generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
eaa955ba98
vpx/vp8/decoder/detokenize.c

795 lines
21 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 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/type_aliases.h"
#include "vp8/common/blockd.h"
Initial WebM release
2010-05-18 17:58:33 +02:00
#include "onyxd_int.h"
#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/mem.h"
framework for assembly version of the detokenizer adds a compile time option: --enable-arm-asm-detok which pulls in vp8/decoder/arm/detokenize.asm currently about break even speed wise, but changes are pending to the fill code (branch and load 3 bytes versus conditionally always load one) and the error handling. Currently it doesn't handle zero runs or overrunning the buffer. this is really just so i don't have to rebase my changes all the time to run benchmarks - now just need to replace one file! Change-Id: I56d0e2354dc0ca3811bffd0e88fe1f952fa6c797
2010-08-12 15:05:37 +02:00
#include "detokenize.h"
Initial WebM release
2010-05-18 17:58:33 +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
#include "vp8/common/seg_common.h"
Initial WebM release
2010-05-18 17:58:33 +02:00
#define BOOL_DATA UINT8
#define OCB_X PREV_COEF_CONTEXTS * ENTROPY_NODES
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
DECLARE_ALIGNED(16, static const unsigned char, coef_bands_x[16]) =
{
0 * OCB_X, 1 * OCB_X, 2 * OCB_X, 3 * OCB_X,
6 * OCB_X, 4 * OCB_X, 5 * OCB_X, 6 * OCB_X,
6 * OCB_X, 6 * OCB_X, 6 * OCB_X, 6 * OCB_X,
6 * OCB_X, 6 * OCB_X, 6 * OCB_X, 7 * OCB_X
};
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
DECLARE_ALIGNED(64, static const unsigned char, coef_bands_x_8x8[64]) = {
0 * OCB_X, 1 * OCB_X, 2 * OCB_X, 3 * OCB_X, 5 * OCB_X, 4 * OCB_X, 4 * OCB_X, 5 * OCB_X,
5 * OCB_X, 3 * OCB_X, 6 * OCB_X, 3 * OCB_X, 5 * OCB_X, 4 * OCB_X, 6 * OCB_X, 6 * OCB_X,
6 * OCB_X, 5 * OCB_X, 5 * OCB_X, 6 * OCB_X, 6 * OCB_X, 6 * OCB_X, 6 * OCB_X, 6 * OCB_X,
6 * OCB_X, 6 * OCB_X, 6 * OCB_X, 6 * OCB_X, 6 * OCB_X, 6 * OCB_X, 6 * OCB_X, 6 * OCB_X,
6 * OCB_X, 6 * OCB_X, 6 * OCB_X, 6 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X,
7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X,
7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X,
7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X, 7 * OCB_X,
};
Merge t8x8 experiments Change-Id: I8e9b6b154e1a0d0cb42d596366380d69c00ac15f
2012-02-29 02:11:12 +01:00
Initial WebM release
2010-05-18 17:58:33 +02:00
#define EOB_CONTEXT_NODE 0
#define ZERO_CONTEXT_NODE 1
#define ONE_CONTEXT_NODE 2
#define LOW_VAL_CONTEXT_NODE 3
#define TWO_CONTEXT_NODE 4
#define THREE_CONTEXT_NODE 5
#define HIGH_LOW_CONTEXT_NODE 6
#define CAT_ONE_CONTEXT_NODE 7
#define CAT_THREEFOUR_CONTEXT_NODE 8
#define CAT_THREE_CONTEXT_NODE 9
#define CAT_FIVE_CONTEXT_NODE 10
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
#define CAT1_MIN_VAL 5
#define CAT2_MIN_VAL 7
#define CAT3_MIN_VAL 11
#define CAT4_MIN_VAL 19
#define CAT5_MIN_VAL 35
#define CAT6_MIN_VAL 67
#define CAT1_PROB0 159
#define CAT2_PROB0 145
#define CAT2_PROB1 165
#define CAT3_PROB0 140
#define CAT3_PROB1 148
#define CAT3_PROB2 173
#define CAT4_PROB0 135
#define CAT4_PROB1 140
#define CAT4_PROB2 155
#define CAT4_PROB3 176
#define CAT5_PROB0 130
#define CAT5_PROB1 134
#define CAT5_PROB2 141
#define CAT5_PROB3 157
#define CAT5_PROB4 180
Merge remote branch 'internal/upstream' into HEAD Conflicts: vp8/decoder/detokenize.c vp8/decoder/onyxd_int.h Change-Id: Ib9b516b939358ac8bf694200a8425fdd62c8d149
2011-07-26 16:22:42 +02:00
static const unsigned char cat6_prob[14] =
{ 129, 130, 133, 140, 153, 177, 196, 230, 243, 249, 252, 254, 254, 0 };
Initial WebM release
2010-05-18 17:58:33 +02:00
void vp8_reset_mb_tokens_context(MACROBLOCKD *x)
{
Remove costly memory reads/writes in vp8_reset_mb_tokens_context() Tests on x86 showed this function costed 2.7% of total decoding time because of all the memory reads/writes. After modification, it only costs about 0.7% of decoding time, which gives a 2% gain. Change-Id: I5003ee30b6dc6dea0bfa42a6ad7e7c22fcc7b215
2010-05-28 20:34:39 +02:00
/* Clear entropy contexts for Y2 blocks */
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
if (x->mode_info_context->mbmi.mode != B_PRED &&
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
x->mode_info_context->mbmi.mode != I8X8_PRED &&
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
x->mode_info_context->mbmi.mode != SPLITMV)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Changed above and left context data layout The main reason for the change was to reduce cycles in the token decoder. (~1.5% gain for 32 bit) This layout should be more cache friendly. As a result of this change, the encoder had to be updated. Change-Id: Id5e804169d8889da0378b3a519ac04dabd28c837 Note: dixie uses a similar layout
