-~.18%. TODO: the entropy used to encode the mode choice, i.e. the mv counts based conditional distribution of modes should be re-collected to reflect this change, it is expected that there is some further gain from that. Change-Id: Ief1e284a36d8aa56b49ae5b360c91419ec494fa4
2011-11-16 01:16:30 +01:00
FILE *f;
Add 8x8 transform to experimental branch Please refer to previous commit messages for detailed info: https://on2-git.corp.google.com/g/#change,5940 https://on2-git.corp.google.com/g/#change,6045 Change-Id: I8b16992f2f69c5a808ad40a3e32ef589cce7c59d
2011-02-14 23:18:18 +01:00
double dr = (double)cpi->oxcf.frame_rate * (double)cpi->bytes * (double)8 / (double)cpi->count / (double)1000 ;
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
sprintf(modes_stats_file, "modes_q%03d.stt",cpi->common.base_qindex);
f = fopen(modes_stats_file, "w");
Initial WebM release
2010-05-18 17:58:33 +02:00
fprintf(f, "intra_mode in Intra Frames:\n");
make uv intra mode coding adaptive to Y mode This commit tries to do UV intra mode coding adaptive to Y intra mode. Entropy context is defined as conditional PDF of uv intra mode given the Y mode. All constants are normalized with 256 to be fit in 8 bits. This provides further coding efficiency beyond the quantizer adaptive y intra mode coding. Consistent gains were observed on all clips and all bit rates for HD all key encoding tests. To test, configure with --enable-experimental --enable-uvintra Change-Id: I2d78d73f143127f063e19bd0bac3b68c418d756a
2011-08-05 01:30:27 +02:00
{
int i;
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
fprintf(f, "Y: ");
for (i=0;i<VP8_YMODES;i++) fprintf(f, " %8d,", y_modes[i]);
fprintf(f, "\n");
}
{
int i;
fprintf(f, "I8: ");
for (i=0;i<VP8_I8X8_MODES;i++) fprintf(f, " %8d,", i8x8_modes[i]);
fprintf(f, "\n");
}
{
int i;
fprintf(f, "UV: ");
for (i=0;i<VP8_UV_MODES;i++) fprintf(f, " %8d,", uv_modes[i]);
fprintf(f, "\n");
}
{
int i, j;
fprintf(f, "KeyFrame Y-UV:\n");
make uv intra mode coding adaptive to Y mode This commit tries to do UV intra mode coding adaptive to Y intra mode. Entropy context is defined as conditional PDF of uv intra mode given the Y mode. All constants are normalized with 256 to be fit in 8 bits. This provides further coding efficiency beyond the quantizer adaptive y intra mode coding. Consistent gains were observed on all clips and all bit rates for HD all key encoding tests. To test, configure with --enable-experimental --enable-uvintra Change-Id: I2d78d73f143127f063e19bd0bac3b68c418d756a
2011-08-05 01:30:27 +02:00
for(i=0;i<VP8_YMODES;i++)
{
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
fprintf(f, "%2d:", i);
for (j=0; j<VP8_UV_MODES;j++) fprintf(f, "%8d, ",uv_modes_y[i][j]);
fprintf(f, "\n");
make uv intra mode coding adaptive to Y mode This commit tries to do UV intra mode coding adaptive to Y intra mode. Entropy context is defined as conditional PDF of uv intra mode given the Y mode. All constants are normalized with 256 to be fit in 8 bits. This provides further coding efficiency beyond the quantizer adaptive y intra mode coding. Consistent gains were observed on all clips and all bit rates for HD all key encoding tests. To test, configure with --enable-experimental --enable-uvintra Change-Id: I2d78d73f143127f063e19bd0bac3b68c418d756a
2011-08-05 01:30:27 +02:00
}
}
Enabled adaptive UV intra coding for inter frames Previously, Y-adaptive UV intra coding only enabled on key frames in UVINTRA experiment. This commit enabled the same coding for inter frames, so the encoding of UV intra modes are consistent cross all frame types. Tests on derf set showed a very small overall gain around .04%: http://www.corp.google.com/~yaowu/no_crawl/interUVintra.html The gain looks to be reasonable given inta coded MBs is only a small portion of MBs in inter frames. Change-Id: Ic6fc261923f2c253f4a0c9f8bccf4797557b9e16
2011-12-08 20:43:09 +01:00
{
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
int i, j;
fprintf(f, "Inter Y-UV:\n");
Enabled adaptive UV intra coding for inter frames Previously, Y-adaptive UV intra coding only enabled on key frames in UVINTRA experiment. This commit enabled the same coding for inter frames, so the encoding of UV intra modes are consistent cross all frame types. Tests on derf set showed a very small overall gain around .04%: http://www.corp.google.com/~yaowu/no_crawl/interUVintra.html The gain looks to be reasonable given inta coded MBs is only a small portion of MBs in inter frames. Change-Id: Ic6fc261923f2c253f4a0c9f8bccf4797557b9e16
2011-12-08 20:43:09 +01:00
for(i=0;i<VP8_YMODES;i++)
{
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
fprintf(f, "%2d:", i);
for (j=0; j<VP8_UV_MODES;j++) fprintf(f, "%8d, ",cpi->y_uv_mode_count[i][j]);
fprintf(f, "\n");
Enabled adaptive UV intra coding for inter frames Previously, Y-adaptive UV intra coding only enabled on key frames in UVINTRA experiment. This commit enabled the same coding for inter frames, so the encoding of UV intra modes are consistent cross all frame types. Tests on derf set showed a very small overall gain around .04%: http://www.corp.google.com/~yaowu/no_crawl/interUVintra.html The gain looks to be reasonable given inta coded MBs is only a small portion of MBs in inter frames. Change-Id: Ic6fc261923f2c253f4a0c9f8bccf4797557b9e16
2011-12-08 20:43:09 +01:00
}
}
Initial WebM release
2010-05-18 17:58:33 +02:00
{
int i;
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
fprintf(f, "B: ");
for (i = 0; i < VP8_BINTRAMODES; i++)
Initial WebM release
2010-05-18 17:58:33 +02:00
fprintf(f, "%8d, ", b_modes[i]);
fprintf(f, "\n");
}
fprintf(f, "Modes in Inter Frames:\n");
{
int i;
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
fprintf(f, "Y: ");
for (i=0;i<MB_MODE_COUNT;i++) fprintf(f, " %8d,", inter_y_modes[i]);
fprintf(f, "\n");
}
{
int i;
fprintf(f, "UV: ");
for (i=0;i<VP8_UV_MODES;i++) fprintf(f, " %8d,", inter_uv_modes[i]);
fprintf(f, "\n");
}
{
int i;
fprintf(f, "B: ");
for (i = 0; i < B_MODE_COUNT; i++) fprintf(f, "%8d, ", inter_b_modes[i]);
Initial WebM release
2010-05-18 17:58:33 +02:00
fprintf(f, "\n");
}
fprintf(f, "P:%8d, %8d, %8d, %8d\n", count_mb_seg[0], count_mb_seg[1], count_mb_seg[2], count_mb_seg[3]);
fprintf(f, "PB:%8d, %8d, %8d, %8d\n", inter_b_modes[LEFT4X4], inter_b_modes[ABOVE4X4], inter_b_modes[ZERO4X4], inter_b_modes[NEW4X4]);
fclose(f);
}
#endif
#ifdef ENTROPY_STATS
{
int i, j, k;
FILE *fmode = fopen("modecontext.c", "w");
fprintf(fmode, "\n#include \"entropymode.h\"\n\n");
fprintf(fmode, "const unsigned int vp8_kf_default_bmode_counts ");
fprintf(fmode, "[VP8_BINTRAMODES] [VP8_BINTRAMODES] [VP8_BINTRAMODES] =\n{\n");
for (i = 0; i < 10; i++)
{
fprintf(fmode, " { //Above Mode : %d\n", i);
for (j = 0; j < 10; j++)
{
fprintf(fmode, " {");
Adds new Directional Intra prediction modes. Adds 6 directional intra predictiom modes for 16x16 and 8x8 blocks. Change-Id: I25eccc0836f28d8d74922e4e9231568a648b47d1
2012-05-15 01:21:01 +02:00
for (k = 0; k < VP8_BINTRAMODES; k++)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
if (!intra_mode_stats[i][j][k])
fprintf(fmode, " %5d, ", 1);
else
fprintf(fmode, " %5d, ", intra_mode_stats[i][j][k]);
}
fprintf(fmode, "}, // left_mode %d\n", j);
}
fprintf(fmode, " },\n");
}
fprintf(fmode, "};\n");
Fix leaked file descriptor with ENTROPY_STATS cppcheck found a leaked file descriptor in the debugging code enabled by defining ENTROPY_STATS. Fixes issue #60. Change-Id: I0c1d0669cb94d44fed77860f97b82763be06b7cb
2010-10-25 19:14:11 +02:00
fclose(fmode);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
#endif
#if defined(SECTIONBITS_OUTPUT)
if (0)
{
int i;
FILE *f = fopen("tokenbits.stt", "a");
for (i = 0; i < 28; i++)
fprintf(f, "%8d", (int)(Sectionbits[i] / 256));
fprintf(f, "\n");
fclose(f);
}
#endif
#if 0
{
printf("\n_pick_loop_filter_level:%d\n", cpi->time_pick_lpf / 1000);
printf("\n_frames recive_data encod_mb_row compress_frame Total\n");
printf("%6d %10ld %10ld %10ld %10ld\n", cpi->common.current_video_frame, cpi->time_receive_data / 1000, cpi->time_encode_mb_row / 1000, cpi->time_compress_data / 1000, (cpi->time_receive_data + cpi->time_compress_data) / 1000);
}
#endif
}
Increase static linkage, remove unused functions A large number of functions were defined with external linkage, even though they were only used from within one file. This patch changes their linkage to static and removes the vp8_ prefix from their names, which should make it more obvious to the reader that the function is contained within the current translation unit. Functions that were not referenced were removed. These symbols were identified by: $ nm -A libvpx.a | sort -k3 | uniq -c -f2 | grep ' [A-Z] ' \ | sort | grep '^ *1 ' Change-Id: I59609f58ab65312012c047036ae1e0634f795779
2011-03-17 22:07:59 +01:00
dealloc_compressor_data(cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
vpx_free(cpi->mb.ss);
vpx_free(cpi->tok);
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
for (i = 0; i < sizeof(cpi->mbgraph_stats) / sizeof(cpi->mbgraph_stats[0]); i++)
{
vpx_free(cpi->mbgraph_stats[i].mb_stats);
}
Initial WebM release
2010-05-18 17:58:33 +02:00
vp8_remove_common(&cpi->common);
vpx_free(cpi);
*ptr = 0;
#ifdef OUTPUT_YUV_SRC
fclose(yuv_file);
#endif
Add 8x8 transform to experimental branch Please refer to previous commit messages for detailed info: https://on2-git.corp.google.com/g/#change,5940 https://on2-git.corp.google.com/g/#change,6045 Change-Id: I8b16992f2f69c5a808ad40a3e32ef589cce7c59d
2011-02-14 23:18:18 +01:00
#ifdef OUTPUT_YUV_REC
fclose(yuv_rec_file);
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
#if 0
if (keyfile)
fclose(keyfile);
if (framepsnr)
fclose(framepsnr);
if (kf_list)
fclose(kf_list);
#endif
}
static uint64_t calc_plane_error(unsigned char *orig, int orig_stride,
unsigned char *recon, int recon_stride,
unsigned int cols, unsigned int rows,
vp8_variance_rtcd_vtable_t *rtcd)
{
unsigned int row, col;
uint64_t total_sse = 0;
int diff;
for (row = 0; row + 16 <= rows; row += 16)
{
for (col = 0; col + 16 <= cols; col += 16)
{
unsigned int sse;
VARIANCE_INVOKE(rtcd, mse16x16)(orig + col, orig_stride,
recon + col, recon_stride,
&sse);
total_sse += sse;
}
/* Handle odd-sized width */
if (col < cols)
{
unsigned int border_row, border_col;
unsigned char *border_orig = orig;
unsigned char *border_recon = recon;
for (border_row = 0; border_row < 16; border_row++)
{
for (border_col = col; border_col < cols; border_col++)
{
diff = border_orig[border_col] - border_recon[border_col];
total_sse += diff * diff;
}
border_orig += orig_stride;
border_recon += recon_stride;
}
}
orig += orig_stride * 16;
recon += recon_stride * 16;
}
/* Handle odd-sized height */
for (; row < rows; row++)
{
for (col = 0; col < cols; col++)
{
diff = orig[col] - recon[col];
total_sse += diff * diff;
}
orig += orig_stride;
recon += recon_stride;
}
return total_sse;
}
static void generate_psnr_packet(VP8_COMP *cpi)
{
YV12_BUFFER_CONFIG *orig = cpi->Source;
YV12_BUFFER_CONFIG *recon = cpi->common.frame_to_show;
struct vpx_codec_cx_pkt pkt;
uint64_t sse;
int i;
unsigned int width = cpi->common.Width;
unsigned int height = cpi->common.Height;
pkt.kind = VPX_CODEC_PSNR_PKT;
sse = calc_plane_error(orig->y_buffer, orig->y_stride,
recon->y_buffer, recon->y_stride,
width, height,
IF_RTCD(&cpi->rtcd.variance));
pkt.data.psnr.sse[0] = sse;
pkt.data.psnr.sse[1] = sse;
pkt.data.psnr.samples[0] = width * height;
pkt.data.psnr.samples[1] = width * height;
width = (width + 1) / 2;
height = (height + 1) / 2;
sse = calc_plane_error(orig->u_buffer, orig->uv_stride,
recon->u_buffer, recon->uv_stride,
width, height,
IF_RTCD(&cpi->rtcd.variance));
pkt.data.psnr.sse[0] += sse;
pkt.data.psnr.sse[2] = sse;
pkt.data.psnr.samples[0] += width * height;
pkt.data.psnr.samples[2] = width * height;
sse = calc_plane_error(orig->v_buffer, orig->uv_stride,
recon->v_buffer, recon->uv_stride,
width, height,
IF_RTCD(&cpi->rtcd.variance));
pkt.data.psnr.sse[0] += sse;
pkt.data.psnr.sse[3] = sse;
pkt.data.psnr.samples[0] += width * height;
pkt.data.psnr.samples[3] = width * height;
for (i = 0; i < 4; i++)
pkt.data.psnr.psnr[i] = vp8_mse2psnr(pkt.data.psnr.samples[i], 255.0,
pkt.data.psnr.sse[i]);
vpx_codec_pkt_list_add(cpi->output_pkt_list, &pkt);
}
int vp8_use_as_reference(VP8_PTR ptr, int ref_frame_flags)
{
VP8_COMP *cpi = (VP8_COMP *)(ptr);
if (ref_frame_flags > 7)
return -1 ;
cpi->ref_frame_flags = ref_frame_flags;
return 0;
}
int vp8_update_reference(VP8_PTR ptr, int ref_frame_flags)
{
VP8_COMP *cpi = (VP8_COMP *)(ptr);
if (ref_frame_flags > 7)
return -1 ;
cpi->common.refresh_golden_frame = 0;
cpi->common.refresh_alt_ref_frame = 0;
cpi->common.refresh_last_frame = 0;
if (ref_frame_flags & VP8_LAST_FLAG)
cpi->common.refresh_last_frame = 1;
if (ref_frame_flags & VP8_GOLD_FLAG)
cpi->common.refresh_golden_frame = 1;
if (ref_frame_flags & VP8_ALT_FLAG)
cpi->common.refresh_alt_ref_frame = 1;
return 0;
}
int vp8_get_reference(VP8_PTR ptr, VP8_REFFRAME ref_frame_flag, YV12_BUFFER_CONFIG *sd)
{
VP8_COMP *cpi = (VP8_COMP *)(ptr);
VP8_COMMON *cm = &cpi->common;
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
int ref_fb_idx;
Initial WebM release
2010-05-18 17:58:33 +02:00
if (ref_frame_flag == VP8_LAST_FLAG)
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
ref_fb_idx = cm->lst_fb_idx;
Initial WebM release
2010-05-18 17:58:33 +02:00
else if (ref_frame_flag == VP8_GOLD_FLAG)
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
ref_fb_idx = cm->gld_fb_idx;
Initial WebM release
2010-05-18 17:58:33 +02:00
else if (ref_frame_flag == VP8_ALT_FLAG)
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
ref_fb_idx = cm->alt_fb_idx;
Initial WebM release
2010-05-18 17:58:33 +02:00
else
return -1;
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
vp8_yv12_copy_frame_ptr(&cm->yv12_fb[ref_fb_idx], sd);
Initial WebM release
2010-05-18 17:58:33 +02:00
return 0;
}
int vp8_set_reference(VP8_PTR ptr, VP8_REFFRAME ref_frame_flag, YV12_BUFFER_CONFIG *sd)
{
VP8_COMP *cpi = (VP8_COMP *)(ptr);
VP8_COMMON *cm = &cpi->common;
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
int ref_fb_idx;
Initial WebM release
2010-05-18 17:58:33 +02:00
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
if (ref_frame_flag == VP8_LAST_FLAG)
ref_fb_idx = cm->lst_fb_idx;
Initial WebM release
2010-05-18 17:58:33 +02:00
else if (ref_frame_flag == VP8_GOLD_FLAG)
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
ref_fb_idx = cm->gld_fb_idx;
Initial WebM release
2010-05-18 17:58:33 +02:00
else if (ref_frame_flag == VP8_ALT_FLAG)
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
ref_fb_idx = cm->alt_fb_idx;
Initial WebM release
2010-05-18 17:58:33 +02:00
else
return -1;
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
vp8_yv12_copy_frame_ptr(sd, &cm->yv12_fb[ref_fb_idx]);
Initial WebM release
2010-05-18 17:58:33 +02:00
return 0;
}
int vp8_update_entropy(VP8_PTR comp, int update)
{
VP8_COMP *cpi = (VP8_COMP *) comp;
VP8_COMMON *cm = &cpi->common;
cm->refresh_entropy_probs = update;
return 0;
}
disable compilation of debugging code This patch avoids compiling some debugging code in onyx_if.c. The most significant fix is to avoid generating code for vp8_write_yuv_frame, which is never called. Some other code was removed by the dead code elimination performed by the compiler, and this patch does it with the preprocessor instead. There are advantages both ways. Change-Id: I044fd43179d2e947553f0d6f2cad5b40907ac458
2010-09-24 17:10:25 +02:00
Add 8x8 transform to experimental branch Please refer to previous commit messages for detailed info: https://on2-git.corp.google.com/g/#change,5940 https://on2-git.corp.google.com/g/#change,6045 Change-Id: I8b16992f2f69c5a808ad40a3e32ef589cce7c59d
2011-02-14 23:18:18 +01:00
#ifdef OUTPUT_YUV_SRC
void vp8_write_yuv_frame(YV12_BUFFER_CONFIG *s)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
unsigned char *src = s->y_buffer;
int h = s->y_height;
do
{
fwrite(src, s->y_width, 1, yuv_file);
src += s->y_stride;
}
while (--h);
src = s->u_buffer;
h = s->uv_height;
do
{
fwrite(src, s->uv_width, 1, yuv_file);
src += s->uv_stride;
}
while (--h);
src = s->v_buffer;
h = s->uv_height;
do
{
fwrite(src, s->uv_width, 1, yuv_file);
src += s->uv_stride;
}
while (--h);
Add 8x8 transform to experimental branch Please refer to previous commit messages for detailed info: https://on2-git.corp.google.com/g/#change,5940 https://on2-git.corp.google.com/g/#change,6045 Change-Id: I8b16992f2f69c5a808ad40a3e32ef589cce7c59d
2011-02-14 23:18:18 +01:00
}
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
Add 8x8 transform to experimental branch Please refer to previous commit messages for detailed info: https://on2-git.corp.google.com/g/#change,5940 https://on2-git.corp.google.com/g/#change,6045 Change-Id: I8b16992f2f69c5a808ad40a3e32ef589cce7c59d
2011-02-14 23:18:18 +01:00
#ifdef OUTPUT_YUV_REC
void vp8_write_yuv_rec_frame(VP8_COMMON *cm)
{
YV12_BUFFER_CONFIG *s = cm->frame_to_show;
unsigned char *src = s->y_buffer;
int h = cm->Height;
do
{
fwrite(src, s->y_width, 1, yuv_rec_file);
src += s->y_stride;
}
while (--h);
src = s->u_buffer;
h = (cm->Height+1)/2;
do
{
fwrite(src, s->uv_width, 1, yuv_rec_file);
src += s->uv_stride;
}
while (--h);
src = s->v_buffer;
h = (cm->Height+1)/2;
do
{
fwrite(src, s->uv_width, 1, yuv_rec_file);
src += s->uv_stride;
}
while (--h);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
disable compilation of debugging code This patch avoids compiling some debugging code in onyx_if.c. The most significant fix is to avoid generating code for vp8_write_yuv_frame, which is never called. Some other code was removed by the dead code elimination performed by the compiler, and this patch does it with the preprocessor instead. There are advantages both ways. Change-Id: I044fd43179d2e947553f0d6f2cad5b40907ac458
2010-09-24 17:10:25 +02:00
#endif
Do not copy data between encoder reference buffers. Golden and ALT reference buffers were refreshed by copying from the new buffer. Replaced this by index manipulation. Also moved all the reference frame updates to one function for easier tracking. Change-Id: Icd3e534e7e2c8c5567168d222e6a64a96aae24a1
2011-04-20 14:12:23 +02:00
static void update_alt_ref_frame_stats(VP8_COMP *cpi)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
VP8_COMMON *cm = &cpi->common;
// Update data structure that monitors level of reference to last GF
Moved gf_active code to encoder only The gf_active code is only used by the encoder, so it was moved from common and decoder. Change-Id: Iada15acd5b2b33ff70c34668ca87d4cfd0d05025
2010-08-11 17:02:31 +02:00
vpx_memset(cpi->gf_active_flags, 1, (cm->mb_rows * cm->mb_cols));
cpi->gf_active_count = cm->mb_rows * cm->mb_cols;
Initial WebM release
2010-05-18 17:58:33 +02:00
Fix encoder real-time only configuration. Remove allocation/deallocation of stats storage. Remove full search functions in machine specific encoder inits. Remove last pass validation in validate_config. Change-Id: I7f29be69273981a4fef6e80ecdb6217c68cbad4e
2011-01-10 10:14:10 +01:00
// this frame refreshes means next frames don't unless specified by user
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->common.frames_since_golden = 0;
// Clear the alternate reference update pending flag.
cpi->source_alt_ref_pending = FALSE;
// Set the alternate refernce frame active flag
cpi->source_alt_ref_active = TRUE;
}
Do not copy data between encoder reference buffers. Golden and ALT reference buffers were refreshed by copying from the new buffer. Replaced this by index manipulation. Also moved all the reference frame updates to one function for easier tracking. Change-Id: Icd3e534e7e2c8c5567168d222e6a64a96aae24a1
2011-04-20 14:12:23 +02:00
static void update_golden_frame_stats(VP8_COMP *cpi)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
VP8_COMMON *cm = &cpi->common;
Do not copy data between encoder reference buffers. Golden and ALT reference buffers were refreshed by copying from the new buffer. Replaced this by index manipulation. Also moved all the reference frame updates to one function for easier tracking. Change-Id: Icd3e534e7e2c8c5567168d222e6a64a96aae24a1
2011-04-20 14:12:23 +02:00
// Update the Golden frame usage counts.
