Mercurial > libavcodec.hg
annotate h264.c @ 4975:9a6a0818e93f libavcodec
split h264.c to move parser in its own file
author | aurel |
---|---|
date | Thu, 10 May 2007 22:26:44 +0000 |
parents | 0d1cc37d9430 |
children | 7df30f671791 |
rev | line source |
---|---|
1168 | 1 /* |
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder | |
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at> | |
4 * | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3927
diff
changeset
|
5 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3927
diff
changeset
|
6 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3927
diff
changeset
|
7 * FFmpeg is free software; you can redistribute it and/or |
1168 | 8 * modify it under the terms of the GNU Lesser General Public |
9 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3927
diff
changeset
|
10 * version 2.1 of the License, or (at your option) any later version. |
1168 | 11 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3927
diff
changeset
|
12 * FFmpeg is distributed in the hope that it will be useful, |
1168 | 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 * Lesser General Public License for more details. | |
16 * | |
17 * You should have received a copy of the GNU Lesser General Public | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3927
diff
changeset
|
18 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3029
diff
changeset
|
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
1168 | 20 * |
21 */ | |
2967 | 22 |
1168 | 23 /** |
24 * @file h264.c | |
25 * H.264 / AVC / MPEG4 part10 codec. | |
26 * @author Michael Niedermayer <michaelni@gmx.at> | |
27 */ | |
28 | |
29 #include "common.h" | |
30 #include "dsputil.h" | |
31 #include "avcodec.h" | |
32 #include "mpegvideo.h" | |
4975 | 33 #include "h264.h" |
1168 | 34 #include "h264data.h" |
4975 | 35 #include "h264_parser.h" |
1168 | 36 #include "golomb.h" |
37 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
38 #include "cabac.h" |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
39 |
3284
a224d9752912
don't force asserts in release builds. 2% faster h264.
lorenm
parents:
3219
diff
changeset
|
40 //#undef NDEBUG |
1168 | 41 #include <assert.h> |
42 | |
43 static VLC coeff_token_vlc[4]; | |
44 static VLC chroma_dc_coeff_token_vlc; | |
45 | |
46 static VLC total_zeros_vlc[15]; | |
47 static VLC chroma_dc_total_zeros_vlc[3]; | |
48 | |
49 static VLC run_vlc[6]; | |
50 static VLC run7_vlc; | |
51 | |
1234 | 52 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp); |
53 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc); | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
54 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize); |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
55 static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize); |
1234 | 56 |
4283
d6f83e2f8804
rename always_inline to av_always_inline and move to common.h
mru
parents:
4277
diff
changeset
|
57 static av_always_inline uint32_t pack16to32(int a, int b){ |
1269 | 58 #ifdef WORDS_BIGENDIAN |
59 return (b&0xFFFF) + (a<<16); | |
60 #else | |
61 return (a&0xFFFF) + (b<<16); | |
62 #endif | |
63 } | |
64 | |
4277
113f3b395bac
Making rem6 and div6 globally visible and thus adding prefixes.
takis
parents:
4276
diff
changeset
|
65 const uint8_t ff_rem6[52]={ |
113f3b395bac
Making rem6 and div6 globally visible and thus adding prefixes.
takis
parents:
4276
diff
changeset
|
66 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, |
113f3b395bac
Making rem6 and div6 globally visible and thus adding prefixes.
takis
parents:
4276
diff
changeset
|
67 }; |
113f3b395bac
Making rem6 and div6 globally visible and thus adding prefixes.
takis
parents:
4276
diff
changeset
|
68 |
113f3b395bac
Making rem6 and div6 globally visible and thus adding prefixes.
takis
parents:
4276
diff
changeset
|
69 const uint8_t ff_div6[52]={ |
113f3b395bac
Making rem6 and div6 globally visible and thus adding prefixes.
takis
parents:
4276
diff
changeset
|
70 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, |
113f3b395bac
Making rem6 and div6 globally visible and thus adding prefixes.
takis
parents:
4276
diff
changeset
|
71 }; |
113f3b395bac
Making rem6 and div6 globally visible and thus adding prefixes.
takis
parents:
4276
diff
changeset
|
72 |
113f3b395bac
Making rem6 and div6 globally visible and thus adding prefixes.