2010-08-31 16:49:57 +02:00
vpx_memset(x->above_context, 0, sizeof(ENTROPY_CONTEXT_PLANES));
vpx_memset(x->left_context, 0, sizeof(ENTROPY_CONTEXT_PLANES));
}
else
{
vpx_memset(x->above_context, 0, sizeof(ENTROPY_CONTEXT_PLANES)-1);
vpx_memset(x->left_context, 0, sizeof(ENTROPY_CONTEXT_PLANES)-1);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
}
framework for assembly version of the detokenizer adds a compile time option: --enable-arm-asm-detok which pulls in vp8/decoder/arm/detokenize.asm currently about break even speed wise, but changes are pending to the fill code (branch and load 3 bytes versus conditionally always load one) and the error handling. Currently it doesn't handle zero runs or overrunning the buffer. this is really just so i don't have to rebase my changes all the time to run benchmarks - now just need to replace one file! Change-Id: I56d0e2354dc0ca3811bffd0e88fe1f952fa6c797
2010-08-12 15:05:37 +02:00
Moved vp8_encode_bool into boolhuff.h allowing the compiler to inline this function. For real-time encodes, this gave a boost of 1% to 2.5%, depending on the speed setting. Change-Id: I3929d176cca086b4261267b848419d5bcff21c02
2011-07-19 15:17:25 +02:00
DECLARE_ALIGNED(16, extern const unsigned char, vp8_norm[256]);
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
#define FILL \
if(count < 0) \
VP8DX_BOOL_DECODER_FILL(count, value, bufptr, bufend);
Initial WebM release
2010-05-18 17:58:33 +02:00
#define NORMALIZE \
/*if(range < 0x80)*/ \
{ \
Moved vp8_encode_bool into boolhuff.h allowing the compiler to inline this function. For real-time encodes, this gave a boost of 1% to 2.5%, depending on the speed setting. Change-Id: I3929d176cca086b4261267b848419d5bcff21c02
2011-07-19 15:17:25 +02:00
shift = vp8_norm[range]; \
Initial WebM release
2010-05-18 17:58:33 +02:00
range <<= shift; \
value <<= shift; \
count -= shift; \
}
#define DECODE_AND_APPLYSIGN(value_to_sign) \
split = (range + 1) >> 1; \
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
bigsplit = (VP8_BD_VALUE)split << (VP8_BD_VALUE_SIZE - 8); \
FILL \
if ( value < bigsplit ) \
Initial WebM release
2010-05-18 17:58:33 +02:00
{ \
range = split; \
v= value_to_sign; \
} \
else \
{ \
range = range-split; \
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
value = value-bigsplit; \
Initial WebM release
2010-05-18 17:58:33 +02:00
v = -value_to_sign; \
} \
range +=range; \
value +=value; \
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
count--;
Initial WebM release
2010-05-18 17:58:33 +02:00
#define DECODE_AND_BRANCH_IF_ZERO(probability,branch) \
{ \
split = 1 + ((( probability*(range-1) ) )>> 8); \
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
bigsplit = (VP8_BD_VALUE)split << (VP8_BD_VALUE_SIZE - 8); \
FILL \
if ( value < bigsplit ) \
Initial WebM release
2010-05-18 17:58:33 +02:00
{ \
range = split; \
NORMALIZE \
goto branch; \
} \
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
value -= bigsplit; \
Initial WebM release
2010-05-18 17:58:33 +02:00
range = range - split; \
NORMALIZE \
}
#define DECODE_AND_LOOP_IF_ZERO(probability,branch) \
{ \
split = 1 + ((( probability*(range-1) ) ) >> 8); \
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
bigsplit = (VP8_BD_VALUE)split << (VP8_BD_VALUE_SIZE - 8); \
FILL \
if ( value < bigsplit ) \
Initial WebM release
2010-05-18 17:58:33 +02:00
{ \
range = split; \
NORMALIZE \
Prob = coef_probs; \
if(c<15) {\
++c; \
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
Prob += coef_bands_x[c]; \
Initial WebM release
2010-05-18 17:58:33 +02:00
goto branch; \
} goto BLOCK_FINISHED; /*for malformed input */\
} \
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
value -= bigsplit; \
Initial WebM release
2010-05-18 17:58:33 +02:00
range = range - split; \
NORMALIZE \
}
Merge t8x8 experiments Change-Id: I8e9b6b154e1a0d0cb42d596366380d69c00ac15f
2012-02-29 02:11:12 +01:00
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
#define DECODE_AND_LOOP_IF_ZERO_8x8_2(probability,branch) \
{ \
split = 1 + ((( probability*(range-1) ) ) >> 8); \
bigsplit = (VP8_BD_VALUE)split << (VP8_BD_VALUE_SIZE - 8); \
FILL \
if ( value < bigsplit ) \
{ \
range = split; \
NORMALIZE \
Prob = coef_probs; \
if(c<3) {\
++c; \
Prob += coef_bands_x[c]; \
goto branch; \
} goto BLOCK_FINISHED_8x8; /*for malformed input */\
} \
value -= bigsplit; \
range = range - split; \
NORMALIZE \
}
#define DECODE_AND_LOOP_IF_ZERO_8X8(probability,branch) \
{ \
split = 1 + ((( probability*(range-1) ) ) >> 8); \
bigsplit = (VP8_BD_VALUE)split << (VP8_BD_VALUE_SIZE - 8); \
FILL \
