vpx/vp8/encoder/tokenize.c

968 lines
25 KiB
C
Raw Normal View History

2010-05-18 17:58:33 +02:00
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
2010-05-18 17:58:33 +02:00
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
2010-05-18 17:58:33 +02:00
*/
#include <math.h>
#include <stdio.h>
#include <string.h>
#include <assert.h>
#include "onyx_int.h"
#include "tokenize.h"
#include "vpx_mem/vpx_mem.h"
2011-10-05 12:26:00 +02:00
#include "vp8/common/seg_common.h"
2010-05-18 17:58:33 +02:00
/* Global event counters used for accumulating statistics across several
compressions, then generating context.c = initial stats. */
#ifdef ENTROPY_STATS
_int64 context_counters[BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [MAX_ENTROPY_TOKENS];
_int64 context_counters_8x8[BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [MAX_ENTROPY_TOKENS];
#endif
2010-05-18 17:58:33 +02:00
void vp8_stuff_mb(VP8_COMP *cpi, MACROBLOCKD *x, TOKENEXTRA **t) ;
void vp8_stuff_mb_8x8(VP8_COMP *cpi, MACROBLOCKD *x, TOKENEXTRA **t) ;
void vp8_fix_contexts(MACROBLOCKD *x);
2010-05-18 17:58:33 +02:00
static TOKENVALUE dct_value_tokens[DCT_MAX_VALUE*2];
const TOKENVALUE *vp8_dct_value_tokens_ptr;
static int dct_value_cost[DCT_MAX_VALUE*2];
Add trellis quantization. Replace the exponential search for optimal rounding during quantization with a linear Viterbi trellis and enable it by default when using --best. Right now this operates on top of the output of the adaptive zero-bin quantizer in vp8_regular_quantize_b() and gives a small gain. It can be tested as a replacement for that quantizer by enabling the call to vp8_strict_quantize_b(), which uses normal rounding and no zero bin offset. Ultimately, the quantizer will have to become a function of lambda in order to take advantage of activity masking, since there is limited ability to change the quantization factor itself. However, currently vp8_strict_quantize_b() plus the trellis quantizer (which is lambda-dependent) loses to vp8_regular_quantize_b() alone (which is not) on my test clip. Patch Set 3: Fix an issue related to the cost evaluation of successor states when a coefficient is reduced to zero. With this issue fixed, now the trellis search almost exactly matches the exponential search. Patch Set 2: Overall, the goal of this patch set is to make "trellis" search to produce encodings that match the exponential search version. There are three main differences between Patch Set 2 and 1: a. Patch set 1 did not properly account for the scale of 2nd order error, so patch set 2 disable it all together for 2nd blocks. b. Patch set 1 was not consistent on when to enable the the quantization optimization. Patch set 2 restore the condition to be consistent. c. Patch set 1 checks quantized level L-1, and L for any input coefficient was quantized to L. Patch set 2 limits the candidate coefficient to those that were rounded up to L. It is worth noting here that a strategy to check L and L+1 for coefficients that were truncated down to L might work. (a and b get trellis quant to basically match the exponential search on all mid/low rate encodings on cif set, without a, b, trellis quant can hurt the psnr by 0.2 to .3db at 200kbps for some cif clips) (c gets trellis quant to match the exponential search to match at Q0 encoding, without c, trellis quant can be 1.5 to 2db lower for encodings with fixed Q at 0 on most derf cif clips) Change-Id: Ib1a043b665d75fbf00cb0257b7c18e90eebab95e
2010-07-02 23:35:53 +02:00
const int *vp8_dct_value_cost_ptr;
#ifdef ENC_DEBUG
extern int mb_row_debug;
extern int mb_col_debug;
extern int enc_debug;
2010-05-18 17:58:33 +02:00
#endif
2010-05-18 17:58:33 +02:00
static void fill_value_tokens()
{
TOKENVALUE *const t = dct_value_tokens + DCT_MAX_VALUE;
2010-05-18 17:58:33 +02:00
vp8_extra_bit_struct *const e = vp8_extra_bits;
int i = -DCT_MAX_VALUE;
int sign = 1;
do
{
if (!i)
sign = 0;
{
const int a = sign ? -i : i;
int eb = sign;
if (a > 4)
{
int j = 4;
while (++j < 11 && e[j].base_val <= a) {}
t[i].Token = --j;
eb |= (a - e[j].base_val) << 1;
}
else
t[i].Token = a;
t[i].Extra = eb;
}
// initialize the cost for extra bits for all possible coefficient value.