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cm->refresh_golden_frame)
{
// Update data structure that monitors level of reference to last GF
Moved gf_active code to encoder only The gf_active code is only used by the encoder, so it was moved from common and decoder. Change-Id: Iada15acd5b2b33ff70c34668ca87d4cfd0d05025
2010-08-11 17:02:31 +02:00
vpx_memset(cpi->gf_active_flags, 1, (cm->mb_rows * cm->mb_cols));
cpi->gf_active_count = cm->mb_rows * cm->mb_cols;
Initial WebM release
2010-05-18 17:58:33 +02:00
// this frame refreshes means next frames don't unless specified by user
cm->refresh_golden_frame = 0;
cpi->common.frames_since_golden = 0;
//if ( cm->frame_type == KEY_FRAME )
//{
cpi->recent_ref_frame_usage[INTRA_FRAME] = 1;
cpi->recent_ref_frame_usage[LAST_FRAME] = 1;
cpi->recent_ref_frame_usage[GOLDEN_FRAME] = 1;
cpi->recent_ref_frame_usage[ALTREF_FRAME] = 1;
//}
//else
//{
// // Carry a potrtion of count over to begining of next gf sequence
// cpi->recent_ref_frame_usage[INTRA_FRAME] >>= 5;
// cpi->recent_ref_frame_usage[LAST_FRAME] >>= 5;
// cpi->recent_ref_frame_usage[GOLDEN_FRAME] >>= 5;
// cpi->recent_ref_frame_usage[ALTREF_FRAME] >>= 5;
//}
// ******** Fixed Q test code only ************
// If we are going to use the ALT reference for the next group of frames set a flag to say so.
if (cpi->oxcf.fixed_q >= 0 &&
cpi->oxcf.play_alternate && !cpi->common.refresh_alt_ref_frame)
{
cpi->source_alt_ref_pending = TRUE;
cpi->frames_till_gf_update_due = cpi->baseline_gf_interval;
}
if (!cpi->source_alt_ref_pending)
cpi->source_alt_ref_active = FALSE;
// Decrement count down till next gf
if (cpi->frames_till_gf_update_due > 0)
cpi->frames_till_gf_update_due--;
}
else if (!cpi->common.refresh_alt_ref_frame)
{
// Decrement count down till next gf
if (cpi->frames_till_gf_update_due > 0)
cpi->frames_till_gf_update_due--;
if (cpi->common.frames_till_alt_ref_frame)
cpi->common.frames_till_alt_ref_frame --;
cpi->common.frames_since_golden ++;
if (cpi->common.frames_since_golden > 1)
{
cpi->recent_ref_frame_usage[INTRA_FRAME] += cpi->count_mb_ref_frame_usage[INTRA_FRAME];
cpi->recent_ref_frame_usage[LAST_FRAME] += cpi->count_mb_ref_frame_usage[LAST_FRAME];
cpi->recent_ref_frame_usage[GOLDEN_FRAME] += cpi->count_mb_ref_frame_usage[GOLDEN_FRAME];
cpi->recent_ref_frame_usage[ALTREF_FRAME] += cpi->count_mb_ref_frame_usage[ALTREF_FRAME];
}
}
}
Further work on extended Q range. Fixed some further QIndex related issues and replaced some tables (eg zbin and rounding) Also Added function (currently disabled by default) to populate the main AC and DC quantizer tables. Using the original AC range the resulting computed DC values give behavior broadly comparable on the DERF set. That is not to say that the equations will hold good over a more extended range. The purpose of this code is to make it easier to experiment with further alterations to the Q range and distribution of Q values plus the relative weights given to AC and DC. The function find_fp_qindex() ensures that changes to the Q tables are reflected in the value passed in to the first pass code. Slight experimental adjustment to static segment Q offset. Change-Id: I36186267d55dfc2a3d565d0cff7218ef300d1cd5
2011-12-02 15:57:21 +01:00
int find_fp_qindex()
{
int i;
for ( i = 0; i < QINDEX_RANGE; i++ )
{
if ( vp8_convert_qindex_to_q(i) >= 30.0 )
{
break;
}
}
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
Further work on extended Q range. Fixed some further QIndex related issues and replaced some tables (eg zbin and rounding) Also Added function (currently disabled by default) to populate the main AC and DC quantizer tables. Using the original AC range the resulting computed DC values give behavior broadly comparable on the DERF set. That is not to say that the equations will hold good over a more extended range. The purpose of this code is to make it easier to experiment with further alterations to the Q range and distribution of Q values plus the relative weights given to AC and DC. The function find_fp_qindex() ensures that changes to the Q tables are reflected in the value passed in to the first pass code. Slight experimental adjustment to static segment Q offset. Change-Id: I36186267d55dfc2a3d565d0cff7218ef300d1cd5
2011-12-02 15:57:21 +01:00
if ( i == QINDEX_RANGE )
i--;
return i;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
static void Pass1Encode(VP8_COMP *cpi, unsigned long *size, unsigned char *dest, unsigned int *frame_flags)
{
(void) size;
(void) dest;
(void) frame_flags;
Further work on extended Q range. Fixed some further QIndex related issues and replaced some tables (eg zbin and rounding) Also Added function (currently disabled by default) to populate the main AC and DC quantizer tables. Using the original AC range the resulting computed DC values give behavior broadly comparable on the DERF set. That is not to say that the equations will hold good over a more extended range. The purpose of this code is to make it easier to experiment with further alterations to the Q range and distribution of Q values plus the relative weights given to AC and DC. The function find_fp_qindex() ensures that changes to the Q tables are reflected in the value passed in to the first pass code. Slight experimental adjustment to static segment Q offset. Change-Id: I36186267d55dfc2a3d565d0cff7218ef300d1cd5
2011-12-02 15:57:21 +01:00
vp8_set_quantizer(cpi, find_fp_qindex());
Initial WebM release
2010-05-18 17:58:33 +02:00
vp8_first_pass(cpi);
}
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
#if 1
void write_yuv_frame_to_file(YV12_BUFFER_CONFIG *frame)
{
// write the frame
int i;
FILE *fp = fopen("encode_recon.yuv", "a");
for (i = 0; i < frame->y_height; i++)
fwrite(frame->y_buffer + i * frame->y_stride,
frame->y_width, 1, fp);
for (i = 0; i < frame->uv_height; i++)
fwrite(frame->u_buffer + i * frame->uv_stride,
frame->uv_width, 1, fp);
for (i = 0; i < frame->uv_height; i++)
fwrite(frame->v_buffer + i * frame->uv_stride,
frame->uv_width, 1, fp);
fclose(fp);
}
#endif
added separate entropy context for alt_ref This commit added code to keep track of separate entropy contexts for normal frames and alt ref frames. The underly assumption was that the two type of frames have different entropy characteristics given they typically have quite different quantization levels. By keeping entropy contexts separate, it helps the entropy context distribution to be more closely adapted to each frame type. Tests on derf set showed a good and very consistent gain on all clips on all metrics, avg psnr: 0.89%, overall psnr: 0.84% and ssim 0.93%. http://www.corp.google.com/~yaowu/no_crawl/mulcontext.html Change-Id: I15bc9697f6ff7829042911fe0c62930585d7e65d
2011-12-01 01:36:46 +01:00
//#define WRITE_RECON_BUFFER 1
changed find_near_mvs search to include a mb from last frame This is an experiment to include a mv contribution from last frame to nearest and near mv definition. Initial test showed some small though consistent gain. latest patch slightly better result ~.13%-~.18%. TODO: the entropy used to encode the mode choice, i.e. the mv counts based conditional distribution of modes should be re-collected to reflect this change, it is expected that there is some further gain from that. Change-Id: Ief1e284a36d8aa56b49ae5b360c91419ec494fa4
2011-11-16 01:16:30 +01:00
#if WRITE_RECON_BUFFER
Initial WebM release
2010-05-18 17:58:33 +02:00
void write_cx_frame_to_file(YV12_BUFFER_CONFIG *frame, int this_frame)
{
// write the frame
FILE *yframe;
int i;
char filename[255];
sprintf(filename, "cx\\y%04d.raw", this_frame);
yframe = fopen(filename, "wb");
for (i = 0; i < frame->y_height; i++)
added separate entropy context for alt_ref This commit added code to keep track of separate entropy contexts for normal frames and alt ref frames. The underly assumption was that the two type of frames have different entropy characteristics given they typically have quite different quantization levels. By keeping entropy contexts separate, it helps the entropy context distribution to be more closely adapted to each frame type. Tests on derf set showed a good and very consistent gain on all clips on all metrics, avg psnr: 0.89%, overall psnr: 0.84% and ssim 0.93%. http://www.corp.google.com/~yaowu/no_crawl/mulcontext.html Change-Id: I15bc9697f6ff7829042911fe0c62930585d7e65d
2011-12-01 01:36:46 +01:00
fwrite(frame->y_buffer + i * frame->y_stride,
frame->y_width, 1, yframe);
Initial WebM release
2010-05-18 17:58:33 +02:00
fclose(yframe);
sprintf(filename, "cx\\u%04d.raw", this_frame);
yframe = fopen(filename, "wb");
for (i = 0; i < frame->uv_height; i++)
added separate entropy context for alt_ref This commit added code to keep track of separate entropy contexts for normal frames and alt ref frames. The underly assumption was that the two type of frames have different entropy characteristics given they typically have quite different quantization levels. By keeping entropy contexts separate, it helps the entropy context distribution to be more closely adapted to each frame type. Tests on derf set showed a good and very consistent gain on all clips on all metrics, avg psnr: 0.89%, overall psnr: 0.84% and ssim 0.93%. http://www.corp.google.com/~yaowu/no_crawl/mulcontext.html Change-Id: I15bc9697f6ff7829042911fe0c62930585d7e65d
2011-12-01 01:36:46 +01:00
fwrite(frame->u_buffer + i * frame->uv_stride,
frame->uv_width, 1, yframe);
Initial WebM release
2010-05-18 17:58:33 +02:00
fclose(yframe);
sprintf(filename, "cx\\v%04d.raw", this_frame);
yframe = fopen(filename, "wb");
for (i = 0; i < frame->uv_height; i++)
added separate entropy context for alt_ref This commit added code to keep track of separate entropy contexts for normal frames and alt ref frames. The underly assumption was that the two type of frames have different entropy characteristics given they typically have quite different quantization levels. By keeping entropy contexts separate, it helps the entropy context distribution to be more closely adapted to each frame type. Tests on derf set showed a good and very consistent gain on all clips on all metrics, avg psnr: 0.89%, overall psnr: 0.84% and ssim 0.93%. http://www.corp.google.com/~yaowu/no_crawl/mulcontext.html Change-Id: I15bc9697f6ff7829042911fe0c62930585d7e65d
2011-12-01 01:36:46 +01:00
fwrite(frame->v_buffer + i * frame->uv_stride,
frame->uv_width, 1, yframe);
Initial WebM release
2010-05-18 17:58:33 +02:00
fclose(yframe);
}
#endif
Making subpel filters switchable at frame level Various refactoring to make the subpel motion compensation filters switchable by a frame level field. Two types of 8-tap filters are supported in addition to the existing bilinar and sixtap filters. One is the default 8-tap and the other has a sharper cut-off for use with frames with substantial edge content. Patch 2: Added a preliminary strategy for filter selection based on edginess detecton. Also includes some filter changes. Change-Id: I866085bda5ae143cfdf2ec88157feaabdf7bd63a
2012-03-19 15:53:05 +01:00
static double compute_edge_pixel_proportion(YV12_BUFFER_CONFIG *frame)
{
#define EDGE_THRESH 128
int i, j;
int num_edge_pels = 0;
int num_pels = (frame->y_height - 2) * (frame->y_width - 2);
unsigned char *prev = frame->y_buffer + 1;
unsigned char *curr = frame->y_buffer + 1 + frame->y_stride;
unsigned char *next = frame->y_buffer + 1 + 2*frame->y_stride;
for (i = 1; i < frame->y_height - 1; i++)
{
for (j = 1; j < frame->y_width - 1; j++)
{
/* Sobel hor and ver gradients */
int v = 2*(curr[1] - curr[-1]) + (prev[1] - prev[-1]) + (next[1] - next[-1]);
int h = 2*(prev[0] - next[0]) + (prev[1] - next[1]) + (prev[-1] - next[-1]);
h = (h < 0 ? -h : h);
v = (v < 0 ? -v : v);
if (h > EDGE_THRESH || v > EDGE_THRESH) num_edge_pels++;
curr++;
prev++;
next++;
}
curr += frame->y_stride - frame->y_width + 2;
prev += frame->y_stride - frame->y_width + 2;
next += frame->y_stride - frame->y_width + 2;
}
return (double)num_edge_pels/(double)num_pels;
}
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
// Function to test for conditions that indicate we should loop
Replaced recode loop test with a function call Replaced existing code to decide if a frame recode is required with a function call. This is to simplify addition of extra clauses that may be needed for the planned constrained quality mode. Also fixed a bug where by alt ref not considered in the test. Change-Id: I3d40bb21abe3e19f8456761e6849deb171738b60
2010-11-17 16:12:04 +01:00
// back and recode a frame.
static BOOL recode_loop_test( VP8_COMP *cpi,
int high_limit, int low_limit,
int q, int maxq, int minq )
{
BOOL force_recode = FALSE;
VP8_COMMON *cm = &cpi->common;
// Is frame recode allowed at all
// Yes if either recode mode 1 is selected or mode two is selcted
// and the frame is a key frame. golden frame or alt_ref_frame
if ( (cpi->sf.recode_loop == 1) ||
( (cpi->sf.recode_loop == 2) &&
( (cm->frame_type == KEY_FRAME) ||
cm->refresh_golden_frame ||
cm->refresh_alt_ref_frame ) ) )
{
// General over and under shoot tests
if ( ((cpi->projected_frame_size > high_limit) && (q < maxq)) ||
KF/GF Pulsing This change is designed to try and reduce pulsing effects when moving with a complex transition like a fade, into an easy or static section in an otherwise difficult clip in CQ mode. The active CQ level is relaxed down to the user entered level for frames that are generating less than the passed in minimum bandwidth. Change-Id: Id6d8b551daad4f489c087bd742bc95418a95f3f0
2011-01-14 12:34:53 +01:00
((cpi->projected_frame_size < low_limit) && (q > minq)) )
Replaced recode loop test with a function call Replaced existing code to decide if a frame recode is required with a function call. This is to simplify addition of extra clauses that may be needed for the planned constrained quality mode. Also fixed a bug where by alt ref not considered in the test. Change-Id: I3d40bb21abe3e19f8456761e6849deb171738b60
2010-11-17 16:12:04 +01:00
{
force_recode = TRUE;
}
KF/GF Pulsing This change is designed to try and reduce pulsing effects when moving with a complex transition like a fade, into an easy or static section in an otherwise difficult clip in CQ mode. The active CQ level is relaxed down to the user entered level for frames that are generating less than the passed in minimum bandwidth. Change-Id: Id6d8b551daad4f489c087bd742bc95418a95f3f0
2011-01-14 12:34:53 +01:00
// Special Constrained quality tests
else if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY)
{
// Undershoot and below auto cq level
if ( (q > cpi->cq_target_quality) &&
(cpi->projected_frame_size <
((cpi->this_frame_target * 7) >> 3)))
{
force_recode = TRUE;
}
// Severe undershoot and between auto and user cq level
else if ( (q > cpi->oxcf.cq_level) &&
(cpi->projected_frame_size < cpi->min_frame_bandwidth) &&
(cpi->active_best_quality > cpi->oxcf.cq_level))
{
force_recode = TRUE;
cpi->active_best_quality = cpi->oxcf.cq_level;
}
}
Replaced recode loop test with a function call Replaced existing code to decide if a frame recode is required with a function call. This is to simplify addition of extra clauses that may be needed for the planned constrained quality mode. Also fixed a bug where by alt ref not considered in the test. Change-Id: I3d40bb21abe3e19f8456761e6849deb171738b60
2010-11-17 16:12:04 +01:00
}
return force_recode;
}
Do not copy data between encoder reference buffers. Golden and ALT reference buffers were refreshed by copying from the new buffer. Replaced this by index manipulation. Also moved all the reference frame updates to one function for easier tracking. Change-Id: Icd3e534e7e2c8c5567168d222e6a64a96aae24a1
2011-04-20 14:12:23 +02:00
void update_reference_frames(VP8_COMMON *cm)
{
YV12_BUFFER_CONFIG *yv12_fb = cm->yv12_fb;
// At this point the new frame has been encoded.
// If any buffer copy / swapping is signaled it should be done here.
if (cm->frame_type == KEY_FRAME)
{
yv12_fb[cm->new_fb_idx].flags |= VP8_GOLD_FLAG | VP8_ALT_FLAG ;
yv12_fb[cm->gld_fb_idx].flags &= ~VP8_GOLD_FLAG;
yv12_fb[cm->alt_fb_idx].flags &= ~VP8_ALT_FLAG;
cm->alt_fb_idx = cm->gld_fb_idx = cm->new_fb_idx;
}
else /* For non key frames */
{
if (cm->refresh_alt_ref_frame)
{
assert(!cm->copy_buffer_to_arf);
cm->yv12_fb[cm->new_fb_idx].flags |= VP8_ALT_FLAG;
cm->yv12_fb[cm->alt_fb_idx].flags &= ~VP8_ALT_FLAG;
cm->alt_fb_idx = cm->new_fb_idx;
}
else if (cm->copy_buffer_to_arf)
{
assert(!(cm->copy_buffer_to_arf & ~0x3));
if (cm->copy_buffer_to_arf == 1)
{
bug fix check frame buffer index before copy in onyx_if.c update_reference_frames() make sure that frame buffer indexes are not equal before preforming a buffer copy. If two frames share the same buffer the flags will already be set correctly. Change-Id: Ida9b5516d08e3435c90f131d2dc19d842cfb536e
2011-05-27 20:57:25 +02:00
if(cm->alt_fb_idx != cm->lst_fb_idx)
{
yv12_fb[cm->lst_fb_idx].flags |= VP8_ALT_FLAG;
yv12_fb[cm->alt_fb_idx].flags &= ~VP8_ALT_FLAG;
cm->alt_fb_idx = cm->lst_fb_idx;
}
Do not copy data between encoder reference buffers. Golden and ALT reference buffers were refreshed by copying from the new buffer. Replaced this by index manipulation. Also moved all the reference frame updates to one function for easier tracking. Change-Id: Icd3e534e7e2c8c5567168d222e6a64a96aae24a1
2011-04-20 14:12:23 +02:00
}
else /* if (cm->copy_buffer_to_arf == 2) */
{
bug fix check frame buffer index before copy in onyx_if.c update_reference_frames() make sure that frame buffer indexes are not equal before preforming a buffer copy. If two frames share the same buffer the flags will already be set correctly. Change-Id: Ida9b5516d08e3435c90f131d2dc19d842cfb536e
2011-05-27 20:57:25 +02:00
if(cm->alt_fb_idx != cm->gld_fb_idx)
{
yv12_fb[cm->gld_fb_idx].flags |= VP8_ALT_FLAG;
yv12_fb[cm->alt_fb_idx].flags &= ~VP8_ALT_FLAG;
cm->alt_fb_idx = cm->gld_fb_idx;
}
Do not copy data between encoder reference buffers. Golden and ALT reference buffers were refreshed by copying from the new buffer. Replaced this by index manipulation. Also moved all the reference frame updates to one function for easier tracking. Change-Id: Icd3e534e7e2c8c5567168d222e6a64a96aae24a1
2011-04-20 14:12:23 +02:00
}
}
if (cm->refresh_golden_frame)
{
assert(!cm->copy_buffer_to_gf);
cm->yv12_fb[cm->new_fb_idx].flags |= VP8_GOLD_FLAG;
cm->yv12_fb[cm->gld_fb_idx].flags &= ~VP8_GOLD_FLAG;
cm->gld_fb_idx = cm->new_fb_idx;
}
else if (cm->copy_buffer_to_gf)
{
assert(!(cm->copy_buffer_to_arf & ~0x3));
if (cm->copy_buffer_to_gf == 1)
{
bug fix check frame buffer index before copy in onyx_if.c update_reference_frames() make sure that frame buffer indexes are not equal before preforming a buffer copy. If two frames share the same buffer the flags will already be set correctly. Change-Id: Ida9b5516d08e3435c90f131d2dc19d842cfb536e
2011-05-27 20:57:25 +02:00
if(cm->gld_fb_idx != cm->lst_fb_idx)
{
yv12_fb[cm->lst_fb_idx].flags |= VP8_GOLD_FLAG;
yv12_fb[cm->gld_fb_idx].flags &= ~VP8_GOLD_FLAG;
cm->gld_fb_idx = cm->lst_fb_idx;
}
Do not copy data between encoder reference buffers. Golden and ALT reference buffers were refreshed by copying from the new buffer. Replaced this by index manipulation. Also moved all the reference frame updates to one function for easier tracking. Change-Id: Icd3e534e7e2c8c5567168d222e6a64a96aae24a1
2011-04-20 14:12:23 +02:00
}
else /* if (cm->copy_buffer_to_gf == 2) */
{
bug fix check frame buffer index before copy in onyx_if.c update_reference_frames() make sure that frame buffer indexes are not equal before preforming a buffer copy. If two frames share the same buffer the flags will already be set correctly. Change-Id: Ida9b5516d08e3435c90f131d2dc19d842cfb536e
2011-05-27 20:57:25 +02:00
if(cm->alt_fb_idx != cm->gld_fb_idx)
{
yv12_fb[cm->alt_fb_idx].flags |= VP8_GOLD_FLAG;
yv12_fb[cm->gld_fb_idx].flags &= ~VP8_GOLD_FLAG;
cm->gld_fb_idx = cm->alt_fb_idx;
}
Do not copy data between encoder reference buffers. Golden and ALT reference buffers were refreshed by copying from the new buffer. Replaced this by index manipulation. Also moved all the reference frame updates to one function for easier tracking. Change-Id: Icd3e534e7e2c8c5567168d222e6a64a96aae24a1
2011-04-20 14:12:23 +02:00
}
}
}
if (cm->refresh_last_frame)
{
cm->yv12_fb[cm->new_fb_idx].flags |= VP8_LAST_FLAG;
cm->yv12_fb[cm->lst_fb_idx].flags &= ~VP8_LAST_FLAG;
cm->lst_fb_idx = cm->new_fb_idx;
}
}
Encoder loopfilter running in its own thread In multithreaded mode the loopfilter is running in its own thread (filter level calculation and frame filtering). Filtering is mostly done in parallel with the bitstream packing. Before starting the packing the loopfilter level has to be calculated. Also any needed reference frame copying is done in the filter thread. Currently the encoder will create n+1 threads, where n > 1 is the number of threads specified by application and 1 is the extra filter thread. With n = 1 the encoder runs in single thread mode. There will never be more than n threads running concurrently. Change-Id: I4fb29b559a40275d6d3babb8727245c40fba931b
2011-02-25 12:42:05 +01:00
void loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm)
{
if (cm->no_lpf)
{
cm->filter_level = 0;
}
else
{
struct vpx_usec_timer timer;
vp8_clear_system_state();
vpx_usec_timer_start(&timer);
if (cpi->sf.auto_filter == 0)
vp8cx_pick_filter_level_fast(cpi->Source, cpi);
else
vp8cx_pick_filter_level(cpi->Source, cpi);
vpx_usec_timer_mark(&timer);
cpi->time_pick_lpf += vpx_usec_timer_elapsed(&timer);
}
if (cm->filter_level > 0)
{
vp8cx_set_alt_lf_level(cpi, cm->filter_level);
fix sharpness bug and clean up sharpness was not recalculated in vp8cx_pick_filter_level_fast remove last_filter_type. all values are calculated, don't need to update the lfi data when it changes. always use cm->sharpness_level. the extra indirection was annoying. don't track last frame_type or sharpness_level manually. frame type only matters for motion search and sharpness_level is taken care of in frame_init move function declarations to their proper header Change-Id: I7ef037bd4bf8cf5e37d2d36bd03b5e22a2ad91db
2011-07-20 21:53:42 +02:00
vp8_loop_filter_frame(cm, &cpi->mb.e_mbd);
Encoder loopfilter running in its own thread In multithreaded mode the loopfilter is running in its own thread (filter level calculation and frame filtering). Filtering is mostly done in parallel with the bitstream packing. Before starting the packing the loopfilter level has to be calculated. Also any needed reference frame copying is done in the filter thread. Currently the encoder will create n+1 threads, where n > 1 is the number of threads specified by application and 1 is the extra filter thread. With n = 1 the encoder runs in single thread mode. There will never be more than n threads running concurrently. Change-Id: I4fb29b559a40275d6d3babb8727245c40fba931b
2011-02-25 12:42:05 +01:00
}
vp8_yv12_extend_frame_borders_ptr(cm->frame_to_show);
}
Made AltRef filter adaptive & added motion compensation Modified AltRef temporal filter to adapt filter length based on macroblock coding modes selected during first-pass encode. Also added sub-pixel motion compensation to the AltRef filter.
2010-09-28 16:23:41 +02:00
static void encode_frame_to_data_rate
Initial WebM release
2010-05-18 17:58:33 +02:00
(
Made AltRef filter adaptive & added motion compensation Modified AltRef temporal filter to adapt filter length based on macroblock coding modes selected during first-pass encode. Also added sub-pixel motion compensation to the AltRef filter.