if ( value < bigsplit ) \
{ \
range = split; \
NORMALIZE \
Prob = coef_probs; \
if(c<63) {\
++c; \
Prob += coef_bands_x_8x8[c]; \
goto branch; \
} goto BLOCK_FINISHED_8x8; /*for malformed input */\
} \
value -= bigsplit; \
range = range - split; \
NORMALIZE \
}
Merge t8x8 experiments Change-Id: I8e9b6b154e1a0d0cb42d596366380d69c00ac15f
2012-02-29 02:11:12 +01:00
Initial WebM release
2010-05-18 17:58:33 +02:00
#define DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT(val) \
DECODE_AND_APPLYSIGN(val) \
Prob = coef_probs + (ENTROPY_NODES*2); \
if(c < 15){\
qcoeff_ptr [ scan[c] ] = (INT16) v; \
++c; \
goto DO_WHILE; }\
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
qcoeff_ptr [ 15 ] = (INT16) v; \
Initial WebM release
2010-05-18 17:58:33 +02:00
goto BLOCK_FINISHED;
Merge t8x8 experiments Change-Id: I8e9b6b154e1a0d0cb42d596366380d69c00ac15f
2012-02-29 02:11:12 +01:00
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
#define DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8_2(val) \
DECODE_AND_APPLYSIGN(val) \
Prob = coef_probs + (ENTROPY_NODES*2); \
if(c < 3){\
qcoeff_ptr [ scan[c] ] = (INT16) v; \
++c; \
goto DO_WHILE_8x8; }\
qcoeff_ptr [ scan[3] ] = (INT16) v; \
goto BLOCK_FINISHED_8x8;
#define DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8(val) \
DECODE_AND_APPLYSIGN(val) \
Prob = coef_probs + (ENTROPY_NODES*2); \
if(c < 63){\
qcoeff_ptr [ scan[c] ] = (INT16) v; \
++c; \
goto DO_WHILE_8x8; }\
qcoeff_ptr [ scan[63] ] = (INT16) v; \
goto BLOCK_FINISHED_8x8;
Merge t8x8 experiments Change-Id: I8e9b6b154e1a0d0cb42d596366380d69c00ac15f
2012-02-29 02:11:12 +01:00
Initial WebM release
2010-05-18 17:58:33 +02:00
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
#define DECODE_EXTRABIT_AND_ADJUST_VAL(prob, bits_count)\
split = 1 + (((range-1) * prob) >> 8); \
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
bigsplit = (VP8_BD_VALUE)split << (VP8_BD_VALUE_SIZE - 8); \
FILL \
if(value >= bigsplit)\
Initial WebM release
2010-05-18 17:58:33 +02:00
{\
range = range-split;\
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
value = value-bigsplit;\
Initial WebM release
2010-05-18 17:58:33 +02:00
val += ((UINT16)1<<bits_count);\
}\
else\
{\
range = split;\
}\
NORMALIZE
Merge t8x8 experiments Change-Id: I8e9b6b154e1a0d0cb42d596366380d69c00ac15f
2012-02-29 02:11:12 +01:00
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
int vp8_decode_mb_tokens_8x8(VP8D_COMP *dx, MACROBLOCKD *x)
{
ENTROPY_CONTEXT *A = (ENTROPY_CONTEXT *)x->above_context;
ENTROPY_CONTEXT *L = (ENTROPY_CONTEXT *)x->left_context;
const VP8_COMMON *const oc = & dx->common;
BOOL_DECODER *bc = x->current_bc;
char *eobs = x->eobs;
ENTROPY_CONTEXT *a, *a1;
ENTROPY_CONTEXT *l, *l1;
int i;
int eobtotal = 0;
register int count;
const BOOL_DATA *bufptr;
const BOOL_DATA *bufend;
register unsigned int range;
VP8_BD_VALUE value;
const int *scan;//
register unsigned int shift;
UINT32 split;
VP8_BD_VALUE bigsplit;
INT16 *qcoeff_ptr;
const vp8_prob *coef_probs;//
int type;
int stop;
INT16 val, bits_count;
INT16 c;
INT16 v;
const vp8_prob *Prob;//
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;
int segment_id = x->mode_info_context->mbmi.segment_id;
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
type = 3;
i = 0;
stop = 16;
scan = vp8_default_zig_zag1d_8x8;
qcoeff_ptr = &x->qcoeff[0];
if (x->mode_info_context->mbmi.mode != B_PRED && x->mode_info_context->mbmi.mode != SPLITMV)
{
i = 24;
stop = 24;
type = 1;
qcoeff_ptr += 24*16;
eobtotal -= 4;
scan = vp8_default_zig_zag1d;
}
bufend = bc->user_buffer_end;
bufptr = bc->user_buffer;
value = bc->value;
count = bc->count;
range = bc->range;
coef_probs = oc->fc.coef_probs_8x8 [type] [ 0 ] [0];
BLOCK_LOOP_8x8:
Reworked context conversion between 8x8 and 4x4 The commit rationized and simplified the entropy context conversion betwen MB using 8x8 transform and MB using 4x4 transform. The old version had a number of weirdness in how 4x4 transform MB's context is used for 8x8 blocks other than the first 8x8 within a MB. Test showed the change has a gain ~.1% for avg psnr, glb psnr and ssim on the limited HD set. Change-Id: I774536c416baa6845aa741f956d8a69fa40e5d47
2012-01-21 00:30:31 +01:00
a = A + vp8_block2above_8x8[i];
l = L + vp8_block2left_8x8[i];
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
c = (INT16)(!type);
// Dest = ((A)!=0) + ((B)!=0);
if(i==24)
{
VP8_COMBINEENTROPYCONTEXTS(v, *a, *l);
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
if ( segfeature_active( x, segment_id, SEG_LVL_EOB ) )
{
seg_eob = get_segdata( x, segment_id, SEG_LVL_EOB );
}
else