{
int cost = 0;
vp8_extra_bit_struct *p = vp8_extra_bits + t[i].Token;
if (p->base_val)
{
const int extra = t[i].Extra;
const int Length = p->Len;
if (Length)
cost += vp8_treed_cost(p->tree, p->prob, extra >> 1, Length);
cost += vp8_cost_bit(vp8_prob_half, extra & 1); /* sign */
dct_value_cost[i + DCT_MAX_VALUE] = cost;
2010-05-18 17:58:33 +02:00
}
}
}
while (++i < DCT_MAX_VALUE);
vp8_dct_value_tokens_ptr = dct_value_tokens + DCT_MAX_VALUE;
vp8_dct_value_cost_ptr = dct_value_cost + DCT_MAX_VALUE;
2010-05-18 17:58:33 +02:00
}
static void tokenize2nd_order_b_8x8
(
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
MACROBLOCKD *xd,
const BLOCKD *const b,
TOKENEXTRA **tp,
const int type, /* which plane: 0=Y no DC, 1=Y2, 2=UV, 3=Y with DC */
const FRAME_TYPE frametype,
ENTROPY_CONTEXT *a,
ENTROPY_CONTEXT *l,
VP8_COMP *cpi
)
{
int pt; /* near block/prev token context index */
int c = 0; /* start at DC */
const int eob = b->eob; /* one beyond last nonzero coeff */
TOKENEXTRA *t = *tp; /* store tokens starting here */
int x;
const short *qcoeff_ptr = b->qcoeff;
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
int seg_eob = 4;
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
int segment_id = xd->mode_info_context->mbmi.segment_id;
if ( segfeature_active( xd, segment_id, SEG_LVL_EOB ) )
{
seg_eob = get_segdata( xd, segment_id, SEG_LVL_EOB );
}
VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
assert(eob<=4);
do
{
const int band = vp8_coef_bands[c];
if (c < eob)
{
int rc = vp8_default_zig_zag1d[c];
const int v = qcoeff_ptr[rc];
assert(-DCT_MAX_VALUE <= v && v < (DCT_MAX_VALUE));
t->Extra = vp8_dct_value_tokens_ptr[v].Extra;
x = vp8_dct_value_tokens_ptr[v].Token;
}
else
x = DCT_EOB_TOKEN;
t->Token = x;
//printf("Token : %d\n", x);
t->context_tree = cpi->common.fc.coef_probs_8x8 [type] [band] [pt];
t->skip_eob_node = pt == 0 && ((band > 0 && type > 0) || (band > 1 && type == 0));
#ifdef ENC_DEBUG
if (t->skip_eob_node && vp8_coef_encodings[x].Len==1)
printf("Trouble 2 x=%d Len=%d skip=%d eob=%d c=%d band=%d type=%d: [%d %d %d]\n",
x, vp8_coef_encodings[x].Len, t->skip_eob_node, eob, c, band, type,
cpi->count, mb_row_debug, mb_col_debug);
#endif
++cpi->coef_counts_8x8 [type] [band] [pt] [x];
}
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
while (pt = vp8_prev_token_class[x], ++t, c < eob && ++c <seg_eob);
*tp = t;
pt = (c != !type); /* 0 <-> all coeff data is zero */
*a = *l = pt;
}
2010-05-18 17:58:33 +02:00
static void tokenize2nd_order_b
(
2011-10-05 12:26:00 +02:00
MACROBLOCKD *xd,
2010-05-18 17:58:33 +02:00
TOKENEXTRA **tp,
VP8_COMP *cpi
)
{
int pt; /* near block/prev token context index */
int c; /* start at DC */
TOKENEXTRA *t = *tp;/* store tokens starting here */
const BLOCKD *b;
const short *qcoeff_ptr;
