virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-11-26 19:01:44 +02:00
Code Issues Releases Activity
985f34b756
FFmpeg/libavcodec/h264_cabac.c

2419 lines
112 KiB
C
Raw Normal View History

Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
/*
* H.26L/H.264/AVC/JVT/14496-10/... cabac decoding
* Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
*
Replace FFmpeg with Libav in licence headers Signed-off-by: Mans Rullgard <mans@mansr.com>
2011-03-18 19:35:10 +02:00
* This file is part of Libav.
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
*
Replace FFmpeg with Libav in licence headers Signed-off-by: Mans Rullgard <mans@mansr.com>
2011-03-18 19:35:10 +02:00
* Libav is free software; you can redistribute it and/or
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
Replace FFmpeg with Libav in licence headers Signed-off-by: Mans Rullgard <mans@mansr.com>
2011-03-18 19:35:10 +02:00
* Libav is distributed in the hope that it will be useful,
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
Replace FFmpeg with Libav in licence headers Signed-off-by: Mans Rullgard <mans@mansr.com>
2011-03-18 19:35:10 +02:00
* License along with Libav; if not, write to the Free Software
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
Remove explicit filename from Doxygen @file commands. Passing an explicit filename to this command is only necessary if the documentation in the @file block refers to a file different from the one the block resides in. Originally committed as revision 22921 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-04-20 17:45:34 +03:00
* @file
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
* H.264 / AVC / MPEG4 part10 cabac decoding.
* @author Michael Niedermayer <michaelni@gmx.at>
*/
h264: add a parameter to the CABAC macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
#define CABAC(h) 1
Replace cabac checks in inline functions from h264.h with constants. No benchmark because its just replacing variables with litteral constants (so no risk for slowdown outside gcc silliness) and i need sleep. Originally committed as revision 21237 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-16 07:41:33 +02:00
cabac: split cabac.h into declarations and function definitions This fixes standalone compilation of some decoders with --disable-optimizations. cabac.h defines some inline functions that use symbols from cabac.c. Without optimizations these inline functions are not eliminated and linking fails with references to non-existing symbols. Splitting the inline functions off into their own header and only #including it in the places where the inline functions are used allows #including cabac.h from anywhere without ill effects.
2012-01-12 23:56:02 +03:00
#include "config.h"
#include "cabac.h"
#include "cabac_functions.h"
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
#include "internal.h"
#include "avcodec.h"
#include "h264.h"
#include "h264data.h"
#include "h264_mvpred.h"
#include "golomb.h"
#if ARCH_X86
#include "x86/h264_i386.h"
#endif
//#undef NDEBUG
#include <assert.h>
/* Cabac pre state table */
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
static const int8_t cabac_context_init_I[1024][2] =
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
{
/* 0 - 10 */
{ 20, -15 }, { 2, 54 }, { 3, 74 }, { 20, -15 },
{ 2, 54 }, { 3, 74 }, { -28,127 }, { -23, 104 },
{ -6, 53 }, { -1, 54 }, { 7, 51 },
/* 11 - 23 unsused for I */
{ 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
{ 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
{ 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
{ 0, 0 },
/* 24- 39 */
{ 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
{ 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
{ 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
{ 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
/* 40 - 53 */
{ 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
{ 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
{ 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
{ 0, 0 }, { 0, 0 },
/* 54 - 59 */
{ 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
{ 0, 0 }, { 0, 0 },
/* 60 - 69 */
{ 0, 41 }, { 0, 63 }, { 0, 63 }, { 0, 63 },
{ -9, 83 }, { 4, 86 }, { 0, 97 }, { -7, 72 },
{ 13, 41 }, { 3, 62 },
/* 70 -> 87 */
{ 0, 11 }, { 1, 55 }, { 0, 69 }, { -17, 127 },
{ -13, 102 },{ 0, 82 }, { -7, 74 }, { -21, 107 },
{ -27, 127 },{ -31, 127 },{ -24, 127 }, { -18, 95 },
{ -27, 127 },{ -21, 114 },{ -30, 127 }, { -17, 123 },
{ -12, 115 },{ -16, 122 },
/* 88 -> 104 */
{ -11, 115 },{ -12, 63 }, { -2, 68 }, { -15, 84 },
{ -13, 104 },{ -3, 70 }, { -8, 93 }, { -10, 90 },
{ -30, 127 },{ -1, 74 }, { -6, 97 }, { -7, 91 },
{ -20, 127 },{ -4, 56 }, { -5, 82 }, { -7, 76 },
{ -22, 125 },
/* 105 -> 135 */
{ -7, 93 }, { -11, 87 }, { -3, 77 }, { -5, 71 },
{ -4, 63 }, { -4, 68 }, { -12, 84 }, { -7, 62 },
{ -7, 65 }, { 8, 61 }, { 5, 56 }, { -2, 66 },
{ 1, 64 }, { 0, 61 }, { -2, 78 }, { 1, 50 },
{ 7, 52 }, { 10, 35 }, { 0, 44 }, { 11, 38 },
{ 1, 45 }, { 0, 46 }, { 5, 44 }, { 31, 17 },
{ 1, 51 }, { 7, 50 }, { 28, 19 }, { 16, 33 },
{ 14, 62 }, { -13, 108 },{ -15, 100 },
/* 136 -> 165 */
{ -13, 101 },{ -13, 91 }, { -12, 94 }, { -10, 88 },
{ -16, 84 }, { -10, 86 }, { -7, 83 }, { -13, 87 },
{ -19, 94 }, { 1, 70 }, { 0, 72 }, { -5, 74 },
{ 18, 59 }, { -8, 102 }, { -15, 100 }, { 0, 95 },
{ -4, 75 }, { 2, 72 }, { -11, 75 }, { -3, 71 },
{ 15, 46 }, { -13, 69 }, { 0, 62 }, { 0, 65 },
{ 21, 37 }, { -15, 72 }, { 9, 57 }, { 16, 54 },
{ 0, 62 }, { 12, 72 },
/* 166 -> 196 */
{ 24, 0 }, { 15, 9 }, { 8, 25 }, { 13, 18 },
{ 15, 9 }, { 13, 19 }, { 10, 37 }, { 12, 18 },
{ 6, 29 }, { 20, 33 }, { 15, 30 }, { 4, 45 },
{ 1, 58 }, { 0, 62 }, { 7, 61 }, { 12, 38 },
{ 11, 45 }, { 15, 39 }, { 11, 42 }, { 13, 44 },
{ 16, 45 }, { 12, 41 }, { 10, 49 }, { 30, 34 },
{ 18, 42 }, { 10, 55 }, { 17, 51 }, { 17, 46 },
{ 0, 89 }, { 26, -19 }, { 22, -17 },
/* 197 -> 226 */
{ 26, -17 }, { 30, -25 }, { 28, -20 }, { 33, -23 },
{ 37, -27 }, { 33, -23 }, { 40, -28 }, { 38, -17 },
{ 33, -11 }, { 40, -15 }, { 41, -6 }, { 38, 1 },
{ 41, 17 }, { 30, -6 }, { 27, 3 }, { 26, 22 },
{ 37, -16 }, { 35, -4 }, { 38, -8 }, { 38, -3 },
{ 37, 3 }, { 38, 5 }, { 42, 0 }, { 35, 16 },
{ 39, 22 }, { 14, 48 }, { 27, 37 }, { 21, 60 },
{ 12, 68 }, { 2, 97 },
/* 227 -> 251 */
{ -3, 71 }, { -6, 42 }, { -5, 50 }, { -3, 54 },
{ -2, 62 }, { 0, 58 }, { 1, 63 }, { -2, 72 },
{ -1, 74 }, { -9, 91 }, { -5, 67 }, { -5, 27 },
{ -3, 39 }, { -2, 44 }, { 0, 46 }, { -16, 64 },
{ -8, 68 }, { -10, 78 }, { -6, 77 }, { -10, 86 },
{ -12, 92 }, { -15, 55 }, { -10, 60 }, { -6, 62 },
{ -4, 65 },
/* 252 -> 275 */
{ -12, 73 }, { -8, 76 }, { -7, 80 }, { -9, 88 },
{ -17, 110 },{ -11, 97 }, { -20, 84 }, { -11, 79 },
{ -6, 73 }, { -4, 74 }, { -13, 86 }, { -13, 96 },
{ -11, 97 }, { -19, 117 },{ -8, 78 }, { -5, 33 },
{ -4, 48 }, { -2, 53 }, { -3, 62 }, { -13, 71 },
{ -10, 79 }, { -12, 86 }, { -13, 90 }, { -14, 97 },
/* 276 a bit special (not used, bypass is used instead) */
{ 0, 0 },
/* 277 -> 307 */
{ -6, 93 }, { -6, 84 }, { -8, 79 }, { 0, 66 },
{ -1, 71 }, { 0, 62 }, { -2, 60 }, { -2, 59 },
{ -5, 75 }, { -3, 62 }, { -4, 58 }, { -9, 66 },
{ -1, 79 }, { 0, 71 }, { 3, 68 }, { 10, 44 },
{ -7, 62 }, { 15, 36 }, { 14, 40 }, { 16, 27 },
{ 12, 29 }, { 1, 44 }, { 20, 36 }, { 18, 32 },
{ 5, 42 }, { 1, 48 }, { 10, 62 }, { 17, 46 },
{ 9, 64 }, { -12, 104 },{ -11, 97 },
/* 308 -> 337 */
{ -16, 96 }, { -7, 88 }, { -8, 85 }, { -7, 85 },
{ -9, 85 }, { -13, 88 }, { 4, 66 }, { -3, 77 },
{ -3, 76 }, { -6, 76 }, { 10, 58 }, { -1, 76 },
{ -1, 83 }, { -7, 99 }, { -14, 95 }, { 2, 95 },
{ 0, 76 }, { -5, 74 }, { 0, 70 }, { -11, 75 },
{ 1, 68 }, { 0, 65 }, { -14, 73 }, { 3, 62 },
{ 4, 62 }, { -1, 68 }, { -13, 75 }, { 11, 55 },
{ 5, 64 }, { 12, 70 },
/* 338 -> 368 */
{ 15, 6 }, { 6, 19 }, { 7, 16 }, { 12, 14 },
{ 18, 13 }, { 13, 11 }, { 13, 15 }, { 15, 16 },
{ 12, 23 }, { 13, 23 }, { 15, 20 }, { 14, 26 },
{ 14, 44 }, { 17, 40 }, { 17, 47 }, { 24, 17 },
{ 21, 21 }, { 25, 22 }, { 31, 27 }, { 22, 29 },
{ 19, 35 }, { 14, 50 }, { 10, 57 }, { 7, 63 },
{ -2, 77 }, { -4, 82 }, { -3, 94 }, { 9, 69 },
{ -12, 109 },{ 36, -35 }, { 36, -34 },
/* 369 -> 398 */
{ 32, -26 }, { 37, -30 }, { 44, -32 }, { 34, -18 },
{ 34, -15 }, { 40, -15 }, { 33, -7 }, { 35, -5 },
{ 33, 0 }, { 38, 2 }, { 33, 13 }, { 23, 35 },
{ 13, 58 }, { 29, -3 }, { 26, 0 }, { 22, 30 },
{ 31, -7 }, { 35, -15 }, { 34, -3 }, { 34, 3 },
{ 36, -1 }, { 34, 5 }, { 32, 11 }, { 35, 5 },
{ 34, 12 }, { 39, 11 }, { 30, 29 }, { 34, 26 },
{ 29, 39 }, { 19, 66 },
/* 399 -> 435 */
{ 31, 21 }, { 31, 31 }, { 25, 50 },
{ -17, 120 }, { -20, 112 }, { -18, 114 }, { -11, 85 },
{ -15, 92 }, { -14, 89 }, { -26, 71 }, { -15, 81 },
{ -14, 80 }, { 0, 68 }, { -14, 70 }, { -24, 56 },
{ -23, 68 }, { -24, 50 }, { -11, 74 }, { 23, -13 },
{ 26, -13 }, { 40, -15 }, { 49, -14 }, { 44, 3 },
{ 45, 6 }, { 44, 34 }, { 33, 54 }, { 19, 82 },
{ -3, 75 }, { -1, 23 }, { 1, 34 }, { 1, 43 },
{ 0, 54 }, { -2, 55 }, { 0, 61 }, { 1, 64 },
{ 0, 68 }, { -9, 92 },
/* 436 -> 459 */
{ -14, 106 }, { -13, 97 }, { -15, 90 }, { -12, 90 },
{ -18, 88 }, { -10, 73 }, { -9, 79 }, { -14, 86 },
{ -10, 73 }, { -10, 70 }, { -10, 69 }, { -5, 66 },
{ -9, 64 }, { -5, 58 }, { 2, 59 }, { 21, -10 },
{ 24, -11 }, { 28, -8 }, { 28, -1 }, { 29, 3 },
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
{ 29, 9 }, { 35, 20 }, { 29, 36 }, { 14, 67 },
/* 460 -> 1024 */
{ -17, 123 }, { -12, 115 }, { -16, 122 }, { -11, 115 },
{ -12, 63 }, { -2, 68 }, { -15, 84 }, { -13, 104 },
{ -3, 70 }, { -8, 93 }, { -10, 90 }, { -30, 127 },
{ -17, 123 }, { -12, 115 }, { -16, 122 }, { -11, 115 },
{ -12, 63 }, { -2, 68 }, { -15, 84 }, { -13, 104 },
{ -3, 70 }, { -8, 93 }, { -10, 90 }, { -30, 127 },
{ -7, 93 }, { -11, 87 }, { -3, 77 }, { -5, 71 },
{ -4, 63 }, { -4, 68 }, { -12, 84 }, { -7, 62 },
{ -7, 65 }, { 8, 61 }, { 5, 56 }, { -2, 66 },
{ 1, 64 }, { 0, 61 }, { -2, 78 }, { 1, 50 },
{ 7, 52 }, { 10, 35 }, { 0, 44 }, { 11, 38 },
{ 1, 45 }, { 0, 46 }, { 5, 44 }, { 31, 17 },
{ 1, 51 }, { 7, 50 }, { 28, 19 }, { 16, 33 },
{ 14, 62 }, { -13, 108 }, { -15, 100 }, { -13, 101 },
{ -13, 91 }, { -12, 94 }, { -10, 88 }, { -16, 84 },
{ -10, 86 }, { -7, 83 }, { -13, 87 }, { -19, 94 },
{ 1, 70 }, { 0, 72 }, { -5, 74 }, { 18, 59 },
{ -7, 93 }, { -11, 87 }, { -3, 77 }, { -5, 71 },
{ -4, 63 }, { -4, 68 }, { -12, 84 }, { -7, 62 },
{ -7, 65 }, { 8, 61 }, { 5, 56 }, { -2, 66 },
{ 1, 64 }, { 0, 61 }, { -2, 78 }, { 1, 50 },
{ 7, 52 }, { 10, 35 }, { 0, 44 }, { 11, 38 },
{ 1, 45 }, { 0, 46 }, { 5, 44 }, { 31, 17 },
{ 1, 51 }, { 7, 50 }, { 28, 19 }, { 16, 33 },
{ 14, 62 }, { -13, 108 }, { -15, 100 }, { -13, 101 },
{ -13, 91 }, { -12, 94 }, { -10, 88 }, { -16, 84 },
{ -10, 86 }, { -7, 83 }, { -13, 87 }, { -19, 94 },
{ 1, 70 }, { 0, 72 }, { -5, 74 }, { 18, 59 },
{ 24, 0 }, { 15, 9 }, { 8, 25 }, { 13, 18 },
{ 15, 9 }, { 13, 19 }, { 10, 37 }, { 12, 18 },
{ 6, 29 }, { 20, 33 }, { 15, 30 }, { 4, 45 },
{ 1, 58 }, { 0, 62 }, { 7, 61 }, { 12, 38 },
{ 11, 45 }, { 15, 39 }, { 11, 42 }, { 13, 44 },
{ 16, 45 }, { 12, 41 }, { 10, 49 }, { 30, 34 },
{ 18, 42 }, { 10, 55 }, { 17, 51 }, { 17, 46 },
{ 0, 89 }, { 26, -19 }, { 22, -17 }, { 26, -17 },
{ 30, -25 }, { 28, -20 }, { 33, -23 }, { 37, -27 },
{ 33, -23 }, { 40, -28 }, { 38, -17 }, { 33, -11 },
{ 40, -15 }, { 41, -6 }, { 38, 1 }, { 41, 17 },
{ 24, 0 }, { 15, 9 }, { 8, 25 }, { 13, 18 },
{ 15, 9 }, { 13, 19 }, { 10, 37 }, { 12, 18 },
{ 6, 29 }, { 20, 33 }, { 15, 30 }, { 4, 45 },
{ 1, 58 }, { 0, 62 }, { 7, 61 }, { 12, 38 },
{ 11, 45 }, { 15, 39 }, { 11, 42 }, { 13, 44 },
{ 16, 45 }, { 12, 41 }, { 10, 49 }, { 30, 34 },
{ 18, 42 }, { 10, 55 }, { 17, 51 }, { 17, 46 },
{ 0, 89 }, { 26, -19 }, { 22, -17 }, { 26, -17 },
{ 30, -25 }, { 28, -20 }, { 33, -23 }, { 37, -27 },
{ 33, -23 }, { 40, -28 }, { 38, -17 }, { 33, -11 },
{ 40, -15 }, { 41, -6 }, { 38, 1 }, { 41, 17 },
{ -17, 120 }, { -20, 112 }, { -18, 114 }, { -11, 85 },
{ -15, 92 }, { -14, 89 }, { -26, 71 }, { -15, 81 },
{ -14, 80 }, { 0, 68 }, { -14, 70 }, { -24, 56 },
{ -23, 68 }, { -24, 50 }, { -11, 74 }, { -14, 106 },
{ -13, 97 }, { -15, 90 }, { -12, 90 }, { -18, 88 },
{ -10, 73 }, { -9, 79 }, { -14, 86 }, { -10, 73 },
{ -10, 70 }, { -10, 69 }, { -5, 66 }, { -9, 64 },
{ -5, 58 }, { 2, 59 }, { 23, -13 }, { 26, -13 },
{ 40, -15 }, { 49, -14 }, { 44, 3 }, { 45, 6 },
{ 44, 34 }, { 33, 54 }, { 19, 82 }, { 21, -10 },
{ 24, -11 }, { 28, -8 }, { 28, -1 }, { 29, 3 },
{ 29, 9 }, { 35, 20 }, { 29, 36 }, { 14, 67 },
{ -3, 75 }, { -1, 23 }, { 1, 34 }, { 1, 43 },
{ 0, 54 }, { -2, 55 }, { 0, 61 }, { 1, 64 },
{ 0, 68 }, { -9, 92 }, { -17, 120 }, { -20, 112 },
{ -18, 114 }, { -11, 85 }, { -15, 92 }, { -14, 89 },
{ -26, 71 }, { -15, 81 }, { -14, 80 }, { 0, 68 },