fixed an issue related to 2nd order size due to merge artifacts. For 8x8 transformed macroblock, the 2nd order transform is a 2x2 haar transform, here there is only 4 coefficients total. A previous merge changed these to 64, causing crashes when encoding with 8x8 transform enabled. (i.e. when input video image size > 640x360 ) This commit reverts them back to 4 and fixes the crashes. Change-Id: I3290b81f8c0d32c7efec03093a61ea57736c0550
2012-02-10 20:49:22 +01:00
seg_eob = 4;
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
}
else
{
Reworked context conversion between 8x8 and 4x4 The commit rationized and simplified the entropy context conversion betwen MB using 8x8 transform and MB using 4x4 transform. The old version had a number of weirdness in how 4x4 transform MB's context is used for 8x8 blocks other than the first 8x8 within a MB. Test showed the change has a gain ~.1% for avg psnr, glb psnr and ssim on the limited HD set. Change-Id: I774536c416baa6845aa741f956d8a69fa40e5d47
2012-01-21 00:30:31 +01:00
VP8_COMBINEENTROPYCONTEXTS(v, *a, *l);
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
if ( segfeature_active( x, segment_id, SEG_LVL_EOB ) )
{
Fixed typo on #define name SE_LVL_EOB => SEG_LVL_EOB Change-Id: I6d10169878a709bc9b82f03e5d5903c629fa7679
2012-02-13 13:06:18 +01:00
seg_eob = get_segdata( x, segment_id, SEG_LVL_EOB );
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
}
else
seg_eob = 64;
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
}
Prob = coef_probs;
Prob += v * ENTROPY_NODES;
DO_WHILE_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
if ( c == seg_eob )
goto BLOCK_FINISHED_8x8;
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(i==24)
Prob += coef_bands_x[c];
else
Prob += coef_bands_x_8x8[c];
DECODE_AND_BRANCH_IF_ZERO(Prob[EOB_CONTEXT_NODE], BLOCK_FINISHED_8x8);
CHECK_0_8x8_:
if (i==24)
{
DECODE_AND_LOOP_IF_ZERO_8x8_2(Prob[ZERO_CONTEXT_NODE], CHECK_0_8x8_);
}
else
{
DECODE_AND_LOOP_IF_ZERO_8X8(Prob[ZERO_CONTEXT_NODE], CHECK_0_8x8_);
}
DECODE_AND_BRANCH_IF_ZERO(Prob[ONE_CONTEXT_NODE], ONE_CONTEXT_NODE_0_8x8_);
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
DECODE_AND_BRANCH_IF_ZERO(Prob[LOW_VAL_CONTEXT_NODE],
LOW_VAL_CONTEXT_NODE_0_8x8_);
DECODE_AND_BRANCH_IF_ZERO(Prob[HIGH_LOW_CONTEXT_NODE],
HIGH_LOW_CONTEXT_NODE_0_8x8_);
DECODE_AND_BRANCH_IF_ZERO(Prob[CAT_THREEFOUR_CONTEXT_NODE],
CAT_THREEFOUR_CONTEXT_NODE_0_8x8_);
DECODE_AND_BRANCH_IF_ZERO(Prob[CAT_FIVE_CONTEXT_NODE],
CAT_FIVE_CONTEXT_NODE_0_8x8_);
Experimental: segfeature added. New setting added to configure script
2011-09-30 17:08:37 +02:00
val = CAT6_MIN_VAL;
Merge Extended Q experiment. Merge the extended Q experiment as indicated by the Change-Id: I02d9e654fff9998cc7e9e2f1f5cd838dad8fb431
2012-02-09 17:44:46 +01:00
bits_count = 12;
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
do
{
Experimental: segfeature added. New setting added to configure script
2011-09-30 17:08:37 +02:00
DECODE_EXTRABIT_AND_ADJUST_VAL(cat6_prob[bits_count], bits_count);
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
bits_count -- ;
}
while (bits_count >= 0);
if(i==24)
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8_2(val);
}
else
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8(val);
}
CAT_FIVE_CONTEXT_NODE_0_8x8_:
Experimental: segfeature added. New setting added to configure script
2011-09-30 17:08:37 +02:00
val = CAT5_MIN_VAL;
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT5_PROB4, 4);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT5_PROB3, 3);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT5_PROB2, 2);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT5_PROB1, 1);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT5_PROB0, 0);
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(i==24)
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8_2(val);
}
else
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8(val);
}
CAT_THREEFOUR_CONTEXT_NODE_0_8x8_:
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
DECODE_AND_BRANCH_IF_ZERO(Prob[CAT_THREE_CONTEXT_NODE],
CAT_THREE_CONTEXT_NODE_0_8x8_);
Experimental: segfeature added. New setting added to configure script
2011-09-30 17:08:37 +02:00
val = CAT4_MIN_VAL;
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT4_PROB3, 3);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT4_PROB2, 2);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT4_PROB1, 1);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT4_PROB0, 0);
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(i==24)
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8_2(val);
}
else
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8(val);
}
CAT_THREE_CONTEXT_NODE_0_8x8_:
Experimental: segfeature added. New setting added to configure script
2011-09-30 17:08:37 +02:00