ENTROPY_CONTEXT * a;
ENTROPY_CONTEXT * l;
int band, rc, v, token;
2011-10-05 12:26:00 +02:00
int seg_eob = 16;
int segment_id = xd->mode_info_context->mbmi.segment_id;
if ( segfeature_active( xd, segment_id, SEG_LVL_EOB ) )
{
seg_eob = get_segdata( xd, segment_id, SEG_LVL_EOB );
2011-10-05 12:26:00 +02:00
}
b = xd->block + 24;
qcoeff_ptr = b->qcoeff;
2011-10-05 12:26:00 +02:00
a = (ENTROPY_CONTEXT *)xd->above_context + 8;
l = (ENTROPY_CONTEXT *)xd->left_context + 8;
2010-05-18 17:58:33 +02:00
VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
for (c = 0; c < b->eob; c++)
2010-05-18 17:58:33 +02:00
{
rc = vp8_default_zig_zag1d[c];
band = vp8_coef_bands[c];
v = qcoeff_ptr[rc];
2010-05-18 17:58:33 +02:00
t->Extra = vp8_dct_value_tokens_ptr[v].Extra;
token = vp8_dct_value_tokens_ptr[v].Token;
t->Token = token;
t->context_tree = cpi->common.fc.coef_probs [1] [band] [pt];
t->skip_eob_node = ((pt == 0) && (band > 0));
++cpi->coef_counts [1] [band] [pt] [token];
pt = vp8_prev_token_class[token];
t++;
}
2011-10-05 12:26:00 +02:00
if (c < seg_eob)
{
band = vp8_coef_bands[c];
t->Token = DCT_EOB_TOKEN;
t->context_tree = cpi->common.fc.coef_probs [1] [band] [pt];
2010-05-18 17:58:33 +02:00
t->skip_eob_node = ((pt == 0) && (band > 0));
++cpi->coef_counts [1] [band] [pt] [DCT_EOB_TOKEN];
2010-05-18 17:58:33 +02:00
t++;
2010-05-18 17:58:33 +02:00
}
*tp = t;
pt = (c != 0); /* 0 <-> all coeff data is zero */
2010-05-18 17:58:33 +02:00
*a = *l = pt;
}
static void tokenize1st_order_b_8x8
(
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
MACROBLOCKD *xd,
const BLOCKD *const b,
TOKENEXTRA **tp,
const int type, /* which plane: 0=Y no DC, 1=Y2, 2=UV, 3=Y with DC */
const FRAME_TYPE frametype,
ENTROPY_CONTEXT *a,
ENTROPY_CONTEXT *l,
VP8_COMP *cpi
)
{
int pt; /* near block/prev token context index */
int c = type ? 0 : 1; /* start at DC unless type 0 */
const int eob = b->eob; /* one beyond last nonzero coeff */
TOKENEXTRA *t = *tp; /* store tokens starting here */
int x;
const short *qcoeff_ptr = b->qcoeff;
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
int seg_eob = 64;
int segment_id = xd->mode_info_context->mbmi.segment_id;
if ( segfeature_active( xd, segment_id, SEG_LVL_EOB ) )
{
seg_eob = get_segdata( xd, segment_id, SEG_LVL_EOB );
}
VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
do
{
const int band = vp8_coef_bands_8x8[c];
x = DCT_EOB_TOKEN;
if (c < eob)
{
int rc = vp8_default_zig_zag1d_8x8[c];
const int v = qcoeff_ptr[rc];
assert(-DCT_MAX_VALUE <= v && v < (DCT_MAX_VALUE));
t->Extra = vp8_dct_value_tokens_ptr[v].Extra;
x = vp8_dct_value_tokens_ptr[v].Token;
}
t->Token = x;
t->context_tree = cpi->common.fc.coef_probs_8x8 [type] [band] [pt];
t->skip_eob_node = pt == 0 && ((band > 0 && type > 0) || (band > 1 && type == 0));