{ -14, 70 }, { -24, 56 }, { -23, 68 }, { -24, 50 },
{ -11, 74 }, { -14, 106 }, { -13, 97 }, { -15, 90 },
{ -12, 90 }, { -18, 88 }, { -10, 73 }, { -9, 79 },
{ -14, 86 }, { -10, 73 }, { -10, 70 }, { -10, 69 },
{ -5, 66 }, { -9, 64 }, { -5, 58 }, { 2, 59 },
{ 23, -13 }, { 26, -13 }, { 40, -15 }, { 49, -14 },
{ 44, 3 }, { 45, 6 }, { 44, 34 }, { 33, 54 },
{ 19, 82 }, { 21, -10 }, { 24, -11 }, { 28, -8 },
{ 28, -1 }, { 29, 3 }, { 29, 9 }, { 35, 20 },
{ 29, 36 }, { 14, 67 }, { -3, 75 }, { -1, 23 },
{ 1, 34 }, { 1, 43 }, { 0, 54 }, { -2, 55 },
{ 0, 61 }, { 1, 64 }, { 0, 68 }, { -9, 92 },
{ -6, 93 }, { -6, 84 }, { -8, 79 }, { 0, 66 },
{ -1, 71 }, { 0, 62 }, { -2, 60 }, { -2, 59 },
{ -5, 75 }, { -3, 62 }, { -4, 58 }, { -9, 66 },
{ -1, 79 }, { 0, 71 }, { 3, 68 }, { 10, 44 },
{ -7, 62 }, { 15, 36 }, { 14, 40 }, { 16, 27 },
{ 12, 29 }, { 1, 44 }, { 20, 36 }, { 18, 32 },
{ 5, 42 }, { 1, 48 }, { 10, 62 }, { 17, 46 },
{ 9, 64 }, { -12, 104 }, { -11, 97 }, { -16, 96 },
{ -7, 88 }, { -8, 85 }, { -7, 85 }, { -9, 85 },
{ -13, 88 }, { 4, 66 }, { -3, 77 }, { -3, 76 },
{ -6, 76 }, { 10, 58 }, { -1, 76 }, { -1, 83 },
{ -6, 93 }, { -6, 84 }, { -8, 79 }, { 0, 66 },
{ -1, 71 }, { 0, 62 }, { -2, 60 }, { -2, 59 },
{ -5, 75 }, { -3, 62 }, { -4, 58 }, { -9, 66 },
{ -1, 79 }, { 0, 71 }, { 3, 68 }, { 10, 44 },
{ -7, 62 }, { 15, 36 }, { 14, 40 }, { 16, 27 },
{ 12, 29 }, { 1, 44 }, { 20, 36 }, { 18, 32 },
{ 5, 42 }, { 1, 48 }, { 10, 62 }, { 17, 46 },
{ 9, 64 }, { -12, 104 }, { -11, 97 }, { -16, 96 },
{ -7, 88 }, { -8, 85 }, { -7, 85 }, { -9, 85 },
{ -13, 88 }, { 4, 66 }, { -3, 77 }, { -3, 76 },
{ -6, 76 }, { 10, 58 }, { -1, 76 }, { -1, 83 },
{ 15, 6 }, { 6, 19 }, { 7, 16 }, { 12, 14 },
{ 18, 13 }, { 13, 11 }, { 13, 15 }, { 15, 16 },
{ 12, 23 }, { 13, 23 }, { 15, 20 }, { 14, 26 },
{ 14, 44 }, { 17, 40 }, { 17, 47 }, { 24, 17 },
{ 21, 21 }, { 25, 22 }, { 31, 27 }, { 22, 29 },
{ 19, 35 }, { 14, 50 }, { 10, 57 }, { 7, 63 },
{ -2, 77 }, { -4, 82 }, { -3, 94 }, { 9, 69 },
{ -12, 109 }, { 36, -35 }, { 36, -34 }, { 32, -26 },
{ 37, -30 }, { 44, -32 }, { 34, -18 }, { 34, -15 },
{ 40, -15 }, { 33, -7 }, { 35, -5 }, { 33, 0 },
{ 38, 2 }, { 33, 13 }, { 23, 35 }, { 13, 58 },
{ 15, 6 }, { 6, 19 }, { 7, 16 }, { 12, 14 },
{ 18, 13 }, { 13, 11 }, { 13, 15 }, { 15, 16 },
{ 12, 23 }, { 13, 23 }, { 15, 20 }, { 14, 26 },
{ 14, 44 }, { 17, 40 }, { 17, 47 }, { 24, 17 },
{ 21, 21 }, { 25, 22 }, { 31, 27 }, { 22, 29 },
{ 19, 35 }, { 14, 50 }, { 10, 57 }, { 7, 63 },
{ -2, 77 }, { -4, 82 }, { -3, 94 }, { 9, 69 },
{ -12, 109 }, { 36, -35 }, { 36, -34 }, { 32, -26 },
{ 37, -30 }, { 44, -32 }, { 34, -18 }, { 34, -15 },
{ 40, -15 }, { 33, -7 }, { 35, -5 }, { 33, 0 },
{ 38, 2 }, { 33, 13 }, { 23, 35 }, { 13, 58 },
{ -3, 71 }, { -6, 42 }, { -5, 50 }, { -3, 54 },
{ -2, 62 }, { 0, 58 }, { 1, 63 }, { -2, 72 },
{ -1, 74 }, { -9, 91 }, { -5, 67 }, { -5, 27 },
{ -3, 39 }, { -2, 44 }, { 0, 46 }, { -16, 64 },
{ -8, 68 }, { -10, 78 }, { -6, 77 }, { -10, 86 },
{ -12, 92 }, { -15, 55 }, { -10, 60 }, { -6, 62 },
{ -4, 65 }, { -12, 73 }, { -8, 76 }, { -7, 80 },
{ -9, 88 }, { -17, 110 }, { -3, 71 }, { -6, 42 },
{ -5, 50 }, { -3, 54 }, { -2, 62 }, { 0, 58 },
{ 1, 63 }, { -2, 72 }, { -1, 74 }, { -9, 91 },
{ -5, 67 }, { -5, 27 }, { -3, 39 }, { -2, 44 },
{ 0, 46 }, { -16, 64 }, { -8, 68 }, { -10, 78 },
{ -6, 77 }, { -10, 86 }, { -12, 92 }, { -15, 55 },
{ -10, 60 }, { -6, 62 }, { -4, 65 }, { -12, 73 },
{ -8, 76 }, { -7, 80 }, { -9, 88 }, { -17, 110 },
{ -3, 70 }, { -8, 93 }, { -10, 90 }, { -30, 127 },
{ -3, 70 }, { -8, 93 }, { -10, 90 }, { -30, 127 },
{ -3, 70 }, { -8, 93 }, { -10, 90 }, { -30, 127 }
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
};
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
static const int8_t cabac_context_init_PB[3][1024][2] =
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
{
/* i_cabac_init_idc == 0 */
{
/* 0 - 10 */
{ 20, -15 }, { 2, 54 }, { 3, 74 }, { 20, -15 },
{ 2, 54 }, { 3, 74 }, { -28, 127 }, { -23, 104 },
{ -6, 53 }, { -1, 54 }, { 7, 51 },
/* 11 - 23 */
{ 23, 33 }, { 23, 2 }, { 21, 0 }, { 1, 9 },
{ 0, 49 }, { -37, 118 }, { 5, 57 }, { -13, 78 },
{ -11, 65 }, { 1, 62 }, { 12, 49 }, { -4, 73 },
{ 17, 50 },
/* 24 - 39 */
{ 18, 64 }, { 9, 43 }, { 29, 0 }, { 26, 67 },
{ 16, 90 }, { 9, 104 }, { -46, 127 }, { -20, 104 },
{ 1, 67 }, { -13, 78 }, { -11, 65 }, { 1, 62 },
{ -6, 86 }, { -17, 95 }, { -6, 61 }, { 9, 45 },
/* 40 - 53 */
{ -3, 69 }, { -6, 81 }, { -11, 96 }, { 6, 55 },
{ 7, 67 }, { -5, 86 }, { 2, 88 }, { 0, 58 },
{ -3, 76 }, { -10, 94 }, { 5, 54 }, { 4, 69 },
{ -3, 81 }, { 0, 88 },
/* 54 - 59 */
{ -7, 67 }, { -5, 74 }, { -4, 74 }, { -5, 80 },
{ -7, 72 }, { 1, 58 },
/* 60 - 69 */
{ 0, 41 }, { 0, 63 }, { 0, 63 }, { 0, 63 },
{ -9, 83 }, { 4, 86 }, { 0, 97 }, { -7, 72 },
{ 13, 41 }, { 3, 62 },
/* 70 - 87 */
{ 0, 45 }, { -4, 78 }, { -3, 96 }, { -27, 126 },
{ -28, 98 }, { -25, 101 }, { -23, 67 }, { -28, 82 },
{ -20, 94 }, { -16, 83 }, { -22, 110 }, { -21, 91 },
{ -18, 102 }, { -13, 93 }, { -29, 127 }, { -7, 92 },
{ -5, 89 }, { -7, 96 }, { -13, 108 }, { -3, 46 },
{ -1, 65 }, { -1, 57 }, { -9, 93 }, { -3, 74 },
{ -9, 92 }, { -8, 87 }, { -23, 126 }, { 5, 54 },
{ 6, 60 }, { 6, 59 }, { 6, 69 }, { -1, 48 },
{ 0, 68 }, { -4, 69 }, { -8, 88 },
/* 105 -> 165 */
{ -2, 85 }, { -6, 78 }, { -1, 75 }, { -7, 77 },
{ 2, 54 }, { 5, 50 }, { -3, 68 }, { 1, 50 },
{ 6, 42 }, { -4, 81 }, { 1, 63 }, { -4, 70 },
{ 0, 67 }, { 2, 57 }, { -2, 76 }, { 11, 35 },
{ 4, 64 }, { 1, 61 }, { 11, 35 }, { 18, 25 },
{ 12, 24 }, { 13, 29 }, { 13, 36 }, { -10, 93 },
{ -7, 73 }, { -2, 73 }, { 13, 46 }, { 9, 49 },
{ -7, 100 }, { 9, 53 }, { 2, 53 }, { 5, 53 },
{ -2, 61 }, { 0, 56 }, { 0, 56 }, { -13, 63 },
{ -5, 60 }, { -1, 62 }, { 4, 57 }, { -6, 69 },
{ 4, 57 }, { 14, 39 }, { 4, 51 }, { 13, 68 },
{ 3, 64 }, { 1, 61 }, { 9, 63 }, { 7, 50 },
{ 16, 39 }, { 5, 44 }, { 4, 52 }, { 11, 48 },
{ -5, 60 }, { -1, 59 }, { 0, 59 }, { 22, 33 },
{ 5, 44 }, { 14, 43 }, { -1, 78 }, { 0, 60 },
{ 9, 69 },
/* 166 - 226 */
{ 11, 28 }, { 2, 40 }, { 3, 44 }, { 0, 49 },
{ 0, 46 }, { 2, 44 }, { 2, 51 }, { 0, 47 },
{ 4, 39 }, { 2, 62 }, { 6, 46 }, { 0, 54 },
{ 3, 54 }, { 2, 58 }, { 4, 63 }, { 6, 51 },
{ 6, 57 }, { 7, 53 }, { 6, 52 }, { 6, 55 },
{ 11, 45 }, { 14, 36 }, { 8, 53 }, { -1, 82 },
{ 7, 55 }, { -3, 78 }, { 15, 46 }, { 22, 31 },
{ -1, 84 }, { 25, 7 }, { 30, -7 }, { 28, 3 },
{ 28, 4 }, { 32, 0 }, { 34, -1 }, { 30, 6 },
{ 30, 6 }, { 32, 9 }, { 31, 19 }, { 26, 27 },
{ 26, 30 }, { 37, 20 }, { 28, 34 }, { 17, 70 },
{ 1, 67 }, { 5, 59 }, { 9, 67 }, { 16, 30 },
{ 18, 32 }, { 18, 35 }, { 22, 29 }, { 24, 31 },
{ 23, 38 }, { 18, 43 }, { 20, 41 }, { 11, 63 },
{ 9, 59 }, { 9, 64 }, { -1, 94 }, { -2, 89 },
{ -9, 108 },
/* 227 - 275 */
{ -6, 76 }, { -2, 44 }, { 0, 45 }, { 0, 52 },
{ -3, 64 }, { -2, 59 }, { -4, 70 }, { -4, 75 },
{ -8, 82 }, { -17, 102 }, { -9, 77 }, { 3, 24 },
{ 0, 42 }, { 0, 48 }, { 0, 55 }, { -6, 59 },
{ -7, 71 }, { -12, 83 }, { -11, 87 }, { -30, 119 },
{ 1, 58 }, { -3, 29 }, { -1, 36 }, { 1, 38 },
{ 2, 43 }, { -6, 55 }, { 0, 58 }, { 0, 64 },
{ -3, 74 }, { -10, 90 }, { 0, 70 }, { -4, 29 },
{ 5, 31 }, { 7, 42 }, { 1, 59 }, { -2, 58 },
{ -3, 72 }, { -3, 81 }, { -11, 97 }, { 0, 58 },
{ 8, 5 }, { 10, 14 }, { 14, 18 }, { 13, 27 },
{ 2, 40 }, { 0, 58 }, { -3, 70 }, { -6, 79 },
{ -8, 85 },
/* 276 a bit special (not used, bypass is used instead) */
{ 0, 0 },
/* 277 - 337 */
{ -13, 106 }, { -16, 106 }, { -10, 87 }, { -21, 114 },
{ -18, 110 }, { -14, 98 }, { -22, 110 }, { -21, 106 },
{ -18, 103 }, { -21, 107 }, { -23, 108 }, { -26, 112 },
{ -10, 96 }, { -12, 95 }, { -5, 91 }, { -9, 93 },
{ -22, 94 }, { -5, 86 }, { 9, 67 }, { -4, 80 },
{ -10, 85 }, { -1, 70 }, { 7, 60 }, { 9, 58 },
{ 5, 61 }, { 12, 50 }, { 15, 50 }, { 18, 49 },
{ 17, 54 }, { 10, 41 }, { 7, 46 }, { -1, 51 },
{ 7, 49 }, { 8, 52 }, { 9, 41 }, { 6, 47 },
{ 2, 55 }, { 13, 41 }, { 10, 44 }, { 6, 50 },
{ 5, 53 }, { 13, 49 }, { 4, 63 }, { 6, 64 },
{ -2, 69 }, { -2, 59 }, { 6, 70 }, { 10, 44 },
{ 9, 31 }, { 12, 43 }, { 3, 53 }, { 14, 34 },
{ 10, 38 }, { -3, 52 }, { 13, 40 }, { 17, 32 },
{ 7, 44 }, { 7, 38 }, { 13, 50 }, { 10, 57 },
{ 26, 43 },
/* 338 - 398 */
{ 14, 11 }, { 11, 14 }, { 9, 11 }, { 18, 11 },
{ 21, 9 }, { 23, -2 }, { 32, -15 }, { 32, -15 },
{ 34, -21 }, { 39, -23 }, { 42, -33 }, { 41, -31 },
{ 46, -28 }, { 38, -12 }, { 21, 29 }, { 45, -24 },
{ 53, -45 }, { 48, -26 }, { 65, -43 }, { 43, -19 },
{ 39, -10 }, { 30, 9 }, { 18, 26 }, { 20, 27 },
{ 0, 57 }, { -14, 82 }, { -5, 75 }, { -19, 97 },
{ -35, 125 }, { 27, 0 }, { 28, 0 }, { 31, -4 },
{ 27, 6 }, { 34, 8 }, { 30, 10 }, { 24, 22 },
{ 33, 19 }, { 22, 32 }, { 26, 31 }, { 21, 41 },
{ 26, 44 }, { 23, 47 }, { 16, 65 }, { 14, 71 },
{ 8, 60 }, { 6, 63 }, { 17, 65 }, { 21, 24 },
{ 23, 20 }, { 26, 23 }, { 27, 32 }, { 28, 23 },
{ 28, 24 }, { 23, 40 }, { 24, 32 }, { 28, 29 },
{ 23, 42 }, { 19, 57 }, { 22, 53 }, { 22, 61 },
{ 11, 86 },
/* 399 - 435 */
{ 12, 40 }, { 11, 51 }, { 14, 59 },
{ -4, 79 }, { -7, 71 }, { -5, 69 }, { -9, 70 },
{ -8, 66 }, { -10, 68 }, { -19, 73 }, { -12, 69 },
{ -16, 70 }, { -15, 67 }, { -20, 62 }, { -19, 70 },
{ -16, 66 }, { -22, 65 }, { -20, 63 }, { 9, -2 },
{ 26, -9 }, { 33, -9 }, { 39, -7 }, { 41, -2 },
{ 45, 3 }, { 49, 9 }, { 45, 27 }, { 36, 59 },
{ -6, 66 }, { -7, 35 }, { -7, 42 }, { -8, 45 },
{ -5, 48 }, { -12, 56 }, { -6, 60 }, { -5, 62 },
{ -8, 66 }, { -8, 76 },
/* 436 - 459 */
{ -5, 85 }, { -6, 81 }, { -10, 77 }, { -7, 81 },
{ -17, 80 }, { -18, 73 }, { -4, 74 }, { -10, 83 },
{ -9, 71 }, { -9, 67 }, { -1, 61 }, { -8, 66 },
{ -14, 66 }, { 0, 59 }, { 2, 59 }, { 21, -13 },
{ 33, -14 }, { 39, -7 }, { 46, -2 }, { 51, 2 },
{ 60, 6 }, { 61, 17 }, { 55, 34 }, { 42, 62 },
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
/* 460 - 1024 */
{ -7, 92 }, { -5, 89 }, { -7, 96 }, { -13, 108 },
{ -3, 46 }, { -1, 65 }, { -1, 57 }, { -9, 93 },
{ -3, 74 }, { -9, 92 }, { -8, 87 }, { -23, 126 },
{ -7, 92 }, { -5, 89 }, { -7, 96 }, { -13, 108 },
{ -3, 46 }, { -1, 65 }, { -1, 57 }, { -9, 93 },
{ -3, 74 }, { -9, 92 }, { -8, 87 }, { -23, 126 },
{ -2, 85 }, { -6, 78 }, { -1, 75 }, { -7, 77 },
{ 2, 54 }, { 5, 50 }, { -3, 68 }, { 1, 50 },
{ 6, 42 }, { -4, 81 }, { 1, 63 }, { -4, 70 },
{ 0, 67 }, { 2, 57 }, { -2, 76 }, { 11, 35 },
{ 4, 64 }, { 1, 61 }, { 11, 35 }, { 18, 25 },
{ 12, 24 }, { 13, 29 }, { 13, 36 }, { -10, 93 },
{ -7, 73 }, { -2, 73 }, { 13, 46 }, { 9, 49 },
{ -7, 100 }, { 9, 53 }, { 2, 53 }, { 5, 53 },
{ -2, 61 }, { 0, 56 }, { 0, 56 }, { -13, 63 },
{ -5, 60 }, { -1, 62 }, { 4, 57 }, { -6, 69 },
{ 4, 57 }, { 14, 39 }, { 4, 51 }, { 13, 68 },
{ -2, 85 }, { -6, 78 }, { -1, 75 }, { -7, 77 },
{ 2, 54 }, { 5, 50 }, { -3, 68 }, { 1, 50 },
{ 6, 42 }, { -4, 81 }, { 1, 63 }, { -4, 70 },
{ 0, 67 }, { 2, 57 }, { -2, 76 }, { 11, 35 },
{ 4, 64 }, { 1, 61 }, { 11, 35 }, { 18, 25 },
{ 12, 24 }, { 13, 29 }, { 13, 36 }, { -10, 93 },
{ -7, 73 }, { -2, 73 }, { 13, 46 }, { 9, 49 },
{ -7, 100 }, { 9, 53 }, { 2, 53 }, { 5, 53 },
{ -2, 61 }, { 0, 56 }, { 0, 56 }, { -13, 63 },
{ -5, 60 }, { -1, 62 }, { 4, 57 }, { -6, 69 },
{ 4, 57 }, { 14, 39 }, { 4, 51 }, { 13, 68 },
{ 11, 28 }, { 2, 40 }, { 3, 44 }, { 0, 49 },
{ 0, 46 }, { 2, 44 }, { 2, 51 }, { 0, 47 },
{ 4, 39 }, { 2, 62 }, { 6, 46 }, { 0, 54 },
{ 3, 54 }, { 2, 58 }, { 4, 63 }, { 6, 51 },
{ 6, 57 }, { 7, 53 }, { 6, 52 }, { 6, 55 },
{ 11, 45 }, { 14, 36 }, { 8, 53 }, { -1, 82 },
{ 7, 55 }, { -3, 78 }, { 15, 46 }, { 22, 31 },
{ -1, 84 }, { 25, 7 }, { 30, -7 }, { 28, 3 },
{ 28, 4 }, { 32, 0 }, { 34, -1 }, { 30, 6 },
{ 30, 6 }, { 32, 9 }, { 31, 19 }, { 26, 27 },
{ 26, 30 }, { 37, 20 }, { 28, 34 }, { 17, 70 },
{ 11, 28 }, { 2, 40 }, { 3, 44 }, { 0, 49 },
{ 0, 46 }, { 2, 44 }, { 2, 51 }, { 0, 47 },
{ 4, 39 }, { 2, 62 }, { 6, 46 }, { 0, 54 },
{ 3, 54 }, { 2, 58 }, { 4, 63 }, { 6, 51 },
{ 6, 57 }, { 7, 53 }, { 6, 52 }, { 6, 55 },
{ 11, 45 }, { 14, 36 }, { 8, 53 }, { -1, 82 },
{ 7, 55 }, { -3, 78 }, { 15, 46 }, { 22, 31 },
{ -1, 84 }, { 25, 7 }, { 30, -7 }, { 28, 3 },
{ 28, 4 }, { 32, 0 }, { 34, -1 }, { 30, 6 },
{ 30, 6 }, { 32, 9 }, { 31, 19 }, { 26, 27 },
{ 26, 30 }, { 37, 20 }, { 28, 34 }, { 17, 70 },
{ -4, 79 }, { -7, 71 }, { -5, 69 }, { -9, 70 },
{ -8, 66 }, { -10, 68 }, { -19, 73 }, { -12, 69 },
{ -16, 70 }, { -15, 67 }, { -20, 62 }, { -19, 70 },
{ -16, 66 }, { -22, 65 }, { -20, 63 }, { -5, 85 },
{ -6, 81 }, { -10, 77 }, { -7, 81 }, { -17, 80 },
{ -18, 73 }, { -4, 74 }, { -10, 83 }, { -9, 71 },
{ -9, 67 }, { -1, 61 }, { -8, 66 }, { -14, 66 },
{ 0, 59 }, { 2, 59 }, { 9, -2 }, { 26, -9 },
{ 33, -9 }, { 39, -7 }, { 41, -2 }, { 45, 3 },
{ 49, 9 }, { 45, 27 }, { 36, 59 }, { 21, -13 },
{ 33, -14 }, { 39, -7 }, { 46, -2 }, { 51, 2 },
{ 60, 6 }, { 61, 17 }, { 55, 34 }, { 42, 62 },
{ -6, 66 }, { -7, 35 }, { -7, 42 }, { -8, 45 },