val = CAT3_MIN_VAL;
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT3_PROB2, 2);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT3_PROB1, 1);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT3_PROB0, 0);
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(i==24)
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8_2(val);
}
else
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8(val);
}
HIGH_LOW_CONTEXT_NODE_0_8x8_:
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
DECODE_AND_BRANCH_IF_ZERO(Prob[CAT_ONE_CONTEXT_NODE],
CAT_ONE_CONTEXT_NODE_0_8x8_);
Experimental: segfeature added. New setting added to configure script
2011-09-30 17:08:37 +02:00
val = CAT2_MIN_VAL;
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT2_PROB1, 1);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT2_PROB0, 0);
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(i==24)
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8_2(val);
}
else
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8(val);
}
CAT_ONE_CONTEXT_NODE_0_8x8_:
Experimental: segfeature added. New setting added to configure script
2011-09-30 17:08:37 +02:00
val = CAT1_MIN_VAL;
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT1_PROB0, 0);
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(i==24)
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8_2(val);
}
else
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8(val);
}
LOW_VAL_CONTEXT_NODE_0_8x8_:
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
DECODE_AND_BRANCH_IF_ZERO(Prob[TWO_CONTEXT_NODE],
TWO_CONTEXT_NODE_0_8x8_);
DECODE_AND_BRANCH_IF_ZERO(Prob[THREE_CONTEXT_NODE],
THREE_CONTEXT_NODE_0_8x8_);
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(i==24)
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8_2(4);
}
else
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8(4);
}
THREE_CONTEXT_NODE_0_8x8_:
if(i==24)
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8_2(3);
}
else
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8(3);
}
TWO_CONTEXT_NODE_0_8x8_:
if(i==24)
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8_2(2);
}
else
{
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT_8x8(2);
}
ONE_CONTEXT_NODE_0_8x8_:
DECODE_AND_APPLYSIGN(1);
Prob = coef_probs + ENTROPY_NODES;
if (i==24)
{
if (c < 3)//15
{
qcoeff_ptr [ scan[c] ] = (INT16) v;
++c;
goto DO_WHILE_8x8;
}
}
else
{
if (c < 63)
{
qcoeff_ptr [ scan[c] ] = (INT16) v;
++c;
goto DO_WHILE_8x8;
}
}
if(i==24)
qcoeff_ptr [ scan[3] ] = (INT16) v;//15
else
qcoeff_ptr [ scan[63] ] = (INT16) v;
BLOCK_FINISHED_8x8:
*a = *l = ((eobs[i] = c) != !type); // any nonzero data?
if (i!=24)
{
Reworked context conversion between 8x8 and 4x4 The commit rationized and simplified the entropy context conversion betwen MB using 8x8 transform and MB using 4x4 transform. The old version had a number of weirdness in how 4x4 transform MB's context is used for 8x8 blocks other than the first 8x8 within a MB. Test showed the change has a gain ~.1% for avg psnr, glb psnr and ssim on the limited HD set. Change-Id: I774536c416baa6845aa741f956d8a69fa40e5d47
2012-01-21 00:30:31 +01:00
*(a + 1) = *a;
*(l + 1) = *l;
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
}
eobtotal += c;
qcoeff_ptr += (i==24 ? 16 : 64);
i+=4;
if (i < stop)
goto BLOCK_LOOP_8x8;
if (i > 24)
{
type = 0;
i = 0;
stop = 16;
coef_probs = oc->fc.coef_probs_8x8 [type] [ 0 ] [0];
qcoeff_ptr -= (24*16 + 16);
scan = vp8_default_zig_zag1d_8x8;
goto BLOCK_LOOP_8x8;
}
if (i == 16)
{
type = 2;
coef_probs = oc->fc.coef_probs_8x8 [type] [ 0 ] [0];
stop = 24;
goto BLOCK_LOOP_8x8;
}
FILL
bc->user_buffer = bufptr;
bc->value = value;
bc->count = count;
bc->range = range;
return eobtotal;
}
Merge t8x8 experiments Change-Id: I8e9b6b154e1a0d0cb42d596366380d69c00ac15f
2012-02-29 02:11:12 +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
int vp8_decode_mb_tokens(VP8D_COMP *dx, MACROBLOCKD *xd)
Initial WebM release
2010-05-18 17:58:33 +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
ENTROPY_CONTEXT *A = (ENTROPY_CONTEXT *)xd->above_context;
ENTROPY_CONTEXT *L = (ENTROPY_CONTEXT *)xd->left_context;
Slightly simplify vp8_decode_mb_tokens. Change-Id: I0058ba7dcfc50a3374b712197639ac337f8726be
2011-04-04 17:47:22 +02:00
const FRAME_CONTEXT * const fc = &dx->common.fc;
Initial WebM release
2010-05-18 17:58:33 +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
BOOL_DECODER *bc = xd->current_bc;
Initial WebM release
2010-05-18 17:58:33 +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
char *eobs = xd->eobs;