#ifdef ENC_DEBUG
if (t->skip_eob_node && vp8_coef_encodings[x].Len==1)
printf("Trouble 1 x=%d Len=%d skip=%d eob=%d c=%d band=%d type=%d: [%d %d %d]\n", x, vp8_coef_encodings[x].Len, t->skip_eob_node, eob, c, band, type, cpi->count, mb_row_debug, mb_col_debug);
#endif
++cpi->coef_counts_8x8 [type] [band] [pt] [x];
}
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
while (pt = vp8_prev_token_class[x], ++t, c < eob && ++c < seg_eob);
*tp = t;
pt = (c != !type); /* 0 <-> all coeff data is zero */
*a = *l = pt;
}
2010-05-18 17:58:33 +02:00
static void tokenize1st_order_b
(
2011-10-05 12:26:00 +02:00
MACROBLOCKD *xd,
2010-05-18 17:58:33 +02:00
TOKENEXTRA **tp,
int type, /* which plane: 0=Y no DC, 1=Y2, 2=UV, 3=Y with DC */
2010-05-18 17:58:33 +02:00
VP8_COMP *cpi
)
{
unsigned int block;
const BLOCKD *b;
int pt; /* near block/prev token context index */
int c;
int token;
TOKENEXTRA *t = *tp;/* store tokens starting here */
const short *qcoeff_ptr;
ENTROPY_CONTEXT * a;
ENTROPY_CONTEXT * l;
int band, rc, v;
int tmp1, tmp2;
2011-10-05 12:26:00 +02:00
int seg_eob = 16;
int segment_id = xd->mode_info_context->mbmi.segment_id;
if ( segfeature_active( xd, segment_id, SEG_LVL_EOB ) )
{
seg_eob = get_segdata( xd, segment_id, SEG_LVL_EOB );
2011-10-05 12:26:00 +02:00
}
b = xd->block;
/* Luma */
for (block = 0; block < 16; block++, b++)
2010-05-18 17:58:33 +02:00
{
tmp1 = vp8_block2above[block];
tmp2 = vp8_block2left[block];
qcoeff_ptr = b->qcoeff;
2011-10-05 12:26:00 +02:00
a = (ENTROPY_CONTEXT *)xd->above_context + tmp1;
l = (ENTROPY_CONTEXT *)xd->left_context + tmp2;
VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
2010-05-18 17:58:33 +02:00
c = type ? 0 : 1;
2010-05-18 17:58:33 +02:00
for (; c < b->eob; c++)
2010-05-18 17:58:33 +02:00
{
rc = vp8_default_zig_zag1d[c];
band = vp8_coef_bands[c];
v = qcoeff_ptr[rc];
2010-05-18 17:58:33 +02:00
t->Extra = vp8_dct_value_tokens_ptr[v].Extra;
token = vp8_dct_value_tokens_ptr[v].Token;
t->Token = token;
t->context_tree = cpi->common.fc.coef_probs [type] [band] [pt];
t->skip_eob_node = pt == 0 &&
((band > 0 && type > 0) || (band > 1 && type == 0));
++cpi->coef_counts [type] [band] [pt] [token];
pt = vp8_prev_token_class[token];
t++;
2010-05-18 17:58:33 +02:00
}
2011-10-05 12:26:00 +02:00
if (c < seg_eob)
{
band = vp8_coef_bands[c];
t->Token = DCT_EOB_TOKEN;
t->context_tree = cpi->common.fc.coef_probs [type] [band] [pt];
2010-05-18 17:58:33 +02:00
t->skip_eob_node = pt == 0 &&
((band > 0 && type > 0) || (band > 1 && type == 0));
++cpi->coef_counts [type] [band] [pt] [DCT_EOB_TOKEN];
t++;
}
*tp = t;
pt = (c != !type); /* 0 <-> all coeff data is zero */
*a = *l = pt;
2010-05-18 17:58:33 +02:00
}
/* Chroma */
for (block = 16; block < 24; block++, b++)
{
tmp1 = vp8_block2above[block];
tmp2 = vp8_block2left[block];