{ -5, 48 }, { -12, 56 }, { -6, 60 }, { -5, 62 },
{ -8, 66 }, { -8, 76 }, { -4, 79 }, { -7, 71 },
{ -5, 69 }, { -9, 70 }, { -8, 66 }, { -10, 68 },
{ -19, 73 }, { -12, 69 }, { -16, 70 }, { -15, 67 },
{ -20, 62 }, { -19, 70 }, { -16, 66 }, { -22, 65 },
{ -20, 63 }, { -5, 85 }, { -6, 81 }, { -10, 77 },
{ -7, 81 }, { -17, 80 }, { -18, 73 }, { -4, 74 },
{ -10, 83 }, { -9, 71 }, { -9, 67 }, { -1, 61 },
{ -8, 66 }, { -14, 66 }, { 0, 59 }, { 2, 59 },
{ 9, -2 }, { 26, -9 }, { 33, -9 }, { 39, -7 },
{ 41, -2 }, { 45, 3 }, { 49, 9 }, { 45, 27 },
{ 36, 59 }, { 21, -13 }, { 33, -14 }, { 39, -7 },
{ 46, -2 }, { 51, 2 }, { 60, 6 }, { 61, 17 },
{ 55, 34 }, { 42, 62 }, { -6, 66 }, { -7, 35 },
{ -7, 42 }, { -8, 45 }, { -5, 48 }, { -12, 56 },
{ -6, 60 }, { -5, 62 }, { -8, 66 }, { -8, 76 },
{ -13, 106 }, { -16, 106 }, { -10, 87 }, { -21, 114 },
{ -18, 110 }, { -14, 98 }, { -22, 110 }, { -21, 106 },
{ -18, 103 }, { -21, 107 }, { -23, 108 }, { -26, 112 },
{ -10, 96 }, { -12, 95 }, { -5, 91 }, { -9, 93 },
{ -22, 94 }, { -5, 86 }, { 9, 67 }, { -4, 80 },
{ -10, 85 }, { -1, 70 }, { 7, 60 }, { 9, 58 },
{ 5, 61 }, { 12, 50 }, { 15, 50 }, { 18, 49 },
{ 17, 54 }, { 10, 41 }, { 7, 46 }, { -1, 51 },
{ 7, 49 }, { 8, 52 }, { 9, 41 }, { 6, 47 },
{ 2, 55 }, { 13, 41 }, { 10, 44 }, { 6, 50 },
{ 5, 53 }, { 13, 49 }, { 4, 63 }, { 6, 64 },
{ -13, 106 }, { -16, 106 }, { -10, 87 }, { -21, 114 },
{ -18, 110 }, { -14, 98 }, { -22, 110 }, { -21, 106 },
{ -18, 103 }, { -21, 107 }, { -23, 108 }, { -26, 112 },
{ -10, 96 }, { -12, 95 }, { -5, 91 }, { -9, 93 },
{ -22, 94 }, { -5, 86 }, { 9, 67 }, { -4, 80 },
{ -10, 85 }, { -1, 70 }, { 7, 60 }, { 9, 58 },
{ 5, 61 }, { 12, 50 }, { 15, 50 }, { 18, 49 },
{ 17, 54 }, { 10, 41 }, { 7, 46 }, { -1, 51 },
{ 7, 49 }, { 8, 52 }, { 9, 41 }, { 6, 47 },
{ 2, 55 }, { 13, 41 }, { 10, 44 }, { 6, 50 },
{ 5, 53 }, { 13, 49 }, { 4, 63 }, { 6, 64 },
{ 14, 11 }, { 11, 14 }, { 9, 11 }, { 18, 11 },
{ 21, 9 }, { 23, -2 }, { 32, -15 }, { 32, -15 },
{ 34, -21 }, { 39, -23 }, { 42, -33 }, { 41, -31 },
{ 46, -28 }, { 38, -12 }, { 21, 29 }, { 45, -24 },
{ 53, -45 }, { 48, -26 }, { 65, -43 }, { 43, -19 },
{ 39, -10 }, { 30, 9 }, { 18, 26 }, { 20, 27 },
{ 0, 57 }, { -14, 82 }, { -5, 75 }, { -19, 97 },
{ -35, 125 }, { 27, 0 }, { 28, 0 }, { 31, -4 },
{ 27, 6 }, { 34, 8 }, { 30, 10 }, { 24, 22 },
{ 33, 19 }, { 22, 32 }, { 26, 31 }, { 21, 41 },
{ 26, 44 }, { 23, 47 }, { 16, 65 }, { 14, 71 },
{ 14, 11 }, { 11, 14 }, { 9, 11 }, { 18, 11 },
{ 21, 9 }, { 23, -2 }, { 32, -15 }, { 32, -15 },
{ 34, -21 }, { 39, -23 }, { 42, -33 }, { 41, -31 },
{ 46, -28 }, { 38, -12 }, { 21, 29 }, { 45, -24 },
{ 53, -45 }, { 48, -26 }, { 65, -43 }, { 43, -19 },
{ 39, -10 }, { 30, 9 }, { 18, 26 }, { 20, 27 },
{ 0, 57 }, { -14, 82 }, { -5, 75 }, { -19, 97 },
{ -35, 125 }, { 27, 0 }, { 28, 0 }, { 31, -4 },
{ 27, 6 }, { 34, 8 }, { 30, 10 }, { 24, 22 },
{ 33, 19 }, { 22, 32 }, { 26, 31 }, { 21, 41 },
{ 26, 44 }, { 23, 47 }, { 16, 65 }, { 14, 71 },
{ -6, 76 }, { -2, 44 }, { 0, 45 }, { 0, 52 },
{ -3, 64 }, { -2, 59 }, { -4, 70 }, { -4, 75 },
{ -8, 82 }, { -17, 102 }, { -9, 77 }, { 3, 24 },
{ 0, 42 }, { 0, 48 }, { 0, 55 }, { -6, 59 },
{ -7, 71 }, { -12, 83 }, { -11, 87 }, { -30, 119 },
{ 1, 58 }, { -3, 29 }, { -1, 36 }, { 1, 38 },
{ 2, 43 }, { -6, 55 }, { 0, 58 }, { 0, 64 },
{ -3, 74 }, { -10, 90 }, { -6, 76 }, { -2, 44 },
{ 0, 45 }, { 0, 52 }, { -3, 64 }, { -2, 59 },
{ -4, 70 }, { -4, 75 }, { -8, 82 }, { -17, 102 },
{ -9, 77 }, { 3, 24 }, { 0, 42 }, { 0, 48 },
{ 0, 55 }, { -6, 59 }, { -7, 71 }, { -12, 83 },
{ -11, 87 }, { -30, 119 }, { 1, 58 }, { -3, 29 },
{ -1, 36 }, { 1, 38 }, { 2, 43 }, { -6, 55 },
{ 0, 58 }, { 0, 64 }, { -3, 74 }, { -10, 90 },
{ -3, 74 }, { -9, 92 }, { -8, 87 }, { -23, 126 },
{ -3, 74 }, { -9, 92 }, { -8, 87 }, { -23, 126 },
{ -3, 74 }, { -9, 92 }, { -8, 87 }, { -23, 126 }
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
},
/* i_cabac_init_idc == 1 */
{
/* 0 - 10 */
{ 20, -15 }, { 2, 54 }, { 3, 74 }, { 20, -15 },
{ 2, 54 }, { 3, 74 }, { -28, 127 }, { -23, 104 },
{ -6, 53 }, { -1, 54 }, { 7, 51 },
/* 11 - 23 */
{ 22, 25 }, { 34, 0 }, { 16, 0 }, { -2, 9 },
{ 4, 41 }, { -29, 118 }, { 2, 65 }, { -6, 71 },
{ -13, 79 }, { 5, 52 }, { 9, 50 }, { -3, 70 },
{ 10, 54 },
/* 24 - 39 */
{ 26, 34 }, { 19, 22 }, { 40, 0 }, { 57, 2 },
{ 41, 36 }, { 26, 69 }, { -45, 127 }, { -15, 101 },
{ -4, 76 }, { -6, 71 }, { -13, 79 }, { 5, 52 },
{ 6, 69 }, { -13, 90 }, { 0, 52 }, { 8, 43 },
/* 40 - 53 */
{ -2, 69 },{ -5, 82 },{ -10, 96 },{ 2, 59 },
{ 2, 75 },{ -3, 87 },{ -3, 100 },{ 1, 56 },
{ -3, 74 },{ -6, 85 },{ 0, 59 },{ -3, 81 },
{ -7, 86 },{ -5, 95 },
/* 54 - 59 */
{ -1, 66 },{ -1, 77 },{ 1, 70 },{ -2, 86 },
{ -5, 72 },{ 0, 61 },
/* 60 - 69 */
{ 0, 41 }, { 0, 63 }, { 0, 63 }, { 0, 63 },
{ -9, 83 }, { 4, 86 }, { 0, 97 }, { -7, 72 },
{ 13, 41 }, { 3, 62 },
/* 70 - 104 */
{ 13, 15 }, { 7, 51 }, { 2, 80 }, { -39, 127 },
{ -18, 91 }, { -17, 96 }, { -26, 81 }, { -35, 98 },
{ -24, 102 }, { -23, 97 }, { -27, 119 }, { -24, 99 },
{ -21, 110 }, { -18, 102 }, { -36, 127 }, { 0, 80 },
{ -5, 89 }, { -7, 94 }, { -4, 92 }, { 0, 39 },
{ 0, 65 }, { -15, 84 }, { -35, 127 }, { -2, 73 },
{ -12, 104 }, { -9, 91 }, { -31, 127 }, { 3, 55 },
{ 7, 56 }, { 7, 55 }, { 8, 61 }, { -3, 53 },
{ 0, 68 }, { -7, 74 }, { -9, 88 },
/* 105 -> 165 */
{ -13, 103 }, { -13, 91 }, { -9, 89 }, { -14, 92 },
{ -8, 76 }, { -12, 87 }, { -23, 110 }, { -24, 105 },
{ -10, 78 }, { -20, 112 }, { -17, 99 }, { -78, 127 },
{ -70, 127 }, { -50, 127 }, { -46, 127 }, { -4, 66 },
{ -5, 78 }, { -4, 71 }, { -8, 72 }, { 2, 59 },
{ -1, 55 }, { -7, 70 }, { -6, 75 }, { -8, 89 },
{ -34, 119 }, { -3, 75 }, { 32, 20 }, { 30, 22 },
{ -44, 127 }, { 0, 54 }, { -5, 61 }, { 0, 58 },
{ -1, 60 }, { -3, 61 }, { -8, 67 }, { -25, 84 },
{ -14, 74 }, { -5, 65 }, { 5, 52 }, { 2, 57 },
{ 0, 61 }, { -9, 69 }, { -11, 70 }, { 18, 55 },
{ -4, 71 }, { 0, 58 }, { 7, 61 }, { 9, 41 },
{ 18, 25 }, { 9, 32 }, { 5, 43 }, { 9, 47 },
{ 0, 44 }, { 0, 51 }, { 2, 46 }, { 19, 38 },
{ -4, 66 }, { 15, 38 }, { 12, 42 }, { 9, 34 },
{ 0, 89 },
/* 166 - 226 */
{ 4, 45 }, { 10, 28 }, { 10, 31 }, { 33, -11 },
{ 52, -43 }, { 18, 15 }, { 28, 0 }, { 35, -22 },
{ 38, -25 }, { 34, 0 }, { 39, -18 }, { 32, -12 },
{ 102, -94 }, { 0, 0 }, { 56, -15 }, { 33, -4 },
{ 29, 10 }, { 37, -5 }, { 51, -29 }, { 39, -9 },
{ 52, -34 }, { 69, -58 }, { 67, -63 }, { 44, -5 },
{ 32, 7 }, { 55, -29 }, { 32, 1 }, { 0, 0 },
{ 27, 36 }, { 33, -25 }, { 34, -30 }, { 36, -28 },
{ 38, -28 }, { 38, -27 }, { 34, -18 }, { 35, -16 },
{ 34, -14 }, { 32, -8 }, { 37, -6 }, { 35, 0 },
{ 30, 10 }, { 28, 18 }, { 26, 25 }, { 29, 41 },
{ 0, 75 }, { 2, 72 }, { 8, 77 }, { 14, 35 },
{ 18, 31 }, { 17, 35 }, { 21, 30 }, { 17, 45 },
{ 20, 42 }, { 18, 45 }, { 27, 26 }, { 16, 54 },
{ 7, 66 }, { 16, 56 }, { 11, 73 }, { 10, 67 },
{ -10, 116 },
/* 227 - 275 */
{ -23, 112 }, { -15, 71 }, { -7, 61 }, { 0, 53 },
{ -5, 66 }, { -11, 77 }, { -9, 80 }, { -9, 84 },
{ -10, 87 }, { -34, 127 }, { -21, 101 }, { -3, 39 },
{ -5, 53 }, { -7, 61 }, { -11, 75 }, { -15, 77 },
{ -17, 91 }, { -25, 107 }, { -25, 111 }, { -28, 122 },
{ -11, 76 }, { -10, 44 }, { -10, 52 }, { -10, 57 },
{ -9, 58 }, { -16, 72 }, { -7, 69 }, { -4, 69 },
{ -5, 74 }, { -9, 86 }, { 2, 66 }, { -9, 34 },
{ 1, 32 }, { 11, 31 }, { 5, 52 }, { -2, 55 },
{ -2, 67 }, { 0, 73 }, { -8, 89 }, { 3, 52 },
{ 7, 4 }, { 10, 8 }, { 17, 8 }, { 16, 19 },
{ 3, 37 }, { -1, 61 }, { -5, 73 }, { -1, 70 },
{ -4, 78 },
/* 276 a bit special (not used, bypass is used instead) */
{ 0, 0 },
/* 277 - 337 */
{ -21, 126 }, { -23, 124 }, { -20, 110 }, { -26, 126 },
{ -25, 124 }, { -17, 105 }, { -27, 121 }, { -27, 117 },
{ -17, 102 }, { -26, 117 }, { -27, 116 }, { -33, 122 },
{ -10, 95 }, { -14, 100 }, { -8, 95 }, { -17, 111 },
{ -28, 114 }, { -6, 89 }, { -2, 80 }, { -4, 82 },
{ -9, 85 }, { -8, 81 }, { -1, 72 }, { 5, 64 },
{ 1, 67 }, { 9, 56 }, { 0, 69 }, { 1, 69 },
{ 7, 69 }, { -7, 69 }, { -6, 67 }, { -16, 77 },
{ -2, 64 }, { 2, 61 }, { -6, 67 }, { -3, 64 },
{ 2, 57 }, { -3, 65 }, { -3, 66 }, { 0, 62 },
{ 9, 51 }, { -1, 66 }, { -2, 71 }, { -2, 75 },
{ -1, 70 }, { -9, 72 }, { 14, 60 }, { 16, 37 },
{ 0, 47 }, { 18, 35 }, { 11, 37 }, { 12, 41 },
{ 10, 41 }, { 2, 48 }, { 12, 41 }, { 13, 41 },
{ 0, 59 }, { 3, 50 }, { 19, 40 }, { 3, 66 },
{ 18, 50 },
/* 338 - 398 */
{ 19, -6 }, { 18, -6 }, { 14, 0 }, { 26, -12 },
{ 31, -16 }, { 33, -25 }, { 33, -22 }, { 37, -28 },
{ 39, -30 }, { 42, -30 }, { 47, -42 }, { 45, -36 },
{ 49, -34 }, { 41, -17 }, { 32, 9 }, { 69, -71 },
{ 63, -63 }, { 66, -64 }, { 77, -74 }, { 54, -39 },
{ 52, -35 }, { 41, -10 }, { 36, 0 }, { 40, -1 },
{ 30, 14 }, { 28, 26 }, { 23, 37 }, { 12, 55 },
{ 11, 65 }, { 37, -33 }, { 39, -36 }, { 40, -37 },
{ 38, -30 }, { 46, -33 }, { 42, -30 }, { 40, -24 },
{ 49, -29 }, { 38, -12 }, { 40, -10 }, { 38, -3 },
{ 46, -5 }, { 31, 20 }, { 29, 30 }, { 25, 44 },
{ 12, 48 }, { 11, 49 }, { 26, 45 }, { 22, 22 },
{ 23, 22 }, { 27, 21 }, { 33, 20 }, { 26, 28 },
{ 30, 24 }, { 27, 34 }, { 18, 42 }, { 25, 39 },
{ 18, 50 }, { 12, 70 }, { 21, 54 }, { 14, 71 },
{ 11, 83 },
/* 399 - 435 */
{ 25, 32 }, { 21, 49 }, { 21, 54 },
{ -5, 85 }, { -6, 81 }, { -10, 77 }, { -7, 81 },
{ -17, 80 }, { -18, 73 }, { -4, 74 }, { -10, 83 },
{ -9, 71 }, { -9, 67 }, { -1, 61 }, { -8, 66 },
{ -14, 66 }, { 0, 59 }, { 2, 59 }, { 17, -10 },
{ 32, -13 }, { 42, -9 }, { 49, -5 }, { 53, 0 },
{ 64, 3 }, { 68, 10 }, { 66, 27 }, { 47, 57 },
{ -5, 71 }, { 0, 24 }, { -1, 36 }, { -2, 42 },
{ -2, 52 }, { -9, 57 }, { -6, 63 }, { -4, 65 },
{ -4, 67 }, { -7, 82 },
/* 436 - 459 */
{ -3, 81 }, { -3, 76 }, { -7, 72 }, { -6, 78 },
{ -12, 72 }, { -14, 68 }, { -3, 70 }, { -6, 76 },
{ -5, 66 }, { -5, 62 }, { 0, 57 }, { -4, 61 },
{ -9, 60 }, { 1, 54 }, { 2, 58 }, { 17, -10 },
{ 32, -13 }, { 42, -9 }, { 49, -5 }, { 53, 0 },
{ 64, 3 }, { 68, 10 }, { 66, 27 }, { 47, 57 },
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
/* 460 - 1024 */
{ 0, 80 }, { -5, 89 }, { -7, 94 }, { -4, 92 },
{ 0, 39 }, { 0, 65 }, { -15, 84 }, { -35, 127 },
{ -2, 73 }, { -12, 104 }, { -9, 91 }, { -31, 127 },
{ 0, 80 }, { -5, 89 }, { -7, 94 }, { -4, 92 },
{ 0, 39 }, { 0, 65 }, { -15, 84 }, { -35, 127 },
{ -2, 73 }, { -12, 104 }, { -9, 91 }, { -31, 127 },
{ -13, 103 }, { -13, 91 }, { -9, 89 }, { -14, 92 },
{ -8, 76 }, { -12, 87 }, { -23, 110 }, { -24, 105 },
{ -10, 78 }, { -20, 112 }, { -17, 99 }, { -78, 127 },
{ -70, 127 }, { -50, 127 }, { -46, 127 }, { -4, 66 },
{ -5, 78 }, { -4, 71 }, { -8, 72 }, { 2, 59 },
{ -1, 55 }, { -7, 70 }, { -6, 75 }, { -8, 89 },
{ -34, 119 }, { -3, 75 }, { 32, 20 }, { 30, 22 },
{ -44, 127 }, { 0, 54 }, { -5, 61 }, { 0, 58 },
{ -1, 60 }, { -3, 61 }, { -8, 67 }, { -25, 84 },
{ -14, 74 }, { -5, 65 }, { 5, 52 }, { 2, 57 },
{ 0, 61 }, { -9, 69 }, { -11, 70 }, { 18, 55 },
{ -13, 103 }, { -13, 91 }, { -9, 89 }, { -14, 92 },
{ -8, 76 }, { -12, 87 }, { -23, 110 }, { -24, 105 },
{ -10, 78 }, { -20, 112 }, { -17, 99 }, { -78, 127 },
{ -70, 127 }, { -50, 127 }, { -46, 127 }, { -4, 66 },
{ -5, 78 }, { -4, 71 }, { -8, 72 }, { 2, 59 },
{ -1, 55 }, { -7, 70 }, { -6, 75 }, { -8, 89 },
{ -34, 119 }, { -3, 75 }, { 32, 20 }, { 30, 22 },
{ -44, 127 }, { 0, 54 }, { -5, 61 }, { 0, 58 },
{ -1, 60 }, { -3, 61 }, { -8, 67 }, { -25, 84 },
{ -14, 74 }, { -5, 65 }, { 5, 52 }, { 2, 57 },
{ 0, 61 }, { -9, 69 }, { -11, 70 }, { 18, 55 },
{ 4, 45 }, { 10, 28 }, { 10, 31 }, { 33, -11 },
{ 52, -43 }, { 18, 15 }, { 28, 0 }, { 35, -22 },
{ 38, -25 }, { 34, 0 }, { 39, -18 }, { 32, -12 },
{ 102, -94 }, { 0, 0 }, { 56, -15 }, { 33, -4 },
{ 29, 10 }, { 37, -5 }, { 51, -29 }, { 39, -9 },
{ 52, -34 }, { 69, -58 }, { 67, -63 }, { 44, -5 },
{ 32, 7 }, { 55, -29 }, { 32, 1 }, { 0, 0 },
{ 27, 36 }, { 33, -25 }, { 34, -30 }, { 36, -28 },
{ 38, -28 }, { 38, -27 }, { 34, -18 }, { 35, -16 },
{ 34, -14 }, { 32, -8 }, { 37, -6 }, { 35, 0 },
{ 30, 10 }, { 28, 18 }, { 26, 25 }, { 29, 41 },
{ 4, 45 }, { 10, 28 }, { 10, 31 }, { 33, -11 },
{ 52, -43 }, { 18, 15 }, { 28, 0 }, { 35, -22 },
{ 38, -25 }, { 34, 0 }, { 39, -18 }, { 32, -12 },
{ 102, -94 }, { 0, 0 }, { 56, -15 }, { 33, -4 },
{ 29, 10 }, { 37, -5 }, { 51, -29 }, { 39, -9 },
{ 52, -34 }, { 69, -58 }, { 67, -63 }, { 44, -5 },
{ 32, 7 }, { 55, -29 }, { 32, 1 }, { 0, 0 },
{ 27, 36 }, { 33, -25 }, { 34, -30 }, { 36, -28 },
{ 38, -28 }, { 38, -27 }, { 34, -18 }, { 35, -16 },
{ 34, -14 }, { 32, -8 }, { 37, -6 }, { 35, 0 },
{ 30, 10 }, { 28, 18 }, { 26, 25 }, { 29, 41 },
{ -5, 85 }, { -6, 81 }, { -10, 77 }, { -7, 81 },
{ -17, 80 }, { -18, 73 }, { -4, 74 }, { -10, 83 },
{ -9, 71 }, { -9, 67 }, { -1, 61 }, { -8, 66 },
{ -14, 66 }, { 0, 59 }, { 2, 59 }, { -3, 81 },
{ -3, 76 }, { -7, 72 }, { -6, 78 }, { -12, 72 },
{ -14, 68 }, { -3, 70 }, { -6, 76 }, { -5, 66 },
{ -5, 62 }, { 0, 57 }, { -4, 61 }, { -9, 60 },
{ 1, 54 }, { 2, 58 }, { 17, -10 }, { 32, -13 },