2010-09-28 16:23:41 +02:00
VP8_COMP *cpi,
unsigned long *size,
unsigned char *dest,
unsigned int *frame_flags
Initial WebM release
2010-05-18 17:58:33 +02:00
)
{
Merge CONFIGURE_SEGMENTATION experiment. Removal of CONFIGURE_SEGMENTATION ifdefs. Removal of legacy support code fo the old coding mechanism. Use local reference "xd" for MACROBLOCKD structure in encode_frame_to_data_rate() Moved call to choose_segmap_coding_method() out of encode loop as the cost of segmentation is not properly accounted in the loop anyway. If this is desirable in the future it can be moved back. The use of this function to do all the analysis and set the probabilities also removes the need to track segment useage in threading code. Change-Id: I85bc8fd63440e7176c73d26cb742698f9b70cade
2011-11-15 17:15:23 +01:00
VP8_COMMON *cm = &cpi->common;
MACROBLOCKD *xd = &cpi->mb.e_mbd;
Initial WebM release
2010-05-18 17:58:33 +02:00
int Q;
int frame_over_shoot_limit;
int frame_under_shoot_limit;
int Loop = FALSE;
int loop_count;
int this_q;
int last_zbin_oq;
int q_low;
int q_high;
int zbin_oq_high;
int zbin_oq_low = 0;
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
Initial WebM release
2010-05-18 17:58:33 +02:00
int top_index;
int bottom_index;
int active_worst_qchanged = FALSE;
int overshoot_seen = FALSE;
int undershoot_seen = FALSE;
Use of temporal context for encoding delta updates. - Used three probability approach for temporal context as follows: P0 - probability of no change if both above and left not changed P1 - probability of no change if one of above and left has changed P2 - probability of no change if both above and left have changed In addition, a 1 bit/frame has been used to decide whether to use temporal context or to encode directly. The cost of using both the schemes is calculated ahead and the temporal_update flag is set if the cost of using temporal context is lower than encoding the segment ids directly. This approach has given around 20% reduction in cost of bits needed to encode segmentation ids. Change-Id: I44a5509599eded215ae5be9554314280d3d35405
2010-09-24 00:25:33 +02:00
T8x8 zbin and rate control changes. Some adjustments to zbin for t8x8. Changes to rules for sizing forced key frames. Some extra stats output in tmp.stt. Approximate gain on YT-hd set 0.5% There are still issues in sizing key frames and gf/arf frames when the image is largely static. These in part relate to problems with cost estimates in the recode loop. Change-Id: I6f0159dc8a8faeab4115a19c668d442491619a68
2012-03-27 14:01:34 +02:00
int loop_size_estimate = 0;
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
#if CONFIG_ENHANCED_INTERP
SPEED_FEATURES *sf = &cpi->sf;
#if RESET_FOREACH_FILTER
int q_low0;
int q_high0;
int zbin_oq_high0;
int zbin_oq_low0 = 0;
int Q0;
int last_zbin_oq0;
int active_best_quality0;
int active_worst_quality0;
double rate_correction_factor0;
double gf_rate_correction_factor0;
#endif
/* list of filters to search over */
int mcomp_filters_to_search[] = {EIGHTTAP, EIGHTTAP_SHARP, SIXTAP};
int mcomp_filters = sizeof(mcomp_filters_to_search)/sizeof(*mcomp_filters_to_search);
int mcomp_filter_index = 0;
INT64 mcomp_filter_cost[4];
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
// Clear down mmx registers to allow floating point in what follows
vp8_clear_system_state();
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
// For an alt ref frame in 2 pass we skip the call to the second
// pass function that sets the target bandwidth so must set it here
if (cpi->common.refresh_alt_ref_frame)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
cpi->per_frame_bandwidth = cpi->twopass.gf_bits; // Per frame bit target for the alt ref frame
cpi->target_bandwidth = cpi->twopass.gf_bits * cpi->output_frame_rate; // per second target bitrate
Initial WebM release
2010-05-18 17:58:33 +02:00
}
// Default turn off buffer to buffer copying
cm->copy_buffer_to_gf = 0;
cm->copy_buffer_to_arf = 0;
// Clear zbin over-quant value and mode boost values.
cpi->zbin_over_quant = 0;
cpi->zbin_mode_boost = 0;
Tuning for the more exact quantizer. Small changes to the default zero bin and rounding tables. Though the tables are currently the same for the Y1 and Y2 cases I have left them as separate tables in case we want to tune this later. There is now some adjustment of the zbin based on the prediction mode. Previously this was restricted to an adjustment for gf/arf 0,0 MV. The exact quantizer now marginal outperforms and is the default. The overall average gain is about 0.5% Change-Id: I5e4353f3d5326dde4e86823684b236a1e9ea7f47
2010-11-08 16:28:54 +01:00
// Enable or disable mode based tweaking of the zbin
// For 2 Pass Only used where GF/ARF prediction quality
// is above a threshold
cpi->zbin_mode_boost = 0;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->zbin_mode_boost_enabled = TRUE;
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
if ( cpi->gfu_boost <= 400 )
Tuning for the more exact quantizer. Small changes to the default zero bin and rounding tables. Though the tables are currently the same for the Y1 and Y2 cases I have left them as separate tables in case we want to tune this later. There is now some adjustment of the zbin based on the prediction mode. Previously this was restricted to an adjustment for gf/arf 0,0 MV. The exact quantizer now marginal outperforms and is the default. The overall average gain is about 0.5% Change-Id: I5e4353f3d5326dde4e86823684b236a1e9ea7f47
2010-11-08 16:28:54 +01:00
{
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
cpi->zbin_mode_boost_enabled = FALSE;
Tuning for the more exact quantizer. Small changes to the default zero bin and rounding tables. Though the tables are currently the same for the Y1 and Y2 cases I have left them as separate tables in case we want to tune this later. There is now some adjustment of the zbin based on the prediction mode. Previously this was restricted to an adjustment for gf/arf 0,0 MV. The exact quantizer now marginal outperforms and is the default. The overall average gain is about 0.5% Change-Id: I5e4353f3d5326dde4e86823684b236a1e9ea7f47
2010-11-08 16:28:54 +01:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
// Current default encoder behaviour for the altref sign bias
if (cpi->source_alt_ref_active)
cpi->common.ref_frame_sign_bias[ALTREF_FRAME] = 1;
else
cpi->common.ref_frame_sign_bias[ALTREF_FRAME] = 0;
// Check to see if a key frame is signalled
// For two pass with auto key frame enabled cm->frame_type may already be set, but not for one pass.
if ((cm->current_video_frame == 0) ||
(cm->frame_flags & FRAMEFLAGS_KEY) ||
(cpi->oxcf.auto_key && (cpi->frames_since_key % cpi->key_frame_frequency == 0)))
{
// Key frame from VFW/auto-keyframe/first frame
cm->frame_type = KEY_FRAME;
}
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
// Set default state for segment based loop filter update flags
Merge CONFIGURE_SEGMENTATION experiment. Removal of CONFIGURE_SEGMENTATION ifdefs. Removal of legacy support code fo the old coding mechanism. Use local reference "xd" for MACROBLOCKD structure in encode_frame_to_data_rate() Moved call to choose_segmap_coding_method() out of encode loop as the cost of segmentation is not properly accounted in the loop anyway. If this is desirable in the future it can be moved back. The use of this function to do all the analysis and set the probabilities also removes the need to track segment useage in threading code. Change-Id: I85bc8fd63440e7176c73d26cb742698f9b70cade
2011-11-15 17:15:23 +01:00
xd->mode_ref_lf_delta_update = 0;
Initial WebM release
2010-05-18 17:58:33 +02:00
// Set various flags etc to special state if it is a key frame
if (cm->frame_type == KEY_FRAME)
{
int i;
Fix loopfilter delta zero transitions Loopfilter deltas are initialized to zero on keyframes in the decoder. The values then persist from the previous frame unless an update bit is set in the bitstream. This data is not included in the entropy data saved by the 'refresh entropy' bit in the bitstream, so it is effectively an additional contextual element beyond the 3 ref-frames and the entropy data. The encoder was treating this delta update bit as update-if-nonzero, meaning that the value would be refreshed even if it hadn't changed, and more significantly, if the correct value for the delta changed to zero, the update wouldn't be sent, and the decoder would preserve the last (presumably non-zero) value. This patch updates the encoder to send an update only if the value has changed from the previously transmitted value. It also forces the value to be transmitted in error resilient mode, to account for lost context in the event of lost frames. Change-Id: I56671d5b42965d0166ac226765dbfce3e5301868
2010-09-29 19:04:04 +02:00
// Reset the loop filter deltas and segmentation map
setup_features(cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
// If segmentation is enabled force a map update for key frames
Merge CONFIGURE_SEGMENTATION experiment. Removal of CONFIGURE_SEGMENTATION ifdefs. Removal of legacy support code fo the old coding mechanism. Use local reference "xd" for MACROBLOCKD structure in encode_frame_to_data_rate() Moved call to choose_segmap_coding_method() out of encode loop as the cost of segmentation is not properly accounted in the loop anyway. If this is desirable in the future it can be moved back. The use of this function to do all the analysis and set the probabilities also removes the need to track segment useage in threading code. Change-Id: I85bc8fd63440e7176c73d26cb742698f9b70cade
2011-11-15 17:15:23 +01:00
if (xd->segmentation_enabled)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Merge CONFIGURE_SEGMENTATION experiment. Removal of CONFIGURE_SEGMENTATION ifdefs. Removal of legacy support code fo the old coding mechanism. Use local reference "xd" for MACROBLOCKD structure in encode_frame_to_data_rate() Moved call to choose_segmap_coding_method() out of encode loop as the cost of segmentation is not properly accounted in the loop anyway. If this is desirable in the future it can be moved back. The use of this function to do all the analysis and set the probabilities also removes the need to track segment useage in threading code. Change-Id: I85bc8fd63440e7176c73d26cb742698f9b70cade
2011-11-15 17:15:23 +01:00
xd->update_mb_segmentation_map = 1;
xd->update_mb_segmentation_data = 1;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
// The alternate reference frame cannot be active for a key frame
cpi->source_alt_ref_active = FALSE;
// Reset the RD threshold multipliers to default of * 1 (128)
for (i = 0; i < MAX_MODES; i++)
{
cpi->rd_thresh_mult[i] = 128;
}
}
Merge COMPRED Merged in most of the current common prediction changes that were under the #if CONFIG_COMPRED option. Change-Id: If4e6f61dbe7b86dd449f6effbe93b5eb7e893885
2012-02-09 17:10:46 +01:00
//#if !CONFIG_COMPRED
// This function has been deprecated for now but we may want to do
// something here at a late date
//update_rd_ref_frame_probs(cpi);
//#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
Segment Features. Removal of #ifdefs Removal of configure #ifdefs so that segment features always available. Removal of code supporting old segment feature method. Still a good deal of tidying up to do. Change-Id: I397855f086f8c09ab1fae0a5f65d9e06d2e3e39f
2011-11-03 13:50:09 +01:00
// Test code for new segment features
init_seg_features( cpi );
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
Initial WebM release
2010-05-18 17:58:33 +02:00
// Decide how big to make the frame
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
vp8_pick_frame_size(cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
vp8_clear_system_state();
Initial WebM release
2010-05-18 17:58:33 +02:00
// Set an active best quality and if necessary active worst quality
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
Q = cpi->active_worst_quality;
Initial WebM release
2010-05-18 17:58:33 +02:00
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
if ( cm->frame_type == KEY_FRAME )
{
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
int high = 2000;
int low = 400;
if ( cpi->kf_boost > high )
cpi->active_best_quality = kf_low_motion_minq[Q];
else if ( cpi->kf_boost < low )
cpi->active_best_quality = kf_high_motion_minq[Q];
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
else
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
{
int gap = high - low;
int offset = high - cpi->kf_boost;
int qdiff = kf_high_motion_minq[Q] - kf_low_motion_minq[Q];
int adjustment = ((offset * qdiff) + (gap>>1)) / gap;
cpi->active_best_quality = kf_low_motion_minq[Q] + adjustment;
}
// Make an adjustment based on the %s static
// The main impact of this is at lower Q to prevent overly large key
// frames unless a lot of the image is static.
if (cpi->kf_zeromotion_pct < 64 )
cpi->active_best_quality += 4 - (cpi->kf_zeromotion_pct >> 4);
Initial WebM release
2010-05-18 17:58:33 +02:00
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
// Special case for key frames forced because we have reached
// the maximum key frame interval. Here force the Q to a range
// based on the ambient Q to reduce the risk of popping
if ( cpi->this_key_frame_forced )
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
int delta_qindex;
int qindex = cpi->last_boosted_qindex;
Fix CQ range and experimental KF sizing changes. The CQ level was not using the q_trans[] array to convert to a 0-127 range as per min and maxq Experimental change to try and match the reconstruction error for forced key frames approximately to that of the previous frame by means of the recode loop. Though this may cause extra recodes and the recode behavior has not been optimized, it can only happen on forced key frames. Change-Id: I1f7e42d526f1b1cb556dd461eff1a692bd1b5b2f
2011-01-17 18:23:11 +01:00
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
delta_qindex = compute_qdelta( cpi, qindex,
(qindex * 0.75) );
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
cpi->active_best_quality = qindex + delta_qindex;
if (cpi->active_best_quality < cpi->best_quality)
cpi->active_best_quality = cpi->best_quality;
Further work to reduce pulsing. These changes are specifically targeted at fade transitions to static scenes. Here we want to place a GF/ARF immediately after the fade and prevent an ARF just before the fade. Also some code lines and comment lines shortened to 80 chars while I was there. Change-Id: Iefdc09a4fa7b265048fc017246b73e138693950f
2011-01-20 19:01:20 +01:00
}
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
else if (cm->refresh_golden_frame || cpi->common.refresh_alt_ref_frame)
{
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
int high = 2000;
int low = 400;
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
// Use the lower of cpi->active_worst_quality and recent
// average Q as basis for GF/ARF Q limit unless last frame was
// a key frame.
if ( (cpi->frames_since_key > 1) &&
(cpi->avg_frame_qindex < cpi->active_worst_quality) )
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
Q = cpi->avg_frame_qindex;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
// For constrained quality dont allow Q less than the cq level
if ( (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY) &&
(Q < cpi->cq_target_quality) )
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
Q = cpi->cq_target_quality;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
if ( cpi->gfu_boost > high )
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
cpi->active_best_quality = gf_low_motion_minq[Q];
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
else if ( cpi->gfu_boost < low )
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
cpi->active_best_quality = gf_high_motion_minq[Q];
else
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
{
int gap = high - low;
int offset = high - cpi->gfu_boost;
int qdiff = gf_high_motion_minq[Q] - gf_low_motion_minq[Q];
int adjustment = ((offset * qdiff) + (gap>>1)) / gap;
cpi->active_best_quality = gf_low_motion_minq[Q] + adjustment;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
// Constrained quality use slightly lower active best.
if ( cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY )
{
cpi->active_best_quality =
cpi->active_best_quality * 15/16;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
}
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
else
1 Pass CQ and VBR bug fixes Issue 291 highlighted the fact that CQ mode was not working as expected in 1 pass mode, This commit fixes that specific problem but in so doing I also uncovered an overflow issue in the VBR code for 1 pass and some data values not being correctly initialized. For some clips (particularly short clips), the resulting improvement is dramatic. Change-Id: Ieefd6c6e4776eb8f1b0550dbfdfb72f86b33c960
2011-03-10 17:11:39 +01:00
{
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
cpi->active_best_quality = inter_minq[Q];
// For the constant/constrained quality mode we dont want
// q to fall below the cq level.
if ((cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY) &&
(cpi->active_best_quality < cpi->cq_target_quality) )
1 Pass CQ and VBR bug fixes Issue 291 highlighted the fact that CQ mode was not working as expected in 1 pass mode, This commit fixes that specific problem but in so doing I also uncovered an overflow issue in the VBR code for 1 pass and some data values not being correctly initialized. For some clips (particularly short clips), the resulting improvement is dramatic. Change-Id: Ieefd6c6e4776eb8f1b0550dbfdfb72f86b33c960
2011-03-10 17:11:39 +01:00
{
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
// If we are strongly undershooting the target rate in the last
// frames then use the user passed in cq value not the auto
// cq value.
if ( cpi->rolling_actual_bits < cpi->min_frame_bandwidth )
cpi->active_best_quality = cpi->oxcf.cq_level;
else
cpi->active_best_quality = cpi->cq_target_quality;
1 Pass CQ and VBR bug fixes Issue 291 highlighted the fact that CQ mode was not working as expected in 1 pass mode, This commit fixes that specific problem but in so doing I also uncovered an overflow issue in the VBR code for 1 pass and some data values not being correctly initialized. For some clips (particularly short clips), the resulting improvement is dramatic. Change-Id: Ieefd6c6e4776eb8f1b0550dbfdfb72f86b33c960
2011-03-10 17:11:39 +01:00
}
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
// Clip the active best and worst quality values to limits
if (cpi->active_worst_quality > cpi->worst_quality)
cpi->active_worst_quality = cpi->worst_quality;
if (cpi->active_best_quality < cpi->best_quality)
cpi->active_best_quality = cpi->best_quality;
Two pass rate control code changes. This comitt brings accross changes from the public branch commit number Icf74d13af77437c08602571dc7a97e747cce5066. The main puurpose of this comit relates to CQ mode but it also includes some refactoring of the two pass code which I hope will make tuning the experimental branch for the new quantizer range a little less painfull. Change-Id: I278e989436a928fc1fe7761068960048f9d7a376
2011-11-23 18:18:31 +01:00
fixed a crash caused invalid Q choice The commit fixed a problem by capping cpi->active_best_quality to be smaller than cpi->worst_quality. Also fixed a few line of code that was misplaced. Change-Id: Ie908264b72140c669122a0afde5d886619c33474
2011-12-08 05:08:31 +01:00
if (cpi->active_best_quality > cpi->worst_quality)
cpi->active_best_quality = cpi->worst_quality;
Two pass rate control code changes. This comitt brings accross changes from the public branch commit number Icf74d13af77437c08602571dc7a97e747cce5066. The main puurpose of this comit relates to CQ mode but it also includes some refactoring of the two pass code which I hope will make tuning the experimental branch for the new quantizer range a little less painfull. Change-Id: I278e989436a928fc1fe7761068960048f9d7a376
2011-11-23 18:18:31 +01:00
if ( cpi->active_worst_quality < cpi->active_best_quality )
cpi->active_worst_quality = cpi->active_best_quality;
Initial WebM release
2010-05-18 17:58:33 +02:00
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
// Specuial case code to try and match quality with forced key frames
if ( (cm->frame_type == KEY_FRAME) && cpi->this_key_frame_forced )
{
Q = cpi->last_boosted_qindex;
}
else
{
// Determine initial Q to try
Q = vp8_regulate_q(cpi, cpi->this_frame_target);
}
Initial WebM release
2010-05-18 17:58:33 +02:00
last_zbin_oq = cpi->zbin_over_quant;
// Set highest allowed value for Zbin over quant
if (cm->frame_type == KEY_FRAME)
zbin_oq_high = 0; //ZBIN_OQ_MAX/16
else if (cm->refresh_alt_ref_frame || (cm->refresh_golden_frame && !cpi->source_alt_ref_active))
zbin_oq_high = 16;
else
zbin_oq_high = ZBIN_OQ_MAX;
vp8_compute_frame_size_bounds(cpi, &frame_under_shoot_limit, &frame_over_shoot_limit);
Correct q_low and q_high limits for the recode loop Corrected the initial Q range limits for the recode loop to reflect the current allowed range for the frame. In experimental work on constrained quality this bug was causing unnecessary recodes. Change-Id: I7e256fbfa681293b0223fe21ec329933d76c229f
2010-12-09 16:02:04 +01:00
// Limit Q range for the adaptive loop.
Initial WebM release
2010-05-18 17:58:33 +02:00
bottom_index = cpi->active_best_quality;
top_index = cpi->active_worst_quality;
Correct q_low and q_high limits for the recode loop Corrected the initial Q range limits for the recode loop to reflect the current allowed range for the frame. In experimental work on constrained quality this bug was causing unnecessary recodes. Change-Id: I7e256fbfa681293b0223fe21ec329933d76c229f
2010-12-09 16:02:04 +01:00
q_low = cpi->active_best_quality;
q_high = cpi->active_worst_quality;
Initial WebM release
2010-05-18 17:58:33 +02:00
loop_count = 0;
Making subpel filters switchable at frame level Various refactoring to make the subpel motion compensation filters switchable by a frame level field. Two types of 8-tap filters are supported in addition to the existing bilinar and sixtap filters. One is the default 8-tap and the other has a sharper cut-off for use with frames with substantial edge content. Patch 2: Added a preliminary strategy for filter selection based on edginess detecton. Also includes some filter changes. Change-Id: I866085bda5ae143cfdf2ec88157feaabdf7bd63a
2012-03-19 15:53:05 +01:00
#if CONFIG_HIGH_PRECISION_MV || CONFIG_ENHANCED_INTERP
Initial refactoring of high_precision mv code. This is the first patch for refactoring of the code related to high-precision mv, so that 1/4 and 1/8 pel motion vectors can co-exist in the same bit-stream by use of a frame level flag. The current patch works fine for only use of 1/4th and only use of 1/8th pel mv, but there are some issues with the mode switching in between. Subsequent patches on this change Id will fix the remaining issues. Patch 2: Adds fixes to make sure that multiple mv precisions can co-exist in the bit-stream. Frame level switching has been tested to work correctly. Patch 3: Fixes lines exceeding 80 char Patch 4: http://www.corp.google.com/~debargha/vp8_results/enhinterp.html Results on derf after ssse3 bugfix, compared to everything enabled but the 8-tap, 1/8-subpel and 1/16-subpel uv. Overall the gains are about 3% now. Hopefully there are no more bugs lingering. Apparently the sse3 bug affected the quartel subpel results more than the eighth pel ones (which is understandabale because one bad predictor due to the bug, matters less if there are a lot more subpel options available as in the 1/8 subpel case). The results in the 4th column correspond to the current settings. The first two columns correspond to two settings of adaptive switching of the 1/4 or 1/8 subpel mode based on initial Q estimate. These do not work as good as just using 1/8 all the time yet. Change-Id: I3ef392ad338329f4d68a85257a49f2b14f3af472
2012-02-27 19:22:38 +01:00
if (cm->frame_type != KEY_FRAME)
{
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
#if CONFIG_ENHANCED_INTERP
Making subpel filters switchable at frame level Various refactoring to make the subpel motion compensation filters switchable by a frame level field. Two types of 8-tap filters are supported in addition to the existing bilinar and sixtap filters. One is the default 8-tap and the other has a sharper cut-off for use with frames with substantial edge content. Patch 2: Added a preliminary strategy for filter selection based on edginess detecton. Also includes some filter changes. Change-Id: I866085bda5ae143cfdf2ec88157feaabdf7bd63a
2012-03-19 15:53:05 +01:00
/* TODO: Decide this more intelligently */
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
if (sf->search_best_filter)
{
cm->mcomp_filter_type = mcomp_filters_to_search[0];
mcomp_filter_index = 0;
}
else
cm->mcomp_filter_type = EIGHTTAP;
Making subpel filters switchable at frame level Various refactoring to make the subpel motion compensation filters switchable by a frame level field. Two types of 8-tap filters are supported in addition to the existing bilinar and sixtap filters. One is the default 8-tap and the other has a sharper cut-off for use with frames with substantial edge content. Patch 2: Added a preliminary strategy for filter selection based on edginess detecton. Also includes some filter changes. Change-Id: I866085bda5ae143cfdf2ec88157feaabdf7bd63a
2012-03-19 15:53:05 +01:00
#endif
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
#if CONFIG_HIGH_PRECISION_MV
Making subpel filters switchable at frame level Various refactoring to make the subpel motion compensation filters switchable by a frame level field. Two types of 8-tap filters are supported in addition to the existing bilinar and sixtap filters. One is the default 8-tap and the other has a sharper cut-off for use with frames with substantial edge content. Patch 2: Added a preliminary strategy for filter selection based on edginess detecton. Also includes some filter changes. Change-Id: I866085bda5ae143cfdf2ec88157feaabdf7bd63a
2012-03-19 15:53:05 +01:00
/* TODO: Decide this more intelligently */
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
xd->allow_high_precision_mv = (Q < HIGH_PRECISION_MV_QTHRESH);
Making subpel filters switchable at frame level Various refactoring to make the subpel motion compensation filters switchable by a frame level field. Two types of 8-tap filters are supported in addition to the existing bilinar and sixtap filters. One is the default 8-tap and the other has a sharper cut-off for use with frames with substantial edge content. Patch 2: Added a preliminary strategy for filter selection based on edginess detecton. Also includes some filter changes. Change-Id: I866085bda5ae143cfdf2ec88157feaabdf7bd63a
2012-03-19 15:53:05 +01:00
#endif
Initial refactoring of high_precision mv code. This is the first patch for refactoring of the code related to high-precision mv, so that 1/4 and 1/8 pel motion vectors can co-exist in the same bit-stream by use of a frame level flag. The current patch works fine for only use of 1/4th and only use of 1/8th pel mv, but there are some issues with the mode switching in between. Subsequent patches on this change Id will fix the remaining issues. Patch 2: Adds fixes to make sure that multiple mv precisions can co-exist in the bit-stream. Frame level switching has been tested to work correctly. Patch 3: Fixes lines exceeding 80 char Patch 4: http://www.corp.google.com/~debargha/vp8_results/enhinterp.html Results on derf after ssse3 bugfix, compared to everything enabled but the 8-tap, 1/8-subpel and 1/16-subpel uv. Overall the gains are about 3% now. Hopefully there are no more bugs lingering. Apparently the sse3 bug affected the quartel subpel results more than the eighth pel ones (which is understandabale because one bad predictor due to the bug, matters less if there are a lot more subpel options available as in the 1/8 subpel case). The results in the 4th column correspond to the current settings. The first two columns correspond to two settings of adaptive switching of the 1/4 or 1/8 subpel mode based on initial Q estimate. These do not work as good as just using 1/8 all the time yet. Change-Id: I3ef392ad338329f4d68a85257a49f2b14f3af472
2012-02-27 19:22:38 +01:00
}
#endif
Simplification of experimental code base. Removed ~CONFIG_REALTIME_ONLY code. Change-Id: I5fafff29a08acd8928699f9ddce8744787024d8c
2012-02-13 18:05:26 +01:00
#if CONFIG_POSTPROC
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->oxcf.noise_sensitivity > 0)
{
unsigned char *src;
int l = 0;
switch (cpi->oxcf.noise_sensitivity)
{
case 1:
l = 20;
break;
case 2:
l = 40;
break;
case 3:
l = 60;
break;
case 4:
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
Initial WebM release
2010-05-18 17:58:33 +02:00
case 5:
l = 100;
break;
case 6:
l = 150;
break;
}
if (cm->frame_type == KEY_FRAME)
{
vp8_de_noise(cpi->Source, cpi->Source, l , 1, 0, RTCD(postproc));
}
else
{
vp8_de_noise(cpi->Source, cpi->Source, l , 1, 0, RTCD(postproc));
src = cpi->Source->y_buffer;
if (cpi->Source->y_stride < 0)
{
src += cpi->Source->y_stride * (cpi->Source->y_height - 1);
}
}
}
#endif
#ifdef OUTPUT_YUV_SRC
vp8_write_yuv_frame(cpi->Source);
#endif
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
#if CONFIG_ENHANCED_INTERP && RESET_FOREACH_FILTER
if (sf->search_best_filter)
{
q_low0 = q_low;
q_high0 = q_high;
Q0 = Q;
zbin_oq_low0 = zbin_oq_low;
zbin_oq_high0 = zbin_oq_high;
last_zbin_oq0 = last_zbin_oq;
rate_correction_factor0 = cpi->rate_correction_factor;
gf_rate_correction_factor0 = cpi->gf_rate_correction_factor;
active_best_quality0 = cpi->active_best_quality;
active_worst_quality0 = cpi->active_worst_quality;
}
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
do
{
vp8_clear_system_state(); //__asm emms;
Move quantizer init functions to quantize.c Group related functions together. Change-Id: I92fd779225b75a7204650f1decb713142c655d71
2011-05-19 17:04:03 +02:00
vp8_set_quantizer(cpi, Q);
Initial WebM release
2010-05-18 17:58:33 +02:00
this_q = Q;
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
if ( loop_count == 0 )
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
// setup skip prob for costing in mode/mv decision
if (cpi->common.mb_no_coeff_skip)
{
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#if CONFIG_NEWENTROPY
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
int k;
for (k=0; k<MBSKIP_CONTEXTS; k++)
cm->mbskip_pred_probs[k] = cpi->base_skip_false_prob[Q][k];
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#else
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
cpi->prob_skip_false = cpi->base_skip_false_prob[Q];
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
if (cm->frame_type != KEY_FRAME)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
if (cpi->common.refresh_alt_ref_frame)
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
{
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
#if CONFIG_NEWENTROPY
for (k=0; k<MBSKIP_CONTEXTS; k++)
{
if (cpi->last_skip_false_probs[2][k] != 0)
cm->mbskip_pred_probs[k] = cpi->last_skip_false_probs[2][k];
}
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#else
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
if (cpi->last_skip_false_probs[2] != 0)
cpi->prob_skip_false = cpi->last_skip_false_probs[2];
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#endif
}
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
else if (cpi->common.refresh_golden_frame)
{
#if CONFIG_NEWENTROPY
for (k=0; k<MBSKIP_CONTEXTS; k++)
{
if (cpi->last_skip_false_probs[1][k] != 0)
cm->mbskip_pred_probs[k] = cpi->last_skip_false_probs[1][k];
}
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#else
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
if (cpi->last_skip_false_probs[1] != 0)
cpi->prob_skip_false = cpi->last_skip_false_probs[1];
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#endif
}
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
else
{
#if CONFIG_NEWENTROPY
int k;
for (k=0; k<MBSKIP_CONTEXTS; k++)
{
if (cpi->last_skip_false_probs[0][k] != 0)
cm->mbskip_pred_probs[k] = cpi->last_skip_false_probs[0][k];
}
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#else
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
if (cpi->last_skip_false_probs[0] != 0)
cpi->prob_skip_false = cpi->last_skip_false_probs[0];
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#endif
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
// as this is for cost estimate, let's make sure it does not
// get extreme either way
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#if CONFIG_NEWENTROPY
{
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
int k;
for (k=0; k<MBSKIP_CONTEXTS; ++k)
{
if (cm->mbskip_pred_probs[k] < 5)
cm->mbskip_pred_probs[k] = 5;
if (cm->mbskip_pred_probs[k] > 250)
cm->mbskip_pred_probs[k] = 250;
if (cpi->is_src_frame_alt_ref)
cm->mbskip_pred_probs[k] = 1;
}
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
}
#else
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
if (cpi->prob_skip_false < 5)
cpi->prob_skip_false = 5;
Initial WebM release
2010-05-18 17:58:33 +02:00
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
if (cpi->prob_skip_false > 250)
cpi->prob_skip_false = 250;
Initial WebM release
2010-05-18 17:58:33 +02:00
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
if (cpi->is_src_frame_alt_ref)
cpi->prob_skip_false = 1;
Adding contextual coding of mb_skip_coeff flag. Using contextual coding of the mkb_skip_coeff flag using the values of this flag from the left and above. There is a small improvement of about 0.15% on Derf: http://www.corp.google.com/~debargha/vp8_results/mbskipcontext.html Refactored to use pred_common.c by adding a new context type. Results on HD set (about 0.66% improvement): http://www.corp.google.com/~debargha/vp8_results/mbskipcontext_hd.html Incliding missing refactoring to use the pred_common utilities. Change-Id: I95373382d429b5a59610d77f69a0fea2be628278
2012-03-19 19:02:04 +01:00
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
// Set up entropy depending on frame type.