qcoeff_ptr = b->qcoeff;
2011-10-05 12:26:00 +02:00
a = (ENTROPY_CONTEXT *)xd->above_context + tmp1;
l = (ENTROPY_CONTEXT *)xd->left_context + tmp2;
2010-05-18 17:58:33 +02:00
VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
for (c = 0; c < b->eob; c++)
{
rc = vp8_default_zig_zag1d[c];
band = vp8_coef_bands[c];
v = qcoeff_ptr[rc];
t->Extra = vp8_dct_value_tokens_ptr[v].Extra;
token = vp8_dct_value_tokens_ptr[v].Token;
t->Token = token;
t->context_tree = cpi->common.fc.coef_probs [2] [band] [pt];
t->skip_eob_node = ((pt == 0) && (band > 0));
++cpi->coef_counts [2] [band] [pt] [token];
pt = vp8_prev_token_class[token];
t++;
}
2011-10-05 12:26:00 +02:00
if (c < seg_eob)
{
band = vp8_coef_bands[c];
t->Token = DCT_EOB_TOKEN;
t->context_tree = cpi->common.fc.coef_probs [2] [band] [pt];
t->skip_eob_node = ((pt == 0) && (band > 0));
++cpi->coef_counts [2] [band] [pt] [DCT_EOB_TOKEN];
t++;
}
*tp = t;
pt = (c != 0); /* 0 <-> all coeff data is zero */
*a = *l = pt;
}
2010-05-18 17:58:33 +02:00
}
int mby_is_skippable(MACROBLOCKD *x, int has_y2_block)
{
int skip = 1;
int i = 0;
if (has_y2_block)
{
for (i = 0; i < 16; i++)
skip &= (x->block[i].eob < 2);
skip &= (!x->block[24].eob);
}
else
{
for (i = 0; i < 16; i++)
skip &= (!x->block[i].eob);
}
return skip;
}
int mbuv_is_skippable(MACROBLOCKD *x)
{
int skip = 1;
int i;
for (i = 16; i < 24; i++)
skip &= (!x->block[i].eob);
return skip;
}
int mb_is_skippable(MACROBLOCKD *x, int has_y2_block)
{
return (mby_is_skippable(x, has_y2_block) &
mbuv_is_skippable(x));
}
int mby_is_skippable_8x8(MACROBLOCKD *x)
{
int skip = 1;
int i = 0;
for (i = 0; i < 16; i+=4)
skip &= (x->block[i].eob < 2);
skip &= (!x->block[24].eob);
return skip;
}
int mbuv_is_skippable_8x8(MACROBLOCKD *x)
{
return (!x->block[16].eob) & (!x->block[20].eob);
}
int mb_is_skippable_8x8(MACROBLOCKD *x)
{
return (mby_is_skippable_8x8(x) & mbuv_is_skippable_8x8(x));
}
2010-05-18 17:58:33 +02:00
void vp8_tokenize_mb(VP8_COMP *cpi, MACROBLOCKD *x, TOKENEXTRA **t)
{
int plane_type;
int has_y2_block;
int b;
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
int tx_type = x->mode_info_context->mbmi.txfm_size;
// If the MB is going to be skipped because of a segment level flag
// exclude this from the skip count stats used to calculate the
// transmitted skip probability;
int skip_inc;
int segment_id = x->mode_info_context->mbmi.segment_id;
if ( !segfeature_active( x, segment_id, SEG_LVL_EOB ) ||
( get_segdata( x, segment_id, SEG_LVL_EOB ) != 0) )
{
skip_inc = 1;
}
else
skip_inc = 0;
has_y2_block = (x->mode_info_context->mbmi.mode != B_PRED
&& x->mode_info_context->mbmi.mode != I8X8_PRED
&& x->mode_info_context->mbmi.mode != SPLITMV);
2010-05-18 17:58:33 +02:00
x->mode_info_context->mbmi.mb_skip_coeff =
(( tx_type == TX_8X8 ) ?