{ 42, -9 }, { 49, -5 }, { 53, 0 }, { 64, 3 },
{ 68, 10 }, { 66, 27 }, { 47, 57 }, { 17, -10 },
{ 32, -13 }, { 42, -9 }, { 49, -5 }, { 53, 0 },
{ 64, 3 }, { 68, 10 }, { 66, 27 }, { 47, 57 },
{ -5, 71 }, { 0, 24 }, { -1, 36 }, { -2, 42 },
{ -2, 52 }, { -9, 57 }, { -6, 63 }, { -4, 65 },
{ -4, 67 }, { -7, 82 }, { -5, 85 }, { -6, 81 },
{ -10, 77 }, { -7, 81 }, { -17, 80 }, { -18, 73 },
{ -4, 74 }, { -10, 83 }, { -9, 71 }, { -9, 67 },
{ -1, 61 }, { -8, 66 }, { -14, 66 }, { 0, 59 },
{ 2, 59 }, { -3, 81 }, { -3, 76 }, { -7, 72 },
{ -6, 78 }, { -12, 72 }, { -14, 68 }, { -3, 70 },
{ -6, 76 }, { -5, 66 }, { -5, 62 }, { 0, 57 },
{ -4, 61 }, { -9, 60 }, { 1, 54 }, { 2, 58 },
{ 17, -10 }, { 32, -13 }, { 42, -9 }, { 49, -5 },
{ 53, 0 }, { 64, 3 }, { 68, 10 }, { 66, 27 },
{ 47, 57 }, { 17, -10 }, { 32, -13 }, { 42, -9 },
{ 49, -5 }, { 53, 0 }, { 64, 3 }, { 68, 10 },
{ 66, 27 }, { 47, 57 }, { -5, 71 }, { 0, 24 },
{ -1, 36 }, { -2, 42 }, { -2, 52 }, { -9, 57 },
{ -6, 63 }, { -4, 65 }, { -4, 67 }, { -7, 82 },
{ -21, 126 }, { -23, 124 }, { -20, 110 }, { -26, 126 },
{ -25, 124 }, { -17, 105 }, { -27, 121 }, { -27, 117 },
{ -17, 102 }, { -26, 117 }, { -27, 116 }, { -33, 122 },
{ -10, 95 }, { -14, 100 }, { -8, 95 }, { -17, 111 },
{ -28, 114 }, { -6, 89 }, { -2, 80 }, { -4, 82 },
{ -9, 85 }, { -8, 81 }, { -1, 72 }, { 5, 64 },
{ 1, 67 }, { 9, 56 }, { 0, 69 }, { 1, 69 },
{ 7, 69 }, { -7, 69 }, { -6, 67 }, { -16, 77 },
{ -2, 64 }, { 2, 61 }, { -6, 67 }, { -3, 64 },
{ 2, 57 }, { -3, 65 }, { -3, 66 }, { 0, 62 },
{ 9, 51 }, { -1, 66 }, { -2, 71 }, { -2, 75 },
{ -21, 126 }, { -23, 124 }, { -20, 110 }, { -26, 126 },
{ -25, 124 }, { -17, 105 }, { -27, 121 }, { -27, 117 },
{ -17, 102 }, { -26, 117 }, { -27, 116 }, { -33, 122 },
{ -10, 95 }, { -14, 100 }, { -8, 95 }, { -17, 111 },
{ -28, 114 }, { -6, 89 }, { -2, 80 }, { -4, 82 },
{ -9, 85 }, { -8, 81 }, { -1, 72 }, { 5, 64 },
{ 1, 67 }, { 9, 56 }, { 0, 69 }, { 1, 69 },
{ 7, 69 }, { -7, 69 }, { -6, 67 }, { -16, 77 },
{ -2, 64 }, { 2, 61 }, { -6, 67 }, { -3, 64 },
{ 2, 57 }, { -3, 65 }, { -3, 66 }, { 0, 62 },
{ 9, 51 }, { -1, 66 }, { -2, 71 }, { -2, 75 },
{ 19, -6 }, { 18, -6 }, { 14, 0 }, { 26, -12 },
{ 31, -16 }, { 33, -25 }, { 33, -22 }, { 37, -28 },
{ 39, -30 }, { 42, -30 }, { 47, -42 }, { 45, -36 },
{ 49, -34 }, { 41, -17 }, { 32, 9 }, { 69, -71 },
{ 63, -63 }, { 66, -64 }, { 77, -74 }, { 54, -39 },
{ 52, -35 }, { 41, -10 }, { 36, 0 }, { 40, -1 },
{ 30, 14 }, { 28, 26 }, { 23, 37 }, { 12, 55 },
{ 11, 65 }, { 37, -33 }, { 39, -36 }, { 40, -37 },
{ 38, -30 }, { 46, -33 }, { 42, -30 }, { 40, -24 },
{ 49, -29 }, { 38, -12 }, { 40, -10 }, { 38, -3 },
{ 46, -5 }, { 31, 20 }, { 29, 30 }, { 25, 44 },
{ 19, -6 }, { 18, -6 }, { 14, 0 }, { 26, -12 },
{ 31, -16 }, { 33, -25 }, { 33, -22 }, { 37, -28 },
{ 39, -30 }, { 42, -30 }, { 47, -42 }, { 45, -36 },
{ 49, -34 }, { 41, -17 }, { 32, 9 }, { 69, -71 },
{ 63, -63 }, { 66, -64 }, { 77, -74 }, { 54, -39 },
{ 52, -35 }, { 41, -10 }, { 36, 0 }, { 40, -1 },
{ 30, 14 }, { 28, 26 }, { 23, 37 }, { 12, 55 },
{ 11, 65 }, { 37, -33 }, { 39, -36 }, { 40, -37 },
{ 38, -30 }, { 46, -33 }, { 42, -30 }, { 40, -24 },
{ 49, -29 }, { 38, -12 }, { 40, -10 }, { 38, -3 },
{ 46, -5 }, { 31, 20 }, { 29, 30 }, { 25, 44 },
{ -23, 112 }, { -15, 71 }, { -7, 61 }, { 0, 53 },
{ -5, 66 }, { -11, 77 }, { -9, 80 }, { -9, 84 },
{ -10, 87 }, { -34, 127 }, { -21, 101 }, { -3, 39 },
{ -5, 53 }, { -7, 61 }, { -11, 75 }, { -15, 77 },
{ -17, 91 }, { -25, 107 }, { -25, 111 }, { -28, 122 },
{ -11, 76 }, { -10, 44 }, { -10, 52 }, { -10, 57 },
{ -9, 58 }, { -16, 72 }, { -7, 69 }, { -4, 69 },
{ -5, 74 }, { -9, 86 }, { -23, 112 }, { -15, 71 },
{ -7, 61 }, { 0, 53 }, { -5, 66 }, { -11, 77 },
{ -9, 80 }, { -9, 84 }, { -10, 87 }, { -34, 127 },
{ -21, 101 }, { -3, 39 }, { -5, 53 }, { -7, 61 },
{ -11, 75 }, { -15, 77 }, { -17, 91 }, { -25, 107 },
{ -25, 111 }, { -28, 122 }, { -11, 76 }, { -10, 44 },
{ -10, 52 }, { -10, 57 }, { -9, 58 }, { -16, 72 },
{ -7, 69 }, { -4, 69 }, { -5, 74 }, { -9, 86 },
{ -2, 73 }, { -12, 104 }, { -9, 91 }, { -31, 127 },
{ -2, 73 }, { -12, 104 }, { -9, 91 }, { -31, 127 },
{ -2, 73 }, { -12, 104 }, { -9, 91 }, { -31, 127 }
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
},
/* i_cabac_init_idc == 2 */
{
/* 0 - 10 */
{ 20, -15 }, { 2, 54 }, { 3, 74 }, { 20, -15 },
{ 2, 54 }, { 3, 74 }, { -28, 127 }, { -23, 104 },
{ -6, 53 }, { -1, 54 }, { 7, 51 },
/* 11 - 23 */
{ 29, 16 }, { 25, 0 }, { 14, 0 }, { -10, 51 },
{ -3, 62 }, { -27, 99 }, { 26, 16 }, { -4, 85 },
{ -24, 102 }, { 5, 57 }, { 6, 57 }, { -17, 73 },
{ 14, 57 },
/* 24 - 39 */
{ 20, 40 }, { 20, 10 }, { 29, 0 }, { 54, 0 },
{ 37, 42 }, { 12, 97 }, { -32, 127 }, { -22, 117 },
{ -2, 74 }, { -4, 85 }, { -24, 102 }, { 5, 57 },
{ -6, 93 }, { -14, 88 }, { -6, 44 }, { 4, 55 },
/* 40 - 53 */
{ -11, 89 },{ -15, 103 },{ -21, 116 },{ 19, 57 },
{ 20, 58 },{ 4, 84 },{ 6, 96 },{ 1, 63 },
{ -5, 85 },{ -13, 106 },{ 5, 63 },{ 6, 75 },
{ -3, 90 },{ -1, 101 },
/* 54 - 59 */
{ 3, 55 },{ -4, 79 },{ -2, 75 },{ -12, 97 },
{ -7, 50 },{ 1, 60 },
/* 60 - 69 */
{ 0, 41 }, { 0, 63 }, { 0, 63 }, { 0, 63 },
{ -9, 83 }, { 4, 86 }, { 0, 97 }, { -7, 72 },
{ 13, 41 }, { 3, 62 },
/* 70 - 104 */
{ 7, 34 }, { -9, 88 }, { -20, 127 }, { -36, 127 },
{ -17, 91 }, { -14, 95 }, { -25, 84 }, { -25, 86 },
{ -12, 89 }, { -17, 91 }, { -31, 127 }, { -14, 76 },
{ -18, 103 }, { -13, 90 }, { -37, 127 }, { 11, 80 },
{ 5, 76 }, { 2, 84 }, { 5, 78 }, { -6, 55 },
{ 4, 61 }, { -14, 83 }, { -37, 127 }, { -5, 79 },
{ -11, 104 }, { -11, 91 }, { -30, 127 }, { 0, 65 },
{ -2, 79 }, { 0, 72 }, { -4, 92 }, { -6, 56 },
{ 3, 68 }, { -8, 71 }, { -13, 98 },
/* 105 -> 165 */
{ -4, 86 }, { -12, 88 }, { -5, 82 }, { -3, 72 },
{ -4, 67 }, { -8, 72 }, { -16, 89 }, { -9, 69 },
{ -1, 59 }, { 5, 66 }, { 4, 57 }, { -4, 71 },
{ -2, 71 }, { 2, 58 }, { -1, 74 }, { -4, 44 },
{ -1, 69 }, { 0, 62 }, { -7, 51 }, { -4, 47 },
{ -6, 42 }, { -3, 41 }, { -6, 53 }, { 8, 76 },
{ -9, 78 }, { -11, 83 }, { 9, 52 }, { 0, 67 },
{ -5, 90 }, { 1, 67 }, { -15, 72 }, { -5, 75 },
{ -8, 80 }, { -21, 83 }, { -21, 64 }, { -13, 31 },
{ -25, 64 }, { -29, 94 }, { 9, 75 }, { 17, 63 },
{ -8, 74 }, { -5, 35 }, { -2, 27 }, { 13, 91 },
{ 3, 65 }, { -7, 69 }, { 8, 77 }, { -10, 66 },
{ 3, 62 }, { -3, 68 }, { -20, 81 }, { 0, 30 },
{ 1, 7 }, { -3, 23 }, { -21, 74 }, { 16, 66 },
{ -23, 124 }, { 17, 37 }, { 44, -18 }, { 50, -34 },
{ -22, 127 },
/* 166 - 226 */
{ 4, 39 }, { 0, 42 }, { 7, 34 }, { 11, 29 },
{ 8, 31 }, { 6, 37 }, { 7, 42 }, { 3, 40 },
{ 8, 33 }, { 13, 43 }, { 13, 36 }, { 4, 47 },
{ 3, 55 }, { 2, 58 }, { 6, 60 }, { 8, 44 },
{ 11, 44 }, { 14, 42 }, { 7, 48 }, { 4, 56 },
{ 4, 52 }, { 13, 37 }, { 9, 49 }, { 19, 58 },
{ 10, 48 }, { 12, 45 }, { 0, 69 }, { 20, 33 },
{ 8, 63 }, { 35, -18 }, { 33, -25 }, { 28, -3 },
{ 24, 10 }, { 27, 0 }, { 34, -14 }, { 52, -44 },
{ 39, -24 }, { 19, 17 }, { 31, 25 }, { 36, 29 },
{ 24, 33 }, { 34, 15 }, { 30, 20 }, { 22, 73 },
{ 20, 34 }, { 19, 31 }, { 27, 44 }, { 19, 16 },
{ 15, 36 }, { 15, 36 }, { 21, 28 }, { 25, 21 },
{ 30, 20 }, { 31, 12 }, { 27, 16 }, { 24, 42 },
{ 0, 93 }, { 14, 56 }, { 15, 57 }, { 26, 38 },
{ -24, 127 },
/* 227 - 275 */
{ -24, 115 }, { -22, 82 }, { -9, 62 }, { 0, 53 },
{ 0, 59 }, { -14, 85 }, { -13, 89 }, { -13, 94 },
{ -11, 92 }, { -29, 127 }, { -21, 100 }, { -14, 57 },
{ -12, 67 }, { -11, 71 }, { -10, 77 }, { -21, 85 },
{ -16, 88 }, { -23, 104 }, { -15, 98 }, { -37, 127 },
{ -10, 82 }, { -8, 48 }, { -8, 61 }, { -8, 66 },
{ -7, 70 }, { -14, 75 }, { -10, 79 }, { -9, 83 },
{ -12, 92 }, { -18, 108 }, { -4, 79 }, { -22, 69 },
{ -16, 75 }, { -2, 58 }, { 1, 58 }, { -13, 78 },
{ -9, 83 }, { -4, 81 }, { -13, 99 }, { -13, 81 },
{ -6, 38 }, { -13, 62 }, { -6, 58 }, { -2, 59 },
{ -16, 73 }, { -10, 76 }, { -13, 86 }, { -9, 83 },
{ -10, 87 },
/* 276 a bit special (not used, bypass is used instead) */
{ 0, 0 },
/* 277 - 337 */
{ -22, 127 }, { -25, 127 }, { -25, 120 }, { -27, 127 },
{ -19, 114 }, { -23, 117 }, { -25, 118 }, { -26, 117 },
{ -24, 113 }, { -28, 118 }, { -31, 120 }, { -37, 124 },
{ -10, 94 }, { -15, 102 }, { -10, 99 }, { -13, 106 },
{ -50, 127 }, { -5, 92 }, { 17, 57 }, { -5, 86 },
{ -13, 94 }, { -12, 91 }, { -2, 77 }, { 0, 71 },
{ -1, 73 }, { 4, 64 }, { -7, 81 }, { 5, 64 },
{ 15, 57 }, { 1, 67 }, { 0, 68 }, { -10, 67 },
{ 1, 68 }, { 0, 77 }, { 2, 64 }, { 0, 68 },
{ -5, 78 }, { 7, 55 }, { 5, 59 }, { 2, 65 },
{ 14, 54 }, { 15, 44 }, { 5, 60 }, { 2, 70 },
{ -2, 76 }, { -18, 86 }, { 12, 70 }, { 5, 64 },
{ -12, 70 }, { 11, 55 }, { 5, 56 }, { 0, 69 },
{ 2, 65 }, { -6, 74 }, { 5, 54 }, { 7, 54 },
{ -6, 76 }, { -11, 82 }, { -2, 77 }, { -2, 77 },
{ 25, 42 },
/* 338 - 398 */
{ 17, -13 }, { 16, -9 }, { 17, -12 }, { 27, -21 },
{ 37, -30 }, { 41, -40 }, { 42, -41 }, { 48, -47 },
{ 39, -32 }, { 46, -40 }, { 52, -51 }, { 46, -41 },
{ 52, -39 }, { 43, -19 }, { 32, 11 }, { 61, -55 },
{ 56, -46 }, { 62, -50 }, { 81, -67 }, { 45, -20 },
{ 35, -2 }, { 28, 15 }, { 34, 1 }, { 39, 1 },
{ 30, 17 }, { 20, 38 }, { 18, 45 }, { 15, 54 },
{ 0, 79 }, { 36, -16 }, { 37, -14 }, { 37, -17 },
{ 32, 1 }, { 34, 15 }, { 29, 15 }, { 24, 25 },
{ 34, 22 }, { 31, 16 }, { 35, 18 }, { 31, 28 },
{ 33, 41 }, { 36, 28 }, { 27, 47 }, { 21, 62 },
{ 18, 31 }, { 19, 26 }, { 36, 24 }, { 24, 23 },
{ 27, 16 }, { 24, 30 }, { 31, 29 }, { 22, 41 },
{ 22, 42 }, { 16, 60 }, { 15, 52 }, { 14, 60 },
{ 3, 78 }, { -16, 123 }, { 21, 53 }, { 22, 56 },
{ 25, 61 },
/* 399 - 435 */
{ 21, 33 }, { 19, 50 }, { 17, 61 },
{ -3, 78 }, { -8, 74 }, { -9, 72 }, { -10, 72 },
{ -18, 75 }, { -12, 71 }, { -11, 63 }, { -5, 70 },
{ -17, 75 }, { -14, 72 }, { -16, 67 }, { -8, 53 },
{ -14, 59 }, { -9, 52 }, { -11, 68 }, { 9, -2 },
{ 30, -10 }, { 31, -4 }, { 33, -1 }, { 33, 7 },
{ 31, 12 }, { 37, 23 }, { 31, 38 }, { 20, 64 },
{ -9, 71 }, { -7, 37 }, { -8, 44 }, { -11, 49 },
{ -10, 56 }, { -12, 59 }, { -8, 63 }, { -9, 67 },
{ -6, 68 }, { -10, 79 },
/* 436 - 459 */
{ -3, 78 }, { -8, 74 }, { -9, 72 }, { -10, 72 },
{ -18, 75 }, { -12, 71 }, { -11, 63 }, { -5, 70 },
{ -17, 75 }, { -14, 72 }, { -16, 67 }, { -8, 53 },
{ -14, 59 }, { -9, 52 }, { -11, 68 }, { 9, -2 },
{ 30, -10 }, { 31, -4 }, { 33, -1 }, { 33, 7 },
{ 31, 12 }, { 37, 23 }, { 31, 38 }, { 20, 64 },
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
/* 460 - 1024 */
{ 11, 80 }, { 5, 76 }, { 2, 84 }, { 5, 78 },
{ -6, 55 }, { 4, 61 }, { -14, 83 }, { -37, 127 },
{ -5, 79 }, { -11, 104 }, { -11, 91 }, { -30, 127 },
{ 11, 80 }, { 5, 76 }, { 2, 84 }, { 5, 78 },
{ -6, 55 }, { 4, 61 }, { -14, 83 }, { -37, 127 },
{ -5, 79 }, { -11, 104 }, { -11, 91 }, { -30, 127 },
{ -4, 86 }, { -12, 88 }, { -5, 82 }, { -3, 72 },
{ -4, 67 }, { -8, 72 }, { -16, 89 }, { -9, 69 },
{ -1, 59 }, { 5, 66 }, { 4, 57 }, { -4, 71 },
{ -2, 71 }, { 2, 58 }, { -1, 74 }, { -4, 44 },
{ -1, 69 }, { 0, 62 }, { -7, 51 }, { -4, 47 },
{ -6, 42 }, { -3, 41 }, { -6, 53 }, { 8, 76 },
{ -9, 78 }, { -11, 83 }, { 9, 52 }, { 0, 67 },
{ -5, 90 }, { 1, 67 }, { -15, 72 }, { -5, 75 },
{ -8, 80 }, { -21, 83 }, { -21, 64 }, { -13, 31 },
{ -25, 64 }, { -29, 94 }, { 9, 75 }, { 17, 63 },
{ -8, 74 }, { -5, 35 }, { -2, 27 }, { 13, 91 },
{ -4, 86 }, { -12, 88 }, { -5, 82 }, { -3, 72 },
{ -4, 67 }, { -8, 72 }, { -16, 89 }, { -9, 69 },
{ -1, 59 }, { 5, 66 }, { 4, 57 }, { -4, 71 },
{ -2, 71 }, { 2, 58 }, { -1, 74 }, { -4, 44 },
{ -1, 69 }, { 0, 62 }, { -7, 51 }, { -4, 47 },
{ -6, 42 }, { -3, 41 }, { -6, 53 }, { 8, 76 },
{ -9, 78 }, { -11, 83 }, { 9, 52 }, { 0, 67 },
{ -5, 90 }, { 1, 67 }, { -15, 72 }, { -5, 75 },
{ -8, 80 }, { -21, 83 }, { -21, 64 }, { -13, 31 },
{ -25, 64 }, { -29, 94 }, { 9, 75 }, { 17, 63 },
{ -8, 74 }, { -5, 35 }, { -2, 27 }, { 13, 91 },
{ 4, 39 }, { 0, 42 }, { 7, 34 }, { 11, 29 },
{ 8, 31 }, { 6, 37 }, { 7, 42 }, { 3, 40 },
{ 8, 33 }, { 13, 43 }, { 13, 36 }, { 4, 47 },
{ 3, 55 }, { 2, 58 }, { 6, 60 }, { 8, 44 },
{ 11, 44 }, { 14, 42 }, { 7, 48 }, { 4, 56 },
{ 4, 52 }, { 13, 37 }, { 9, 49 }, { 19, 58 },
{ 10, 48 }, { 12, 45 }, { 0, 69 }, { 20, 33 },
{ 8, 63 }, { 35, -18 }, { 33, -25 }, { 28, -3 },
{ 24, 10 }, { 27, 0 }, { 34, -14 }, { 52, -44 },
{ 39, -24 }, { 19, 17 }, { 31, 25 }, { 36, 29 },
{ 24, 33 }, { 34, 15 }, { 30, 20 }, { 22, 73 },
{ 4, 39 }, { 0, 42 }, { 7, 34 }, { 11, 29 },
{ 8, 31 }, { 6, 37 }, { 7, 42 }, { 3, 40 },
{ 8, 33 }, { 13, 43 }, { 13, 36 }, { 4, 47 },
{ 3, 55 }, { 2, 58 }, { 6, 60 }, { 8, 44 },
{ 11, 44 }, { 14, 42 }, { 7, 48 }, { 4, 56 },
{ 4, 52 }, { 13, 37 }, { 9, 49 }, { 19, 58 },
{ 10, 48 }, { 12, 45 }, { 0, 69 }, { 20, 33 },
{ 8, 63 }, { 35, -18 }, { 33, -25 }, { 28, -3 },
{ 24, 10 }, { 27, 0 }, { 34, -14 }, { 52, -44 },
{ 39, -24 }, { 19, 17 }, { 31, 25 }, { 36, 29 },
{ 24, 33 }, { 34, 15 }, { 30, 20 }, { 22, 73 },
{ -3, 78 }, { -8, 74 }, { -9, 72 }, { -10, 72 },
{ -18, 75 }, { -12, 71 }, { -11, 63 }, { -5, 70 },