if (cm->frame_type == KEY_FRAME)
vp8_setup_key_frame(cpi);
else
vp8_setup_inter_frame(cpi);
}
Revised coding using adaptive mode context to depend on frame type A previous commit 76feb965 made the vp8_mode_context adaptive on a frame frame basis, this commit further made the coding context adaptive to two frame types separately. Tests on derf set showed a further small gain on all metrics: avg psnr 0.10%, glb psnr: 0.11%, ssim: 0.08% http://www.corp.google.com/~yaowu/no_crawl/newNearMode_1209.html Change-Id: I7b3e32ec8729de1903d14a3f1213f1624b78cdee
2011-12-08 20:43:09 +01:00
Initial WebM release
2010-05-18 17:58:33 +02:00
// transform / motion compensation build reconstruction frame
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
Initial WebM release
2010-05-18 17:58:33 +02:00
vp8_encode_frame(cpi);
Encoder loopfilter running in its own thread In multithreaded mode the loopfilter is running in its own thread (filter level calculation and frame filtering). Filtering is mostly done in parallel with the bitstream packing. Before starting the packing the loopfilter level has to be calculated. Also any needed reference frame copying is done in the filter thread. Currently the encoder will create n+1 threads, where n > 1 is the number of threads specified by application and 1 is the extra filter thread. With n = 1 the encoder runs in single thread mode. There will never be more than n threads running concurrently. Change-Id: I4fb29b559a40275d6d3babb8727245c40fba931b
2011-02-25 12:42:05 +01:00
Changes to costing of skip. Update the costing of skip in the recode loop and rd code. Change-Id: I2e5ebbd7ddf201212b32441321e12626cd0423e9
2012-04-11 15:37:48 +02:00
// Update the skip mb flag probabilities based on the distribution
// seen in the last encoder iteration.
update_base_skip_probs( cpi );
Initial WebM release
2010-05-18 17:58:33 +02:00
vp8_clear_system_state(); //__asm emms;
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
// Dummy pack of the bitstream using up to date stats to get an
// accurate estimate of output frame size to determine if we need
// to recode.
vp8_save_coding_context(cpi);
a number of fixes to entropy stats collection 1. block types There are only three types of blocks for 8x8 transformed MBs, i.e. Y block with DC does not exist for 8x8 transformed MBs as all MB using 8x8 transform have 2nd order haar transform. This commit introduced a new macro BLOCK_TYPES_8X8 to reflect such fact. 2. context counters This commit also fixed the mixed of context_counters between 4x4 and 8x8 transformed MBs. The mixed use of the counters leads me to think the existing the context probabilities were not properly generated from 8x8 transformed MBs. 3. redundant collecting in recoding The commit also corrected the code that accumulates entropy stats by making sure stats only collected for final packing, not during the recode loop Change-Id: I029f09f8f60bd0c3240cc392ff5c6d05435e322c
2012-05-02 22:53:15 +02:00
cpi->dummy_packing = 1;
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
vp8_pack_bitstream(cpi, dest, size);
cpi->projected_frame_size = (*size) << 3;
vp8_restore_coding_context(cpi);
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
if (frame_over_shoot_limit == 0)
frame_over_shoot_limit = 1;
active_worst_qchanged = FALSE;
Fix CQ range and experimental KF sizing changes. The CQ level was not using the q_trans[] array to convert to a 0-127 range as per min and maxq Experimental change to try and match the reconstruction error for forced key frames approximately to that of the previous frame by means of the recode loop. Though this may cause extra recodes and the recode behavior has not been optimized, it can only happen on forced key frames. Change-Id: I1f7e42d526f1b1cb556dd461eff1a692bd1b5b2f
2011-01-17 18:23:11 +01:00
// Special case handling for forced key frames
if ( (cm->frame_type == KEY_FRAME) && cpi->this_key_frame_forced )
{
int last_q = Q;
int kf_err = vp8_calc_ss_err(cpi->Source,
&cm->yv12_fb[cm->new_fb_idx],
IF_RTCD(&cpi->rtcd.variance));
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
int high_err_target = cpi->ambient_err;
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
int low_err_target = (cpi->ambient_err >> 1);
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
Possible divide by 0 error. Put traps to prevent two possible divide by 0 errors. Change-Id: Ia415b945244253dcdd12f54f1f157f9ca8c94d6b
2012-01-18 12:10:51 +01:00
// Prevent possible divide by zero error below for perfect KF
kf_err += (!kf_err);
T8x8 zbin and rate control changes. Some adjustments to zbin for t8x8. Changes to rules for sizing forced key frames. Some extra stats output in tmp.stt. Approximate gain on YT-hd set 0.5% There are still issues in sizing key frames and gf/arf frames when the image is largely static. These in part relate to problems with cost estimates in the recode loop. Change-Id: I6f0159dc8a8faeab4115a19c668d442491619a68
2012-03-27 14:01:34 +02:00
// The key frame is not good enough or we can afford
// to make it better without undue risk of popping.
if ( ( (kf_err > high_err_target) &&
(cpi->projected_frame_size <= frame_over_shoot_limit) ) ||
( (kf_err > low_err_target) &&
(cpi->projected_frame_size <= frame_under_shoot_limit) ) )
Fix CQ range and experimental KF sizing changes. The CQ level was not using the q_trans[] array to convert to a 0-127 range as per min and maxq Experimental change to try and match the reconstruction error for forced key frames approximately to that of the previous frame by means of the recode loop. Though this may cause extra recodes and the recode behavior has not been optimized, it can only happen on forced key frames. Change-Id: I1f7e42d526f1b1cb556dd461eff1a692bd1b5b2f
2011-01-17 18:23:11 +01:00
{
// Lower q_high
q_high = (Q > q_low) ? (Q - 1) : q_low;
// Adjust Q
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
Q = (Q * high_err_target) / kf_err;
if ( Q < ((q_high + q_low) >> 1))
Q = (q_high + q_low) >> 1;
Fix CQ range and experimental KF sizing changes. The CQ level was not using the q_trans[] array to convert to a 0-127 range as per min and maxq Experimental change to try and match the reconstruction error for forced key frames approximately to that of the previous frame by means of the recode loop. Though this may cause extra recodes and the recode behavior has not been optimized, it can only happen on forced key frames. Change-Id: I1f7e42d526f1b1cb556dd461eff1a692bd1b5b2f
2011-01-17 18:23:11 +01:00
}
// The key frame is much better than the previous frame
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
else if ( (kf_err < low_err_target) &&
(cpi->projected_frame_size >= frame_under_shoot_limit) )
Fix CQ range and experimental KF sizing changes. The CQ level was not using the q_trans[] array to convert to a 0-127 range as per min and maxq Experimental change to try and match the reconstruction error for forced key frames approximately to that of the previous frame by means of the recode loop. Though this may cause extra recodes and the recode behavior has not been optimized, it can only happen on forced key frames. Change-Id: I1f7e42d526f1b1cb556dd461eff1a692bd1b5b2f
2011-01-17 18:23:11 +01:00
{
// Raise q_low
q_low = (Q < q_high) ? (Q + 1) : q_high;
// Adjust Q
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
Q = (Q * low_err_target) / kf_err;
if ( Q > ((q_high + q_low + 1) >> 1))
Q = (q_high + q_low + 1) >> 1;
Fix CQ range and experimental KF sizing changes. The CQ level was not using the q_trans[] array to convert to a 0-127 range as per min and maxq Experimental change to try and match the reconstruction error for forced key frames approximately to that of the previous frame by means of the recode loop. Though this may cause extra recodes and the recode behavior has not been optimized, it can only happen on forced key frames. Change-Id: I1f7e42d526f1b1cb556dd461eff1a692bd1b5b2f
2011-01-17 18:23:11 +01:00
}
// Clamp Q to upper and lower limits:
if (Q > q_high)
Q = q_high;
else if (Q < q_low)
Q = q_low;
Loop = ((Q != last_q)) ? TRUE : FALSE;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
// Is the projected frame size out of range and are we allowed to attempt to recode.
Fix CQ range and experimental KF sizing changes. The CQ level was not using the q_trans[] array to convert to a 0-127 range as per min and maxq Experimental change to try and match the reconstruction error for forced key frames approximately to that of the previous frame by means of the recode loop. Though this may cause extra recodes and the recode behavior has not been optimized, it can only happen on forced key frames. Change-Id: I1f7e42d526f1b1cb556dd461eff1a692bd1b5b2f
2011-01-17 18:23:11 +01:00
else if ( recode_loop_test( cpi,
Replaced recode loop test with a function call Replaced existing code to decide if a frame recode is required with a function call. This is to simplify addition of extra clauses that may be needed for the planned constrained quality mode. Also fixed a bug where by alt ref not considered in the test. Change-Id: I3d40bb21abe3e19f8456761e6849deb171738b60
2010-11-17 16:12:04 +01:00
frame_over_shoot_limit, frame_under_shoot_limit,
Q, top_index, bottom_index ) )
Initial WebM release
2010-05-18 17:58:33 +02:00
{
int last_q = Q;
int Retries = 0;
// Frame size out of permitted range:
// Update correction factor & compute new Q to try...
Replaced recode loop test with a function call Replaced existing code to decide if a frame recode is required with a function call. This is to simplify addition of extra clauses that may be needed for the planned constrained quality mode. Also fixed a bug where by alt ref not considered in the test. Change-Id: I3d40bb21abe3e19f8456761e6849deb171738b60
2010-11-17 16:12:04 +01:00
// Frame is too large
if (cpi->projected_frame_size > cpi->this_frame_target)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
q_low = (Q < q_high) ? (Q + 1) : q_high; // Raise Qlow as to at least the current value
Fix CQ range and experimental KF sizing changes. The CQ level was not using the q_trans[] array to convert to a 0-127 range as per min and maxq Experimental change to try and match the reconstruction error for forced key frames approximately to that of the previous frame by means of the recode loop. Though this may cause extra recodes and the recode behavior has not been optimized, it can only happen on forced key frames. Change-Id: I1f7e42d526f1b1cb556dd461eff1a692bd1b5b2f
2011-01-17 18:23:11 +01:00
if (cpi->zbin_over_quant > 0) // If we are using over quant do the same for zbin_oq_low
Initial WebM release
2010-05-18 17:58:33 +02:00
zbin_oq_low = (cpi->zbin_over_quant < zbin_oq_high) ? (cpi->zbin_over_quant + 1) : zbin_oq_high;
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
if ( undershoot_seen || (loop_count > 1) )
Initial WebM release
2010-05-18 17:58:33 +02:00
{
// Update rate_correction_factor unless cpi->active_worst_quality has changed.
if (!active_worst_qchanged)
vp8_update_rate_correction_factors(cpi, 1);
Q = (q_high + q_low + 1) / 2;
// Adjust cpi->zbin_over_quant (only allowed when Q is max)
if (Q < MAXQ)
cpi->zbin_over_quant = 0;
else
{
zbin_oq_low = (cpi->zbin_over_quant < zbin_oq_high) ? (cpi->zbin_over_quant + 1) : zbin_oq_high;
cpi->zbin_over_quant = (zbin_oq_high + zbin_oq_low) / 2;
}
}
else
{
// Update rate_correction_factor unless cpi->active_worst_quality has changed.
if (!active_worst_qchanged)
vp8_update_rate_correction_factors(cpi, 0);
Q = vp8_regulate_q(cpi, cpi->this_frame_target);
while (((Q < q_low) || (cpi->zbin_over_quant < zbin_oq_low)) && (Retries < 10))
{
vp8_update_rate_correction_factors(cpi, 0);
Q = vp8_regulate_q(cpi, cpi->this_frame_target);
Retries ++;
}
}
overshoot_seen = TRUE;
}
Replaced recode loop test with a function call Replaced existing code to decide if a frame recode is required with a function call. This is to simplify addition of extra clauses that may be needed for the planned constrained quality mode. Also fixed a bug where by alt ref not considered in the test. Change-Id: I3d40bb21abe3e19f8456761e6849deb171738b60
2010-11-17 16:12:04 +01:00
// Frame is too small
Initial WebM release
2010-05-18 17:58:33 +02:00
else
{
if (cpi->zbin_over_quant == 0)
q_high = (Q > q_low) ? (Q - 1) : q_low; // Lower q_high if not using over quant
else // else lower zbin_oq_high
zbin_oq_high = (cpi->zbin_over_quant > zbin_oq_low) ? (cpi->zbin_over_quant - 1) : zbin_oq_low;
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
if ( overshoot_seen || (loop_count > 1) )
Initial WebM release
2010-05-18 17:58:33 +02:00
{
// Update rate_correction_factor unless cpi->active_worst_quality has changed.
if (!active_worst_qchanged)
vp8_update_rate_correction_factors(cpi, 1);
Q = (q_high + q_low) / 2;
// Adjust cpi->zbin_over_quant (only allowed when Q is max)
if (Q < MAXQ)
cpi->zbin_over_quant = 0;
else
cpi->zbin_over_quant = (zbin_oq_high + zbin_oq_low) / 2;
}
else
{
// Update rate_correction_factor unless cpi->active_worst_quality has changed.
if (!active_worst_qchanged)
vp8_update_rate_correction_factors(cpi, 0);
Q = vp8_regulate_q(cpi, cpi->this_frame_target);
KF/GF Pulsing This change is designed to try and reduce pulsing effects when moving with a complex transition like a fade, into an easy or static section in an otherwise difficult clip in CQ mode. The active CQ level is relaxed down to the user entered level for frames that are generating less than the passed in minimum bandwidth. Change-Id: Id6d8b551daad4f489c087bd742bc95418a95f3f0
2011-01-14 12:34:53 +01:00
// Special case reset for qlow for constrained quality.
// This should only trigger where there is very substantial
// undershoot on a frame and the auto cq level is above
// the user passsed in value.
if ( (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY) &&
(Q < q_low) )
{
q_low = Q;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
while (((Q > q_high) || (cpi->zbin_over_quant > zbin_oq_high)) && (Retries < 10))
{
vp8_update_rate_correction_factors(cpi, 0);
Q = vp8_regulate_q(cpi, cpi->this_frame_target);
Retries ++;
}
}
undershoot_seen = TRUE;
}
// Clamp Q to upper and lower limits:
if (Q > q_high)
Q = q_high;
else if (Q < q_low)
Q = q_low;
// Clamp cpi->zbin_over_quant
cpi->zbin_over_quant = (cpi->zbin_over_quant < zbin_oq_low) ? zbin_oq_low : (cpi->zbin_over_quant > zbin_oq_high) ? zbin_oq_high : cpi->zbin_over_quant;
//Loop = ((Q != last_q) || (last_zbin_oq != cpi->zbin_over_quant)) ? TRUE : FALSE;
Loop = ((Q != last_q)) ? TRUE : FALSE;
last_zbin_oq = cpi->zbin_over_quant;
}
else
Loop = FALSE;
if (cpi->is_src_frame_alt_ref)
Loop = FALSE;
Adds search option for best interpolation filter. Adds a speed feature to conduct a brute force search among a set of available interpolation filters for the best one in an RD sense. There is a gain of 0.4% on derf, 1.0% on Std-HD. Patch 2: A macro added to determine if the encoder state is reset for each new filter tried. Patch 3: rebase, also fixes a bug (decodframe.c) introduced by a couple of missing function pointer assignements. Patch 4: rebase. Change-Id: Ic9ccca9d8c35c6af557449ae867391a2f996cc29
2012-04-16 23:53:37 +02:00
#if CONFIG_ENHANCED_INTERP
if (Loop == FALSE && cm->frame_type != KEY_FRAME && sf->search_best_filter)
{
if (mcomp_filter_index < mcomp_filters)
{
INT64 err = vp8_calc_ss_err(cpi->Source,
&cm->yv12_fb[cm->new_fb_idx],
IF_RTCD(&cpi->rtcd.variance));
INT64 rate = cpi->projected_frame_size << 8;
mcomp_filter_cost[mcomp_filter_index] =
(RDCOST(cpi->RDMULT, cpi->RDDIV, rate, err));
mcomp_filter_index++;
if (mcomp_filter_index < mcomp_filters)
{
cm->mcomp_filter_type = mcomp_filters_to_search[mcomp_filter_index];
loop_count = -1;
Loop = TRUE;
}
else
{
int f;
INT64 best_cost = mcomp_filter_cost[0];
int mcomp_best_filter = mcomp_filters_to_search[0];
for (f = 1; f < mcomp_filters; f++)
{
if (mcomp_filter_cost[f] < best_cost)
{
mcomp_best_filter = mcomp_filters_to_search[f];
best_cost = mcomp_filter_cost[f];
}
}
if (mcomp_best_filter != mcomp_filters_to_search[mcomp_filters-1])
{
loop_count = -1;
Loop = TRUE;
cm->mcomp_filter_type = mcomp_best_filter;
}
/*
printf(" best filter = %d, ( ", mcomp_best_filter);
for (f=0;f<mcomp_filters; f++) printf("%d ", mcomp_filter_cost[f]);
printf(")\n");
*/
}
#if RESET_FOREACH_FILTER
if (Loop == TRUE)
{
overshoot_seen = FALSE;
undershoot_seen = FALSE;
zbin_oq_low = zbin_oq_low0;
zbin_oq_high = zbin_oq_high0;
q_low = q_low0;
q_high = q_high0;
Q = Q0;
cpi->zbin_over_quant = last_zbin_oq = last_zbin_oq0;
cpi->rate_correction_factor = rate_correction_factor0;
cpi->gf_rate_correction_factor = gf_rate_correction_factor0;
cpi->active_best_quality = active_best_quality0;
cpi->active_worst_quality = active_worst_quality0;
}
#endif
}
}
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
if (Loop == TRUE)
{
loop_count++;
changed configure option name to reduce confusion Renamed configure option "enable-psnr" to "enable-internal-stats" to better reflect the purpose of the option and eliminate the confusion reported in http://code.google.com/p/webm/issues/detail?id=35 Change-Id: If72df6fdb9f1e33dab1329240ba4d8911d2f1f7a
2011-04-29 18:37:59 +02:00
#if CONFIG_INTERNAL_STATS
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->tot_recode_hits++;
#endif
}
}
while (Loop == TRUE);
Fix CQ range and experimental KF sizing changes. The CQ level was not using the q_trans[] array to convert to a 0-127 range as per min and maxq Experimental change to try and match the reconstruction error for forced key frames approximately to that of the previous frame by means of the recode loop. Though this may cause extra recodes and the recode behavior has not been optimized, it can only happen on forced key frames. Change-Id: I1f7e42d526f1b1cb556dd461eff1a692bd1b5b2f
2011-01-17 18:23:11 +01:00
// Special case code to reduce pulsing when key frames are forced at a
// fixed interval. Note the reconstruction error if it is the frame before
// the force key frame
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
if ( cpi->next_key_frame_forced && (cpi->twopass.frames_to_key == 0) )
Fix CQ range and experimental KF sizing changes. The CQ level was not using the q_trans[] array to convert to a 0-127 range as per min and maxq Experimental change to try and match the reconstruction error for forced key frames approximately to that of the previous frame by means of the recode loop. Though this may cause extra recodes and the recode behavior has not been optimized, it can only happen on forced key frames. Change-Id: I1f7e42d526f1b1cb556dd461eff1a692bd1b5b2f
2011-01-17 18:23:11 +01:00
{
cpi->ambient_err = vp8_calc_ss_err(cpi->Source,
&cm->yv12_fb[cm->new_fb_idx],
IF_RTCD(&cpi->rtcd.variance));
}
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
// This frame's MVs are saved and will be used in next frame's MV
// prediction. Last frame has one more line(add to bottom) and one
// more column(add to right) than cm->mip. The edge elements are
// initialized to 0.
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
if(cm->show_frame) //do not save for altref frame
{
Refine motion vector prediction for NEWMV mode Adjust checking points in motion vector prediction to better cover possible movements, and get a better prediction. Tests on test clips showed a 0.1% improvement in SSIM, and no change in PSNR and performance. Change-Id: Ifdab05d35e10faea1445c61bb73debf888c9d2f8
2011-01-25 21:54:34 +01:00
int mb_row;
int mb_col;
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
MODE_INFO *tmp = cm->mip;
Refine motion vector prediction for NEWMV mode Adjust checking points in motion vector prediction to better cover possible movements, and get a better prediction. Tests on test clips showed a 0.1% improvement in SSIM, and no change in PSNR and performance. Change-Id: Ifdab05d35e10faea1445c61bb73debf888c9d2f8
2011-01-25 21:54:34 +01:00
if(cm->frame_type != KEY_FRAME)
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
{
Refine motion vector prediction for NEWMV mode Adjust checking points in motion vector prediction to better cover possible movements, and get a better prediction. Tests on test clips showed a 0.1% improvement in SSIM, and no change in PSNR and performance. Change-Id: Ifdab05d35e10faea1445c61bb73debf888c9d2f8
2011-01-25 21:54:34 +01:00
for (mb_row = 0; mb_row < cm->mb_rows+1; mb_row ++)
{
for (mb_col = 0; mb_col < cm->mb_cols+1; mb_col ++)
{
if(tmp->mbmi.ref_frame != INTRA_FRAME)
cpi->lfmv[mb_col + mb_row*(cm->mode_info_stride+1)].as_int = tmp->mbmi.mv.as_int;
cpi->lf_ref_frame_sign_bias[mb_col + mb_row*(cm->mode_info_stride+1)] = cm->ref_frame_sign_bias[tmp->mbmi.ref_frame];
cpi->lf_ref_frame[mb_col + mb_row*(cm->mode_info_stride+1)] = tmp->mbmi.ref_frame;
tmp++;
}
}
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
}
}
Merge "Refine motion vector prediction for NEWMV mode"
2011-01-27 17:10:53 +01:00
// Update the GF useage maps.