takis
parents:
4276
diff
changeset
|
73 |
1168 | 74 /** |
75 * fill a rectangle. | |
2392 | 76 * @param h height of the rectangle, should be a constant |
77 * @param w width of the rectangle, should be a constant | |
1168 | 78 * @param size the size of val (1 or 4), should be a constant |
79 */ | |
4283
d6f83e2f8804
rename always_inline to av_always_inline and move to common.h
mru
parents:
4277
diff
changeset
|
80 static av_always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ |
1187 | 81 uint8_t *p= (uint8_t*)vp; |
1168 | 82 assert(size==1 || size==4); |
3315 | 83 assert(w<=4); |
2967 | 84 |
1168 | 85 w *= size; |
86 stride *= size; | |
2967 | 87 |
2962 | 88 assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0); |
2669 | 89 assert((stride&(w-1))==0); |
3315 | 90 if(w==2){ |
91 const uint16_t v= size==4 ? val : val*0x0101; | |
92 *(uint16_t*)(p + 0*stride)= v; | |
93 if(h==1) return; | |
94 *(uint16_t*)(p + 1*stride)= v; | |
95 if(h==2) return; | |
1168 | 96 *(uint16_t*)(p + 2*stride)= |
3315 | 97 *(uint16_t*)(p + 3*stride)= v; |
98 }else if(w==4){ | |
99 const uint32_t v= size==4 ? val : val*0x01010101; | |
100 *(uint32_t*)(p + 0*stride)= v; | |
101 if(h==1) return; | |
102 *(uint32_t*)(p + 1*stride)= v; | |
103 if(h==2) return; | |
1168 | 104 *(uint32_t*)(p + 2*stride)= |
3315 | 105 *(uint32_t*)(p + 3*stride)= v; |
106 }else if(w==8){ | |
107 //gcc can't optimize 64bit math on x86_32 | |
108 #if defined(ARCH_X86_64) || (defined(MP_WORDSIZE) && MP_WORDSIZE >= 64) | |
109 const uint64_t v= val*0x0100000001ULL; | |
110 *(uint64_t*)(p + 0*stride)= v; | |
111 if(h==1) return; | |
112 *(uint64_t*)(p + 1*stride)= v; | |
113 if(h==2) return; | |
1168 | 114 *(uint64_t*)(p + 2*stride)= |
3315 | 115 *(uint64_t*)(p + 3*stride)= v; |
116 }else if(w==16){ | |
117 const uint64_t v= val*0x0100000001ULL; | |
1168 | 118 *(uint64_t*)(p + 0+0*stride)= |
119 *(uint64_t*)(p + 8+0*stride)= | |
120 *(uint64_t*)(p + 0+1*stride)= | |
3315 | 121 *(uint64_t*)(p + 8+1*stride)= v; |
122 if(h==2) return; | |
1168 | 123 *(uint64_t*)(p + 0+2*stride)= |
124 *(uint64_t*)(p + 8+2*stride)= | |
125 *(uint64_t*)(p + 0+3*stride)= | |
3315 | 126 *(uint64_t*)(p + 8+3*stride)= v; |
127 #else | |
128 *(uint32_t*)(p + 0+0*stride)= | |
129 *(uint32_t*)(p + 4+0*stride)= val; | |
130 if(h==1) return; | |
131 *(uint32_t*)(p + 0+1*stride)= | |
132 *(uint32_t*)(p + 4+1*stride)= val; | |
133 if(h==2) return; | |
134 *(uint32_t*)(p + 0+2*stride)= | |
135 *(uint32_t*)(p + 4+2*stride)= | |
136 *(uint32_t*)(p + 0+3*stride)= | |
137 *(uint32_t*)(p + 4+3*stride)= val; | |
138 }else if(w==16){ | |
139 *(uint32_t*)(p + 0+0*stride)= | |
140 *(uint32_t*)(p + 4+0*stride)= | |
141 *(uint32_t*)(p + 8+0*stride)= | |
142 *(uint32_t*)(p +12+0*stride)= | |
143 *(uint32_t*)(p + 0+1*stride)= | |
144 *(uint32_t*)(p + 4+1*stride)= | |
145 *(uint32_t*)(p + 8+1*stride)= | |
146 *(uint32_t*)(p +12+1*stride)= val; | |
147 if(h==2) return; | |
148 *(uint32_t*)(p + 0+2*stride)= | |
149 *(uint32_t*)(p + 4+2*stride)= | |
150 *(uint32_t*)(p + 8+2*stride)= | |
151 *(uint32_t*)(p +12+2*stride)= | |
152 *(uint32_t*)(p + 0+3*stride)= | |
153 *(uint32_t*)(p + 4+3*stride)= | |
154 *(uint32_t*)(p + 8+3*stride)= | |
155 *(uint32_t*)(p +12+3*stride)= val; | |
156 #endif | |
1168 | 157 }else |
158 assert(0); | |
3315 | 159 assert(h==4); |
1168 | 160 } |
161 | |
3028 | 162 static void fill_caches(H264Context *h, int mb_type, int for_deblock){ |
1168 | 163 MpegEncContext * const s = &h->s; |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
164 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 165 int topleft_xy, top_xy, topright_xy, left_xy[2]; |
166 int topleft_type, top_type, topright_type, left_type[2]; | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
167 int left_block[8]; |
1168 | 168 int i; |
169 | |
3482 | 170 //FIXME deblocking could skip the intra and nnz parts. |
171 if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[mb_xy-s->mb_stride]) && !FRAME_MBAFF) | |
2623
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
172 return; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
173 |
2967 | 174 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it |
175 | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
176 top_xy = mb_xy - s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
177 topleft_xy = top_xy - 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
178 topright_xy= top_xy + 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
179 left_xy[1] = left_xy[0] = mb_xy-1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
180 left_block[0]= 0; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
181 left_block[1]= 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
182 left_block[2]= 2; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
183 left_block[3]= 3; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
184 left_block[4]= 7; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
185 left_block[5]= 10; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
186 left_block[6]= 8; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
187 left_block[7]= 11; |
3316 | 188 if(FRAME_MBAFF){ |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
189 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
190 const int top_pair_xy = pair_xy - s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
191 const int topleft_pair_xy = top_pair_xy - 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
192 const int topright_pair_xy = top_pair_xy + 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