{ -17, 75 }, { -14, 72 }, { -16, 67 }, { -8, 53 },
{ -14, 59 }, { -9, 52 }, { -11, 68 }, { -3, 78 },
{ -8, 74 }, { -9, 72 }, { -10, 72 }, { -18, 75 },
{ -12, 71 }, { -11, 63 }, { -5, 70 }, { -17, 75 },
{ -14, 72 }, { -16, 67 }, { -8, 53 }, { -14, 59 },
{ -9, 52 }, { -11, 68 }, { 9, -2 }, { 30, -10 },
{ 31, -4 }, { 33, -1 }, { 33, 7 }, { 31, 12 },
{ 37, 23 }, { 31, 38 }, { 20, 64 }, { 9, -2 },
{ 30, -10 }, { 31, -4 }, { 33, -1 }, { 33, 7 },
{ 31, 12 }, { 37, 23 }, { 31, 38 }, { 20, 64 },
{ -9, 71 }, { -7, 37 }, { -8, 44 }, { -11, 49 },
{ -10, 56 }, { -12, 59 }, { -8, 63 }, { -9, 67 },
{ -6, 68 }, { -10, 79 }, { -3, 78 }, { -8, 74 },
{ -9, 72 }, { -10, 72 }, { -18, 75 }, { -12, 71 },
{ -11, 63 }, { -5, 70 }, { -17, 75 }, { -14, 72 },
{ -16, 67 }, { -8, 53 }, { -14, 59 }, { -9, 52 },
{ -11, 68 }, { -3, 78 }, { -8, 74 }, { -9, 72 },
{ -10, 72 }, { -18, 75 }, { -12, 71 }, { -11, 63 },
{ -5, 70 }, { -17, 75 }, { -14, 72 }, { -16, 67 },
{ -8, 53 }, { -14, 59 }, { -9, 52 }, { -11, 68 },
{ 9, -2 }, { 30, -10 }, { 31, -4 }, { 33, -1 },
{ 33, 7 }, { 31, 12 }, { 37, 23 }, { 31, 38 },
{ 20, 64 }, { 9, -2 }, { 30, -10 }, { 31, -4 },
{ 33, -1 }, { 33, 7 }, { 31, 12 }, { 37, 23 },
{ 31, 38 }, { 20, 64 }, { -9, 71 }, { -7, 37 },
{ -8, 44 }, { -11, 49 }, { -10, 56 }, { -12, 59 },
{ -8, 63 }, { -9, 67 }, { -6, 68 }, { -10, 79 },
{ -22, 127 }, { -25, 127 }, { -25, 120 }, { -27, 127 },
{ -19, 114 }, { -23, 117 }, { -25, 118 }, { -26, 117 },
{ -24, 113 }, { -28, 118 }, { -31, 120 }, { -37, 124 },
{ -10, 94 }, { -15, 102 }, { -10, 99 }, { -13, 106 },
{ -50, 127 }, { -5, 92 }, { 17, 57 }, { -5, 86 },
{ -13, 94 }, { -12, 91 }, { -2, 77 }, { 0, 71 },
{ -1, 73 }, { 4, 64 }, { -7, 81 }, { 5, 64 },
{ 15, 57 }, { 1, 67 }, { 0, 68 }, { -10, 67 },
{ 1, 68 }, { 0, 77 }, { 2, 64 }, { 0, 68 },
{ -5, 78 }, { 7, 55 }, { 5, 59 }, { 2, 65 },
{ 14, 54 }, { 15, 44 }, { 5, 60 }, { 2, 70 },
{ -22, 127 }, { -25, 127 }, { -25, 120 }, { -27, 127 },
{ -19, 114 }, { -23, 117 }, { -25, 118 }, { -26, 117 },
{ -24, 113 }, { -28, 118 }, { -31, 120 }, { -37, 124 },
{ -10, 94 }, { -15, 102 }, { -10, 99 }, { -13, 106 },
{ -50, 127 }, { -5, 92 }, { 17, 57 }, { -5, 86 },
{ -13, 94 }, { -12, 91 }, { -2, 77 }, { 0, 71 },
{ -1, 73 }, { 4, 64 }, { -7, 81 }, { 5, 64 },
{ 15, 57 }, { 1, 67 }, { 0, 68 }, { -10, 67 },
{ 1, 68 }, { 0, 77 }, { 2, 64 }, { 0, 68 },
{ -5, 78 }, { 7, 55 }, { 5, 59 }, { 2, 65 },
{ 14, 54 }, { 15, 44 }, { 5, 60 }, { 2, 70 },
{ 17, -13 }, { 16, -9 }, { 17, -12 }, { 27, -21 },
{ 37, -30 }, { 41, -40 }, { 42, -41 }, { 48, -47 },
{ 39, -32 }, { 46, -40 }, { 52, -51 }, { 46, -41 },
{ 52, -39 }, { 43, -19 }, { 32, 11 }, { 61, -55 },
{ 56, -46 }, { 62, -50 }, { 81, -67 }, { 45, -20 },
{ 35, -2 }, { 28, 15 }, { 34, 1 }, { 39, 1 },
{ 30, 17 }, { 20, 38 }, { 18, 45 }, { 15, 54 },
{ 0, 79 }, { 36, -16 }, { 37, -14 }, { 37, -17 },
{ 32, 1 }, { 34, 15 }, { 29, 15 }, { 24, 25 },
{ 34, 22 }, { 31, 16 }, { 35, 18 }, { 31, 28 },
{ 33, 41 }, { 36, 28 }, { 27, 47 }, { 21, 62 },
{ 17, -13 }, { 16, -9 }, { 17, -12 }, { 27, -21 },
{ 37, -30 }, { 41, -40 }, { 42, -41 }, { 48, -47 },
{ 39, -32 }, { 46, -40 }, { 52, -51 }, { 46, -41 },
{ 52, -39 }, { 43, -19 }, { 32, 11 }, { 61, -55 },
{ 56, -46 }, { 62, -50 }, { 81, -67 }, { 45, -20 },
{ 35, -2 }, { 28, 15 }, { 34, 1 }, { 39, 1 },
{ 30, 17 }, { 20, 38 }, { 18, 45 }, { 15, 54 },
{ 0, 79 }, { 36, -16 }, { 37, -14 }, { 37, -17 },
{ 32, 1 }, { 34, 15 }, { 29, 15 }, { 24, 25 },
{ 34, 22 }, { 31, 16 }, { 35, 18 }, { 31, 28 },
{ 33, 41 }, { 36, 28 }, { 27, 47 }, { 21, 62 },
{ -24, 115 }, { -22, 82 }, { -9, 62 }, { 0, 53 },
{ 0, 59 }, { -14, 85 }, { -13, 89 }, { -13, 94 },
{ -11, 92 }, { -29, 127 }, { -21, 100 }, { -14, 57 },
{ -12, 67 }, { -11, 71 }, { -10, 77 }, { -21, 85 },
{ -16, 88 }, { -23, 104 }, { -15, 98 }, { -37, 127 },
{ -10, 82 }, { -8, 48 }, { -8, 61 }, { -8, 66 },
{ -7, 70 }, { -14, 75 }, { -10, 79 }, { -9, 83 },
{ -12, 92 }, { -18, 108 }, { -24, 115 }, { -22, 82 },
{ -9, 62 }, { 0, 53 }, { 0, 59 }, { -14, 85 },
{ -13, 89 }, { -13, 94 }, { -11, 92 }, { -29, 127 },
{ -21, 100 }, { -14, 57 }, { -12, 67 }, { -11, 71 },
{ -10, 77 }, { -21, 85 }, { -16, 88 }, { -23, 104 },
{ -15, 98 }, { -37, 127 }, { -10, 82 }, { -8, 48 },
{ -8, 61 }, { -8, 66 }, { -7, 70 }, { -14, 75 },
{ -10, 79 }, { -9, 83 }, { -12, 92 }, { -18, 108 },
{ -5, 79 }, { -11, 104 }, { -11, 91 }, { -30, 127 },
{ -5, 79 }, { -11, 104 }, { -11, 91 }, { -30, 127 },
{ -5, 79 }, { -11, 104 }, { -11, 91 }, { -30, 127 }
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
};
void ff_h264_init_cabac_states(H264Context *h) {
int i;
2x faster ff_h264_init_cabac_states(), 4k cpu cycles less. Sadly this is just per slice so the speedup with normal files should be negligible. Originally committed as revision 21859 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-17 01:43:08 +02:00
const int8_t (*tab)[2];
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
const int slice_qp = av_clip(h->qscale - 6*(h->sps.bit_depth_luma-8), 0, 51);
2x faster ff_h264_init_cabac_states(), 4k cpu cycles less. Sadly this is just per slice so the speedup with normal files should be negligible. Originally committed as revision 21859 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-17 01:43:08 +02:00
Replace deprecated FF_*_TYPE symbols with AV_PICTURE_TYPE_*. Signed-off-by: Diego Biurrun <diego@biurrun.de>
2011-04-29 19:53:57 +03:00
if( h->slice_type_nos == AV_PICTURE_TYPE_I ) tab = cabac_context_init_I;
2x faster ff_h264_init_cabac_states(), 4k cpu cycles less. Sadly this is just per slice so the speedup with normal files should be negligible. Originally committed as revision 21859 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-17 01:43:08 +02:00
else tab = cabac_context_init_PB[h->cabac_init_idc];
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
/* calculate pre-state */
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
for( i= 0; i < 1024; i++ ) {
Add support for higher QP values in h264. In high bit depth, the QP values may now be up to (51 + 6*(bit_depth-8)). Preparatory patch for high bit depth h264 decoding support. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-03-29 18:48:58 +03:00
int pre = 2*(((tab[i][0] * slice_qp) >>4 ) + tab[i][1]) - 127;
2x faster ff_h264_init_cabac_states(), 4k cpu cycles less. Sadly this is just per slice so the speedup with normal files should be negligible. Originally committed as revision 21859 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-17 01:43:08 +02:00
pre^= pre>>31;
if(pre > 124)
pre= 124 + (pre&1);
h->cabac_state[i] = pre;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
}
static int decode_cabac_field_decoding_flag(H264Context *h) {
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
const long mbb_xy = h->mb_xy - 2L*h->mb_stride;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
Optimize decode_cabac_field_decoding_flag(). ~4 cpu cycles faster Originally committed as revision 21447 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-25 04:44:34 +02:00
unsigned long ctx = 0;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
lavc decoders: work with refcounted frames.
2012-11-21 23:34:46 +03:00
ctx += h->mb_field_decoding_flag & !!h->mb_x; //for FMO:(s->current_picture.mb_type[mba_xy] >> 7) & (h->slice_table[mba_xy] == h->slice_num);
ctx += (h->cur_pic.mb_type[mbb_xy] >> 7) & (h->slice_table[mbb_xy] == h->slice_num);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
Optimize decode_cabac_field_decoding_flag(). ~4 cpu cycles faster Originally committed as revision 21447 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-25 04:44:34 +02:00
return get_cabac_noinline( &h->cabac, &(h->cabac_state+70)[ctx] );
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
uint8_t *state= &h->cabac_state[ctx_base];
int mb_type;
if(intra_slice){
int ctx=0;
H.264: template left MB handling Faster H.264 decoding with ALLOW_INTERLACE off.
2011-06-30 01:38:39 +03:00
if( h->left_type[LTOP] & (MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM))
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ctx++;
H.264: template left MB handling Faster H.264 decoding with ALLOW_INTERLACE off.
2011-06-30 01:38:39 +03:00
if( h->top_type & (MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM))
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ctx++;
if( get_cabac_noinline( &h->cabac, &state[ctx] ) == 0 )
return 0; /* I4x4 */
state += 2;
}else{
Cosmetic, get rid of &x[0] Originally committed as revision 21309 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-19 01:55:19 +02:00
if( get_cabac_noinline( &h->cabac, state ) == 0 )
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
return 0; /* I4x4 */
}
if( get_cabac_terminate( &h->cabac ) )
return 25; /* PCM */
mb_type = 1; /* I16x16 */
mb_type += 12 * get_cabac_noinline( &h->cabac, &state[1] ); /* cbp_luma != 0 */
if( get_cabac_noinline( &h->cabac, &state[2] ) ) /* cbp_chroma */
mb_type += 4 + 4 * get_cabac_noinline( &h->cabac, &state[2+intra_slice] );
mb_type += 2 * get_cabac_noinline( &h->cabac, &state[3+intra_slice] );
mb_type += 1 * get_cabac_noinline( &h->cabac, &state[3+2*intra_slice] );
return mb_type;
}
static int decode_cabac_mb_skip( H264Context *h, int mb_x, int mb_y ) {
int mba_xy, mbb_xy;
int ctx = 0;
h264: add a parameter to the FRAME_MBAFF macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
if (FRAME_MBAFF(h)) { //FIXME merge with the stuff in fill_caches?
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
int mb_xy = mb_x + (mb_y&~1)*h->mb_stride;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
mba_xy = mb_xy - 1;
if( (mb_y&1)
&& h->slice_table[mba_xy] == h->slice_num
h264: add a parameter to the MB_FIELD macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
&& MB_FIELD(h) == !!IS_INTERLACED( h->cur_pic.mb_type[mba_xy] ) )
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
mba_xy += h->mb_stride;
h264: add a parameter to the MB_FIELD macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
if (MB_FIELD(h)) {
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
mbb_xy = mb_xy - h->mb_stride;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
if( !(mb_y&1)
&& h->slice_table[mbb_xy] == h->slice_num
lavc decoders: work with refcounted frames.
2012-11-21 23:34:46 +03:00
&& IS_INTERLACED( h->cur_pic.mb_type[mbb_xy] ) )
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
mbb_xy -= h->mb_stride;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}else
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
mbb_xy = mb_x + (mb_y-1)*h->mb_stride;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}else{
int mb_xy = h->mb_xy;
mba_xy = mb_xy - 1;
h264: add a parameter to the FIELD_PICTURE macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
mbb_xy = mb_xy - (h->mb_stride << FIELD_PICTURE(h));
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
lavc decoders: work with refcounted frames.
2012-11-21 23:34:46 +03:00
if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP(h->cur_pic.mb_type[mba_xy] ))
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ctx++;
lavc decoders: work with refcounted frames.
2012-11-21 23:34:46 +03:00
if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP(h->cur_pic.mb_type[mbb_xy] ))
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ctx++;
Replace deprecated FF_*_TYPE symbols with AV_PICTURE_TYPE_*. Signed-off-by: Diego Biurrun <diego@biurrun.de>
2011-04-29 19:53:57 +03:00
if( h->slice_type_nos == AV_PICTURE_TYPE_B )
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ctx += 13;
return get_cabac_noinline( &h->cabac, &h->cabac_state[11+ctx] );
}
static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
int mode = 0;
if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
return pred_mode;
mode += 1 * get_cabac( &h->cabac, &h->cabac_state[69] );
mode += 2 * get_cabac( &h->cabac, &h->cabac_state[69] );
mode += 4 * get_cabac( &h->cabac, &h->cabac_state[69] );
Simplify decode_cabac_mb_intra4x4_pred_mode(). same speed Originally committed as revision 21839 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 21:22:09 +02:00
return mode + ( mode >= pred_mode );
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
const int mba_xy = h->left_mb_xy[0];
const int mbb_xy = h->top_mb_xy;
int ctx = 0;
/* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
H.264: template left MB handling Faster H.264 decoding with ALLOW_INTERLACE off.
2011-06-30 01:38:39 +03:00
if( h->left_type[LTOP] && h->chroma_pred_mode_table[mba_xy] != 0 )
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ctx++;
H.264: template left MB handling Faster H.264 decoding with ALLOW_INTERLACE off.