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
// This is done after completing the compression of a frame when all modes
// etc. are finalized but before loop filter
Moved gf_active code to encoder only The gf_active code is only used by the encoder, so it was moved from common and decoder. Change-Id: Iada15acd5b2b33ff70c34668ca87d4cfd0d05025
2010-08-11 17:02:31 +02:00
vp8_update_gf_useage_maps(cpi, cm, &cpi->mb);
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cm->frame_type == KEY_FRAME)
cm->refresh_last_frame = 1;
disable compilation of debugging code This patch avoids compiling some debugging code in onyx_if.c. The most significant fix is to avoid generating code for vp8_write_yuv_frame, which is never called. Some other code was removed by the dead code elimination performed by the compiler, and this patch does it with the preprocessor instead. There are advantages both ways. Change-Id: I044fd43179d2e947553f0d6f2cad5b40907ac458
2010-09-24 17:10:25 +02:00
#if 0
Initial WebM release
2010-05-18 17:58:33 +02:00
{
FILE *f = fopen("gfactive.stt", "a");
Moved gf_active code to encoder only The gf_active code is only used by the encoder, so it was moved from common and decoder. Change-Id: Iada15acd5b2b33ff70c34668ca87d4cfd0d05025
2010-08-11 17:02:31 +02:00
fprintf(f, "%8d %8d %8d %8d %8d\n", cm->current_video_frame, (100 * cpi->gf_active_count) / (cpi->common.mb_rows * cpi->common.mb_cols), cpi->this_iiratio, cpi->next_iiratio, cm->refresh_golden_frame);
Initial WebM release
2010-05-18 17:58:33 +02:00
fclose(f);
}
disable compilation of debugging code This patch avoids compiling some debugging code in onyx_if.c. The most significant fix is to avoid generating code for vp8_write_yuv_frame, which is never called. Some other code was removed by the dead code elimination performed by the compiler, and this patch does it with the preprocessor instead. There are advantages both ways. Change-Id: I044fd43179d2e947553f0d6f2cad5b40907ac458
2010-09-24 17:10:25 +02:00
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
Do not copy data between encoder reference buffers. Golden and ALT reference buffers were refreshed by copying from the new buffer. Replaced this by index manipulation. Also moved all the reference frame updates to one function for easier tracking. Change-Id: Icd3e534e7e2c8c5567168d222e6a64a96aae24a1
2011-04-20 14:12:23 +02:00
cm->frame_to_show = &cm->yv12_fb[cm->new_fb_idx];
Encoder loopfilter running in its own thread In multithreaded mode the loopfilter is running in its own thread (filter level calculation and frame filtering). Filtering is mostly done in parallel with the bitstream packing. Before starting the packing the loopfilter level has to be calculated. Also any needed reference frame copying is done in the filter thread. Currently the encoder will create n+1 threads, where n > 1 is the number of threads specified by application and 1 is the extra filter thread. With n = 1 the encoder runs in single thread mode. There will never be more than n threads running concurrently. Change-Id: I4fb29b559a40275d6d3babb8727245c40fba931b
2011-02-25 12:42:05 +01:00
changed find_near_mvs search to include a mb from last frame This is an experiment to include a mv contribution from last frame to nearest and near mv definition. Initial test showed some small though consistent gain. latest patch slightly better result ~.13%-~.18%. TODO: the entropy used to encode the mode choice, i.e. the mv counts based conditional distribution of modes should be re-collected to reflect this change, it is expected that there is some further gain from that. Change-Id: Ief1e284a36d8aa56b49ae5b360c91419ec494fa4
2011-11-16 01:16:30 +01:00
#if WRITE_RECON_BUFFER
if(cm->show_frame)
added separate entropy context for alt_ref This commit added code to keep track of separate entropy contexts for normal frames and alt ref frames. The underly assumption was that the two type of frames have different entropy characteristics given they typically have quite different quantization levels. By keeping entropy contexts separate, it helps the entropy context distribution to be more closely adapted to each frame type. Tests on derf set showed a good and very consistent gain on all clips on all metrics, avg psnr: 0.89%, overall psnr: 0.84% and ssim 0.93%. http://www.corp.google.com/~yaowu/no_crawl/mulcontext.html Change-Id: I15bc9697f6ff7829042911fe0c62930585d7e65d
2011-12-01 01:36:46 +01:00
write_cx_frame_to_file(cm->frame_to_show,
cm->current_video_frame);
else
write_cx_frame_to_file(cm->frame_to_show,
cm->current_video_frame+1000);
changed find_near_mvs search to include a mb from last frame This is an experiment to include a mv contribution from last frame to nearest and near mv definition. Initial test showed some small though consistent gain. latest patch slightly better result ~.13%-~.18%. TODO: the entropy used to encode the mode choice, i.e. the mv counts based conditional distribution of modes should be re-collected to reflect this change, it is expected that there is some further gain from that. Change-Id: Ief1e284a36d8aa56b49ae5b360c91419ec494fa4
2011-11-16 01:16:30 +01:00
#endif
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
// Pick the loop filter level for the frame.
loopfilter_frame(cpi, cm);
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
cosmetic: remove unnecessary scope Clean up some unnecessary scoping around pick_filter_level. Change-Id: Ic57fa33e3fcae37fe6beae977e5743783399d5af
2011-02-18 14:46:07 +01:00
// build the bitstream
a number of fixes to entropy stats collection 1. block types There are only three types of blocks for 8x8 transformed MBs, i.e. Y block with DC does not exist for 8x8 transformed MBs as all MB using 8x8 transform have 2nd order haar transform. This commit introduced a new macro BLOCK_TYPES_8X8 to reflect such fact. 2. context counters This commit also fixed the mixed of context_counters between 4x4 and 8x8 transformed MBs. The mixed use of the counters leads me to think the existing the context probabilities were not properly generated from 8x8 transformed MBs. 3. redundant collecting in recoding The commit also corrected the code that accumulates entropy stats by making sure stats only collected for final packing, not during the recode loop Change-Id: I029f09f8f60bd0c3240cc392ff5c6d05435e322c
2012-05-02 22:53:15 +02:00
cpi->dummy_packing = 0;
cosmetic: remove unnecessary scope Clean up some unnecessary scoping around pick_filter_level. Change-Id: Ic57fa33e3fcae37fe6beae977e5743783399d5af
2011-02-18 14:46:07 +01:00
vp8_pack_bitstream(cpi, dest, size);
Refactoring of encode loop and bitstream packing Some code re-factored / moved to allow the main pack operation inside the recode loop so that the size estimate is accurate. Deletion of some redundant code relating to one pass. Aproximate improvement over March 27 code base: Derf 0.0%, YT 0.5%, YThd 0.3% Std_hd 0.25% Change-Id: Id2d071794ab44f0b52935f6fcdb5733d09a6bb86
2012-04-11 16:44:14 +02:00
update_reference_frames(cm);
Adaptive entropy coding of coefficients, modes, mv. This patch incorporates adaptive entropy coding of coefficient tokens, and mode/mv information based on distributions encountered in a frame. Specifically, there is an initial forward update to the probabilities in the bitstream as before for coding the symbols in the frame, however at the end of decoding each frame, the forward update to the probabilities is reverted and instead the probabilities are updated towards the actual distributions encountered within the frame. The amount of update is weighted by the number of hits within each context. Results on derf/hd/std-hd are all up by 1.6%. On derf, the most of the gains come from coefficients, however for the hd and std-hd sets, the most of the gains come from the mode/mv information updates. Change-Id: I708c0e11fdacafee04940fe7ae159ba6844005fd
2012-06-06 00:25:07 +02:00
#if CONFIG_ADAPTIVE_ENTROPY
vp8_copy(cpi->common.fc.coef_counts, cpi->coef_counts);
vp8_copy(cpi->common.fc.coef_counts_8x8, cpi->coef_counts_8x8);
vp8_adapt_coef_probs(&cpi->common);
if (cpi->common.frame_type != KEY_FRAME)
{
vp8_copy(cpi->common.fc.ymode_counts, cpi->ymode_count);
vp8_copy(cpi->common.fc.uv_mode_counts, cpi->y_uv_mode_count);
vp8_copy(cpi->common.fc.bmode_counts, cpi->bmode_count);
vp8_copy(cpi->common.fc.i8x8_mode_counts, cpi->i8x8_mode_count);
vp8_adapt_mode_probs(&cpi->common);
vp8_copy(cpi->common.fc.MVcount, cpi->MVcount);
#if CONFIG_HIGH_PRECISION_MV
vp8_copy(cpi->common.fc.MVcount_hp, cpi->MVcount_hp);
#endif
vp8_adapt_mv_probs(&cpi->common);
}
#endif /* CONFIG_ADAPTIVE_ENTROPY */
Initial refactoring of high_precision mv code. This is the first patch for refactoring of the code related to high-precision mv, so that 1/4 and 1/8 pel motion vectors can co-exist in the same bit-stream by use of a frame level flag. The current patch works fine for only use of 1/4th and only use of 1/8th pel mv, but there are some issues with the mode switching in between. Subsequent patches on this change Id will fix the remaining issues. Patch 2: Adds fixes to make sure that multiple mv precisions can co-exist in the bit-stream. Frame level switching has been tested to work correctly. Patch 3: Fixes lines exceeding 80 char Patch 4: http://www.corp.google.com/~debargha/vp8_results/enhinterp.html Results on derf after ssse3 bugfix, compared to everything enabled but the 8-tap, 1/8-subpel and 1/16-subpel uv. Overall the gains are about 3% now. Hopefully there are no more bugs lingering. Apparently the sse3 bug affected the quartel subpel results more than the eighth pel ones (which is understandabale because one bad predictor due to the bug, matters less if there are a lot more subpel options available as in the 1/8 subpel case). The results in the 4th column correspond to the current settings. The first two columns correspond to two settings of adaptive switching of the 1/4 or 1/8 subpel mode based on initial Q estimate. These do not work as good as just using 1/8 all the time yet. Change-Id: I3ef392ad338329f4d68a85257a49f2b14f3af472
2012-02-27 19:22:38 +01:00
Encoder loopfilter running in its own thread In multithreaded mode the loopfilter is running in its own thread (filter level calculation and frame filtering). Filtering is mostly done in parallel with the bitstream packing. Before starting the packing the loopfilter level has to be calculated. Also any needed reference frame copying is done in the filter thread. Currently the encoder will create n+1 threads, where n > 1 is the number of threads specified by application and 1 is the extra filter thread. With n = 1 the encoder runs in single thread mode. There will never be more than n threads running concurrently. Change-Id: I4fb29b559a40275d6d3babb8727245c40fba931b
2011-02-25 12:42:05 +01:00
/* Move storing frame_type out of the above loop since it is also
* needed in motion search besides loopfilter */
Do not copy data between encoder reference buffers. Golden and ALT reference buffers were refreshed by copying from the new buffer. Replaced this by index manipulation. Also moved all the reference frame updates to one function for easier tracking. Change-Id: Icd3e534e7e2c8c5567168d222e6a64a96aae24a1
2011-04-20 14:12:23 +02:00
cm->last_frame_type = cm->frame_type;
Initial WebM release
2010-05-18 17:58:33 +02:00
T8x8 zbin and rate control changes. Some adjustments to zbin for t8x8. Changes to rules for sizing forced key frames. Some extra stats output in tmp.stt. Approximate gain on YT-hd set 0.5% There are still issues in sizing key frames and gf/arf frames when the image is largely static. These in part relate to problems with cost estimates in the recode loop. Change-Id: I6f0159dc8a8faeab4115a19c668d442491619a68
2012-03-27 14:01:34 +02:00
// Keep a copy of the size estimate used in the loop
loop_size_estimate = cpi->projected_frame_size;
Initial WebM release
2010-05-18 17:58:33 +02:00
// Update rate control heuristics
cpi->total_byte_count += (*size);
cpi->projected_frame_size = (*size) << 3;
if (!active_worst_qchanged)
vp8_update_rate_correction_factors(cpi, 2);
cpi->last_q[cm->frame_type] = cm->base_qindex;
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
// Keep record of last boosted (KF/KF/ARF) Q value.
// If the current frame is coded at a lower Q then we also update it.
// If all mbs in this group are skipped only update if the Q value is
// better than that already stored.
// This is used to help set quality in forced key frames to reduce popping
if ( (cm->base_qindex < cpi->last_boosted_qindex) ||
( (cpi->static_mb_pct < 100) &&
( (cm->frame_type == KEY_FRAME) ||
cm->refresh_alt_ref_frame ||
(cm->refresh_golden_frame && !cpi->is_src_frame_alt_ref) ) ) )
{
cpi->last_boosted_qindex = cm->base_qindex;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cm->frame_type == KEY_FRAME)
{
vp8_adjust_key_frame_context(cpi);
}
// Keep a record of ambient average Q.
Further CQ, Key frame and ARF changes This code fixes a bug in the calculation of the minimum Q for alt ref frames. It also allows an extended gf/arf interval for sections of clips that completely static (or nearly so). Change-Id: I1a21aaa16d4f0578e5f99b13bebd78d59403c73b
2011-01-18 16:19:05 +01:00
if (cm->frame_type != KEY_FRAME)
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->avg_frame_qindex = (2 + 3 * cpi->avg_frame_qindex + cm->base_qindex) >> 2;
// Keep a record from which we can calculate the average Q excluding GF updates and key frames
if ((cm->frame_type != KEY_FRAME) && !cm->refresh_golden_frame && !cm->refresh_alt_ref_frame)
{
cpi->ni_frames++;
Some further QIndex issues with extended Q Resolved or factored out some further issues with Q index. Put in a 3rd order polynomial instead of less accurate power function as the best fit on gf and kf boost adjustment. Added avg_q value to use instead of ni_av_qi. Compute segment delta Q values based on avg_q. Fixed bug in adjust_maxq_qrange(). The extended range Q on the derf set, using standard data rates (which do not extend high enough to get big benefits) still show a shortfall of between 0.5 and 1% though so there would appear to be further issues that need to be tracked down. Change-Id: Icfd49b9f401906ba487ef1bef7d397048295d959
2011-12-06 15:48:52 +01:00
cpi->tot_q += vp8_convert_qindex_to_q(Q);
cpi->avg_q = cpi->tot_q / (double)cpi->ni_frames;
Initial WebM release
2010-05-18 17:58:33 +02:00
Two Pass VBR change Further experiment with restriction of the Q range. This uses the average non KF/GF/ARF quantizer, instead of just relying on the initial value. It is not such a strong constraint but there may be a reduced risk of rate misses. Change-Id: I424fe782a37a2f4e18c70805e240db55bfaa25ec
2011-01-10 17:41:53 +01:00
// Calculate the average Q for normal inter frames (not key or GFU
// frames).
Code base simplification. Removal of most code to do with 1 pass. Removal of cyclic refresh code. Change-Id: I74971082bc19dd76e795d4d2e781a0424cec5c8c
2012-02-17 17:29:03 +01:00
cpi->ni_tot_qi += Q;
cpi->ni_av_qi = (cpi->ni_tot_qi / cpi->ni_frames);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
// Update the buffer level variable.
Correction to buffer update for non-viewable frames. The code previously tested cpi->common.refresh_alt_ref_frame but there are situations where this flag may be set for viewable frames. The correct test should be !cm->show_frame. Change-Id: Ia1a600622992a4a68fe1d38ac23bf6b34b133688
2011-01-26 13:52:31 +01:00
// Non-viewable frames are a special case and are treated as pure overhead.
Revert "Improved 1-pass CBR rate control" This reverts commit b5ea2fbc2c1554769848774c836aad262af95072. Further testing showed noticable keyframe popping in some cases, reverting this for now to give time for a proper fix. Conflicts: vp8/encoder/onyx_if.c vp8/encoder/ratectrl.c Change-Id: I159f53d1bf0e24c035754ab3ded8ccfd58fd04af
2011-08-12 20:51:36 +02:00
if ( !cm->show_frame )
cpi->bits_off_target -= cpi->projected_frame_size;
else
cpi->bits_off_target += cpi->av_per_frame_bandwidth - cpi->projected_frame_size;
Initial WebM release
2010-05-18 17:58:33 +02:00
Clip buffer level to the maximum buffer size in CBR The buffer level was able to increase indefinitely rather than being clipped to the maximum buffer size specified by the user. This change checks the buffrer level and prevents it from going beyond the upper limit of the buffer. Change-Id: Ifff55f79d3c018e4d3d77e554b11ada543cc1654
2011-11-18 00:57:37 +01:00
// Clip the buffer level at the maximum buffer size
if (cpi->bits_off_target > cpi->oxcf.maximum_buffer_size)
cpi->bits_off_target = cpi->oxcf.maximum_buffer_size;
Initial WebM release
2010-05-18 17:58:33 +02:00
// Rolling monitors of whether we are over or underspending used to help regulate min and Max Q in two pass.
cpi->rolling_target_bits = ((cpi->rolling_target_bits * 3) + cpi->this_frame_target + 2) / 4;
cpi->rolling_actual_bits = ((cpi->rolling_actual_bits * 3) + cpi->projected_frame_size + 2) / 4;
cpi->long_rolling_target_bits = ((cpi->long_rolling_target_bits * 31) + cpi->this_frame_target + 16) / 32;
cpi->long_rolling_actual_bits = ((cpi->long_rolling_actual_bits * 31) + cpi->projected_frame_size + 16) / 32;
// Actual bits spent
cpi->total_actual_bits += cpi->projected_frame_size;
// Debug stats
cpi->total_target_vs_actual += (cpi->this_frame_target - cpi->projected_frame_size);
Revert "Improved 1-pass CBR rate control" This reverts commit b5ea2fbc2c1554769848774c836aad262af95072. Further testing showed noticable keyframe popping in some cases, reverting this for now to give time for a proper fix. Conflicts: vp8/encoder/onyx_if.c vp8/encoder/ratectrl.c Change-Id: I159f53d1bf0e24c035754ab3ded8ccfd58fd04af
2011-08-12 20:51:36 +02:00
cpi->buffer_level = cpi->bits_off_target;
Initial WebM release
2010-05-18 17:58:33 +02:00
// Update bits left to the kf and gf groups to account for overshoot or undershoot on these frames
if (cm->frame_type == KEY_FRAME)
{
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
cpi->twopass.kf_group_bits += cpi->this_frame_target - cpi->projected_frame_size;
Initial WebM release
2010-05-18 17:58:33 +02:00
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
if (cpi->twopass.kf_group_bits < 0)
cpi->twopass.kf_group_bits = 0 ;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
else if (cm->refresh_golden_frame || cm->refresh_alt_ref_frame)
{
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
cpi->twopass.gf_group_bits += cpi->this_frame_target - cpi->projected_frame_size;
Initial WebM release
2010-05-18 17:58:33 +02:00
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
if (cpi->twopass.gf_group_bits < 0)
cpi->twopass.gf_group_bits = 0 ;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Changes to costing of skip. Update the costing of skip in the recode loop and rd code. Change-Id: I2e5ebbd7ddf201212b32441321e12626cd0423e9
2012-04-11 15:37:48 +02:00
// Update the skip mb flag probabilities based on the distribution seen
// in this frame.