mb_is_skippable_8x8(x) :
mb_is_skippable(x, has_y2_block));
if (x->mode_info_context->mbmi.mb_skip_coeff)
2010-05-18 17:58:33 +02:00
{
cpi->skip_true_count += skip_inc;
2010-05-18 17:58:33 +02:00
if (!cpi->common.mb_no_coeff_skip)
{
if ( tx_type == TX_8X8 )
vp8_stuff_mb_8x8(cpi, x, t) ;
else
vp8_stuff_mb(cpi, x, t) ;
}
2010-05-18 17:58:33 +02:00
else
{
vp8_fix_contexts(x);
2010-05-18 17:58:33 +02:00
}
return;
}
cpi->skip_false_count += skip_inc;
plane_type = 3;
if(has_y2_block)
2010-05-18 17:58:33 +02:00
{
if ( tx_type == TX_8X8 )
{
ENTROPY_CONTEXT * A = (ENTROPY_CONTEXT *)x->above_context;
ENTROPY_CONTEXT * L = (ENTROPY_CONTEXT *)x->left_context;
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
tokenize2nd_order_b_8x8(x,
x->block + 24, t, 1, x->frame_type,
A + vp8_block2above_8x8[24],
L + vp8_block2left_8x8[24], cpi);
}
else
tokenize2nd_order_b(x, t, cpi);
plane_type = 0;
2010-05-18 17:58:33 +02:00
}
if ( tx_type == TX_8X8 )
{
ENTROPY_CONTEXT * A = (ENTROPY_CONTEXT *)x->above_context;
ENTROPY_CONTEXT * L = (ENTROPY_CONTEXT *)x->left_context;
for (b = 0; b < 16; b+=4)
{
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
tokenize1st_order_b_8x8(x,
x->block + b, t, plane_type, x->frame_type,
A + vp8_block2above_8x8[b],
L + vp8_block2left_8x8[b],
cpi);
*(A + vp8_block2above_8x8[b] + 1) = *(A + vp8_block2above_8x8[b]);
*(L + vp8_block2left_8x8[b] + 1) = *(L + vp8_block2left_8x8[b] );
}
for (b = 16; b < 24; b+=4)
{
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
tokenize1st_order_b_8x8(x,
x->block + b, t, 2, x->frame_type,
A + vp8_block2above_8x8[b],
L + vp8_block2left_8x8[b],
cpi);
*(A + vp8_block2above_8x8[b]+1) = *(A + vp8_block2above_8x8[b]);
*(L + vp8_block2left_8x8[b]+1 ) = *(L + vp8_block2left_8x8[b]);
}
}
else
tokenize1st_order_b(x, t, plane_type, cpi);
2010-05-18 17:58:33 +02:00
}
2010-05-18 17:58:33 +02:00
#ifdef ENTROPY_STATS
void init_context_counters(void)
{
vpx_memset(context_counters, 0, sizeof(context_counters));
vpx_memset(context_counters_8x8, 0, sizeof(context_counters_8x8));
2010-05-18 17:58:33 +02:00
}
void print_context_counters()
{
int type, band, pt, t;
FILE *const f = fopen("context.c", "w");
fprintf(f, "#include \"entropy.h\"\n");
fprintf(f, "\n/* *** GENERATED FILE: DO NOT EDIT *** */\n\n");
fprintf(f, "int Contexts[BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [MAX_ENTROPY_TOKENS];\n\n");
2010-05-18 17:58:33 +02:00
fprintf(f, "const int default_contexts[BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [MAX_ENTROPY_TOKENS] = {");
2010-05-18 17:58:33 +02:00
# define Comma( X) (X? ",":"")
type = 0;
do
{
fprintf(f, "%s\n { /* block Type %d */", Comma(type), type);
band = 0;
do
{
fprintf(f, "%s\n { /* Coeff Band %d */", Comma(band), band);
pt = 0;
do
{
fprintf(f, "%s\n {", Comma(pt));
t = 0;
do
{
const _int64 x = context_counters [type] [band] [pt] [t];
const int y = (int) x;
assert(x == (INT64) y); /* no overflow handling yet */
fprintf(f, "%s %d", Comma(t), y);
}
while (++t < MAX_ENTROPY_TOKENS);
fprintf(f, "}");
}
while (++pt < PREV_COEF_CONTEXTS);
fprintf(f, "\n }");
}
while (++band < COEF_BANDS);
fprintf(f, "\n }");
}
while (++type < BLOCK_TYPES);
fprintf(f, "int Contexts_8x8[BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [MAX_ENTROPY_TOKENS];\n\n");