2011-06-30 01:38:39 +03:00
if( h->top_type && h->chroma_pred_mode_table[mbb_xy] != 0 )
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ctx++;
if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
return 0;
if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+3] ) == 0 )
return 1;
if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+3] ) == 0 )
return 2;
else
return 3;
}
static int decode_cabac_mb_cbp_luma( H264Context *h) {
int cbp_b, cbp_a, ctx, cbp = 0;
Remove slice_table checks from decode_cabac_mb_cbp_luma() and set left/top_cbp so these checks arent needed. Originally committed as revision 21819 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-14 04:08:48 +02:00
cbp_a = h->left_cbp;
cbp_b = h->top_cbp;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ctx = !(cbp_a & 0x02) + 2 * !(cbp_b & 0x04);
h264: Use + instead of | in some places 6 insns less on x86-64/gcc 4.2. Originally committed as revision 22692 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-26 07:04:03 +02:00
cbp += get_cabac_noinline(&h->cabac, &h->cabac_state[73 + ctx]);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ctx = !(cbp & 0x01) + 2 * !(cbp_b & 0x08);
h264: Use + instead of | in some places 6 insns less on x86-64/gcc 4.2. Originally committed as revision 22692 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-26 07:04:03 +02:00
cbp += get_cabac_noinline(&h->cabac, &h->cabac_state[73 + ctx]) << 1;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ctx = !(cbp_a & 0x08) + 2 * !(cbp & 0x01);
h264: Use + instead of | in some places 6 insns less on x86-64/gcc 4.2. Originally committed as revision 22692 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-26 07:04:03 +02:00
cbp += get_cabac_noinline(&h->cabac, &h->cabac_state[73 + ctx]) << 2;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ctx = !(cbp & 0x04) + 2 * !(cbp & 0x02);
h264: Use + instead of | in some places 6 insns less on x86-64/gcc 4.2. Originally committed as revision 22692 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-26 07:04:03 +02:00
cbp += get_cabac_noinline(&h->cabac, &h->cabac_state[73 + ctx]) << 3;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
return cbp;
}
static int decode_cabac_mb_cbp_chroma( H264Context *h) {
int ctx;
int cbp_a, cbp_b;
cbp_a = (h->left_cbp>>4)&0x03;
cbp_b = (h-> top_cbp>>4)&0x03;
ctx = 0;
if( cbp_a > 0 ) ctx++;
if( cbp_b > 0 ) ctx += 2;
if( get_cabac_noinline( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
return 0;
ctx = 4;
if( cbp_a == 2 ) ctx++;
if( cbp_b == 2 ) ctx += 2;
return 1 + get_cabac_noinline( &h->cabac, &h->cabac_state[77 + ctx] );
}
static int decode_cabac_p_mb_sub_type( H264Context *h ) {
if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
return 0; /* 8x8 */
if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
return 1; /* 8x4 */
if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
return 2; /* 4x8 */
return 3; /* 4x4 */
}
static int decode_cabac_b_mb_sub_type( H264Context *h ) {
int type;
if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )
return 0; /* B_Direct_8x8 */
if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )
return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */
type = 3;
if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {
if( get_cabac( &h->cabac, &h->cabac_state[39] ) )
return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */
type += 4;
}
type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );
type += get_cabac( &h->cabac, &h->cabac_state[39] );
return type;
}
static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
int refa = h->ref_cache[list][scan8[n] - 1];
int refb = h->ref_cache[list][scan8[n] - 8];
int ref = 0;
int ctx = 0;
Replace deprecated FF_*_TYPE symbols with AV_PICTURE_TYPE_*. Signed-off-by: Diego Biurrun <diego@biurrun.de>
2011-04-29 19:53:57 +03:00
if( h->slice_type_nos == AV_PICTURE_TYPE_B) {
Store sub_mb_type in direct_cache/direct_table. This is equal complexity but could be more usefull. Originally committed as revision 21821 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-14 16:41:27 +02:00
if( refa > 0 && !(h->direct_cache[scan8[n] - 1]&(MB_TYPE_DIRECT2>>1)) )
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ctx++;
Store sub_mb_type in direct_cache/direct_table. This is equal complexity but could be more usefull. Originally committed as revision 21821 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-14 16:41:27 +02:00
if( refb > 0 && !(h->direct_cache[scan8[n] - 8]&(MB_TYPE_DIRECT2>>1)) )
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ctx += 2;
} else {
if( refa > 0 )
ctx++;
if( refb > 0 )
ctx += 2;
}
while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
ref++;
ctx = (ctx>>2)+4;
if(ref >= 32 /*h->ref_list[list]*/){
return -1;
}
}
return ref;
}
Calculate mvd without abs() same speed (ask gcc why, i dont know) Originally committed as revision 22035 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:50:02 +02:00
static int decode_cabac_mb_mvd( H264Context *h, int ctxbase, int amvd, int *mvda) {
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
int mvd;
Optimize (amvd>2)+(amvd>32), about 1 cpu cycles faster. patch by Zhou Zongyi @ zhouzy () os punkt pku dot edu speck cn Originally committed as revision 22084 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-27 00:45:35 +02:00
if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+((amvd-3)>>(INT_BIT-1))+((amvd-33)>>(INT_BIT-1))+2])){
// if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+(amvd>2)+(amvd>32)])){
Calculate mvd without abs() same speed (ask gcc why, i dont know) Originally committed as revision 22035 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:50:02 +02:00
*mvda= 0;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
return 0;
Calculate mvd without abs() same speed (ask gcc why, i dont know) Originally committed as revision 22035 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:50:02 +02:00
}
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
mvd= 1;
Get rid of a local variable, 10 cpu cycles faster. Originally committed as revision 21888 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-19 04:37:11 +02:00
ctxbase+= 3;
while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase] ) ) {
if( mvd < 4 )
ctxbase++;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
mvd++;
}
if( mvd >= 9 ) {
int k = 3;
while( get_cabac_bypass( &h->cabac ) ) {
mvd += 1 << k;
k++;
if(k>24){
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
av_log(h->avctx, AV_LOG_ERROR, "overflow in decode_cabac_mb_mvd\n");
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
return INT_MIN;
}
}
while( k-- ) {
get rid of an if() 1 cpu cycle faster. Originally committed as revision 21889 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-19 05:10:26 +02:00
mvd += get_cabac_bypass( &h->cabac )<<k;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster for high resolution videos. about 20cycles faster per MB for cathederal. Originally committed as revision 22038 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 22:43:06 +02:00
*mvda=mvd < 70 ? mvd : 70;
}else
*mvda=mvd;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
return get_cabac_bypass_sign( &h->cabac, -mvd );
}
Factorize common code from the top of decode_cabac_mb_mvd() 10-15 cpu cycles faster. Originally committed as revision 22029 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:06:02 +02:00
#define DECODE_CABAC_MB_MVD( h, list, n )\
{\
int amvd0 = h->mvd_cache[list][scan8[n] - 1][0] +\
h->mvd_cache[list][scan8[n] - 8][0];\
int amvd1 = h->mvd_cache[list][scan8[n] - 1][1] +\
h->mvd_cache[list][scan8[n] - 8][1];\
\
Calculate mvd without abs() same speed (ask gcc why, i dont know) Originally committed as revision 22035 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:50:02 +02:00
mx += decode_cabac_mb_mvd( h, 40, amvd0, &mpx );\
my += decode_cabac_mb_mvd( h, 47, amvd1, &mpy );\
Factorize common code from the top of decode_cabac_mb_mvd() 10-15 cpu cycles faster. Originally committed as revision 22029 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:06:02 +02:00
}
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
static av_always_inline int get_cabac_cbf_ctx( H264Context *h, int cat, int idx, int max_coeff, int is_dc ) {
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
int nza, nzb;
int ctx = 0;
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
static const uint16_t base_ctx[14] = {85,89,93,97,101,1012,460,464,468,1016,472,476,480,1020};
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
if( is_dc ) {
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
if( cat == 3 ) {
H.264: switch to x264-style tracking of luma/chroma DC NNZ Useful so that we don't have to run the hierarchical DC iDCT if there aren't any coefficients. Opens up some future opportunities for optimization as well. Originally committed as revision 26337 to svn://svn.ffmpeg.org/ffmpeg/trunk
2011-01-14 23:36:16 +02:00
idx -= CHROMA_DC_BLOCK_INDEX;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
nza = (h->left_cbp>>(6+idx))&0x01;
nzb = (h-> top_cbp>>(6+idx))&0x01;
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
} else {
idx -= LUMA_DC_BLOCK_INDEX;
nza = h->left_cbp&(0x100<<idx);
nzb = h-> top_cbp&(0x100<<idx);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
} else {
nza = h->non_zero_count_cache[scan8[idx] - 1];
nzb = h->non_zero_count_cache[scan8[idx] - 8];
}
if( nza > 0 )
ctx++;
if( nzb > 0 )
ctx += 2;
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
return base_ctx[cat] + ctx;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
static av_always_inline void
Drop DCTELEM typedef It does not help as an abstraction and adds dsputil dependencies. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2013-01-20 03:02:29 +03:00
decode_cabac_residual_internal(H264Context *h, int16_t *block,
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
int cat, int n, const uint8_t *scantable,
const uint32_t *qmul, int max_coeff,
int is_dc, int chroma422)
{
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
static const int significant_coeff_flag_offset[2][14] = {
{ 105+0, 105+15, 105+29, 105+44, 105+47, 402, 484+0, 484+15, 484+29, 660, 528+0, 528+15, 528+29, 718 },
{ 277+0, 277+15, 277+29, 277+44, 277+47, 436, 776+0, 776+15, 776+29, 675, 820+0, 820+15, 820+29, 733 }
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
};
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
static const int last_coeff_flag_offset[2][14] = {
{ 166+0, 166+15, 166+29, 166+44, 166+47, 417, 572+0, 572+15, 572+29, 690, 616+0, 616+15, 616+29, 748 },
{ 338+0, 338+15, 338+29, 338+44, 338+47, 451, 864+0, 864+15, 864+29, 699, 908+0, 908+15, 908+29, 757 }
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
};
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
static const int coeff_abs_level_m1_offset[14] = {
227+0, 227+10, 227+20, 227+30, 227+39, 426, 952+0, 952+10, 952+20, 708, 982+0, 982+10, 982+20, 766
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
};
static const uint8_t significant_coeff_flag_offset_8x8[2][63] = {
{ 0, 1, 2, 3, 4, 5, 5, 4, 4, 3, 3, 4, 4, 4, 5, 5,
4, 4, 4, 4, 3, 3, 6, 7, 7, 7, 8, 9,10, 9, 8, 7,
7, 6,11,12,13,11, 6, 7, 8, 9,14,10, 9, 8, 6,11,
12,13,11, 6, 9,14,10, 9,11,12,13,11,14,10,12 },
{ 0, 1, 1, 2, 2, 3, 3, 4, 5, 6, 7, 7, 7, 8, 4, 5,
6, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,11,12,11,
9, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,13,13, 9,
9,10,10, 8,13,13, 9, 9,10,10,14,14,14,14,14 }
};
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
static const uint8_t sig_coeff_offset_dc[7] = { 0, 0, 1, 1, 2, 2, 2 };
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
/* node ctx: 0..3: abslevel1 (with abslevelgt1 == 0).
* 4..7: abslevelgt1 + 3 (and abslevel1 doesn't matter).
* map node ctx => cabac ctx for level=1 */
static const uint8_t coeff_abs_level1_ctx[8] = { 1, 2, 3, 4, 0, 0, 0, 0 };
/* map node ctx => cabac ctx for level>1 */
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
static const uint8_t coeff_abs_levelgt1_ctx[2][8] = {
{ 5, 5, 5, 5, 6, 7, 8, 9 },
{ 5, 5, 5, 5, 6, 7, 8, 8 }, // 422/dc case
};
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
static const uint8_t coeff_abs_level_transition[2][8] = {
/* update node ctx after decoding a level=1 */
{ 1, 2, 3, 3, 4, 5, 6, 7 },
/* update node ctx after decoding a level>1 */
{ 4, 4, 4, 4, 5, 6, 7, 7 }
};
int index[64];
int av_unused last;
int coeff_count = 0;
int node_ctx = 0;
uint8_t *significant_coeff_ctx_base;
uint8_t *last_coeff_ctx_base;
uint8_t *abs_level_m1_ctx_base;
#if !ARCH_X86
#define CABAC_ON_STACK
#endif
#ifdef CABAC_ON_STACK
#define CC &cc
CABACContext cc;
cc.range = h->cabac.range;
cc.low = h->cabac.low;
cc.bytestream= h->cabac.bytestream;
h264: fix cabac-on-stack after safe cabac reader.
2012-03-29 02:32:27 +03:00
cc.bytestream_end = h->cabac.bytestream_end;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
#else
#define CC &h->cabac
#endif
significant_coeff_ctx_base = h->cabac_state
h264: add a parameter to the MB_FIELD macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
+ significant_coeff_flag_offset[MB_FIELD(h)][cat];
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
last_coeff_ctx_base = h->cabac_state
h264: add a parameter to the MB_FIELD macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
+ last_coeff_flag_offset[MB_FIELD(h)][cat];
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
abs_level_m1_ctx_base = h->cabac_state
+ coeff_abs_level_m1_offset[cat];
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
if( !is_dc && max_coeff == 64 ) {
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
#define DECODE_SIGNIFICANCE( coefs, sig_off, last_off ) \
for(last= 0; last < coefs; last++) { \
uint8_t *sig_ctx = significant_coeff_ctx_base + sig_off; \
if( get_cabac( CC, sig_ctx )) { \
uint8_t *last_ctx = last_coeff_ctx_base + last_off; \
index[coeff_count++] = last; \
if( get_cabac( CC, last_ctx ) ) { \
last= max_coeff; \
break; \
} \
} \
}\
if( last == max_coeff -1 ) {\
index[coeff_count++] = last;\
}
h264: add a parameter to the MB_FIELD macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
const uint8_t *sig_off = significant_coeff_flag_offset_8x8[MB_FIELD(h)];
h264: use asm cabac reader under a generic condition This removes a dependency on implementation details from generic code and allows easy addition of the equivalent optimisation for other architectures than x86. Signed-off-by: Mans Rullgard <mans@mansr.com>
2012-06-22 20:52:27 +03:00
#ifdef decode_significance
coeff_count = decode_significance_8x8(CC, significant_coeff_ctx_base, index,
H.264: optimize CABAC x86 asm for Atom
2011-07-27 05:08:05 +03:00
last_coeff_ctx_base, sig_off);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
} else {
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
if (is_dc && chroma422) { // dc 422
DECODE_SIGNIFICANCE(7, sig_coeff_offset_dc[last], sig_coeff_offset_dc[last]);
} else {
h264: use asm cabac reader under a generic condition This removes a dependency on implementation details from generic code and allows easy addition of the equivalent optimisation for other architectures than x86. Signed-off-by: Mans Rullgard <mans@mansr.com>
2012-06-22 20:52:27 +03:00
coeff_count = decode_significance(CC, max_coeff, significant_coeff_ctx_base, index,
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
last_coeff_ctx_base-significant_coeff_ctx_base);
}
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
#else
h264: use one table instead of several for cabac functions The reason is this is easier for PIC code (in particular on darwin...). Keep the old names as pointers (static in cabac_functions.h so gcc knows these are just immediate offsets) so the c code can nicely stay the same (alternatively could use offsets directly in the functions needing the tables). This should produce the same code as before with non-pic and better code (confirmed) with pic. The assembly uses the new table but still won't work for PIC case. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2012-04-27 23:19:38 +03:00
DECODE_SIGNIFICANCE( 63, sig_off[last], ff_h264_last_coeff_flag_offset_8x8[last] );
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
} else {
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
if (is_dc && chroma422) { // dc 422
DECODE_SIGNIFICANCE(7, sig_coeff_offset_dc[last], sig_coeff_offset_dc[last]);
} else {
DECODE_SIGNIFICANCE(max_coeff - 1, last, last);
}
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
#endif
}
assert(coeff_count > 0);
if( is_dc ) {
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
if( cat == 3 )
Roll back 4:4:4 H.264 for now Needs some ARM/PPC asm modifications.
2011-06-13 23:38:46 +03:00
h->cbp_table[h->mb_xy] |= 0x40 << (n - CHROMA_DC_BLOCK_INDEX);
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
else
h->cbp_table[h->mb_xy] |= 0x100 << (n - LUMA_DC_BLOCK_INDEX);
H.264: switch to x264-style tracking of luma/chroma DC NNZ Useful so that we don't have to run the hierarchical DC iDCT if there aren't any coefficients. Opens up some future opportunities for optimization as well. Originally committed as revision 26337 to svn://svn.ffmpeg.org/ffmpeg/trunk
2011-01-14 23:36:16 +02:00
h->non_zero_count_cache[scan8[n]] = coeff_count;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
} else {
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
if( max_coeff == 64 )
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
fill_rectangle(&h->non_zero_count_cache[scan8[n]], 2, 2, 8, coeff_count, 1);
else {
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
assert( cat == 1 || cat == 2 || cat == 4 || cat == 7 || cat == 8 || cat == 11 || cat == 12 );
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
h->non_zero_count_cache[scan8[n]] = coeff_count;
}
}
Add the notion of pixel size in h264 related functions. In high bit depth the pixels will not be stored in uint8_t like in the normal case, but in uint16_t. The pixel size is thus 1 in normal bit depth and 2 in high bit depth. Preparatory patch for high bit depth h264 decoding support. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-03-29 18:48:57 +03:00
#define STORE_BLOCK(type) \
do { \
uint8_t *ctx = coeff_abs_level1_ctx[node_ctx] + abs_level_m1_ctx_base; \
\
int j= scantable[index[--coeff_count]]; \
\
if( get_cabac( CC, ctx ) == 0 ) { \
node_ctx = coeff_abs_level_transition[0][node_ctx]; \
if( is_dc ) { \
((type*)block)[j] = get_cabac_bypass_sign( CC, -1); \
}else{ \
((type*)block)[j] = (get_cabac_bypass_sign( CC, -qmul[j]) + 32) >> 6; \
} \
} else { \
int coeff_abs = 2; \
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
ctx = coeff_abs_levelgt1_ctx[is_dc && chroma422][node_ctx] + abs_level_m1_ctx_base; \
Add the notion of pixel size in h264 related functions. In high bit depth the pixels will not be stored in uint8_t like in the normal case, but in uint16_t. The pixel size is thus 1 in normal bit depth and 2 in high bit depth. Preparatory patch for high bit depth h264 decoding support. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-03-29 18:48:57 +03:00
node_ctx = coeff_abs_level_transition[1][node_ctx]; \
\
while( coeff_abs < 15 && get_cabac( CC, ctx ) ) { \
coeff_abs++; \
} \
\
if( coeff_abs >= 15 ) { \
int j = 0; \
while( get_cabac_bypass( CC ) ) { \
j++; \
} \
\
coeff_abs=1; \
while( j-- ) { \
coeff_abs += coeff_abs + get_cabac_bypass( CC ); \
} \
coeff_abs+= 14; \
} \
\
if( is_dc ) { \
((type*)block)[j] = get_cabac_bypass_sign( CC, -coeff_abs ); \
}else{ \
((type*)block)[j] = ((int)(get_cabac_bypass_sign( CC, -coeff_abs ) * qmul[j] + 32)) >> 6; \
} \
} \
} while ( coeff_count );
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
Add the notion of pixel size in h264 related functions. In high bit depth the pixels will not be stored in uint8_t like in the normal case, but in uint16_t. The pixel size is thus 1 in normal bit depth and 2 in high bit depth. Preparatory patch for high bit depth h264 decoding support. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-03-29 18:48:57 +03:00
if (h->pixel_shift) {
STORE_BLOCK(int32_t)
} else {
STORE_BLOCK(int16_t)
}
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
#ifdef CABAC_ON_STACK
h->cabac.range = cc.range ;
h->cabac.low = cc.low ;
h->cabac.bytestream= cc.bytestream;
#endif
}
Drop DCTELEM typedef It does not help as an abstraction and adds dsputil dependencies. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2013-01-20 03:02:29 +03:00
static void decode_cabac_residual_dc_internal(H264Context *h, int16_t *block,
int cat, int n,
const uint8_t *scantable,
int max_coeff)
{
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
decode_cabac_residual_internal(h, block, cat, n, scantable, NULL, max_coeff, 1, 0);
}
Drop DCTELEM typedef It does not help as an abstraction and adds dsputil dependencies. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2013-01-20 03:02:29 +03:00
static void decode_cabac_residual_dc_internal_422(H264Context *h, int16_t *block,
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
int cat, int n, const uint8_t *scantable,
int max_coeff)
{
decode_cabac_residual_internal(h, block, cat, n, scantable, NULL, max_coeff, 1, 1);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
Drop DCTELEM typedef It does not help as an abstraction and adds dsputil dependencies. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2013-01-20 03:02:29 +03:00
static void decode_cabac_residual_nondc_internal(H264Context *h, int16_t *block,
int cat, int n,
const uint8_t *scantable,
const uint32_t *qmul,
int max_coeff)
{
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
decode_cabac_residual_internal(h, block, cat, n, scantable, qmul, max_coeff, 0, 0);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
H.264: Partially inline CABAC residual decoding Improves CABAC performance about ~1.2%. Trick originates from x264 and has also been used in ffvp8. It's useful because coded block flags are usually zero, so it helps to have the early termination inlined into the main function. Originally committed as revision 26375 to svn://svn.ffmpeg.org/ffmpeg/trunk
2011-01-15 19:52:48 +02:00
/* cat: 0-> DC 16x16 n = 0
* 1-> AC 16x16 n = luma4x4idx
* 2-> Luma4x4 n = luma4x4idx
* 3-> DC Chroma n = iCbCr
* 4-> AC Chroma n = 16 + 4 * iCbCr + chroma4x4idx
* 5-> Luma8x8 n = 4 * luma8x8idx */
/* Partially inline the CABAC residual decode: inline the coded block flag.