update_base_skip_probs( cpi );
Initial WebM release
2010-05-18 17:58:33 +02:00
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
#if 1 && CONFIG_INTERNAL_STATS
Initial WebM release
2010-05-18 17:58:33 +02:00
{
FILE *f = fopen("tmp.stt", "a");
T8x8 zbin and rate control changes. Some adjustments to zbin for t8x8. Changes to rules for sizing forced key frames. Some extra stats output in tmp.stt. Approximate gain on YT-hd set 0.5% There are still issues in sizing key frames and gf/arf frames when the image is largely static. These in part relate to problems with cost estimates in the recode loop. Change-Id: I6f0159dc8a8faeab4115a19c668d442491619a68
2012-03-27 14:01:34 +02:00
int recon_err;
Initial WebM release
2010-05-18 17:58:33 +02:00
vp8_clear_system_state(); //__asm emms;
T8x8 zbin and rate control changes. Some adjustments to zbin for t8x8. Changes to rules for sizing forced key frames. Some extra stats output in tmp.stt. Approximate gain on YT-hd set 0.5% There are still issues in sizing key frames and gf/arf frames when the image is largely static. These in part relate to problems with cost estimates in the recode loop. Change-Id: I6f0159dc8a8faeab4115a19c668d442491619a68
2012-03-27 14:01:34 +02:00
recon_err = vp8_calc_ss_err(cpi->Source,
&cm->yv12_fb[cm->new_fb_idx],
IF_RTCD(&cpi->rtcd.variance));
Two pass rate control code changes. This comitt brings accross changes from the public branch commit number Icf74d13af77437c08602571dc7a97e747cce5066. The main puurpose of this comit relates to CQ mode but it also includes some refactoring of the two pass code which I hope will make tuning the experimental branch for the new quantizer range a little less painfull. Change-Id: I278e989436a928fc1fe7761068960048f9d7a376
2011-11-23 18:18:31 +01:00
if (cpi->twopass.total_left_stats->coded_error != 0.0)
T8x8 zbin and rate control changes. Some adjustments to zbin for t8x8. Changes to rules for sizing forced key frames. Some extra stats output in tmp.stt. Approximate gain on YT-hd set 0.5% There are still issues in sizing key frames and gf/arf frames when the image is largely static. These in part relate to problems with cost estimates in the recode loop. Change-Id: I6f0159dc8a8faeab4115a19c668d442491619a68
2012-03-27 14:01:34 +02:00
fprintf(f, "%10d %10d %10d %10d %10d %10d %10d %10d"
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
"%7.2f %7.2f %7.2f %7.2f %7.2f %7.2f %7.2f"
Further work on extended Q range. Fixed bug in firspass.c call to vp8_initialize_rd_consts() This was passing in vp8_dc_quant(cm->base_qindex, cm->y1dc_delta_q) instead of (cm->base_qindex + cm->y1dc_delta_q). It just so happens that for the value 26 used for cm->base_qindex in the unextended Q case, the two give similar results. However, when using the extended Q range the two are very different. Also added more stats output and partly disabled another broken feature. Change-Id: Iddf6cf5ea8467c44b7c133f38e629f6ba6f2581e
2011-11-24 19:25:03 +01:00
"%6d %5d %5d %5d %8d %8.2f %10d %10.3f"
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
"%10.3f %8d %10d %10d %10d\n",
Control of active min quantizer for two pass. Create look up tables for controlling the active quantizer range. Some initial tuning to improve quality circa 0.5% on test set. Clean up of some stats output code Change-Id: Ia698a8525f8b8129a503cadace3ee73fe888f543
2010-09-29 13:03:19 +02:00
cpi->common.current_video_frame, cpi->this_frame_target,
T8x8 zbin and rate control changes. Some adjustments to zbin for t8x8. Changes to rules for sizing forced key frames. Some extra stats output in tmp.stt. Approximate gain on YT-hd set 0.5% There are still issues in sizing key frames and gf/arf frames when the image is largely static. These in part relate to problems with cost estimates in the recode loop. Change-Id: I6f0159dc8a8faeab4115a19c668d442491619a68
2012-03-27 14:01:34 +02:00
cpi->projected_frame_size, loop_size_estimate,
Control of active min quantizer for two pass. Create look up tables for controlling the active quantizer range. Some initial tuning to improve quality circa 0.5% on test set. Clean up of some stats output code Change-Id: Ia698a8525f8b8129a503cadace3ee73fe888f543
2010-09-29 13:03:19 +02:00
(cpi->projected_frame_size - cpi->this_frame_target),
(int)cpi->total_target_vs_actual,
(cpi->oxcf.starting_buffer_level-cpi->bits_off_target),
Further work on extended Q range. Fixed bug in firspass.c call to vp8_initialize_rd_consts() This was passing in vp8_dc_quant(cm->base_qindex, cm->y1dc_delta_q) instead of (cm->base_qindex + cm->y1dc_delta_q). It just so happens that for the value 26 used for cm->base_qindex in the unextended Q case, the two give similar results. However, when using the extended Q range the two are very different. Also added more stats output and partly disabled another broken feature. Change-Id: Iddf6cf5ea8467c44b7c133f38e629f6ba6f2581e
2011-11-24 19:25:03 +01:00
(int)cpi->total_actual_bits,
vp8_convert_qindex_to_q(cm->base_qindex),
(double)vp8_dc_quant(cm->base_qindex,0)/4.0,
vp8_convert_qindex_to_q(cpi->active_best_quality),
vp8_convert_qindex_to_q(cpi->active_worst_quality),
Some further QIndex issues with extended Q Resolved or factored out some further issues with Q index. Put in a 3rd order polynomial instead of less accurate power function as the best fit on gf and kf boost adjustment. Added avg_q value to use instead of ni_av_qi. Compute segment delta Q values based on avg_q. Fixed bug in adjust_maxq_qrange(). The extended range Q on the derf set, using standard data rates (which do not extend high enough to get big benefits) still show a shortfall of between 0.5 and 1% though so there would appear to be further issues that need to be tracked down. Change-Id: Icfd49b9f401906ba487ef1bef7d397048295d959
2011-12-06 15:48:52 +01:00
cpi->avg_q,
Further work on extended Q range. Fixed bug in firspass.c call to vp8_initialize_rd_consts() This was passing in vp8_dc_quant(cm->base_qindex, cm->y1dc_delta_q) instead of (cm->base_qindex + cm->y1dc_delta_q). It just so happens that for the value 26 used for cm->base_qindex in the unextended Q case, the two give similar results. However, when using the extended Q range the two are very different. Also added more stats output and partly disabled another broken feature. Change-Id: Iddf6cf5ea8467c44b7c133f38e629f6ba6f2581e
2011-11-24 19:25:03 +01:00
vp8_convert_qindex_to_q(cpi->ni_av_qi),
vp8_convert_qindex_to_q(cpi->cq_target_quality),
Two pass rate control code changes. This comitt brings accross changes from the public branch commit number Icf74d13af77437c08602571dc7a97e747cce5066. The main puurpose of this comit relates to CQ mode but it also includes some refactoring of the two pass code which I hope will make tuning the experimental branch for the new quantizer range a little less painfull. Change-Id: I278e989436a928fc1fe7761068960048f9d7a376
2011-11-23 18:18:31 +01:00
cpi->zbin_over_quant,
CQ Mode The merge includes hooks to for CQ mode and other code changes merged from the test branch. CQ mode attempts to maintain a more stable quantizer within a clip whilst also trying to adhere to a guidline maximum bitrate. The existing target data rate parameter is used to specify the guideline maximum bitrate. A new parameter allows the user to specify a target CQ level. For normal (non kf/gf/arf) frames, the quantizer will not drop BELOW the user specified value (0-63). However, in some cases the encoder may choose to impose a target CQ that is above that specified by the user, if it estimates that consistent use of the target value is not compatible with guideline maximum bitrate. Change-Id: I2221f9eecae8cc3c431d36caf83503941b25e4c1
2011-01-07 19:29:37 +01:00
//cpi->avg_frame_qindex, cpi->zbin_over_quant,
Control of active min quantizer for two pass. Create look up tables for controlling the active quantizer range. Some initial tuning to improve quality circa 0.5% on test set. Clean up of some stats output code Change-Id: Ia698a8525f8b8129a503cadace3ee73fe888f543
2010-09-29 13:03:19 +02:00
cm->refresh_golden_frame, cm->refresh_alt_ref_frame,
cm->frame_type, cpi->gfu_boost,
Two pass rate control code changes. This comitt brings accross changes from the public branch commit number Icf74d13af77437c08602571dc7a97e747cce5066. The main puurpose of this comit relates to CQ mode but it also includes some refactoring of the two pass code which I hope will make tuning the experimental branch for the new quantizer range a little less painfull. Change-Id: I278e989436a928fc1fe7761068960048f9d7a376
2011-11-23 18:18:31 +01:00
cpi->twopass.est_max_qcorrection_factor,
(int)cpi->twopass.bits_left,
cpi->twopass.total_left_stats->coded_error,
(double)cpi->twopass.bits_left /
cpi->twopass.total_left_stats->coded_error,
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
cpi->tot_recode_hits, recon_err, cpi->kf_boost,
cpi->kf_zeromotion_pct);
Initial WebM release
2010-05-18 17:58:33 +02:00
else
T8x8 zbin and rate control changes. Some adjustments to zbin for t8x8. Changes to rules for sizing forced key frames. Some extra stats output in tmp.stt. Approximate gain on YT-hd set 0.5% There are still issues in sizing key frames and gf/arf frames when the image is largely static. These in part relate to problems with cost estimates in the recode loop. Change-Id: I6f0159dc8a8faeab4115a19c668d442491619a68
2012-03-27 14:01:34 +02:00
fprintf(f, "%10d %10d %10d %10d %10d %10d %10d %10d"
Rate control on static scenes plus Y2dc delta Q fix. A problem can arise on static clips with force key frames where attempts to avoid popping lead to a progressive reduction in key frame Q that ultimately may lead to unexpected overspend against the rate target. The changes in this patch help to insure that in such clips the quality of the key frames across the clip is more uniform (rather than starting bad and getting better - especially at low target rates). This patch also includes a fix that removes a delta on the Y2DC when the baseline q index < 4 as this is no longer needed. There is also a fix to try and prevent repeat single step Q adjustment in the recode loop leading to lots of recodes, especially where the use of forced skips as part of segmentation has made the impact of Q on the number of bits generated much smaller. Patch 2: Amend "last_boosted_qindex" calculation for arf overlay frames. Change-Id: Ia1feeb79ed8ed014e4239994fcf5e58e68fd9459
2012-01-11 15:05:57 +01:00
"%7.2f %7.2f %7.2f %7.2f %7.2f %7.2f %7.2f"
Further work on extended Q range. Fixed bug in firspass.c call to vp8_initialize_rd_consts() This was passing in vp8_dc_quant(cm->base_qindex, cm->y1dc_delta_q) instead of (cm->base_qindex + cm->y1dc_delta_q). It just so happens that for the value 26 used for cm->base_qindex in the unextended Q case, the two give similar results. However, when using the extended Q range the two are very different. Also added more stats output and partly disabled another broken feature. Change-Id: Iddf6cf5ea8467c44b7c133f38e629f6ba6f2581e
2011-11-24 19:25:03 +01:00
"%6d %5d %5d %5d %8d %8.2f %10d %10.3f"
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
"%8d %10d %10d %10d\n",
Control of active min quantizer for two pass. Create look up tables for controlling the active quantizer range. Some initial tuning to improve quality circa 0.5% on test set. Clean up of some stats output code Change-Id: Ia698a8525f8b8129a503cadace3ee73fe888f543
2010-09-29 13:03:19 +02:00
cpi->common.current_video_frame,
cpi->this_frame_target, cpi->projected_frame_size,
T8x8 zbin and rate control changes. Some adjustments to zbin for t8x8. Changes to rules for sizing forced key frames. Some extra stats output in tmp.stt. Approximate gain on YT-hd set 0.5% There are still issues in sizing key frames and gf/arf frames when the image is largely static. These in part relate to problems with cost estimates in the recode loop. Change-Id: I6f0159dc8a8faeab4115a19c668d442491619a68
2012-03-27 14:01:34 +02:00
loop_size_estimate,
Control of active min quantizer for two pass. Create look up tables for controlling the active quantizer range. Some initial tuning to improve quality circa 0.5% on test set. Clean up of some stats output code Change-Id: Ia698a8525f8b8129a503cadace3ee73fe888f543
2010-09-29 13:03:19 +02:00
(cpi->projected_frame_size - cpi->this_frame_target),
(int)cpi->total_target_vs_actual,
(cpi->oxcf.starting_buffer_level-cpi->bits_off_target),
Further work on extended Q range. Fixed bug in firspass.c call to vp8_initialize_rd_consts() This was passing in vp8_dc_quant(cm->base_qindex, cm->y1dc_delta_q) instead of (cm->base_qindex + cm->y1dc_delta_q). It just so happens that for the value 26 used for cm->base_qindex in the unextended Q case, the two give similar results. However, when using the extended Q range the two are very different. Also added more stats output and partly disabled another broken feature. Change-Id: Iddf6cf5ea8467c44b7c133f38e629f6ba6f2581e
2011-11-24 19:25:03 +01:00
(int)cpi->total_actual_bits,
vp8_convert_qindex_to_q(cm->base_qindex),
(double)vp8_dc_quant(cm->base_qindex,0)/4.0,
vp8_convert_qindex_to_q(cpi->active_best_quality),
vp8_convert_qindex_to_q(cpi->active_worst_quality),
Some further QIndex issues with extended Q Resolved or factored out some further issues with Q index. Put in a 3rd order polynomial instead of less accurate power function as the best fit on gf and kf boost adjustment. Added avg_q value to use instead of ni_av_qi. Compute segment delta Q values based on avg_q. Fixed bug in adjust_maxq_qrange(). The extended range Q on the derf set, using standard data rates (which do not extend high enough to get big benefits) still show a shortfall of between 0.5 and 1% though so there would appear to be further issues that need to be tracked down. Change-Id: Icfd49b9f401906ba487ef1bef7d397048295d959
2011-12-06 15:48:52 +01:00
cpi->avg_q,
Further work on extended Q range. Fixed bug in firspass.c call to vp8_initialize_rd_consts() This was passing in vp8_dc_quant(cm->base_qindex, cm->y1dc_delta_q) instead of (cm->base_qindex + cm->y1dc_delta_q). It just so happens that for the value 26 used for cm->base_qindex in the unextended Q case, the two give similar results. However, when using the extended Q range the two are very different. Also added more stats output and partly disabled another broken feature. Change-Id: Iddf6cf5ea8467c44b7c133f38e629f6ba6f2581e
2011-11-24 19:25:03 +01:00
vp8_convert_qindex_to_q(cpi->ni_av_qi),
vp8_convert_qindex_to_q(cpi->cq_target_quality),
Two pass rate control code changes. This comitt brings accross changes from the public branch commit number Icf74d13af77437c08602571dc7a97e747cce5066. The main puurpose of this comit relates to CQ mode but it also includes some refactoring of the two pass code which I hope will make tuning the experimental branch for the new quantizer range a little less painfull. Change-Id: I278e989436a928fc1fe7761068960048f9d7a376
2011-11-23 18:18:31 +01:00
cpi->zbin_over_quant,
CQ Mode The merge includes hooks to for CQ mode and other code changes merged from the test branch. CQ mode attempts to maintain a more stable quantizer within a clip whilst also trying to adhere to a guidline maximum bitrate. The existing target data rate parameter is used to specify the guideline maximum bitrate. A new parameter allows the user to specify a target CQ level. For normal (non kf/gf/arf) frames, the quantizer will not drop BELOW the user specified value (0-63). However, in some cases the encoder may choose to impose a target CQ that is above that specified by the user, if it estimates that consistent use of the target value is not compatible with guideline maximum bitrate. Change-Id: I2221f9eecae8cc3c431d36caf83503941b25e4c1
2011-01-07 19:29:37 +01:00
//cpi->avg_frame_qindex, cpi->zbin_over_quant,
Control of active min quantizer for two pass. Create look up tables for controlling the active quantizer range. Some initial tuning to improve quality circa 0.5% on test set. Clean up of some stats output code Change-Id: Ia698a8525f8b8129a503cadace3ee73fe888f543
2010-09-29 13:03:19 +02:00
cm->refresh_golden_frame, cm->refresh_alt_ref_frame,
cm->frame_type, cpi->gfu_boost,
Two pass rate control code changes. This comitt brings accross changes from the public branch commit number Icf74d13af77437c08602571dc7a97e747cce5066. The main puurpose of this comit relates to CQ mode but it also includes some refactoring of the two pass code which I hope will make tuning the experimental branch for the new quantizer range a little less painfull. Change-Id: I278e989436a928fc1fe7761068960048f9d7a376
2011-11-23 18:18:31 +01:00
cpi->twopass.est_max_qcorrection_factor,
(int)cpi->twopass.bits_left,
cpi->twopass.total_left_stats->coded_error,
Key frame and GF sizing changes. Changes to the equations for kf and gf minQ and to the boost calculations for kg and gf Change-Id: I312031c910e6a575334f49075c32f49a8dfff239
2012-06-07 14:39:20 +02:00
cpi->tot_recode_hits, recon_err, cpi->kf_boost,
cpi->kf_zeromotion_pct);
Initial WebM release
2010-05-18 17:58:33 +02:00
fclose(f);
Two pass rate control code changes. This comitt brings accross changes from the public branch commit number Icf74d13af77437c08602571dc7a97e747cce5066. The main puurpose of this comit relates to CQ mode but it also includes some refactoring of the two pass code which I hope will make tuning the experimental branch for the new quantizer range a little less painfull. Change-Id: I278e989436a928fc1fe7761068960048f9d7a376
2011-11-23 18:18:31 +01:00
if ( 0 )
Initial WebM release
2010-05-18 17:58:33 +02:00
{
FILE *fmodes = fopen("Modes.stt", "a");
int i;
Control of active min quantizer for two pass. Create look up tables for controlling the active quantizer range. Some initial tuning to improve quality circa 0.5% on test set. Clean up of some stats output code Change-Id: Ia698a8525f8b8129a503cadace3ee73fe888f543
2010-09-29 13:03:19 +02:00
fprintf(fmodes, "%6d:%1d:%1d:%1d ",
cpi->common.current_video_frame,
cm->frame_type, cm->refresh_golden_frame,
cm->refresh_alt_ref_frame);
Initial WebM release
2010-05-18 17:58:33 +02:00
for (i = 0; i < MAX_MODES; i++)
fprintf(fmodes, "%5d ", cpi->mode_chosen_counts[i]);
fprintf(fmodes, "\n");
fclose(fmodes);
}
}
Segment coding of mode and reference frame. Proof of concept test code that encodes mode and reference frame data at the segment level. Decode-able bit stream but some issues not yet resolved. As it this helps a little on a couple of clips but hurts on most as the basis for segmentation is unsound. To build and test, configure with --enable-experimental --enable-segfeatures Change-Id: I22a60774f69273523fb152db8c31f4b10b07c7f4
2011-09-30 17:45:16 +02:00
#endif
#if 0
// Debug stats for segment feature experiments.
print_seg_map(cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
// If this was a kf or Gf note the Q
if ((cm->frame_type == KEY_FRAME) || cm->refresh_golden_frame || cm->refresh_alt_ref_frame)
cm->last_kf_gf_q = cm->base_qindex;
if (cm->refresh_golden_frame == 1)
cm->frame_flags = cm->frame_flags | FRAMEFLAGS_GOLDEN;
else
cm->frame_flags = cm->frame_flags&~FRAMEFLAGS_GOLDEN;
if (cm->refresh_alt_ref_frame == 1)
cm->frame_flags = cm->frame_flags | FRAMEFLAGS_ALTREF;
else
cm->frame_flags = cm->frame_flags&~FRAMEFLAGS_ALTREF;
if (cm->refresh_last_frame & cm->refresh_golden_frame) // both refreshed
cpi->gold_is_last = 1;
else if (cm->refresh_last_frame ^ cm->refresh_golden_frame) // 1 refreshed but not the other
cpi->gold_is_last = 0;
if (cm->refresh_last_frame & cm->refresh_alt_ref_frame) // both refreshed
cpi->alt_is_last = 1;
else if (cm->refresh_last_frame ^ cm->refresh_alt_ref_frame) // 1 refreshed but not the other
cpi->alt_is_last = 0;
if (cm->refresh_alt_ref_frame & cm->refresh_golden_frame) // both refreshed
cpi->gold_is_alt = 1;
else if (cm->refresh_alt_ref_frame ^ cm->refresh_golden_frame) // 1 refreshed but not the other
cpi->gold_is_alt = 0;
cpi->ref_frame_flags = VP8_ALT_FLAG | VP8_GOLD_FLAG | VP8_LAST_FLAG;
if (cpi->gold_is_last)
Changed unary operator from ! to ~ Since the intent is to reset the appropriate bit in ref_frame_flags not to test a logic condition. Prior result would always have been ref_frame_flags being set to 0. (Issue reported by dgohman, issue 47) Change-Id: I2c12502ed74c73cf38e98c9680e0249c29e16433
2010-06-18 16:58:18 +02:00
cpi->ref_frame_flags &= ~VP8_GOLD_FLAG;
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->alt_is_last)
Changed unary operator from ! to ~ Since the intent is to reset the appropriate bit in ref_frame_flags not to test a logic condition. Prior result would always have been ref_frame_flags being set to 0. (Issue reported by dgohman, issue 47) Change-Id: I2c12502ed74c73cf38e98c9680e0249c29e16433
2010-06-18 16:58:18 +02:00
cpi->ref_frame_flags &= ~VP8_ALT_FLAG;
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->gold_is_alt)
Changed unary operator from ! to ~ Since the intent is to reset the appropriate bit in ref_frame_flags not to test a logic condition. Prior result would always have been ref_frame_flags being set to 0. (Issue reported by dgohman, issue 47) Change-Id: I2c12502ed74c73cf38e98c9680e0249c29e16433
2010-06-18 16:58:18 +02:00
cpi->ref_frame_flags &= ~VP8_ALT_FLAG;
Initial WebM release
2010-05-18 17:58:33 +02:00
Code Cleanup. Removal of error_resilient_mode features. The interface has been left in place but does nothing. Change-Id: I2407863bd0d3c98407354507423ca48d29f63b17
2012-02-26 02:15:47 +01:00
if (cpi->oxcf.play_alternate && cm->refresh_alt_ref_frame && (cm->frame_type != KEY_FRAME))
// Update the alternate reference frame stats as appropriate.
update_alt_ref_frame_stats(cpi);
else
// Update the Golden frame stats as appropriate.
update_golden_frame_stats(cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cm->frame_type == KEY_FRAME)
{
// Tell the caller that the frame was coded as a key frame
*frame_flags = cm->frame_flags | FRAMEFLAGS_KEY;
// As this frame is a key frame the next defaults to an inter frame.
cm->frame_type = INTER_FRAME;
}
else
{
*frame_flags = cm->frame_flags&~FRAMEFLAGS_KEY;
}
// Clear the one shot update flags for segmentation map and mode/ref loop filter deltas.
Merge CONFIGURE_SEGMENTATION experiment. Removal of CONFIGURE_SEGMENTATION ifdefs. Removal of legacy support code fo the old coding mechanism. Use local reference "xd" for MACROBLOCKD structure in encode_frame_to_data_rate() Moved call to choose_segmap_coding_method() out of encode loop as the cost of segmentation is not properly accounted in the loop anyway. If this is desirable in the future it can be moved back. The use of this function to do all the analysis and set the probabilities also removes the need to track segment useage in threading code. Change-Id: I85bc8fd63440e7176c73d26cb742698f9b70cade
2011-11-15 17:15:23 +01:00
xd->update_mb_segmentation_map = 0;
xd->update_mb_segmentation_data = 0;
xd->mode_ref_lf_delta_update = 0;
Initial WebM release
2010-05-18 17:58:33 +02:00
// Dont increment frame counters if this was an altref buffer update not a real frame
if (cm->show_frame)
{
cm->current_video_frame++;
cpi->frames_since_key++;
}
// reset to normal state now that we are done.
disable compilation of debugging code This patch avoids compiling some debugging code in onyx_if.c. The most significant fix is to avoid generating code for vp8_write_yuv_frame, which is never called. Some other code was removed by the dead code elimination performed by the compiler, and this patch does it with the preprocessor instead. There are advantages both ways. Change-Id: I044fd43179d2e947553f0d6f2cad5b40907ac458
2010-09-24 17:10:25 +02:00
#if 0
Initial WebM release
2010-05-18 17:58:33 +02:00
{
char filename[512];
FILE *recon_file;
sprintf(filename, "enc%04d.yuv", (int) cm->current_video_frame);
recon_file = fopen(filename, "wb");
Swap alt/gold/new/last frame buffer ptrs instead of copying. At the end of the decode, frame buffers were being copied. The frames are not updated after the copy, they are just for reference on later frames. This change allows multiple references to the same frame buffer instead of copying it. Changes needed to be made to the encoder to handle this. The encoder is still doing frame buffer copies in similar places where pointer reference could be done. Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
2010-07-22 14:07:32 +02:00
fwrite(cm->yv12_fb[cm->lst_fb_idx].buffer_alloc,
cm->yv12_fb[cm->lst_fb_idx].frame_size, 1, recon_file);
Initial WebM release
2010-05-18 17:58:33 +02:00
fclose(recon_file);
}
disable compilation of debugging code This patch avoids compiling some debugging code in onyx_if.c. The most significant fix is to avoid generating code for vp8_write_yuv_frame, which is never called. Some other code was removed by the dead code elimination performed by the compiler, and this patch does it with the preprocessor instead. There are advantages both ways. Change-Id: I044fd43179d2e947553f0d6f2cad5b40907ac458
2010-09-24 17:10:25 +02:00
#endif
Initial refactoring of high_precision mv code. This is the first patch for refactoring of the code related to high-precision mv, so that 1/4 and 1/8 pel motion vectors can co-exist in the same bit-stream by use of a frame level flag. The current patch works fine for only use of 1/4th and only use of 1/8th pel mv, but there are some issues with the mode switching in between. Subsequent patches on this change Id will fix the remaining issues. Patch 2: Adds fixes to make sure that multiple mv precisions can co-exist in the bit-stream. Frame level switching has been tested to work correctly. Patch 3: Fixes lines exceeding 80 char Patch 4: http://www.corp.google.com/~debargha/vp8_results/enhinterp.html Results on derf after ssse3 bugfix, compared to everything enabled but the 8-tap, 1/8-subpel and 1/16-subpel uv. Overall the gains are about 3% now. Hopefully there are no more bugs lingering. Apparently the sse3 bug affected the quartel subpel results more than the eighth pel ones (which is understandabale because one bad predictor due to the bug, matters less if there are a lot more subpel options available as in the 1/8 subpel case). The results in the 4th column correspond to the current settings. The first two columns correspond to two settings of adaptive switching of the 1/4 or 1/8 subpel mode based on initial Q estimate. These do not work as good as just using 1/8 all the time yet. Change-Id: I3ef392ad338329f4d68a85257a49f2b14f3af472
2012-02-27 19:22:38 +01:00
#ifdef OUTPUT_YUV_REC
Add 8x8 transform to experimental branch Please refer to previous commit messages for detailed info: https://on2-git.corp.google.com/g/#change,5940 https://on2-git.corp.google.com/g/#change,6045 Change-Id: I8b16992f2f69c5a808ad40a3e32ef589cce7c59d
2011-02-14 23:18:18 +01:00
vp8_write_yuv_rec_frame(cm);
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
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
changed find_near_mvs search to include a mb from last frame This is an experiment to include a mv contribution from last frame to nearest and near mv definition. Initial test showed some small though consistent gain. latest patch slightly better result ~.13%-~.18%. TODO: the entropy used to encode the mode choice, i.e. the mv counts based conditional distribution of modes should be re-collected to reflect this change, it is expected that there is some further gain from that. Change-Id: Ief1e284a36d8aa56b49ae5b360c91419ec494fa4
2011-11-16 01:16:30 +01:00
if(cm->show_frame)
{
vpx_memcpy(cm->prev_mip, cm->mip,
(cm->mb_cols + 1) * (cm->mb_rows + 1)* sizeof(MODE_INFO));
}
else
{
vpx_memset(cm->prev_mip, 0,
(cm->mb_cols + 1) * (cm->mb_rows + 1)* sizeof(MODE_INFO));
}
Initial WebM release
2010-05-18 17:58:33 +02:00
}
static void Pass2Encode(VP8_COMP *cpi, unsigned long *size, unsigned char *dest, unsigned int *frame_flags)
{
if (!cpi->common.refresh_alt_ref_frame)
vp8_second_pass(cpi);
encode_frame_to_data_rate(cpi, size, dest, frame_flags);
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
cpi->twopass.bits_left -= 8 * *size;
Initial WebM release
2010-05-18 17:58:33 +02:00
if (!cpi->common.refresh_alt_ref_frame)
Fixed an encoder debug/relese mismatch in x86_64-win64-vs8 Visual c++ compiler uses xmm registers for floating point operations for 64 bit architecture, therefore its calling convention requires the preservation of xmm6-xmm15 in any function that have used these registers. However, the sse2 functions, that were originally written for 32 bit windows, may have used xmm6 and xmm7 without preserving the content. In this particular case, the compiler used xmm6 to save the variable "two_pass_min_rate", the value of the variable is mucked up by our sse2 optimized loop filter functions, hence the results of release/debug mismatching.
2010-05-20 00:48:00 +02:00
{
Allow for frame overheads in min frame bandwidth. Change-Id: I6ade229ff400fe492709010ac5bada37f8afa73e
2012-03-02 02:24:30 +01:00
double lower_bounds_min_rate = FRAME_OVERHEAD_BITS*cpi->oxcf.frame_rate;
Fixed an encoder debug/relese mismatch in x86_64-win64-vs8 Visual c++ compiler uses xmm registers for floating point operations for 64 bit architecture, therefore its calling convention requires the preservation of xmm6-xmm15 in any function that have used these registers. However, the sse2 functions, that were originally written for 32 bit windows, may have used xmm6 and xmm7 without preserving the content. In this particular case, the compiler used xmm6 to save the variable "two_pass_min_rate", the value of the variable is mucked up by our sse2 optimized loop filter functions, hence the results of release/debug mismatching.