fprintf(f, "const int default_contexts_8x8[BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [MAX_ENTROPY_TOKENS] = {");
2010-05-18 17:58:33 +02:00
type = 0;
do
{
fprintf(f, "%s\n { /* block Type %d */", Comma(type), type);
band = 0;
do
{
fprintf(f, "%s\n { /* Coeff Band %d */", Comma(band), band);
pt = 0;
do
{
fprintf(f, "%s\n {", Comma(pt));
t = 0;
do
{
const _int64 x = context_counters [type] [band] [pt] [t];
const int y = (int) x;
assert(x == (_int64) y); /* no overflow handling yet */
fprintf(f, "%s %d", Comma(t), y);
}
while (++t < MAX_ENTROPY_TOKENS);
2010-05-18 17:58:33 +02:00
fprintf(f, "}");
}
while (++pt < PREV_COEF_CONTEXTS);
fprintf(f, "\n }");
}
while (++band < COEF_BANDS);
fprintf(f, "\n }");
}
while (++type < BLOCK_TYPES);
fprintf(f, "\n};\n");
fclose(f);
}
#endif
void vp8_tokenize_initialize()
{
fill_value_tokens();
}
static __inline void stuff2nd_order_b_8x8
(
const BLOCKD *const b,
TOKENEXTRA **tp,
const int type, /* which plane: 0=Y no DC, 1=Y2, 2=UV, 3=Y with DC */
const FRAME_TYPE frametype,
ENTROPY_CONTEXT *a,
ENTROPY_CONTEXT *l,
VP8_COMP *cpi
)
{
int pt; /* near block/prev token context index */
TOKENEXTRA *t = *tp; /* store tokens starting here */
VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
(void) frametype;
(void) type;
(void) b;
t->Token = DCT_EOB_TOKEN;
t->context_tree = cpi->common.fc.coef_probs_8x8 [1] [0] [pt];
//t->section = 11;
t->skip_eob_node = 0;
++cpi->coef_counts_8x8 [1] [0] [pt] [DCT_EOB_TOKEN];
++t;
*tp = t;
pt = 0;
*a = *l = pt;
}
static __inline void stuff1st_order_b_8x8
(
const BLOCKD *const b,
TOKENEXTRA **tp,
const int type, /* which plane: 0=Y no DC, 1=Y2, 2=UV, 3=Y with DC */
const FRAME_TYPE frametype,
ENTROPY_CONTEXT *a,
ENTROPY_CONTEXT *l,
VP8_COMP *cpi
)
{
int pt; /* near block/prev token context index */
TOKENEXTRA *t = *tp; /* store tokens starting here */
VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
(void) frametype;
(void) type;
(void) b;
t->Token = DCT_EOB_TOKEN;
t->context_tree = cpi->common.fc.coef_probs_8x8 [0] [1] [pt];
//t->section = 8;
t->skip_eob_node = 0;
++cpi->coef_counts_8x8 [0] [1] [pt] [DCT_EOB_TOKEN];
++t;
*tp = t;
pt = 0; /* 0 <-> all coeff data is zero */
*a = *l = pt;
}
static __inline
void stuff1st_order_buv_8x8
(
const BLOCKD *const b,
TOKENEXTRA **tp,
const int type, /* which plane: 0=Y no DC, 1=Y2, 2=UV, 3=Y with DC */
const FRAME_TYPE frametype,
ENTROPY_CONTEXT *a,
ENTROPY_CONTEXT *l,
VP8_COMP *cpi
)
{
int pt; /* near block/prev token context index */
TOKENEXTRA *t = *tp; /* store tokens starting here */
VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
(void) frametype;
(void) type;
(void) b;
t->Token = DCT_EOB_TOKEN;
t->context_tree = cpi->common.fc.coef_probs_8x8 [2] [0] [pt];
//t->section = 13;
t->skip_eob_node = 0;
++cpi->coef_counts_8x8[2] [0] [pt] [DCT_EOB_TOKEN];
++t;
*tp = t;
pt = 0; /* 0 <-> all coeff data is zero */
*a = *l = pt;
}
void vp8_stuff_mb_8x8(VP8_COMP *cpi, MACROBLOCKD *x, TOKENEXTRA **t)
{
ENTROPY_CONTEXT * A = (ENTROPY_CONTEXT *)x->above_context;
ENTROPY_CONTEXT * L = (ENTROPY_CONTEXT *)x->left_context;
int plane_type;
int b;
stuff2nd_order_b_8x8(x->block + 24, t, 1, x->frame_type,
A + vp8_block2above_8x8[24],
L + vp8_block2left_8x8[24], cpi);
plane_type = 0;