Rework idct calling structure. Moving the eob structure allows for a non-struct based function to handle decoding an entire mb of idct/dequant/recon data. This allows for SIMD functions to idct/dequant/recon multiple blocks at once. SSE2 implementation gives 3% gain on Atom. Change-Id: I8a8f3efd546ea4e0535f517d94f347cfb737c9c2
2010-08-20 19:58:19 +02:00
Initial WebM release
2010-05-18 17:58:33 +02:00
ENTROPY_CONTEXT *a;
ENTROPY_CONTEXT *l;
int i;
int eobtotal = 0;
register int count;
const BOOL_DATA *bufptr;
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
const BOOL_DATA *bufend;
Initial WebM release
2010-05-18 17:58:33 +02:00
register unsigned int range;
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
VP8_BD_VALUE value;
Initial WebM release
2010-05-18 17:58:33 +02:00
const int *scan;
register unsigned int shift;
UINT32 split;
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
VP8_BD_VALUE bigsplit;
Initial WebM release
2010-05-18 17:58:33 +02:00
INT16 *qcoeff_ptr;
const vp8_prob *coef_probs;
int type;
int stop;
INT16 val, bits_count;
INT16 c;
INT16 v;
const vp8_prob *Prob;
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 seg_eob = 16;
int segment_id = xd->mode_info_context->mbmi.segment_id;
if ( segfeature_active( xd, segment_id, SEG_LVL_EOB ) )
{
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
seg_eob = get_segdata( xd, segment_id, SEG_LVL_EOB );
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
type = 3;
i = 0;
stop = 16;
Changed above and left context data layout The main reason for the change was to reduce cycles in the token decoder. (~1.5% gain for 32 bit) This layout should be more cache friendly. As a result of this change, the encoder had to be updated. Change-Id: Id5e804169d8889da0378b3a519ac04dabd28c837 Note: dixie uses a similar layout
2010-08-31 16:49:57 +02:00
scan = vp8_default_zig_zag1d;
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
qcoeff_ptr = &xd->qcoeff[0];
if (xd->mode_info_context->mbmi.mode != B_PRED &&
xd->mode_info_context->mbmi.mode != I8X8_PRED &&
xd->mode_info_context->mbmi.mode != SPLITMV)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
i = 24;
stop = 24;
type = 1;
Changed above and left context data layout The main reason for the change was to reduce cycles in the token decoder. (~1.5% gain for 32 bit) This layout should be more cache friendly. As a result of this change, the encoder had to be updated. Change-Id: Id5e804169d8889da0378b3a519ac04dabd28c837 Note: dixie uses a similar layout
2010-08-31 16:49:57 +02:00
qcoeff_ptr += 24*16;
Initial WebM release
2010-05-18 17:58:33 +02:00
eobtotal -= 16;
}
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
bufend = bc->user_buffer_end;
bufptr = bc->user_buffer;
value = bc->value;
Initial WebM release
2010-05-18 17:58:33 +02:00
count = bc->count;
range = bc->range;
Slightly simplify vp8_decode_mb_tokens. Change-Id: I0058ba7dcfc50a3374b712197639ac337f8726be
2011-04-04 17:47:22 +02:00
coef_probs = fc->coef_probs [type] [ 0 ] [0];
Initial WebM release
2010-05-18 17:58:33 +02:00
BLOCK_LOOP:
Changed above and left context data layout The main reason for the change was to reduce cycles in the token decoder. (~1.5% gain for 32 bit) This layout should be more cache friendly. As a result of this change, the encoder had to be updated. Change-Id: Id5e804169d8889da0378b3a519ac04dabd28c837 Note: dixie uses a similar layout
2010-08-31 16:49:57 +02:00
a = A + vp8_block2above[i];
l = L + vp8_block2left[i];
Initial WebM release
2010-05-18 17:58:33 +02:00
c = (INT16)(!type);
Eliminate more warnings. This eliminates a large set of warnings exposed by the Mozilla build system (Use of C++ comments in ISO C90 source, commas at the end of enum lists, a couple incomplete initializers, and signed/unsigned comparisons). It also eliminates many (but not all) of the warnings expose by newer GCC versions and _FORTIFY_SOURCE (e.g., calling fread and fwrite without checking the return values). There are a few spurious warnings left on my system: ../vp8/encoder/encodemb.c:274:9: warning: 'sz' may be used uninitialized in this function gcc seems to be unable to figure out that the value shortcut doesn't change between the two if blocks that test it here. ../vp8/encoder/onyx_if.c:5314:5: warning: comparison of unsigned expression >= 0 is always true ../vp8/encoder/onyx_if.c:5319:5: warning: comparison of unsigned expression >= 0 is always true This is true, so far as it goes, but it's comparing against an enum, and the C standard does not mandate that enums be unsigned, so the checks can't be removed. Change-Id: Iaf689ae3e3d0ddc5ade00faa474debe73b8d3395
2010-10-28 01:04:02 +02:00
/*Dest = ((A)!=0) + ((B)!=0);*/