2010-05-20 00:48:00 +02:00
double two_pass_min_rate = (double)(cpi->oxcf.target_bandwidth
*cpi->oxcf.two_pass_vbrmin_section / 100);
Allow for frame overheads in min frame bandwidth. Change-Id: I6ade229ff400fe492709010ac5bada37f8afa73e
2012-03-02 02:24:30 +01:00
if (two_pass_min_rate < lower_bounds_min_rate)
two_pass_min_rate = lower_bounds_min_rate;
cosmetics: consistently use [u]int64_t Removes mixed usage of (unsigned) long long and INT64. Fixes Issue #208. Change-Id: I220d3ed5ce4bb1280cd38bb3715f208ce23cf83a
2011-07-26 03:44:59 +02:00
cpi->twopass.bits_left += (int64_t)(two_pass_min_rate / cpi->oxcf.frame_rate);
Fixed an encoder debug/relese mismatch in x86_64-win64-vs8 Visual c++ compiler uses xmm registers for floating point operations for 64 bit architecture, therefore its calling convention requires the preservation of xmm6-xmm15 in any function that have used these registers. However, the sse2 functions, that were originally written for 32 bit windows, may have used xmm6 and xmm7 without preserving the content. In this particular case, the compiler used xmm6 to save the variable "two_pass_min_rate", the value of the variable is mucked up by our sse2 optimized loop filter functions, hence the results of release/debug mismatching.
2010-05-20 00:48:00 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
}
//For ARM NEON, d8-d15 are callee-saved registers, and need to be saved by us.
#if HAVE_ARMV7
cosmetics: consistently use [u]int64_t Removes mixed usage of (unsigned) long long and INT64. Fixes Issue #208. Change-Id: I220d3ed5ce4bb1280cd38bb3715f208ce23cf83a
2011-07-26 03:44:59 +02:00
extern void vp8_push_neon(int64_t *store);
extern void vp8_pop_neon(int64_t *store);
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
cosmetics: consistently use [u]int64_t Removes mixed usage of (unsigned) long long and INT64. Fixes Issue #208. Change-Id: I220d3ed5ce4bb1280cd38bb3715f208ce23cf83a
2011-07-26 03:44:59 +02:00
int vp8_receive_raw_frame(VP8_PTR ptr, unsigned int frame_flags, YV12_BUFFER_CONFIG *sd, int64_t time_stamp, int64_t end_time)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Remove unused local variables Removing unused local variables causing compiler warnings in Visual Studio. Change-Id: I0e2096303be1fdbc01428a6e57cca9796bb32c8a
2011-01-11 13:01:24 +01:00
#if HAVE_ARMV7
cosmetics: consistently use [u]int64_t Removes mixed usage of (unsigned) long long and INT64. Fixes Issue #208. Change-Id: I220d3ed5ce4bb1280cd38bb3715f208ce23cf83a
2011-07-26 03:44:59 +02:00
int64_t store_reg[8];
Remove unused local variables Removing unused local variables causing compiler warnings in Visual Studio. Change-Id: I0e2096303be1fdbc01428a6e57cca9796bb32c8a
2011-01-11 13:01:24 +01:00
#endif
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
VP8_COMP *cpi = (VP8_COMP *) ptr;
VP8_COMMON *cm = &cpi->common;
Initial WebM release
2010-05-18 17:58:33 +02:00
struct vpx_usec_timer timer;
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
int res = 0;
Initial WebM release
2010-05-18 17:58:33 +02:00
#if HAVE_ARMV7
Add runtime CPU detection support for ARM. The primary goal is to allow a binary to be built which supports NEON, but can fall back to non-NEON routines, since some Android devices do not have NEON, even if they are otherwise ARMv7 (e.g., Tegra). The configure-generated flags HAVE_ARMV7, etc., are used to decide which versions of each function to build, and when CONFIG_RUNTIME_CPU_DETECT is enabled, the correct version is chosen at run time. In order for this to work, the CFLAGS must be set to something appropriate (e.g., without -mfpu=neon for ARMv7, and with appropriate -march and -mcpu for even earlier configurations), or the native C code will not be able to run. The ASFLAGS must remain set for the most advanced instruction set required at build time, since the ARM assembler will refuse to emit them otherwise. I have not attempted to make any changes to configure to do this automatically. Doing so will probably require the addition of new configure options. Many of the hooks for RTCD on ARM were already there, but a lot of the code had bit-rotted, and a good deal of the ARM-specific code is not integrated into the RTCD structs at all. I did not try to resolve the latter, merely to add the minimal amount of protection around them to allow RTCD to work. Those functions that were called based on an ifdef at the calling site were expanded to check the RTCD flags at that site, but they should be added to an RTCD struct somewhere in the future. The functions invoked with global function pointers still are, but these should be moved into an RTCD struct for thread safety (I believe every platform currently supported has atomic pointer stores, but this is not guaranteed). The encoder's boolhuff functions did not even have _c and armv7 suffixes, and the correct version was resolved at link time. The token packing functions did have appropriate suffixes, but the version was selected with a define, with no associated RTCD struct. However, for both of these, the only armv7 instruction they actually used was rbit, and this was completely superfluous, so I reworked them to avoid it. The only non-ARMv4 instruction remaining in them is clz, which is ARMv5 (not even ARMv5TE is required). Considering that there are no ARM-specific configs which are not at least ARMv5TE, I did not try to detect these at runtime, and simply enable them for ARMv5 and above. Finally, the NEON register saving code was completely non-reentrant, since it saved the registers to a global, static variable. I moved the storage for this onto the stack. A single binary built with this code was tested on an ARM11 (ARMv6) and a Cortex A8 (ARMv7 w/NEON), for both the encoder and decoder, and produced identical output, while using the correct accelerated functions on each. I did not test on any earlier processors. Change-Id: I45cbd63a614f4554c3b325c45d46c0806f009eaa
2010-10-21 00:39:11 +02:00
#if CONFIG_RUNTIME_CPU_DETECT
if (cm->rtcd.flags & HAS_NEON)
#endif
{
vp8_push_neon(store_reg);
}
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
vpx_usec_timer_start(&timer);
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
if(vp8_lookahead_push(cpi->lookahead, sd, time_stamp, end_time,
Copy less when active map is in use When active map is specified and the current frame is not a key frame, golden frame nor a altref frame then copy only those active regions. This significantly reduces encoding time by as much as 19% on the test system where realtime encoding is used. This is particularly useful when the frame size is large (e.g. 2560x1600) and there's only a few action macroblocks. Change-Id: If394a813ec2df5a0201745d1348dbde4278f7ad4
2011-08-09 21:59:45 +02:00
frame_flags, cpi->active_map_enabled ? cpi->active_map : NULL))
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
res = -1;
Initial WebM release
2010-05-18 17:58:33 +02:00
cm->clr_type = sd->clrtype;
vpx_usec_timer_mark(&timer);
cpi->time_receive_data += vpx_usec_timer_elapsed(&timer);
#if HAVE_ARMV7
Add runtime CPU detection support for ARM. The primary goal is to allow a binary to be built which supports NEON, but can fall back to non-NEON routines, since some Android devices do not have NEON, even if they are otherwise ARMv7 (e.g., Tegra). The configure-generated flags HAVE_ARMV7, etc., are used to decide which versions of each function to build, and when CONFIG_RUNTIME_CPU_DETECT is enabled, the correct version is chosen at run time. In order for this to work, the CFLAGS must be set to something appropriate (e.g., without -mfpu=neon for ARMv7, and with appropriate -march and -mcpu for even earlier configurations), or the native C code will not be able to run. The ASFLAGS must remain set for the most advanced instruction set required at build time, since the ARM assembler will refuse to emit them otherwise. I have not attempted to make any changes to configure to do this automatically. Doing so will probably require the addition of new configure options. Many of the hooks for RTCD on ARM were already there, but a lot of the code had bit-rotted, and a good deal of the ARM-specific code is not integrated into the RTCD structs at all. I did not try to resolve the latter, merely to add the minimal amount of protection around them to allow RTCD to work. Those functions that were called based on an ifdef at the calling site were expanded to check the RTCD flags at that site, but they should be added to an RTCD struct somewhere in the future. The functions invoked with global function pointers still are, but these should be moved into an RTCD struct for thread safety (I believe every platform currently supported has atomic pointer stores, but this is not guaranteed). The encoder's boolhuff functions did not even have _c and armv7 suffixes, and the correct version was resolved at link time. The token packing functions did have appropriate suffixes, but the version was selected with a define, with no associated RTCD struct. However, for both of these, the only armv7 instruction they actually used was rbit, and this was completely superfluous, so I reworked them to avoid it. The only non-ARMv4 instruction remaining in them is clz, which is ARMv5 (not even ARMv5TE is required). Considering that there are no ARM-specific configs which are not at least ARMv5TE, I did not try to detect these at runtime, and simply enable them for ARMv5 and above. Finally, the NEON register saving code was completely non-reentrant, since it saved the registers to a global, static variable. I moved the storage for this onto the stack. A single binary built with this code was tested on an ARM11 (ARMv6) and a Cortex A8 (ARMv7 w/NEON), for both the encoder and decoder, and produced identical output, while using the correct accelerated functions on each. I did not test on any earlier processors. Change-Id: I45cbd63a614f4554c3b325c45d46c0806f009eaa
2010-10-21 00:39:11 +02:00
#if CONFIG_RUNTIME_CPU_DETECT
if (cm->rtcd.flags & HAS_NEON)
#endif
{
vp8_pop_neon(store_reg);
}
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
return res;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
Set VPX_FRAME_IS_DROPPABLE Allow the encoder to inform the application that the encoded frame will not be used as a reference. Change-Id: I90e41962325ef73d44da03327deb340d6f7f4860
2011-07-07 16:38:23 +02:00
static int frame_is_reference(const VP8_COMP *cpi)
{
const VP8_COMMON *cm = &cpi->common;
const MACROBLOCKD *xd = &cpi->mb.e_mbd;
return cm->frame_type == KEY_FRAME || cm->refresh_last_frame
|| cm->refresh_golden_frame || cm->refresh_alt_ref_frame
|| cm->copy_buffer_to_gf || cm->copy_buffer_to_arf
|| cm->refresh_entropy_probs
|| xd->mode_ref_lf_delta_update
|| xd->update_mb_segmentation_map || xd->update_mb_segmentation_data;
}
cosmetics: consistently use [u]int64_t Removes mixed usage of (unsigned) long long and INT64. Fixes Issue #208. Change-Id: I220d3ed5ce4bb1280cd38bb3715f208ce23cf83a
2011-07-26 03:44:59 +02:00
int vp8_get_compressed_data(VP8_PTR ptr, unsigned int *frame_flags, unsigned long *size, unsigned char *dest, int64_t *time_stamp, int64_t *time_end, int flush)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Remove unused local variables Removing unused local variables causing compiler warnings in Visual Studio. Change-Id: I0e2096303be1fdbc01428a6e57cca9796bb32c8a
2011-01-11 13:01:24 +01:00
#if HAVE_ARMV7
cosmetics: consistently use [u]int64_t Removes mixed usage of (unsigned) long long and INT64. Fixes Issue #208. Change-Id: I220d3ed5ce4bb1280cd38bb3715f208ce23cf83a
2011-07-26 03:44:59 +02:00
int64_t store_reg[8];
Remove unused local variables Removing unused local variables causing compiler warnings in Visual Studio. Change-Id: I0e2096303be1fdbc01428a6e57cca9796bb32c8a
2011-01-11 13:01:24 +01:00
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
VP8_COMP *cpi = (VP8_COMP *) ptr;
VP8_COMMON *cm = &cpi->common;
struct vpx_usec_timer cmptimer;
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
YV12_BUFFER_CONFIG *force_src_buffer = NULL;
Initial WebM release
2010-05-18 17:58:33 +02:00
if (!cpi)
return -1;
#if HAVE_ARMV7
Add runtime CPU detection support for ARM. The primary goal is to allow a binary to be built which supports NEON, but can fall back to non-NEON routines, since some Android devices do not have NEON, even if they are otherwise ARMv7 (e.g., Tegra). The configure-generated flags HAVE_ARMV7, etc., are used to decide which versions of each function to build, and when CONFIG_RUNTIME_CPU_DETECT is enabled, the correct version is chosen at run time. In order for this to work, the CFLAGS must be set to something appropriate (e.g., without -mfpu=neon for ARMv7, and with appropriate -march and -mcpu for even earlier configurations), or the native C code will not be able to run. The ASFLAGS must remain set for the most advanced instruction set required at build time, since the ARM assembler will refuse to emit them otherwise. I have not attempted to make any changes to configure to do this automatically. Doing so will probably require the addition of new configure options. Many of the hooks for RTCD on ARM were already there, but a lot of the code had bit-rotted, and a good deal of the ARM-specific code is not integrated into the RTCD structs at all. I did not try to resolve the latter, merely to add the minimal amount of protection around them to allow RTCD to work. Those functions that were called based on an ifdef at the calling site were expanded to check the RTCD flags at that site, but they should be added to an RTCD struct somewhere in the future. The functions invoked with global function pointers still are, but these should be moved into an RTCD struct for thread safety (I believe every platform currently supported has atomic pointer stores, but this is not guaranteed). The encoder's boolhuff functions did not even have _c and armv7 suffixes, and the correct version was resolved at link time. The token packing functions did have appropriate suffixes, but the version was selected with a define, with no associated RTCD struct. However, for both of these, the only armv7 instruction they actually used was rbit, and this was completely superfluous, so I reworked them to avoid it. The only non-ARMv4 instruction remaining in them is clz, which is ARMv5 (not even ARMv5TE is required). Considering that there are no ARM-specific configs which are not at least ARMv5TE, I did not try to detect these at runtime, and simply enable them for ARMv5 and above. Finally, the NEON register saving code was completely non-reentrant, since it saved the registers to a global, static variable. I moved the storage for this onto the stack. A single binary built with this code was tested on an ARM11 (ARMv6) and a Cortex A8 (ARMv7 w/NEON), for both the encoder and decoder, and produced identical output, while using the correct accelerated functions on each. I did not test on any earlier processors. Change-Id: I45cbd63a614f4554c3b325c45d46c0806f009eaa
2010-10-21 00:39:11 +02:00
#if CONFIG_RUNTIME_CPU_DETECT
if (cm->rtcd.flags & HAS_NEON)
#endif
{
vp8_push_neon(store_reg);
}
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
vpx_usec_timer_start(&cmptimer);
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
cpi->source = NULL;
Initial WebM release
2010-05-18 17:58:33 +02:00
Initial refactoring of high_precision mv code. This is the first patch for refactoring of the code related to high-precision mv, so that 1/4 and 1/8 pel motion vectors can co-exist in the same bit-stream by use of a frame level flag. The current patch works fine for only use of 1/4th and only use of 1/8th pel mv, but there are some issues with the mode switching in between. Subsequent patches on this change Id will fix the remaining issues. Patch 2: Adds fixes to make sure that multiple mv precisions can co-exist in the bit-stream. Frame level switching has been tested to work correctly. Patch 3: Fixes lines exceeding 80 char Patch 4: http://www.corp.google.com/~debargha/vp8_results/enhinterp.html Results on derf after ssse3 bugfix, compared to everything enabled but the 8-tap, 1/8-subpel and 1/16-subpel uv. Overall the gains are about 3% now. Hopefully there are no more bugs lingering. Apparently the sse3 bug affected the quartel subpel results more than the eighth pel ones (which is understandabale because one bad predictor due to the bug, matters less if there are a lot more subpel options available as in the 1/8 subpel case). The results in the 4th column correspond to the current settings. The first two columns correspond to two settings of adaptive switching of the 1/4 or 1/8 subpel mode based on initial Q estimate. These do not work as good as just using 1/8 all the time yet. Change-Id: I3ef392ad338329f4d68a85257a49f2b14f3af472
2012-02-27 19:22:38 +01:00
#if CONFIG_HIGH_PRECISION_MV
cpi->mb.e_mbd.allow_high_precision_mv = ALTREF_HIGH_PRECISION_MV;
#endif
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
// Should we code an alternate reference frame
Code Cleanup. Removal of error_resilient_mode features. The interface has been left in place but does nothing. Change-Id: I2407863bd0d3c98407354507423ca48d29f63b17
2012-02-26 02:15:47 +01:00
if (cpi->oxcf.play_alternate &&
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
cpi->source_alt_ref_pending)
{
if ((cpi->source = vp8_lookahead_peek(cpi->lookahead,
cpi->frames_till_gf_update_due)))
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
cpi->alt_ref_source = cpi->source;
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->oxcf.arnr_max_frames > 0)
{
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
vp8_temporal_filter_prepare_c(cpi,
cpi->frames_till_gf_update_due);
force_src_buffer = &cpi->alt_ref_buffer;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
cm->frames_till_alt_ref_frame = cpi->frames_till_gf_update_due;
cm->refresh_alt_ref_frame = 1;
cm->refresh_golden_frame = 0;
cm->refresh_last_frame = 0;
cm->show_frame = 0;
cpi->source_alt_ref_pending = FALSE; // Clear Pending altf Ref flag.
cpi->is_src_frame_alt_ref = 0;
}
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
}
if (!cpi->source)
{
if ((cpi->source = vp8_lookahead_pop(cpi->lookahead, flush)))
Initial WebM release
2010-05-18 17:58:33 +02:00
{
cm->show_frame = 1;
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
cpi->is_src_frame_alt_ref = cpi->alt_ref_source
&& (cpi->source == cpi->alt_ref_source);
change to make use of more trellis quantization when a subsequent frame is encoded as an alt reference frame, it is unlikely that any mb in current frame will be used as reference for future frames, so we can enable quantization optimization even when the RD constant is slightly rate-biased. The change has an overall benefit between 0.1% to 0.2% bit savings on the test sets based on vpxssim scores. Change-Id: I9aa7bc5cd573ea84e3ee655d2834c18c4460ceea
2010-10-15 03:58:34 +02:00
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
if(cpi->is_src_frame_alt_ref)
cpi->alt_ref_source = NULL;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
if (cpi->source)
{
cpi->un_scaled_source =
cpi->Source = force_src_buffer ? force_src_buffer : &cpi->source->img;
*time_stamp = cpi->source->ts_start;
*time_end = cpi->source->ts_end;
*frame_flags = cpi->source->flags;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
else
{
*size = 0;
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
if (flush && cpi->pass == 1 && !cpi->twopass.first_pass_done)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
vp8_end_first_pass(cpi); /* get last stats packet */
cleanup: collect twopass variables This patch collects the twopass specific memebers of VP8_COMP into a dedicated struct. This is a first step towards isolating the two pass rate control and aids readability by decorating these variables with the 'twopass.' namespace. This makes it clear to the reader in what contexts the variable will be valid, and is a hint that a section of code might be a good candidate to move to firstpass.c in later refactoring. There likely will be other rate control modes that need their own specific data as well. This notation is probably overly verbose in firstpass.c, so an alternative would be to access this struct through a pointer like 'rc->' instead of 'cpi->firstpass.' in that file. Feel free to make a review comment to that effect if you prefer. Change-Id: I0ab8254647cb4b493a77c16b5d236d0d4a94ca4d
2011-05-19 23:16:39 +02:00
cpi->twopass.first_pass_done = 1;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
#if HAVE_ARMV7
Add runtime CPU detection support for ARM. The primary goal is to allow a binary to be built which supports NEON, but can fall back to non-NEON routines, since some Android devices do not have NEON, even if they are otherwise ARMv7 (e.g., Tegra). The configure-generated flags HAVE_ARMV7, etc., are used to decide which versions of each function to build, and when CONFIG_RUNTIME_CPU_DETECT is enabled, the correct version is chosen at run time. In order for this to work, the CFLAGS must be set to something appropriate (e.g., without -mfpu=neon for ARMv7, and with appropriate -march and -mcpu for even earlier configurations), or the native C code will not be able to run. The ASFLAGS must remain set for the most advanced instruction set required at build time, since the ARM assembler will refuse to emit them otherwise. I have not attempted to make any changes to configure to do this automatically. Doing so will probably require the addition of new configure options. Many of the hooks for RTCD on ARM were already there, but a lot of the code had bit-rotted, and a good deal of the ARM-specific code is not integrated into the RTCD structs at all. I did not try to resolve the latter, merely to add the minimal amount of protection around them to allow RTCD to work. Those functions that were called based on an ifdef at the calling site were expanded to check the RTCD flags at that site, but they should be added to an RTCD struct somewhere in the future. The functions invoked with global function pointers still are, but these should be moved into an RTCD struct for thread safety (I believe every platform currently supported has atomic pointer stores, but this is not guaranteed). The encoder's boolhuff functions did not even have _c and armv7 suffixes, and the correct version was resolved at link time. The token packing functions did have appropriate suffixes, but the version was selected with a define, with no associated RTCD struct. However, for both of these, the only armv7 instruction they actually used was rbit, and this was completely superfluous, so I reworked them to avoid it. The only non-ARMv4 instruction remaining in them is clz, which is ARMv5 (not even ARMv5TE is required). Considering that there are no ARM-specific configs which are not at least ARMv5TE, I did not try to detect these at runtime, and simply enable them for ARMv5 and above. Finally, the NEON register saving code was completely non-reentrant, since it saved the registers to a global, static variable. I moved the storage for this onto the stack. A single binary built with this code was tested on an ARM11 (ARMv6) and a Cortex A8 (ARMv7 w/NEON), for both the encoder and decoder, and produced identical output, while using the correct accelerated functions on each. I did not test on any earlier processors. Change-Id: I45cbd63a614f4554c3b325c45d46c0806f009eaa
2010-10-21 00:39:11 +02:00
#if CONFIG_RUNTIME_CPU_DETECT
if (cm->rtcd.flags & HAS_NEON)
#endif
{
vp8_pop_neon(store_reg);
}
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
return -1;
}
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
if (cpi->source->ts_start < cpi->first_time_stamp_ever)
improve average framerate calculation Change Ice204e86 identified a problem with bitrate undershoot due to low precision in the timestamps passed to the library. This patch takes a different approach by calculating the duration of this frame and passing it to the library, rather than using a fixed duration and letting the library average it out with higher precision timestamps. This part of the fix only applies to vpxenc. This patch also attempts to fix the problem for generic applications that may have made the same mistake vpxenc did. Instead of calculating this frame's duration by the difference of this frame's and the last frame's start time, we use the end times instead. This allows the framerate calculation to scavenge "unclaimed" time from the last frame. For instance: start | end | calculated duration ======+=======+==================== 0ms 33ms 33ms 33ms 66ms 33ms 66ms 99ms 33ms 100ms 133ms 34ms Change-Id: I92be4b3518e0bd530e97f90e69e75330a4c413fc
2010-11-04 20:05:45 +01:00
{
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
cpi->first_time_stamp_ever = cpi->source->ts_start;
cpi->last_end_time_stamp_seen = cpi->source->ts_start;
improve average framerate calculation Change Ice204e86 identified a problem with bitrate undershoot due to low precision in the timestamps passed to the library. This patch takes a different approach by calculating the duration of this frame and passing it to the library, rather than using a fixed duration and letting the library average it out with higher precision timestamps. This part of the fix only applies to vpxenc. This patch also attempts to fix the problem for generic applications that may have made the same mistake vpxenc did. Instead of calculating this frame's duration by the difference of this frame's and the last frame's start time, we use the end times instead. This allows the framerate calculation to scavenge "unclaimed" time from the last frame. For instance: start | end | calculated duration ======+=======+==================== 0ms 33ms 33ms 33ms 66ms 33ms 66ms 99ms 33ms 100ms 133ms 34ms Change-Id: I92be4b3518e0bd530e97f90e69e75330a4c413fc
2010-11-04 20:05:45 +01:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
// adjust frame rates based on timestamps given
if (!cm->refresh_alt_ref_frame)
{
cosmetics: consistently use [u]int64_t Removes mixed usage of (unsigned) long long and INT64. Fixes Issue #208. Change-Id: I220d3ed5ce4bb1280cd38bb3715f208ce23cf83a
2011-07-26 03:44:59 +02:00
int64_t this_duration;
Improve framerate adaptation This patch improves the accuracy of frame rate estimation by using a larger, 1 second window. It also more quickly adapts to step changes in the input frame rate (ie 30fps to 15fps) Change-Id: I39e48a8f5ac880b4c4b2ebd81049259b81a0218e
2011-05-12 21:04:06 +02:00
int step = 0;
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
if (cpi->source->ts_start == cpi->first_time_stamp_ever)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Improve framerate adaptation This patch improves the accuracy of frame rate estimation by using a larger, 1 second window. It also more quickly adapts to step changes in the input frame rate (ie 30fps to 15fps) Change-Id: I39e48a8f5ac880b4c4b2ebd81049259b81a0218e
2011-05-12 21:04:06 +02:00
this_duration = cpi->source->ts_end - cpi->source->ts_start;
step = 1;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
else
{
cosmetics: consistently use [u]int64_t Removes mixed usage of (unsigned) long long and INT64. Fixes Issue #208. Change-Id: I220d3ed5ce4bb1280cd38bb3715f208ce23cf83a
2011-07-26 03:44:59 +02:00
int64_t last_duration;
Improve framerate adaptation This patch improves the accuracy of frame rate estimation by using a larger, 1 second window. It also more quickly adapts to step changes in the input frame rate (ie 30fps to 15fps) Change-Id: I39e48a8f5ac880b4c4b2ebd81049259b81a0218e
2011-05-12 21:04:06 +02:00
this_duration = cpi->source->ts_end - cpi->last_end_time_stamp_seen;
last_duration = cpi->last_end_time_stamp_seen
- cpi->last_time_stamp_seen;
// do a step update if the duration changes by 10%
if (last_duration)
step = ((this_duration - last_duration) * 10 / last_duration);
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Improve framerate adaptation This patch improves the accuracy of frame rate estimation by using a larger, 1 second window. It also more quickly adapts to step changes in the input frame rate (ie 30fps to 15fps) Change-Id: I39e48a8f5ac880b4c4b2ebd81049259b81a0218e
2011-05-12 21:04:06 +02:00
if (this_duration)
{
if (step)
vp8_new_frame_rate(cpi, 10000000.0 / this_duration);
else
Fixing divide by zero Change-Id: I9d8a98a2f7ed1e3116d0bae35164618c41998bac
2011-03-01 23:59:22 +01:00
{
Improve framerate adaptation This patch improves the accuracy of frame rate estimation by using a larger, 1 second window. It also more quickly adapts to step changes in the input frame rate (ie 30fps to 15fps) Change-Id: I39e48a8f5ac880b4c4b2ebd81049259b81a0218e
2011-05-12 21:04:06 +02:00
double avg_duration, interval;
Initial WebM release
2010-05-18 17:58:33 +02:00
Improve framerate adaptation This patch improves the accuracy of frame rate estimation by using a larger, 1 second window. It also more quickly adapts to step changes in the input frame rate (ie 30fps to 15fps) Change-Id: I39e48a8f5ac880b4c4b2ebd81049259b81a0218e
2011-05-12 21:04:06 +02:00
/* Average this frame's rate into the last second's average
* frame rate. If we haven't seen 1 second yet, then average
* over the whole interval seen.