193 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]); |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
194 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]); |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
195 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]); |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
196 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]); |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
197 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type); |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
198 const int bottom = (s->mb_y & 1); |
4600 | 199 tprintf(s->avctx, "fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag); |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
200 if (bottom |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
201 ? !curr_mb_frame_flag // bottom macroblock |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
202 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
203 ) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
204 top_xy -= s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
205 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
206 if (bottom |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
207 ? !curr_mb_frame_flag // bottom macroblock |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
208 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
209 ) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
210 topleft_xy -= s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
211 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
212 if (bottom |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
213 ? !curr_mb_frame_flag // bottom macroblock |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
214 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
215 ) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
216 topright_xy -= s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
217 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
218 if (left_mb_frame_flag != curr_mb_frame_flag) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
219 left_xy[1] = left_xy[0] = pair_xy - 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
220 if (curr_mb_frame_flag) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
221 if (bottom) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
222 left_block[0]= 2; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
223 left_block[1]= 2; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
224 left_block[2]= 3; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
225 left_block[3]= 3; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
226 left_block[4]= 8; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
227 left_block[5]= 11; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
228 left_block[6]= 8; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
229 left_block[7]= 11; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
230 } else { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
231 left_block[0]= 0; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
232 left_block[1]= 0; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
233 left_block[2]= 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
234 left_block[3]= 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
235 left_block[4]= 7; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
236 left_block[5]= 10; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
237 left_block[6]= 7; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
238 left_block[7]= 10; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
239 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
240 } else { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
241 left_xy[1] += s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
242 //left_block[0]= 0; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
243 left_block[1]= 2; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
244 left_block[2]= 0; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
245 left_block[3]= 2; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
246 //left_block[4]= 7; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
247 left_block[5]= 10; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
248 left_block[6]= 7; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
249 left_block[7]= 10; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
250 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
251 } |
1168 | 252 } |
253 | |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
254 h->top_mb_xy = top_xy; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
255 h->left_mb_xy[0] = left_xy[0]; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
256 h->left_mb_xy[1] = left_xy[1]; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
257 if(for_deblock){ |
3482 | 258 topleft_type = 0; |
259 topright_type = 0; | |
2449 | 260 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0; |
261 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0; | |
262 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0; | |
3316 | 263 |
264 if(FRAME_MBAFF && !IS_INTRA(mb_type)){ | |
265 int list; | |
266 int v = *(uint16_t*)&h->non_zero_count[mb_xy][14]; | |
267 for(i=0; i<16; i++) | |
268 h->non_zero_count_cache[scan8[i]] = (v>>i)&1; | |
4533 | 269 for(list=0; list<h->list_count; list++){ |
3316 | 270 if(USES_LIST(mb_type,list)){ |
271 uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]]; | |
272 uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]]; | |