Changed above and left context data layout The main reason for the change was to reduce cycles in the token decoder. (~1.5% gain for 32 bit) This layout should be more cache friendly. As a result of this change, the encoder had to be updated. Change-Id: Id5e804169d8889da0378b3a519ac04dabd28c837 Note: dixie uses a similar layout
2010-08-31 16:49:57 +02:00
VP8_COMBINEENTROPYCONTEXTS(v, *a, *l);
Initial WebM release
2010-05-18 17:58:33 +02:00
Prob = coef_probs;
Changed above and left context data layout The main reason for the change was to reduce cycles in the token decoder. (~1.5% gain for 32 bit) This layout should be more cache friendly. As a result of this change, the encoder had to be updated. Change-Id: Id5e804169d8889da0378b3a519ac04dabd28c837 Note: dixie uses a similar layout
2010-08-31 16:49:57 +02:00
Prob += v * ENTROPY_NODES;
Initial WebM release
2010-05-18 17:58:33 +02:00
DO_WHILE:
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
if ( c == seg_eob )
goto BLOCK_FINISHED;
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
Prob += coef_bands_x[c];
Initial WebM release
2010-05-18 17:58:33 +02:00
DECODE_AND_BRANCH_IF_ZERO(Prob[EOB_CONTEXT_NODE], BLOCK_FINISHED);
CHECK_0_:
DECODE_AND_LOOP_IF_ZERO(Prob[ZERO_CONTEXT_NODE], CHECK_0_);
DECODE_AND_BRANCH_IF_ZERO(Prob[ONE_CONTEXT_NODE], ONE_CONTEXT_NODE_0_);
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
DECODE_AND_BRANCH_IF_ZERO(Prob[LOW_VAL_CONTEXT_NODE],
LOW_VAL_CONTEXT_NODE_0_);
DECODE_AND_BRANCH_IF_ZERO(Prob[HIGH_LOW_CONTEXT_NODE],
HIGH_LOW_CONTEXT_NODE_0_);
DECODE_AND_BRANCH_IF_ZERO(Prob[CAT_THREEFOUR_CONTEXT_NODE],
CAT_THREEFOUR_CONTEXT_NODE_0_);
DECODE_AND_BRANCH_IF_ZERO(Prob[CAT_FIVE_CONTEXT_NODE],
CAT_FIVE_CONTEXT_NODE_0_);
val = CAT6_MIN_VAL;
Merge Extended Q experiment. Merge the extended Q experiment as indicated by the Change-Id: I02d9e654fff9998cc7e9e2f1f5cd838dad8fb431
2012-02-09 17:44:46 +01:00
bits_count = 12;
Initial WebM release
2010-05-18 17:58:33 +02:00
do
{
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
DECODE_EXTRABIT_AND_ADJUST_VAL(cat6_prob[bits_count], bits_count);
Initial WebM release
2010-05-18 17:58:33 +02:00
bits_count -- ;
}
while (bits_count >= 0);
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT(val);
CAT_FIVE_CONTEXT_NODE_0_:
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
val = CAT5_MIN_VAL;
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT5_PROB4, 4);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT5_PROB3, 3);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT5_PROB2, 2);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT5_PROB1, 1);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT5_PROB0, 0);
Initial WebM release
2010-05-18 17:58:33 +02:00
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT(val);
CAT_THREEFOUR_CONTEXT_NODE_0_:
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
DECODE_AND_BRANCH_IF_ZERO(Prob[CAT_THREE_CONTEXT_NODE],
CAT_THREE_CONTEXT_NODE_0_);
val = CAT4_MIN_VAL;
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT4_PROB3, 3);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT4_PROB2, 2);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT4_PROB1, 1);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT4_PROB0, 0);
Initial WebM release
2010-05-18 17:58:33 +02:00
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT(val);
CAT_THREE_CONTEXT_NODE_0_:
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
val = CAT3_MIN_VAL;
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT3_PROB2, 2);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT3_PROB1, 1);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT3_PROB0, 0);
Initial WebM release
2010-05-18 17:58:33 +02:00
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT(val);
HIGH_LOW_CONTEXT_NODE_0_:
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
DECODE_AND_BRANCH_IF_ZERO(Prob[CAT_ONE_CONTEXT_NODE],
CAT_ONE_CONTEXT_NODE_0_);
Initial WebM release
2010-05-18 17:58:33 +02:00
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
val = CAT2_MIN_VAL;
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT2_PROB1, 1);
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT2_PROB0, 0);
Initial WebM release
2010-05-18 17:58:33 +02:00
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT(val);
CAT_ONE_CONTEXT_NODE_0_:
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
val = CAT1_MIN_VAL;
DECODE_EXTRABIT_AND_ADJUST_VAL(CAT1_PROB0, 0);
Initial WebM release
2010-05-18 17:58:33 +02:00
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT(val);
LOW_VAL_CONTEXT_NODE_0_:
DECODE_AND_BRANCH_IF_ZERO(Prob[TWO_CONTEXT_NODE], TWO_CONTEXT_NODE_0_);
DECODE_AND_BRANCH_IF_ZERO(Prob[THREE_CONTEXT_NODE], THREE_CONTEXT_NODE_0_);
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT(4);
THREE_CONTEXT_NODE_0_:
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT(3);