*/
interval = cpi->source->ts_end - cpi->first_time_stamp_ever;
if(interval > 10000000.0)
interval = 10000000;
Initial WebM release
2010-05-18 17:58:33 +02:00
Improve framerate adaptation This patch improves the accuracy of frame rate estimation by using a larger, 1 second window. It also more quickly adapts to step changes in the input frame rate (ie 30fps to 15fps) Change-Id: I39e48a8f5ac880b4c4b2ebd81049259b81a0218e
2011-05-12 21:04:06 +02:00
avg_duration = 10000000.0 / cpi->oxcf.frame_rate;
avg_duration *= (interval - avg_duration + this_duration);
avg_duration /= interval;
vp8_new_frame_rate(cpi, 10000000.0 / avg_duration);
}
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
cpi->last_time_stamp_seen = cpi->source->ts_start;
cpi->last_end_time_stamp_seen = cpi->source->ts_end;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
// start with a 0 size frame
*size = 0;
// Clear down mmx registers
vp8_clear_system_state(); //__asm emms;
cm->frame_type = INTER_FRAME;
cm->frame_flags = *frame_flags;
#if 0
if (cm->refresh_alt_ref_frame)
{
//cm->refresh_golden_frame = 1;
cm->refresh_golden_frame = 0;
cm->refresh_last_frame = 0;
}
else
{
cm->refresh_golden_frame = 0;
cm->refresh_last_frame = 1;
}
#endif
Do not copy data between encoder reference buffers. Golden and ALT reference buffers were refreshed by copying from the new buffer. Replaced this by index manipulation. Also moved all the reference frame updates to one function for easier tracking. Change-Id: Icd3e534e7e2c8c5567168d222e6a64a96aae24a1
2011-04-20 14:12:23 +02:00
/* find a free buffer for the new frame */
{
int i = 0;
for(; i < NUM_YV12_BUFFERS; i++)
{
if(!cm->yv12_fb[i].flags)
{
cm->new_fb_idx = i;
break;
}
}
Initial WebM release
2010-05-18 17:58:33 +02:00
Do not copy data between encoder reference buffers. Golden and ALT reference buffers were refreshed by copying from the new buffer. Replaced this by index manipulation. Also moved all the reference frame updates to one function for easier tracking. Change-Id: Icd3e534e7e2c8c5567168d222e6a64a96aae24a1
2011-04-20 14:12:23 +02:00
assert(i < NUM_YV12_BUFFERS );
}
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->pass == 1)
{
Pass1Encode(cpi, size, dest, frame_flags);
}
else if (cpi->pass == 2)
{
Pass2Encode(cpi, size, dest, frame_flags);
}
else
Simplification of experimental code base. Removed ~CONFIG_REALTIME_ONLY code. Change-Id: I5fafff29a08acd8928699f9ddce8744787024d8c
2012-02-13 18:05:26 +01:00
encode_frame_to_data_rate(cpi, size, dest, frame_flags);
Initial WebM release
2010-05-18 17:58:33 +02:00
added separate entropy context for alt_ref This commit added code to keep track of separate entropy contexts for normal frames and alt ref frames. The underly assumption was that the two type of frames have different entropy characteristics given they typically have quite different quantization levels. By keeping entropy contexts separate, it helps the entropy context distribution to be more closely adapted to each frame type. Tests on derf set showed a good and very consistent gain on all clips on all metrics, avg psnr: 0.89%, overall psnr: 0.84% and ssim 0.93%. http://www.corp.google.com/~yaowu/no_crawl/mulcontext.html Change-Id: I15bc9697f6ff7829042911fe0c62930585d7e65d
2011-12-01 01:36:46 +01:00
if(cm->refresh_entropy_probs)
{
if(cm->refresh_alt_ref_frame)
vpx_memcpy(&cm->lfc_a, &cm->fc, sizeof(cm->fc));
else
vpx_memcpy(&cm->lfc, &cm->fc, sizeof(cm->fc));
}
Initial WebM release
2010-05-18 17:58:33 +02:00
// if its a dropped frame honor the requests on subsequent frames
if (*size > 0)
{
Set VPX_FRAME_IS_DROPPABLE Allow the encoder to inform the application that the encoded frame will not be used as a reference. Change-Id: I90e41962325ef73d44da03327deb340d6f7f4860
2011-07-07 16:38:23 +02:00
cpi->droppable = !frame_is_reference(cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
// return to normal state
cm->refresh_entropy_probs = 1;
cm->refresh_alt_ref_frame = 0;
cm->refresh_golden_frame = 0;
cm->refresh_last_frame = 1;
cm->frame_type = INTER_FRAME;
}
vpx_usec_timer_mark(&cmptimer);
cpi->time_compress_data += vpx_usec_timer_elapsed(&cmptimer);
if (cpi->b_calculate_psnr && cpi->pass != 1 && cm->show_frame)
Encoder loopfilter running in its own thread In multithreaded mode the loopfilter is running in its own thread (filter level calculation and frame filtering). Filtering is mostly done in parallel with the bitstream packing. Before starting the packing the loopfilter level has to be calculated. Also any needed reference frame copying is done in the filter thread. Currently the encoder will create n+1 threads, where n > 1 is the number of threads specified by application and 1 is the extra filter thread. With n = 1 the encoder runs in single thread mode. There will never be more than n threads running concurrently. Change-Id: I4fb29b559a40275d6d3babb8727245c40fba931b
2011-02-25 12:42:05 +01:00
{
Initial WebM release
2010-05-18 17:58:33 +02:00
generate_psnr_packet(cpi);
Encoder loopfilter running in its own thread In multithreaded mode the loopfilter is running in its own thread (filter level calculation and frame filtering). Filtering is mostly done in parallel with the bitstream packing. Before starting the packing the loopfilter level has to be calculated. Also any needed reference frame copying is done in the filter thread. Currently the encoder will create n+1 threads, where n > 1 is the number of threads specified by application and 1 is the extra filter thread. With n = 1 the encoder runs in single thread mode. There will never be more than n threads running concurrently. Change-Id: I4fb29b559a40275d6d3babb8727245c40fba931b
2011-02-25 12:42:05 +01:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
changed configure option name to reduce confusion Renamed configure option "enable-psnr" to "enable-internal-stats" to better reflect the purpose of the option and eliminate the confusion reported in http://code.google.com/p/webm/issues/detail?id=35 Change-Id: If72df6fdb9f1e33dab1329240ba4d8911d2f1f7a
2011-04-29 18:37:59 +02:00
#if CONFIG_INTERNAL_STATS
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->pass != 1)
{
cpi->bytes += *size;
if (cm->show_frame)
{
cpi->count ++;
if (cpi->b_calculate_psnr)
{
vp8cx - psnr converted to call assemblerized sse Change-Id: Ie388d4618c44b131f96b9fe526618b457f020dfa
2011-03-08 21:23:40 +01:00
double ye,ue,ve;
double frame_psnr;
YV12_BUFFER_CONFIG *orig = cpi->Source;
YV12_BUFFER_CONFIG *recon = cpi->common.frame_to_show;
YV12_BUFFER_CONFIG *pp = &cm->post_proc_buffer;
int y_samples = orig->y_height * orig->y_width ;
int uv_samples = orig->uv_height * orig->uv_width ;
int t_samples = y_samples + 2 * uv_samples;
cosmetics: consistently use [u]int64_t Removes mixed usage of (unsigned) long long and INT64. Fixes Issue #208. Change-Id: I220d3ed5ce4bb1280cd38bb3715f208ce23cf83a
2011-07-26 03:44:59 +02:00
int64_t sq_error;
vp8cx - psnr converted to call assemblerized sse Change-Id: Ie388d4618c44b131f96b9fe526618b457f020dfa
2011-03-08 21:23:40 +01:00
ye = calc_plane_error(orig->y_buffer, orig->y_stride,
recon->y_buffer, recon->y_stride, orig->y_width, orig->y_height,
IF_RTCD(&cpi->rtcd.variance));
ue = calc_plane_error(orig->u_buffer, orig->uv_stride,
recon->u_buffer, recon->uv_stride, orig->uv_width, orig->uv_height,
IF_RTCD(&cpi->rtcd.variance));
ve = calc_plane_error(orig->v_buffer, orig->uv_stride,
recon->v_buffer, recon->uv_stride, orig->uv_width, orig->uv_height,
IF_RTCD(&cpi->rtcd.variance));
sq_error = ye + ue + ve;
frame_psnr = vp8_mse2psnr(t_samples, 255.0, sq_error);
cpi->total_y += vp8_mse2psnr(y_samples, 255.0, ye);
cpi->total_u += vp8_mse2psnr(uv_samples, 255.0, ue);
cpi->total_v += vp8_mse2psnr(uv_samples, 255.0, ve);
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->total_sq_error += sq_error;
cpi->total += frame_psnr;
{
Refactor lookahead ring buffer This patch cleans up the source buffer storage and copy mechanism to allow access through a standard push/pop/peek interface. This approach also avoids an extra copy in the case where the source is not a multiple of 16, fixing issue #102. Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9
2011-04-13 20:00:18 +02:00
double frame_psnr2, frame_ssim2 = 0;
Initial WebM release
2010-05-18 17:58:33 +02:00
double weight = 0;
vp8_deblock(cm->frame_to_show, &cm->post_proc_buffer, cm->filter_level * 10 / 6, 1, 0, IF_RTCD(&cm->rtcd.postproc));
vp8_clear_system_state();
vp8cx - psnr converted to call assemblerized sse Change-Id: Ie388d4618c44b131f96b9fe526618b457f020dfa
2011-03-08 21:23:40 +01:00
ye = calc_plane_error(orig->y_buffer, orig->y_stride,
pp->y_buffer, pp->y_stride, orig->y_width, orig->y_height,
IF_RTCD(&cpi->rtcd.variance));
Initial WebM release
2010-05-18 17:58:33 +02:00
vp8cx - psnr converted to call assemblerized sse Change-Id: Ie388d4618c44b131f96b9fe526618b457f020dfa
2011-03-08 21:23:40 +01:00
ue = calc_plane_error(orig->u_buffer, orig->uv_stride,
pp->u_buffer, pp->uv_stride, orig->uv_width, orig->uv_height,
IF_RTCD(&cpi->rtcd.variance));
ve = calc_plane_error(orig->v_buffer, orig->uv_stride,
pp->v_buffer, pp->uv_stride, orig->uv_width, orig->uv_height,
IF_RTCD(&cpi->rtcd.variance));
sq_error = ye + ue + ve;
frame_psnr2 = vp8_mse2psnr(t_samples, 255.0, sq_error);
cpi->totalp_y += vp8_mse2psnr(y_samples, 255.0, ye);
cpi->totalp_u += vp8_mse2psnr(uv_samples, 255.0, ue);
cpi->totalp_v += vp8_mse2psnr(uv_samples, 255.0, ve);
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->total_sq_error2 += sq_error;
vp8cx - psnr converted to call assemblerized sse Change-Id: Ie388d4618c44b131f96b9fe526618b457f020dfa
2011-03-08 21:23:40 +01:00
cpi->totalp += frame_psnr2;
vp8cx- alternate ssim function with optimizations Change-Id: I91921b0a90dbaddc7010380b038955be347964b3
2011-03-08 15:05:18 +01:00
frame_ssim2 = vp8_calc_ssim(cpi->Source,
&cm->post_proc_buffer, 1, &weight,
IF_RTCD(&cpi->rtcd.variance));
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->summed_quality += frame_ssim2 * weight;
cpi->summed_weights += weight;
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
#if 0
{
FILE *f = fopen("q_used.stt", "a");
fprintf(f, "%5d : Y%f7.3:U%f7.3:V%f7.3:F%f7.3:S%7.3f\n",
cpi->common.current_video_frame,y2, u2, v2,
frame_psnr2, frame_ssim2);
fclose(f);
}
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
}
}
if (cpi->b_calculate_ssimg)
{
double y, u, v, frame_all;
change to use fast ssim code for internal ssim calculations The commit also removed the slow ssim calculation that uses a 7x7 kernel, and revised the comments to better describe how sample ssim values are computed and averaged Change-Id: I1d874073cddca00f3c997f4b9a9a3db0aa212276
2011-05-03 00:27:14 +02:00
frame_all = vp8_calc_ssimg(cpi->Source, cm->frame_to_show,
&y, &u, &v, IF_RTCD(&cpi->rtcd.variance));
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->total_ssimg_y += y;
cpi->total_ssimg_u += u;
cpi->total_ssimg_v += v;
cpi->total_ssimg_all += frame_all;
}
}
}
#endif
#if HAVE_ARMV7
Add runtime CPU detection support for ARM. The primary goal is to allow a binary to be built which supports NEON, but can fall back to non-NEON routines, since some Android devices do not have NEON, even if they are otherwise ARMv7 (e.g., Tegra). The configure-generated flags HAVE_ARMV7, etc., are used to decide which versions of each function to build, and when CONFIG_RUNTIME_CPU_DETECT is enabled, the correct version is chosen at run time. In order for this to work, the CFLAGS must be set to something appropriate (e.g., without -mfpu=neon for ARMv7, and with appropriate -march and -mcpu for even earlier configurations), or the native C code will not be able to run. The ASFLAGS must remain set for the most advanced instruction set required at build time, since the ARM assembler will refuse to emit them otherwise. I have not attempted to make any changes to configure to do this automatically. Doing so will probably require the addition of new configure options. Many of the hooks for RTCD on ARM were already there, but a lot of the code had bit-rotted, and a good deal of the ARM-specific code is not integrated into the RTCD structs at all. I did not try to resolve the latter, merely to add the minimal amount of protection around them to allow RTCD to work. Those functions that were called based on an ifdef at the calling site were expanded to check the RTCD flags at that site, but they should be added to an RTCD struct somewhere in the future. The functions invoked with global function pointers still are, but these should be moved into an RTCD struct for thread safety (I believe every platform currently supported has atomic pointer stores, but this is not guaranteed). The encoder's boolhuff functions did not even have _c and armv7 suffixes, and the correct version was resolved at link time. The token packing functions did have appropriate suffixes, but the version was selected with a define, with no associated RTCD struct. However, for both of these, the only armv7 instruction they actually used was rbit, and this was completely superfluous, so I reworked them to avoid it. The only non-ARMv4 instruction remaining in them is clz, which is ARMv5 (not even ARMv5TE is required). Considering that there are no ARM-specific configs which are not at least ARMv5TE, I did not try to detect these at runtime, and simply enable them for ARMv5 and above. Finally, the NEON register saving code was completely non-reentrant, since it saved the registers to a global, static variable. I moved the storage for this onto the stack. A single binary built with this code was tested on an ARM11 (ARMv6) and a Cortex A8 (ARMv7 w/NEON), for both the encoder and decoder, and produced identical output, while using the correct accelerated functions on each. I did not test on any earlier processors. Change-Id: I45cbd63a614f4554c3b325c45d46c0806f009eaa
2010-10-21 00:39:11 +02:00
#if CONFIG_RUNTIME_CPU_DETECT
if (cm->rtcd.flags & HAS_NEON)
#endif
{
vp8_pop_neon(store_reg);
}
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
return 0;
}
postproc : Re-work posproc calling to allow more flags. Debugging in postproc needs more flags to allow for specific block types to be turned on or off in the visualizations. Must be enabled with --enable-postproc-visualizer during configuration time. Change-Id: Ia74f357ddc3ad4fb8082afd3a64f62384e4fcb2d
2010-11-05 00:03:36 +01:00
int vp8_get_preview_raw_frame(VP8_PTR comp, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t *flags)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
VP8_COMP *cpi = (VP8_COMP *) comp;
if (cpi->common.refresh_alt_ref_frame)
return -1;
else
{
int ret;
#if CONFIG_POSTPROC
postproc : Re-work posproc calling to allow more flags. Debugging in postproc needs more flags to allow for specific block types to be turned on or off in the visualizations. Must be enabled with --enable-postproc-visualizer during configuration time. Change-Id: Ia74f357ddc3ad4fb8082afd3a64f62384e4fcb2d
2010-11-05 00:03:36 +01:00
ret = vp8_post_proc_frame(&cpi->common, dest, flags);
Initial WebM release
2010-05-18 17:58:33 +02:00
#else
if (cpi->common.frame_to_show)
{
*dest = *cpi->common.frame_to_show;
dest->y_width = cpi->common.Width;
dest->y_height = cpi->common.Height;
dest->uv_height = cpi->common.Height / 2;
ret = 0;
}
else
{
ret = -1;
}
#endif //!CONFIG_POSTPROC
vp8_clear_system_state();
return ret;
}
}
int vp8_set_roimap(VP8_PTR comp, unsigned char *map, unsigned int rows, unsigned int cols, int delta_q[4], int delta_lf[4], unsigned int threshold[4])
{
VP8_COMP *cpi = (VP8_COMP *) comp;
Segment Features: Some basic plumbing added for a range of segment level features. MB_LVL_* changed to SEG_LVL_* to better reflect meaning. Change-Id: Iac96da36990aa0e40afc0d86e990df337fd0c50b
2011-09-13 12:26:39 +02:00
signed char feature_data[SEG_LVL_MAX][MAX_MB_SEGMENTS];
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
MACROBLOCKD *xd = &cpi->mb.e_mbd;
int i;
Initial WebM release
2010-05-18 17:58:33 +02:00
if (cpi->common.mb_rows != rows || cpi->common.mb_cols != cols)
return -1;
if (!map)
{
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
vp8_disable_segmentation((VP8_PTR)cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
return 0;
}
// Set the segmentation Map
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
vp8_set_segmentation_map((VP8_PTR)cpi, map);
Initial WebM release
2010-05-18 17:58:33 +02:00
// Activate segmentation.
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
vp8_enable_segmentation((VP8_PTR)cpi);
Initial WebM release
2010-05-18 17:58:33 +02:00
// Set up the quant segment data
Segment Features: Some basic plumbing added for a range of segment level features. MB_LVL_* changed to SEG_LVL_* to better reflect meaning. Change-Id: Iac96da36990aa0e40afc0d86e990df337fd0c50b
2011-09-13 12:26:39 +02:00
feature_data[SEG_LVL_ALT_Q][0] = delta_q[0];
feature_data[SEG_LVL_ALT_Q][1] = delta_q[1];
feature_data[SEG_LVL_ALT_Q][2] = delta_q[2];
feature_data[SEG_LVL_ALT_Q][3] = delta_q[3];
Initial WebM release
2010-05-18 17:58:33 +02:00
// Set up the loop segment data s
Segment Features: Some basic plumbing added for a range of segment level features. MB_LVL_* changed to SEG_LVL_* to better reflect meaning. Change-Id: Iac96da36990aa0e40afc0d86e990df337fd0c50b
2011-09-13 12:26:39 +02:00
feature_data[SEG_LVL_ALT_LF][0] = delta_lf[0];
feature_data[SEG_LVL_ALT_LF][1] = delta_lf[1];
feature_data[SEG_LVL_ALT_LF][2] = delta_lf[2];
feature_data[SEG_LVL_ALT_LF][3] = delta_lf[3];
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->segment_encode_breakout[0] = threshold[0];
cpi->segment_encode_breakout[1] = threshold[1];
cpi->segment_encode_breakout[2] = threshold[2];
cpi->segment_encode_breakout[3] = threshold[3];
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
// Enable the loop and quant changes in the feature mask
for ( i = 0; i < 4; i++ )
{
if (delta_q[i])
enable_segfeature(xd, i, SEG_LVL_ALT_Q);
else
disable_segfeature(xd, i, SEG_LVL_ALT_Q);
if (delta_lf[i])
enable_segfeature(xd, i, SEG_LVL_ALT_LF);
else
disable_segfeature(xd, i, SEG_LVL_ALT_LF);
}
Initial WebM release
2010-05-18 17:58:33 +02:00
// Initialise the feature data structure
// SEGMENT_DELTADATA 0, SEGMENT_ABSDATA 1
Further segment feature extensions. This quite large check in includes the following: Merge in some code from Ronald (mbgraph.c) that scans a Gf/arf group. This is used as a basis for a simple segmentation for the normal frames in a gf/arf group. This code also uses satd functions from Yaowu. Adds functionality for coding the latest possible position of an EOB for blocks in the segment. (Currently 0-15 only, hence just for 4x4 dct). Where the EOB position is 0 this acts like "skip" and the normal coding of skip at the per mb level is disabled. Added functions (seg_common.c) for setting and reading segment feature elements. These may want to be optimized away at some point but while the mecahnism is in a state of flux they provide a single location for making changes and keep things a bit cleaner. This is still proof of concept code. Currently the tested feature set:- Quantizer, Loop Filter level, Reference frame, Prediction Mode, EOB end stop. TBD:- Add functions for setting and reading the feature data with range and validity checking. Handling of signed and unsigned feature data. At the moment all is assumed to be signed and a sign bit is coded but many cannot be negative. Correct handling of EOB feature with intra coded blocks. Testing/trapping of legal/illegal ref frame and mode combinations. Transform size switch plus merge and test with 8c8 DCT work Merge and test with Sumans Segmenation coding optimizations Change-Id: Iee12e83661c7abbd1e0ce6810915eb4ec35e2d8e
2011-10-05 12:26:00 +02:00
vp8_set_segment_data((VP8_PTR)cpi, &feature_data[0][0], SEGMENT_DELTADATA);
Initial WebM release
2010-05-18 17:58:33 +02:00
return 0;
}
int vp8_set_active_map(VP8_PTR comp, unsigned char *map, unsigned int rows, unsigned int cols)
{
VP8_COMP *cpi = (VP8_COMP *) comp;
if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols)
{
if (map)
{
vpx_memcpy(cpi->active_map, map, rows * cols);
cpi->active_map_enabled = 1;
}
else
cpi->active_map_enabled = 0;
return 0;
}
else
{
//cpi->active_map_enabled = 0;
return -1 ;
}
}
int vp8_set_internal_size(VP8_PTR comp, VPX_SCALING horiz_mode, VPX_SCALING vert_mode)
{
VP8_COMP *cpi = (VP8_COMP *) comp;
remove warning was having: "vp8/encoder/onyx_if.c:5365: warning: comparison of unsigned expression >= 0 is always true"
2010-11-18 01:50:02 +01:00
if (horiz_mode <= ONETWO)
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->common.horiz_scale = horiz_mode;
else
return -1;
remove warning was having: "vp8/encoder/onyx_if.c:5365: warning: comparison of unsigned expression >= 0 is always true"
2010-11-18 01:50:02 +01:00
if (vert_mode <= ONETWO)
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->common.vert_scale = vert_mode;
else
return -1;
return 0;
}
int vp8_calc_ss_err(YV12_BUFFER_CONFIG *source, YV12_BUFFER_CONFIG *dest, const vp8_variance_rtcd_vtable_t *rtcd)
{
int i, j;
int Total = 0;
unsigned char *src = source->y_buffer;
unsigned char *dst = dest->y_buffer;
(void)rtcd;
// Loop through the Y plane raw and reconstruction data summing (square differences)
for (i = 0; i < source->y_height; i += 16)
{
for (j = 0; j < source->y_width; j += 16)
{
unsigned int sse;
Total += VARIANCE_INVOKE(rtcd, mse16x16)(src + j, source->y_stride, dst + j, dest->y_stride, &sse);
}
src += 16 * source->y_stride;
dst += 16 * dest->y_stride;
}
return Total;
}
Increase static linkage, remove unused functions A large number of functions were defined with external linkage, even though they were only used from within one file. This patch changes their linkage to static and removes the vp8_ prefix from their names, which should make it more obvious to the reader that the function is contained within the current translation unit. Functions that were not referenced were removed. These symbols were identified by: $ nm -A libvpx.a | sort -k3 | uniq -c -f2 | grep ' [A-Z] ' \ | sort | grep '^ *1 ' Change-Id: I59609f58ab65312012c047036ae1e0634f795779
2011-03-17 22:07:59 +01:00
Initial WebM release
2010-05-18 17:58:33 +02:00
int vp8_get_quantizer(VP8_PTR c)
{
VP8_COMP *cpi = (VP8_COMP *) c;
return cpi->common.base_qindex;
}
Reference in New Issue Copy Permalink