* This has very little impact on binary size and improves performance
* because it allows improved constant propagation into get_cabac_cbf_ctx,
* as well as because most blocks have zero CBFs. */
Drop DCTELEM typedef It does not help as an abstraction and adds dsputil dependencies. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2013-01-20 03:02:29 +03:00
static av_always_inline void decode_cabac_residual_dc(H264Context *h,
int16_t *block,
int cat, int n,
const uint8_t *scantable,
int max_coeff)
{
H.264: Partially inline CABAC residual decoding Improves CABAC performance about ~1.2%. Trick originates from x264 and has also been used in ffvp8. It's useful because coded block flags are usually zero, so it helps to have the early termination inlined into the main function. Originally committed as revision 26375 to svn://svn.ffmpeg.org/ffmpeg/trunk
2011-01-15 19:52:48 +02:00
/* read coded block flag */
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
if( get_cabac( &h->cabac, &h->cabac_state[get_cabac_cbf_ctx( h, cat, n, max_coeff, 1 ) ] ) == 0 ) {
H.264: Partially inline CABAC residual decoding Improves CABAC performance about ~1.2%. Trick originates from x264 and has also been used in ffvp8. It's useful because coded block flags are usually zero, so it helps to have the early termination inlined into the main function. Originally committed as revision 26375 to svn://svn.ffmpeg.org/ffmpeg/trunk
2011-01-15 19:52:48 +02:00
h->non_zero_count_cache[scan8[n]] = 0;
return;
}
decode_cabac_residual_dc_internal( h, block, cat, n, scantable, max_coeff );
}
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
static av_always_inline void
Drop DCTELEM typedef It does not help as an abstraction and adds dsputil dependencies. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2013-01-20 03:02:29 +03:00
decode_cabac_residual_dc_422(H264Context *h, int16_t *block,
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
int cat, int n, const uint8_t *scantable,
int max_coeff)
{
/* read coded block flag */
if (get_cabac(&h->cabac, &h->cabac_state[get_cabac_cbf_ctx(h, cat, n, max_coeff, 1)]) == 0) {
h->non_zero_count_cache[scan8[n]] = 0;
return;
}
decode_cabac_residual_dc_internal_422(h, block, cat, n, scantable, max_coeff);
}
Drop DCTELEM typedef It does not help as an abstraction and adds dsputil dependencies. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2013-01-20 03:02:29 +03:00
static av_always_inline void decode_cabac_residual_nondc(H264Context *h,
int16_t *block,
int cat, int n,
const uint8_t *scantable,
const uint32_t *qmul,
int max_coeff)
{
H.264: Partially inline CABAC residual decoding Improves CABAC performance about ~1.2%. Trick originates from x264 and has also been used in ffvp8. It's useful because coded block flags are usually zero, so it helps to have the early termination inlined into the main function. Originally committed as revision 26375 to svn://svn.ffmpeg.org/ffmpeg/trunk
2011-01-15 19:52:48 +02:00
/* read coded block flag */
h264: add a parameter to the CHROMA444 macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
if( (cat != 5 || CHROMA444(h)) && get_cabac( &h->cabac, &h->cabac_state[get_cabac_cbf_ctx( h, cat, n, max_coeff, 0 ) ] ) == 0 ) {
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
if( max_coeff == 64 ) {
fill_rectangle(&h->non_zero_count_cache[scan8[n]], 2, 2, 8, 0, 1);
} else {
h->non_zero_count_cache[scan8[n]] = 0;
}
H.264: Partially inline CABAC residual decoding Improves CABAC performance about ~1.2%. Trick originates from x264 and has also been used in ffvp8. It's useful because coded block flags are usually zero, so it helps to have the early termination inlined into the main function. Originally committed as revision 26375 to svn://svn.ffmpeg.org/ffmpeg/trunk
2011-01-15 19:52:48 +02:00
return;
}
decode_cabac_residual_nondc_internal( h, block, cat, n, scantable, qmul, max_coeff );
}
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
static av_always_inline void decode_cabac_luma_residual( H264Context *h, const uint8_t *scan, const uint8_t *scan8x8, int pixel_shift, int mb_type, int cbp, int p )
{
static const uint8_t ctx_cat[4][3] = {{0,6,10},{1,7,11},{2,8,12},{5,9,13}};
const uint32_t *qmul;
int i8x8, i4x4;
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
int qscale = p == 0 ? h->qscale : h->chroma_qp[p-1];
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
if( IS_INTRA16x16( mb_type ) ) {
AV_ZERO128(h->mb_luma_dc[p]+0);
AV_ZERO128(h->mb_luma_dc[p]+8);
AV_ZERO128(h->mb_luma_dc[p]+16);
AV_ZERO128(h->mb_luma_dc[p]+24);
decode_cabac_residual_dc(h, h->mb_luma_dc[p], ctx_cat[0][p], LUMA_DC_BLOCK_INDEX+p, scan, 16);
if( cbp&15 ) {
qmul = h->dequant4_coeff[p][qscale];
for( i4x4 = 0; i4x4 < 16; i4x4++ ) {
const int index = 16*p + i4x4;
decode_cabac_residual_nondc(h, h->mb + (16*index << pixel_shift), ctx_cat[1][p], index, scan + 1, qmul, 15);
}
} else {
fill_rectangle(&h->non_zero_count_cache[scan8[16*p]], 4, 4, 8, 0, 1);
}
} else {
int cqm = (IS_INTRA( mb_type ) ? 0:3) + p;
for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
if( cbp & (1<<i8x8) ) {
if( IS_8x8DCT(mb_type) ) {
const int index = 16*p + 4*i8x8;
decode_cabac_residual_nondc(h, h->mb + (16*index << pixel_shift), ctx_cat[3][p], index,
scan8x8, h->dequant8_coeff[cqm][qscale], 64);
} else {
qmul = h->dequant4_coeff[cqm][qscale];
for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
const int index = 16*p + 4*i8x8 + i4x4;
//START_TIMER
decode_cabac_residual_nondc(h, h->mb + (16*index << pixel_shift), ctx_cat[2][p], index, scan, qmul, 16);
//STOP_TIMER("decode_residual")
}
}
} else {
H.264: use fill_rectangle in CABAC decoding
2011-07-06 23:25:13 +03:00
fill_rectangle(&h->non_zero_count_cache[scan8[4*i8x8+16*p]], 2, 2, 8, 0, 1);
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
}
}
}
}
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
/**
doxygen: misc consistency, spelling and wording fixes
2011-12-07 15:03:53 +03:00
* Decode a macroblock.
error_resilience: use the ER_ namespace Add the namespace to {AC_,DC_,MV_}{END,ERROR} macros Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
2011-11-28 02:31:51 +03:00
* @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR if an error is noticed
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
*/
int ff_h264_decode_mb_cabac(H264Context *h) {
int mb_xy;
int mb_type, partition_count, cbp = 0;
int dct8x8_allowed= h->pps.transform_8x8_mode;
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
int decode_chroma = h->sps.chroma_format_idc == 1 || h->sps.chroma_format_idc == 2;
Add the notion of pixel size in h264 related functions. In high bit depth the pixels will not be stored in uint8_t like in the normal case, but in uint16_t. The pixel size is thus 1 in normal bit depth and 2 in high bit depth. Preparatory patch for high bit depth h264 decoding support. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-03-29 18:48:57 +03:00
const int pixel_shift = h->pixel_shift;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
mb_xy = h->mb_xy = h->mb_x + h->mb_y*h->mb_stride;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
tprintf(h->avctx, "pic:%d mb:%d/%d\n", h->frame_num, h->mb_x, h->mb_y);
Replace deprecated FF_*_TYPE symbols with AV_PICTURE_TYPE_*. Signed-off-by: Diego Biurrun <diego@biurrun.de>
2011-04-29 19:53:57 +03:00
if( h->slice_type_nos != AV_PICTURE_TYPE_I ) {
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
int skip;
/* a skipped mb needs the aff flag from the following mb */
h264: add a parameter to the FRAME_MBAFF macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
if (FRAME_MBAFF(h) && (h->mb_y & 1) == 1 && h->prev_mb_skipped)
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
skip = h->next_mb_skipped;
else
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
skip = decode_cabac_mb_skip( h, h->mb_x, h->mb_y );
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
/* read skip flags */
if( skip ) {
h264: add a parameter to the FRAME_MBAFF macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
if (FRAME_MBAFF(h) && (h->mb_y & 1) == 0) {
lavc decoders: work with refcounted frames.
2012-11-21 23:34:46 +03:00
h->cur_pic.mb_type[mb_xy] = MB_TYPE_SKIP;
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
h->next_mb_skipped = decode_cabac_mb_skip( h, h->mb_x, h->mb_y+1 );
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
if(!h->next_mb_skipped)
h->mb_mbaff = h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);
}
decode_mb_skip(h);
h->cbp_table[mb_xy] = 0;
h->chroma_pred_mode_table[mb_xy] = 0;
h->last_qscale_diff = 0;
return 0;
}
}
h264: add a parameter to the FRAME_MBAFF macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
if (FRAME_MBAFF(h)) {
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
if( (h->mb_y&1) == 0 )
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
h->mb_mbaff =
h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);
}
h->prev_mb_skipped = 0;
h264: add a parameter to the MB_FIELD macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
fill_decode_neighbors(h, -(MB_FIELD(h)));
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
Replace deprecated FF_*_TYPE symbols with AV_PICTURE_TYPE_*. Signed-off-by: Diego Biurrun <diego@biurrun.de>
2011-04-29 19:53:57 +03:00
if( h->slice_type_nos == AV_PICTURE_TYPE_B ) {
Merge decode_cabac_mb_type_b() into calling code. This avoids a conditional branch and is about 3 cpu cyclues faster. Originally committed as revision 21838 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 21:20:49 +02:00
int ctx = 0;
Replace deprecated FF_*_TYPE symbols with AV_PICTURE_TYPE_*. Signed-off-by: Diego Biurrun <diego@biurrun.de>
2011-04-29 19:53:57 +03:00
assert(h->slice_type_nos == AV_PICTURE_TYPE_B);
Merge decode_cabac_mb_type_b() into calling code. This avoids a conditional branch and is about 3 cpu cyclues faster. Originally committed as revision 21838 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 21:20:49 +02:00
H.264: template left MB handling Faster H.264 decoding with ALLOW_INTERLACE off.
2011-06-30 01:38:39 +03:00
if( !IS_DIRECT( h->left_type[LTOP]-1 ) )
Merge decode_cabac_mb_type_b() into calling code. This avoids a conditional branch and is about 3 cpu cyclues faster. Originally committed as revision 21838 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 21:20:49 +02:00
ctx++;
Drop a few redundant slice_num checks. Originally committed as revision 21844 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-16 02:09:30 +02:00
if( !IS_DIRECT( h->top_type-1 ) )
Merge decode_cabac_mb_type_b() into calling code. This avoids a conditional branch and is about 3 cpu cyclues faster. Originally committed as revision 21838 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 21:20:49 +02:00
ctx++;
if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+ctx] ) ){
mb_type= 0; /* B_Direct_16x16 */
}else if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+3] ) ) {
mb_type= 1 + get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}else{
Merge decode_cabac_mb_type_b() into calling code. This avoids a conditional branch and is about 3 cpu cyclues faster. Originally committed as revision 21838 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 21:20:49 +02:00
int bits;
bits = get_cabac_noinline( &h->cabac, &h->cabac_state[27+4] ) << 3;
h264: Use + instead of | in some places 6 insns less on x86-64/gcc 4.2. Originally committed as revision 22692 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-26 07:04:03 +02:00
bits+= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ) << 2;
bits+= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ) << 1;
bits+= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] );
Merge decode_cabac_mb_type_b() into calling code. This avoids a conditional branch and is about 3 cpu cyclues faster. Originally committed as revision 21838 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 21:20:49 +02:00
if( bits < 8 ){
mb_type= bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
}else if( bits == 13 ){
mb_type= decode_cabac_intra_mb_type(h, 32, 0);
goto decode_intra_mb;
}else if( bits == 14 ){
mb_type= 11; /* B_L1_L0_8x16 */
}else if( bits == 15 ){
mb_type= 22; /* B_8x8 */
}else{
h264: Use + instead of | in some places 6 insns less on x86-64/gcc 4.2. Originally committed as revision 22692 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-26 07:04:03 +02:00
bits= ( bits<<1 ) + get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] );
Merge decode_cabac_mb_type_b() into calling code. This avoids a conditional branch and is about 3 cpu cyclues faster. Originally committed as revision 21838 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 21:20:49 +02:00
mb_type= bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
}
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
Merge decode_cabac_mb_type_b() into calling code. This avoids a conditional branch and is about 3 cpu cyclues faster. Originally committed as revision 21838 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 21:20:49 +02:00
partition_count= b_mb_type_info[mb_type].partition_count;
mb_type= b_mb_type_info[mb_type].type;
Replace deprecated FF_*_TYPE symbols with AV_PICTURE_TYPE_*. Signed-off-by: Diego Biurrun <diego@biurrun.de>
2011-04-29 19:53:57 +03:00
} else if( h->slice_type_nos == AV_PICTURE_TYPE_P ) {
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
if( get_cabac_noinline( &h->cabac, &h->cabac_state[14] ) == 0 ) {
/* P-type */
if( get_cabac_noinline( &h->cabac, &h->cabac_state[15] ) == 0 ) {
/* P_L0_D16x16, P_8x8 */
mb_type= 3 * get_cabac_noinline( &h->cabac, &h->cabac_state[16] );
} else {
/* P_L0_D8x16, P_L0_D16x8 */
mb_type= 2 - get_cabac_noinline( &h->cabac, &h->cabac_state[17] );
}
partition_count= p_mb_type_info[mb_type].partition_count;
mb_type= p_mb_type_info[mb_type].type;
} else {
mb_type= decode_cabac_intra_mb_type(h, 17, 0);
goto decode_intra_mb;
}
} else {
mb_type= decode_cabac_intra_mb_type(h, 3, 1);
Replace deprecated FF_*_TYPE symbols with AV_PICTURE_TYPE_*. Signed-off-by: Diego Biurrun <diego@biurrun.de>
2011-04-29 19:53:57 +03:00
if(h->slice_type == AV_PICTURE_TYPE_SI && mb_type)
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
mb_type--;
Replace deprecated FF_*_TYPE symbols with AV_PICTURE_TYPE_*. Signed-off-by: Diego Biurrun <diego@biurrun.de>
2011-04-29 19:53:57 +03:00
assert(h->slice_type_nos == AV_PICTURE_TYPE_I);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
decode_intra_mb:
partition_count = 0;
cbp= i_mb_type_info[mb_type].cbp;
h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
mb_type= i_mb_type_info[mb_type].type;
}
h264: add a parameter to the MB_FIELD macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
if(MB_FIELD(h))
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
mb_type |= MB_TYPE_INTERLACED;
h->slice_table[ mb_xy ]= h->slice_num;
if(IS_INTRA_PCM(mb_type)) {
h264: Factorize declaration of mb_sizes array.
2012-04-04 15:53:25 +03:00
const int mb_size = ff_h264_mb_sizes[h->sps.chroma_format_idc] *
h->sps.bit_depth_luma >> 3;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
const uint8_t *ptr;
// We assume these blocks are very rare so we do not optimize it.
// FIXME The two following lines get the bitstream position in the cabac
// decode, I think it should be done by a function in cabac.h (or cabac.c).
ptr= h->cabac.bytestream;
if(h->cabac.low&0x1) ptr--;
if(CABAC_BITS==16){
if(h->cabac.low&0x1FF) ptr--;
}
// The pixels are stored in the same order as levels in h->mb array.
h264: prevent overreads in intra PCM decoding. Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind CC: libav-stable@libav.org
2012-02-29 05:48:27 +03:00
if ((int) (h->cabac.bytestream_end - ptr) < mb_size)
return -1;
h264: Don't store intra pcm samples in h->mb Instead, keep them in the bitstream buffer until we read them verbatim, this saves a memcpy() and a subsequent clearing of the target buffer. decode_cabac+decode_mb for a sample file (CAPM3_Sony_D.jsv) goes from 6121.4 to 6095.5 cycles, i.e. 26 cycles faster. Signed-off-by: Martin Storsjö <martin@martin.st>
2013-02-18 01:52:24 +03:00
h->intra_pcm_ptr = ptr;
ptr += mb_size;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);
// All blocks are present
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
h->cbp_table[mb_xy] = 0xf7ef;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
h->chroma_pred_mode_table[mb_xy] = 0;
// In deblocking, the quantizer is 0
lavc decoders: work with refcounted frames.
2012-11-21 23:34:46 +03:00
h->cur_pic.qscale_table[mb_xy] = 0;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
// All coeffs are present
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
memset(h->non_zero_count[mb_xy], 16, 48);
lavc decoders: work with refcounted frames.
2012-11-21 23:34:46 +03:00
h->cur_pic.mb_type[mb_xy] = mb_type;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
h->last_qscale_diff = 0;
return 0;
}
Split fill_caches() between filter and decoder. Originally committed as revision 21271 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-17 23:43:08 +02:00
fill_decode_caches(h, mb_type);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
if( IS_INTRA( mb_type ) ) {
int i, pred_mode;
if( IS_INTRA4x4( mb_type ) ) {
Merge the single line function decode_cabac_mb_transform_size() into the calling code. 8 cpu cycles faster Originally committed as revision 21828 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 03:04:07 +02:00
if( dct8x8_allowed && get_cabac_noinline( &h->cabac, &h->cabac_state[399 + h->neighbor_transform_size] ) ) {
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
mb_type |= MB_TYPE_8x8DCT;
for( i = 0; i < 16; i+=4 ) {
int pred = pred_intra_mode( h, i );
int mode = decode_cabac_mb_intra4x4_pred_mode( h, pred );
fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );
}
} else {
for( i = 0; i < 16; i++ ) {
int pred = pred_intra_mode( h, i );
h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
av_dlog(h->avctx, "i4x4 pred=%d mode=%d\n", pred,
avcodec: Convert some commented-out printf/av_log instances to av_dlog
2012-08-21 13:41:40 +03:00
h->intra4x4_pred_mode_cache[scan8[i]]);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
}
H.264: faster write_back_* Avoid aliasing, unroll loops, and inline more functions.
2011-06-29 23:27:36 +03:00
write_back_intra_pred_mode(h);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
if( ff_h264_check_intra4x4_pred_mode(h) < 0 ) return -1;
} else {
h264: disallow constrained intra prediction modes for luma. Conversion of the luma intra prediction mode to one of the constrained ("alzheimer") ones can happen by crafting special bitstreams, causing a crash because we'll call a NULL function pointer for 16x16 block intra prediction, since constrained intra prediction functions are only implemented for chroma (8x8 blocks). Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind CC: libav-stable@libav.org
2012-02-10 09:57:01 +03:00
h->intra16x16_pred_mode= ff_h264_check_intra_pred_mode( h, h->intra16x16_pred_mode, 0 );
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
if( h->intra16x16_pred_mode < 0 ) return -1;
}
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
if(decode_chroma){
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
h->chroma_pred_mode_table[mb_xy] =
pred_mode = decode_cabac_mb_chroma_pre_mode( h );
h264: disallow constrained intra prediction modes for luma. Conversion of the luma intra prediction mode to one of the constrained ("alzheimer") ones can happen by crafting special bitstreams, causing a crash because we'll call a NULL function pointer for 16x16 block intra prediction, since constrained intra prediction functions are only implemented for chroma (8x8 blocks). Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind CC: libav-stable@libav.org
2012-02-10 09:57:01 +03:00
pred_mode= ff_h264_check_intra_pred_mode( h, pred_mode, 1 );
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
if( pred_mode < 0 ) return -1;
h->chroma_pred_mode= pred_mode;
Set gray (128) U/V planes for chroma-less samples. Fixes two fate samples when played with -flags emu_edge.