TWO_CONTEXT_NODE_0_:
DECODE_SIGN_WRITE_COEFF_AND_CHECK_EXIT(2);
ONE_CONTEXT_NODE_0_:
DECODE_AND_APPLYSIGN(1);
Prob = coef_probs + ENTROPY_NODES;
if (c < 15)
{
qcoeff_ptr [ scan[c] ] = (INT16) v;
++c;
goto DO_WHILE;
}
Eliminated TOKENEXTRABITS Noticed small performance gains, depending on material. Change-Id: I334369f6312bc19aa73481fc3f790ab181e11867
2011-07-25 23:11:24 +02:00
qcoeff_ptr [ 15 ] = (INT16) v;
Initial WebM release
2010-05-18 17:58:33 +02:00
BLOCK_FINISHED:
Eliminate more warnings. This eliminates a large set of warnings exposed by the Mozilla build system (Use of C++ comments in ISO C90 source, commas at the end of enum lists, a couple incomplete initializers, and signed/unsigned comparisons). It also eliminates many (but not all) of the warnings expose by newer GCC versions and _FORTIFY_SOURCE (e.g., calling fread and fwrite without checking the return values). There are a few spurious warnings left on my system: ../vp8/encoder/encodemb.c:274:9: warning: 'sz' may be used uninitialized in this function gcc seems to be unable to figure out that the value shortcut doesn't change between the two if blocks that test it here. ../vp8/encoder/onyx_if.c:5314:5: warning: comparison of unsigned expression >= 0 is always true ../vp8/encoder/onyx_if.c:5319:5: warning: comparison of unsigned expression >= 0 is always true This is true, so far as it goes, but it's comparing against an enum, and the C standard does not mandate that enums be unsigned, so the checks can't be removed. Change-Id: Iaf689ae3e3d0ddc5ade00faa474debe73b8d3395
2010-10-28 01:04:02 +02:00
*a = *l = ((eobs[i] = c) != !type); /* any nonzero data? */
Rework idct calling structure. Moving the eob structure allows for a non-struct based function to handle decoding an entire mb of idct/dequant/recon data. This allows for SIMD functions to idct/dequant/recon multiple blocks at once. SSE2 implementation gives 3% gain on Atom. Change-Id: I8a8f3efd546ea4e0535f517d94f347cfb737c9c2
2010-08-20 19:58:19 +02:00
eobtotal += c;
Initial WebM release
2010-05-18 17:58:33 +02:00
qcoeff_ptr += 16;
i++;
if (i < stop)
goto BLOCK_LOOP;
if (i == 25)
{
type = 0;
i = 0;
stop = 16;
Slightly simplify vp8_decode_mb_tokens. Change-Id: I0058ba7dcfc50a3374b712197639ac337f8726be
2011-04-04 17:47:22 +02:00
coef_probs = fc->coef_probs [type] [ 0 ] [0];
Changed above and left context data layout The main reason for the change was to reduce cycles in the token decoder. (~1.5% gain for 32 bit) This layout should be more cache friendly. As a result of this change, the encoder had to be updated. Change-Id: Id5e804169d8889da0378b3a519ac04dabd28c837 Note: dixie uses a similar layout
2010-08-31 16:49:57 +02:00
qcoeff_ptr -= (24*16 + 16);
Initial WebM release
2010-05-18 17:58:33 +02:00
goto BLOCK_LOOP;
}
if (i == 16)
{
type = 2;
Slightly simplify vp8_decode_mb_tokens. Change-Id: I0058ba7dcfc50a3374b712197639ac337f8726be
2011-04-04 17:47:22 +02:00
coef_probs = fc->coef_probs [type] [ 0 ] [0];
Initial WebM release
2010-05-18 17:58:33 +02:00
stop = 24;
goto BLOCK_LOOP;
}
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
FILL
bc->user_buffer = bufptr;
Initial WebM release
2010-05-18 17:58:33 +02:00
bc->value = value;
Change bitreader to use a larger window. Change bitreading functions to use a larger window which is refilled less often. This makes it cheap enough to do bounds checking each time the window is refilled, which avoids the need to copy the input into a large circular buffer. This uses less memory and speeds up the total decode time by 1.6% on an ARM11, 2.8% on a Cortex A8, and 2.2% on x86-32, but less than 1% on x86-64. Inlining vp8dx_bool_decoder_fill() has a big penalty on x86-32, as does moving the refill loop to the front of vp8dx_decode_bool(). However, having the refill loop between computation of the split values and the branch in vp8_decode_mb_tokens() is a big win on ARM (presumably due to memory latency and code size: refilling after normalization duplicates the code in the DECODE_AND_BRANCH_IF_ZERO and DECODE_AND_LOOP_IF_ZERO cases. Unfortunately, refilling at the end of vp8dx_bool_decoder_fill() and at the beginning of each decode step in vp8_decode_mb_tokens() means the latter requires an extra refill at the end. Platform-specific versions could avoid the problem, but would require most of detokenize.c to be duplicated. Change-Id: I16c782a63376f2a15b78f8086d899b987204c1c7
2010-05-05 23:58:19 +02:00
bc->count = count;
Initial WebM release
2010-05-18 17:58:33 +02:00
bc->range = range;
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
Initial WebM release
2010-05-18 17:58:33 +02:00
return eobtotal;
}
Reference in New Issue Copy Permalink