for (b = 0; b < 16; b+=4)
{
stuff1st_order_b_8x8(x->block + b, t, plane_type, x->frame_type,
A + vp8_block2above_8x8[b],
L + vp8_block2left_8x8[b],
cpi);
*(A + vp8_block2above_8x8[b] + 1) = *(A + vp8_block2above_8x8[b]);
*(L + vp8_block2left_8x8[b] + 1) = *(L + vp8_block2left_8x8[b] );
}
for (b = 16; b < 24; b+=4)
{
stuff1st_order_buv_8x8(x->block + b, t, 2, x->frame_type,
A + vp8_block2above[b],
L + vp8_block2left[b],
cpi);
*(A + vp8_block2above_8x8[b]+1) = *(A + vp8_block2above_8x8[b]);
*(L + vp8_block2left_8x8[b]+1 ) = *(L + vp8_block2left_8x8[b]);
}
}
2010-05-18 17:58:33 +02:00
static __inline void stuff2nd_order_b
(
TOKENEXTRA **tp,
ENTROPY_CONTEXT *a,
ENTROPY_CONTEXT *l,
VP8_COMP *cpi
)
{
int pt; /* near block/prev token context index */
TOKENEXTRA *t = *tp; /* store tokens starting here */
VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
t->Token = DCT_EOB_TOKEN;
t->context_tree = cpi->common.fc.coef_probs [1] [0] [pt];
t->skip_eob_node = 0;
++cpi->coef_counts [1] [0] [pt] [DCT_EOB_TOKEN];
++t;
*tp = t;
pt = 0;
*a = *l = pt;
}
static __inline void stuff1st_order_b
(
TOKENEXTRA **tp,
ENTROPY_CONTEXT *a,
ENTROPY_CONTEXT *l,
VP8_COMP *cpi
)
{
int pt; /* near block/prev token context index */
TOKENEXTRA *t = *tp; /* store tokens starting here */
VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
t->Token = DCT_EOB_TOKEN;
t->context_tree = cpi->common.fc.coef_probs [0] [1] [pt];
t->skip_eob_node = 0;
++cpi->coef_counts [0] [1] [pt] [DCT_EOB_TOKEN];
++t;
*tp = t;
pt = 0; /* 0 <-> all coeff data is zero */
*a = *l = pt;
}
static __inline
void stuff1st_order_buv
(
TOKENEXTRA **tp,
ENTROPY_CONTEXT *a,
ENTROPY_CONTEXT *l,
VP8_COMP *cpi
)
{
int pt; /* near block/prev token context index */
TOKENEXTRA *t = *tp; /* store tokens starting here */
VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
t->Token = DCT_EOB_TOKEN;
t->context_tree = cpi->common.fc.coef_probs [2] [0] [pt];
t->skip_eob_node = 0;
++cpi->coef_counts[2] [0] [pt] [DCT_EOB_TOKEN];
++t;
*tp = t;
pt = 0; /* 0 <-> all coeff data is zero */
*a = *l = pt;
}
void vp8_stuff_mb(VP8_COMP *cpi, MACROBLOCKD *x, TOKENEXTRA **t)
{
ENTROPY_CONTEXT * A = (ENTROPY_CONTEXT *)x->above_context;
ENTROPY_CONTEXT * L = (ENTROPY_CONTEXT *)x->left_context;
2010-05-18 17:58:33 +02:00
int plane_type;
int b;
stuff2nd_order_b(t,
A + vp8_block2above[24], L + vp8_block2left[24], cpi);
2010-05-18 17:58:33 +02:00
plane_type = 0;
for (b = 0; b < 16; b++)
stuff1st_order_b(t,
A + vp8_block2above[b],
L + vp8_block2left[b], cpi);
2010-05-18 17:58:33 +02:00
for (b = 16; b < 24; b++)
stuff1st_order_buv(t,
A + vp8_block2above[b],
L + vp8_block2left[b], cpi);
2010-05-18 17:58:33 +02:00
}
void vp8_fix_contexts(MACROBLOCKD *x)
2010-05-18 17:58:33 +02:00
{
/* Clear entropy contexts for Y2 blocks */
if (x->mode_info_context->mbmi.mode != B_PRED
&& x->mode_info_context->mbmi.mode != I8X8_PRED
&& x->mode_info_context->mbmi.mode != SPLITMV)
2010-05-18 17:58:33 +02:00
{
vpx_memset(x->above_context, 0, sizeof(ENTROPY_CONTEXT_PLANES));
vpx_memset(x->left_context, 0, sizeof(ENTROPY_CONTEXT_PLANES));
2010-05-18 17:58:33 +02:00
}
else
{
vpx_memset(x->above_context, 0, sizeof(ENTROPY_CONTEXT_PLANES)-1);
vpx_memset(x->left_context, 0, sizeof(ENTROPY_CONTEXT_PLANES)-1);
}
2010-05-18 17:58:33 +02:00
}