2011-01-20 21:00:34 +02:00
} else {
h->chroma_pred_mode= DC_128_PRED8x8;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
} else if( partition_count == 4 ) {
int i, j, sub_partition_count[4], list, ref[2][4];
Replace deprecated FF_*_TYPE symbols with AV_PICTURE_TYPE_*. Signed-off-by: Diego Biurrun <diego@biurrun.de>
2011-04-29 19:53:57 +03:00
if( h->slice_type_nos == AV_PICTURE_TYPE_B ) {
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
for( i = 0; i < 4; i++ ) {
h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );
sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
}
if( IS_DIRECT(h->sub_mb_type[0] | h->sub_mb_type[1] |
h->sub_mb_type[2] | h->sub_mb_type[3]) ) {
ff_h264_pred_direct_motion(h, &mb_type);
h->ref_cache[0][scan8[4]] =
h->ref_cache[1][scan8[4]] =
h->ref_cache[0][scan8[12]] =
h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
for( i = 0; i < 4; i++ )
Set sub_mb_type in direct_cache instead of just the direct flag. Simpler, cleaner and faster. Originally committed as revision 21822 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-14 18:51:31 +02:00
fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, (h->sub_mb_type[i]>>1)&0xFF, 1 );
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
} else {
for( i = 0; i < 4; i++ ) {
h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );
sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
}
}
for( list = 0; list < h->list_count; list++ ) {
for( i = 0; i < 4; i++ ) {
if(IS_DIRECT(h->sub_mb_type[i])) continue;
if(IS_DIR(h->sub_mb_type[i], 0, list)){
h264: add a parameter to the MB_MBAFF macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
int rc = h->ref_count[list] << MB_MBAFF(h);
h264: don't touch H264Context->ref_count[] during MB decoding The variable is copied to subsequent threads at the same time, so this may cause wrong ref_count[] values to be copied to subsequent threads. This bug was found using TSAN. Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
2012-10-04 02:25:14 +03:00
if (rc > 1) {
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
h264: don't touch H264Context->ref_count[] during MB decoding The variable is copied to subsequent threads at the same time, so this may cause wrong ref_count[] values to be copied to subsequent threads. This bug was found using TSAN. Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
2012-10-04 02:25:14 +03:00
if (ref[list][i] >= (unsigned) rc) {
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
av_log(h->avctx, AV_LOG_ERROR, "Reference %d >= %d\n", ref[list][i], rc);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
return -1;
}
}else
ref[list][i] = 0;
} else {
ref[list][i] = -1;
}
h->ref_cache[list][ scan8[4*i]+1 ]=
h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
}
}
if(dct8x8_allowed)
dct8x8_allowed = get_dct8x8_allowed(h);
for(list=0; list<h->list_count; list++){
for(i=0; i<4; i++){
h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ];
if(IS_DIRECT(h->sub_mb_type[i])){
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster for high resolution videos. about 20cycles faster per MB for cathederal. Originally committed as revision 22038 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 22:43:06 +02:00
fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 2);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
continue;
}
if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
const int sub_mb_type= h->sub_mb_type[i];
const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
for(j=0; j<sub_partition_count[i]; j++){
int mpx, mpy;
int mx, my;
const int index= 4*i + block_width*j;
int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster for high resolution videos. about 20cycles faster per MB for cathederal. Originally committed as revision 22038 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 22:43:06 +02:00
uint8_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
Calculate mvd without abs() same speed (ask gcc why, i dont know) Originally committed as revision 22035 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:50:02 +02:00
pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
Factorize common code from the top of decode_cabac_mb_mvd() 10-15 cpu cycles faster. Originally committed as revision 22029 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:06:02 +02:00
DECODE_CABAC_MB_MVD( h, list, index)
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
tprintf(h->avctx, "final mv:%d %d\n", mx, my);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
if(IS_SUB_8X8(sub_mb_type)){
mv_cache[ 1 ][0]=
mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
mv_cache[ 1 ][1]=
mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
mvd_cache[ 1 ][0]=
Move abs() from decode_cabac_mb_mvd() to the code that writes mvd_cache. 4-8 cycles faster Originally committed as revision 21887 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-19 01:37:48 +02:00
mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mpx;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
mvd_cache[ 1 ][1]=
Move abs() from decode_cabac_mb_mvd() to the code that writes mvd_cache. 4-8 cycles faster Originally committed as revision 21887 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-19 01:37:48 +02:00
mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= mpy;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}else if(IS_SUB_8X4(sub_mb_type)){
mv_cache[ 1 ][0]= mx;
mv_cache[ 1 ][1]= my;
Move abs() from decode_cabac_mb_mvd() to the code that writes mvd_cache. 4-8 cycles faster Originally committed as revision 21887 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-19 01:37:48 +02:00
mvd_cache[ 1 ][0]= mpx;
mvd_cache[ 1 ][1]= mpy;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}else if(IS_SUB_4X8(sub_mb_type)){
mv_cache[ 8 ][0]= mx;
mv_cache[ 8 ][1]= my;
Move abs() from decode_cabac_mb_mvd() to the code that writes mvd_cache. 4-8 cycles faster Originally committed as revision 21887 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-19 01:37:48 +02:00
mvd_cache[ 8 ][0]= mpx;
mvd_cache[ 8 ][1]= mpy;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
mv_cache[ 0 ][0]= mx;
mv_cache[ 0 ][1]= my;
Move abs() from decode_cabac_mb_mvd() to the code that writes mvd_cache. 4-8 cycles faster Originally committed as revision 21887 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-19 01:37:48 +02:00
mvd_cache[ 0 ][0]= mpx;
mvd_cache[ 0 ][1]= mpy;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
}else{
Replace ad-hoc fill rectangle by fill_rectangle(). Originally committed as revision 22025 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 15:12:09 +02:00
fill_rectangle(h->mv_cache [list][ scan8[4*i] ], 2, 2, 8, 0, 4);
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster for high resolution videos. about 20cycles faster per MB for cathederal. Originally committed as revision 22038 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 22:43:06 +02:00
fill_rectangle(h->mvd_cache[list][ scan8[4*i] ], 2, 2, 8, 0, 2);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
}
}
} else if( IS_DIRECT(mb_type) ) {
ff_h264_pred_direct_motion(h, &mb_type);
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster for high resolution videos. about 20cycles faster per MB for cathederal. Originally committed as revision 22038 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 22:43:06 +02:00
fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 2);
fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 2);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
dct8x8_allowed &= h->sps.direct_8x8_inference_flag;
} else {
Calculate mvd without abs() same speed (ask gcc why, i dont know) Originally committed as revision 22035 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:50:02 +02:00
int list, i;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
if(IS_16X16(mb_type)){
for(list=0; list<h->list_count; list++){
if(IS_DIR(mb_type, 0, list)){
h264: add a parameter to the MB_MBAFF macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
int ref, rc = h->ref_count[list] << MB_MBAFF(h);
h264: don't touch H264Context->ref_count[] during MB decoding The variable is copied to subsequent threads at the same time, so this may cause wrong ref_count[] values to be copied to subsequent threads. This bug was found using TSAN. Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
2012-10-04 02:25:14 +03:00
if (rc > 1) {
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ref= decode_cabac_mb_ref(h, list, 0);
h264: don't touch H264Context->ref_count[] during MB decoding The variable is copied to subsequent threads at the same time, so this may cause wrong ref_count[] values to be copied to subsequent threads. This bug was found using TSAN. Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
2012-10-04 02:25:14 +03:00
if (ref >= (unsigned) rc) {
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
av_log(h->avctx, AV_LOG_ERROR, "Reference %d >= %d\n", ref, rc);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
return -1;
}
}else
ref=0;
fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
Remove some unneeded fill_rectangle() for 16x16 blocks. Originally committed as revision 22124 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-01 01:54:24 +02:00
}
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
for(list=0; list<h->list_count; list++){
if(IS_DIR(mb_type, 0, list)){
Calculate mvd without abs() same speed (ask gcc why, i dont know) Originally committed as revision 22035 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:50:02 +02:00
int mx,my,mpx,mpy;
pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
Factorize common code from the top of decode_cabac_mb_mvd() 10-15 cpu cycles faster. Originally committed as revision 22029 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:06:02 +02:00
DECODE_CABAC_MB_MVD( h, list, 0)
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
tprintf(h->avctx, "final mv:%d %d\n", mx, my);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster for high resolution videos. about 20cycles faster per MB for cathederal. Originally committed as revision 22038 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 22:43:06 +02:00
fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack8to16(mpx,mpy), 2);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
Remove some unneeded fill_rectangle() for 16x16 blocks. Originally committed as revision 22124 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-03-01 01:54:24 +02:00
}
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
}
else if(IS_16X8(mb_type)){
for(list=0; list<h->list_count; list++){
for(i=0; i<2; i++){
if(IS_DIR(mb_type, i, list)){
h264: add a parameter to the MB_MBAFF macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
int ref, rc = h->ref_count[list] << MB_MBAFF(h);
h264: don't touch H264Context->ref_count[] during MB decoding The variable is copied to subsequent threads at the same time, so this may cause wrong ref_count[] values to be copied to subsequent threads. This bug was found using TSAN. Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
2012-10-04 02:25:14 +03:00
if (rc > 1) {
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ref= decode_cabac_mb_ref( h, list, 8*i );
h264: don't touch H264Context->ref_count[] during MB decoding The variable is copied to subsequent threads at the same time, so this may cause wrong ref_count[] values to be copied to subsequent threads. This bug was found using TSAN. Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
2012-10-04 02:25:14 +03:00
if (ref >= (unsigned) rc) {
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
av_log(h->avctx, AV_LOG_ERROR, "Reference %d >= %d\n", ref, rc);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
return -1;
}
}else
ref=0;
fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
}else
fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
}
}
for(list=0; list<h->list_count; list++){
for(i=0; i<2; i++){
if(IS_DIR(mb_type, i, list)){
Calculate mvd without abs() same speed (ask gcc why, i dont know) Originally committed as revision 22035 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:50:02 +02:00
int mx,my,mpx,mpy;
pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
Factorize common code from the top of decode_cabac_mb_mvd() 10-15 cpu cycles faster. Originally committed as revision 22029 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:06:02 +02:00
DECODE_CABAC_MB_MVD( h, list, 8*i)
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
tprintf(h->avctx, "final mv:%d %d\n", mx, my);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster for high resolution videos. about 20cycles faster per MB for cathederal. Originally committed as revision 22038 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 22:43:06 +02:00
fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack8to16(mpx,mpy), 2);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
}else{
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster for high resolution videos. about 20cycles faster per MB for cathederal. Originally committed as revision 22038 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 22:43:06 +02:00
fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 2);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
}
}
}
}else{
assert(IS_8X16(mb_type));
for(list=0; list<h->list_count; list++){
for(i=0; i<2; i++){
if(IS_DIR(mb_type, i, list)){ //FIXME optimize
h264: add a parameter to the MB_MBAFF macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
int ref, rc = h->ref_count[list] << MB_MBAFF(h);
h264: don't touch H264Context->ref_count[] during MB decoding The variable is copied to subsequent threads at the same time, so this may cause wrong ref_count[] values to be copied to subsequent threads. This bug was found using TSAN. Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
2012-10-04 02:25:14 +03:00
if (rc > 1) {
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
ref= decode_cabac_mb_ref( h, list, 4*i );
h264: don't touch H264Context->ref_count[] during MB decoding The variable is copied to subsequent threads at the same time, so this may cause wrong ref_count[] values to be copied to subsequent threads. This bug was found using TSAN. Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
2012-10-04 02:25:14 +03:00
if (ref >= (unsigned) rc) {
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
av_log(h->avctx, AV_LOG_ERROR, "Reference %d >= %d\n", ref, rc);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
return -1;
}
}else
ref=0;
fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
}else
fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
}
}
for(list=0; list<h->list_count; list++){
for(i=0; i<2; i++){
if(IS_DIR(mb_type, i, list)){
Calculate mvd without abs() same speed (ask gcc why, i dont know) Originally committed as revision 22035 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:50:02 +02:00
int mx,my,mpx,mpy;
pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
Factorize common code from the top of decode_cabac_mb_mvd() 10-15 cpu cycles faster. Originally committed as revision 22029 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 20:06:02 +02:00
DECODE_CABAC_MB_MVD( h, list, 4*i)
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
tprintf(h->avctx, "final mv:%d %d\n", mx, my);
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster for high resolution videos. about 20cycles faster per MB for cathederal. Originally committed as revision 22038 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 22:43:06 +02:00
fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack8to16(mpx,mpy), 2);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
}else{
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster for high resolution videos. about 20cycles faster per MB for cathederal. Originally committed as revision 22038 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-24 22:43:06 +02:00
fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 2);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
}
}
}
}
}
if( IS_INTER( mb_type ) ) {
h->chroma_pred_mode_table[mb_xy] = 0;
write_back_motion( h, mb_type );
}
if( !IS_INTRA16x16( mb_type ) ) {
cbp = decode_cabac_mb_cbp_luma( h );
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
if(decode_chroma)
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
}
h->cbp_table[mb_xy] = h->cbp = cbp;
if( dct8x8_allowed && (cbp&15) && !IS_INTRA( mb_type ) ) {
Merge the single line function decode_cabac_mb_transform_size() into the calling code. 8 cpu cycles faster Originally committed as revision 21828 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 03:04:07 +02:00
mb_type |= MB_TYPE_8x8DCT * get_cabac_noinline( &h->cabac, &h->cabac_state[399 + h->neighbor_transform_size] );
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
/* It would be better to do this in fill_decode_caches, but we don't know
* the transform mode of the current macroblock there. */
h264: add a parameter to the CHROMA444 macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
if (CHROMA444(h) && IS_8x8DCT(mb_type)){
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
int i;
H.264: faster write_back_* Avoid aliasing, unroll loops, and inline more functions.
2011-06-29 23:27:36 +03:00
uint8_t *nnz_cache = h->non_zero_count_cache;
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
for (i = 0; i < 2; i++){
H.264: template left MB handling Faster H.264 decoding with ALLOW_INTERLACE off.
2011-06-30 01:38:39 +03:00
if (h->left_type[LEFT(i)] && !IS_8x8DCT(h->left_type[LEFT(i)])){
H.264: faster write_back_* Avoid aliasing, unroll loops, and inline more functions.
2011-06-29 23:27:36 +03:00
nnz_cache[3+8* 1 + 2*8*i]=
nnz_cache[3+8* 2 + 2*8*i]=
nnz_cache[3+8* 6 + 2*8*i]=
nnz_cache[3+8* 7 + 2*8*i]=
nnz_cache[3+8*11 + 2*8*i]=
nnz_cache[3+8*12 + 2*8*i]= IS_INTRA(mb_type) ? 64 : 0;
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
}
}
if (h->top_type && !IS_8x8DCT(h->top_type)){
h264: add a parameter to the CABAC macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
uint32_t top_empty = CABAC(h) && !IS_INTRA(mb_type) ? 0 : 0x40404040;
H.264: faster write_back_* Avoid aliasing, unroll loops, and inline more functions.
2011-06-29 23:27:36 +03:00
AV_WN32A(&nnz_cache[4+8* 0], top_empty);
AV_WN32A(&nnz_cache[4+8* 5], top_empty);
AV_WN32A(&nnz_cache[4+8*10], top_empty);
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
}
}
lavc decoders: work with refcounted frames.
2012-11-21 23:34:46 +03:00
h->cur_pic.mb_type[mb_xy] = mb_type;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
if( cbp || IS_INTRA16x16( mb_type ) ) {
H.264: split luma dc idct out and implement MMX/SSE2 versions About 2.5x the speed. NOTE: the way that the asm code handles large qmuls is a bit suboptimal. If x264-style dequant was used (separate shift and qmul values), it might be possible to get some extra speed. Originally committed as revision 26336 to svn://svn.ffmpeg.org/ffmpeg/trunk
2011-01-14 23:34:25 +02:00
const uint8_t *scan, *scan8x8;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
const uint32_t *qmul;
if(IS_INTERLACED(mb_type)){
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
scan8x8= h->qscale ? h->field_scan8x8 : h->field_scan8x8_q0;
scan= h->qscale ? h->field_scan : h->field_scan_q0;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}else{
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
scan8x8= h->qscale ? h->zigzag_scan8x8 : h->zigzag_scan8x8_q0;
scan= h->qscale ? h->zigzag_scan : h->zigzag_scan_q0;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
Merge decode_cabac_mb_dqp() with surronding code. ~20 cpu cycles faster Originally committed as revision 21826 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 01:06:25 +02:00
// decode_cabac_mb_dqp
if(get_cabac_noinline( &h->cabac, &h->cabac_state[60 + (h->last_qscale_diff != 0)])){
int val = 1;
int ctx= 2;
Add support for higher QP values in h264. In high bit depth, the QP values may now be up to (51 + 6*(bit_depth-8)). Preparatory patch for high bit depth h264 decoding support. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-03-29 18:48:58 +03:00
const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
Merge decode_cabac_mb_dqp() with surronding code. ~20 cpu cycles faster Originally committed as revision 21826 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 01:06:25 +02:00
while( get_cabac_noinline( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
ctx= 3;
val++;
Add support for higher QP values in h264. In high bit depth, the QP values may now be up to (51 + 6*(bit_depth-8)). Preparatory patch for high bit depth h264 decoding support. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-03-29 18:48:58 +03:00
if(val > 2*max_qp){ //prevent infinite loop
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
av_log(h->avctx, AV_LOG_ERROR, "cabac decode of qscale diff failed at %d %d\n", h->mb_x, h->mb_y);
Merge decode_cabac_mb_dqp() with surronding code. ~20 cpu cycles faster Originally committed as revision 21826 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 01:06:25 +02:00
return -1;
}
}
if( val&0x01 )
val= (val + 1)>>1 ;
else
val= -((val + 1)>>1);
h->last_qscale_diff = val;
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
h->qscale += val;
if(((unsigned)h->qscale) > max_qp){
if(h->qscale<0) h->qscale+= max_qp+1;
else h->qscale-= max_qp+1;
indent Originally committed as revision 21827 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 01:10:02 +02:00
}
h264: deMpegEncContextize Most of the changes are just trivial are just trivial replacements of fields from MpegEncContext with equivalent fields in H264Context. Everything in h264* other than h264.c are those trivial changes. The nontrivial parts are: 1) extracting a simplified version of the frame management code from mpegvideo.c. We don't need last/next_picture anymore, since h264 uses its own more complex system already and those were set only to appease the mpegvideo parts. 2) some tables that need to be allocated/freed in appropriate places. 3) hwaccels -- mostly trivial replacements. for dxva, the draw_horiz_band() call is moved from ff_dxva2_common_end_frame() to per-codec end_frame() callbacks, because it's now different for h264 and MpegEncContext-based decoders. 4) svq3 -- it does not use h264 complex reference system, so I just added some very simplistic frame management instead and dropped the use of ff_h264_frame_start(). Because of this I also had to move some initialization code to svq3. Additional fixes for chroma format and bit depth changes by Janne Grunau <janne-libav@jannau.net> Signed-off-by: Anton Khirnov <anton@khirnov.net>
2013-02-03 13:10:05 +03:00
h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
Merge decode_cabac_mb_dqp() with surronding code. ~20 cpu cycles faster Originally committed as revision 21826 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-02-15 01:06:25 +02:00
}else
h->last_qscale_diff=0;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
decode_cabac_luma_residual(h, scan, scan8x8, pixel_shift, mb_type, cbp, 0);
h264: add a parameter to the CHROMA444 macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
if (CHROMA444(h)) {
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
decode_cabac_luma_residual(h, scan, scan8x8, pixel_shift, mb_type, cbp, 1);
decode_cabac_luma_residual(h, scan, scan8x8, pixel_shift, mb_type, cbp, 2);
h264: add a parameter to the CHROMA422 macro. This way it does not look like a constant.
2013-03-09 22:37:11 +03:00
} else if (CHROMA422(h)) {
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
if( cbp&0x30 ){
int c;
avcodec: Drop some silly commented-out av_log() invocations
2012-09-25 18:19:34 +03:00
for (c = 0; c < 2; c++)
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
decode_cabac_residual_dc_422(h, h->mb + ((256 + 16*16*c) << pixel_shift), 3,
CHROMA_DC_BLOCK_INDEX + c,
chroma422_dc_scan, 8);
}
if( cbp&0x20 ) {
int c, i, i8x8;
for( c = 0; c < 2; c++ ) {
Drop DCTELEM typedef It does not help as an abstraction and adds dsputil dependencies. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2013-01-20 03:02:29 +03:00
int16_t *mb = h->mb + (16*(16 + 16*c) << pixel_shift);
h264: 4:2:2 intra decoding support Signed-off-by: Diego Biurrun <diego@biurrun.de> Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2011-08-16 18:05:44 +03:00
qmul = h->dequant4_coeff[c+1+(IS_INTRA( mb_type ) ? 0:3)][h->chroma_qp[c]];
for (i8x8 = 0; i8x8 < 2; i8x8++) {
for (i = 0; i < 4; i++) {
const int index = 16 + 16 * c + 8*i8x8 + i;
decode_cabac_residual_nondc(h, mb, 4, index, scan + 1, qmul, 15);
mb += 16<<pixel_shift;
}
}
}
} else {
fill_rectangle(&h->non_zero_count_cache[scan8[16]], 4, 4, 8, 0, 1);
fill_rectangle(&h->non_zero_count_cache[scan8[32]], 4, 4, 8, 0, 1);
}
} else /* yuv420 */ {
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
if( cbp&0x30 ){
int c;
avcodec: Drop some silly commented-out av_log() invocations
2012-09-25 18:19:34 +03:00
for (c = 0; c < 2; c++)
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
decode_cabac_residual_dc(h, h->mb + ((256 + 16*16*c) << pixel_shift), 3, CHROMA_DC_BLOCK_INDEX+c, chroma_dc_scan, 4);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
if( cbp&0x20 ) {
int c, i;
for( c = 0; c < 2; c++ ) {
qmul = h->dequant4_coeff[c+1+(IS_INTRA( mb_type ) ? 0:3)][h->chroma_qp[c]];
for( i = 0; i < 4; i++ ) {
const int index = 16 + 16 * c + i;
decode_cabac_residual_nondc(h, h->mb + (16*index << pixel_shift), 4, index, scan + 1, qmul, 15);
}
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
} else {
fill_rectangle(&h->non_zero_count_cache[scan8[16]], 4, 4, 8, 0, 1);
fill_rectangle(&h->non_zero_count_cache[scan8[32]], 4, 4, 8, 0, 1);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
}
}
} else {
4:4:4 H.264 decoding support Note: this is 4:4:4 from the 2007 spec revision, not the previous (now deprecated) 4:4:4 mode in H.264.
2011-06-03 11:12:28 +03:00
fill_rectangle(&h->non_zero_count_cache[scan8[ 0]], 4, 4, 8, 0, 1);
fill_rectangle(&h->non_zero_count_cache[scan8[16]], 4, 4, 8, 0, 1);
fill_rectangle(&h->non_zero_count_cache[scan8[32]], 4, 4, 8, 0, 1);
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
h->last_qscale_diff = 0;
}
lavc decoders: work with refcounted frames.
2012-11-21 23:34:46 +03:00
h->cur_pic.qscale_table[mb_xy] = h->qscale;
Split cabac decoding code out of h264.c. not slower according to benchmarks. Originally committed as revision 21181 to svn://svn.ffmpeg.org/ffmpeg/trunk
2010-01-13 04:35:36 +02:00
write_back_non_zero_count(h);
return 0;
}
Reference in New Issue Copy Permalink