3776 | 273 int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]]; |
3316 | 274 for(i=0; i<4; i++, dst+=8, src+=h->b_stride){ |
275 dst[0] = src[0]; | |
276 dst[1] = src[1]; | |
277 dst[2] = src[2]; | |
278 dst[3] = src[3]; | |
279 } | |
280 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] = | |
281 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101; | |
282 ref += h->b8_stride; | |
283 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] = | |
284 *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101; | |
285 }else{ | |
286 fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4); | |
287 fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1); | |
288 } | |
289 } | |
290 } | |
2449 | 291 }else{ |
292 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0; | |
293 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0; | |
294 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0; | |
295 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0; | |
296 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0; | |
297 } | |
1168 | 298 |
299 if(IS_INTRA(mb_type)){ | |
2967 | 300 h->topleft_samples_available= |
301 h->top_samples_available= | |
1168 | 302 h->left_samples_available= 0xFFFF; |
303 h->topright_samples_available= 0xEEEA; | |
304 | |
305 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){ | |
306 h->topleft_samples_available= 0xB3FF; | |
307 h->top_samples_available= 0x33FF; | |
308 h->topright_samples_available= 0x26EA; | |
309 } | |
310 for(i=0; i<2; i++){ | |
311 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){ | |
312 h->topleft_samples_available&= 0xDF5F; | |
313 h->left_samples_available&= 0x5F5F; | |
314 } | |
315 } | |
2967 | 316 |
1168 | 317 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred)) |
318 h->topleft_samples_available&= 0x7FFF; | |
2967 | 319 |
1168 | 320 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred)) |
321 h->topright_samples_available&= 0xFBFF; | |
2967 | 322 |
1168 | 323 if(IS_INTRA4x4(mb_type)){ |
324 if(IS_INTRA4x4(top_type)){ | |
325 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4]; | |
326 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5]; | |
327 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6]; | |
328 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3]; | |
329 }else{ | |
330 int pred; | |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
331 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred)) |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
332 pred= -1; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
333 else{ |
1168 | 334 pred= 2; |
335 } | |
336 h->intra4x4_pred_mode_cache[4+8*0]= | |
337 h->intra4x4_pred_mode_cache[5+8*0]= | |
338 h->intra4x4_pred_mode_cache[6+8*0]= | |
339 h->intra4x4_pred_mode_cache[7+8*0]= pred; | |
340 } | |
341 for(i=0; i<2; i++){ | |
342 if(IS_INTRA4x4(left_type[i])){ | |
343 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]]; | |
344 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]]; | |
345 }else{ | |
346 int pred; | |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
347 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred)) |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
348 pred= -1; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
349 else{ |
1168 | 350 pred= 2; |
351 } | |
352 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= | |
353 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred; | |
354 } | |
355 } | |
356 } | |
357 } | |
2967 | 358 |
359 | |
1168 | 360 /* |
2967 | 361 0 . T T. T T T T |
362 1 L . .L . . . . | |
363 2 L . .L . . . . | |
364 3 . T TL . . . . | |
365 4 L . .L . . . . | |
366 5 L . .. . . . . | |
1168 | 367 */ |
368 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec) | |
369 if(top_type){ | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
370 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
371 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
372 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
373 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3]; |
2967 | 374 |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
375 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
376 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8]; |
2967 | 377 |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
378 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
379 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11]; |
2967 | 380 |
1168 | 381 }else{ |
2967 | 382 h->non_zero_count_cache[4+8*0]= |
1168 | 383 h->non_zero_count_cache[5+8*0]= |
384 h->non_zero_count_cache[6+8*0]= | |
385 h->non_zero_count_cache[7+8*0]= | |
2967 | 386 |
1168 | 387 h->non_zero_count_cache[1+8*0]= |
388 h->non_zero_count_cache[2+8*0]= | |
2967 | 389 |
1168 | 390 h->non_zero_count_cache[1+8*3]= |
2314 | 391 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64; |
2967 | 392 |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
393 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
394 |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
395 for (i=0; i<2; i++) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
396 if(left_type[i]){ |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
397 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
398 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
399 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
400 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
401 }else{ |
2967 | 402 h->non_zero_count_cache[3+8*1 + 2*8*i]= |
403 h->non_zero_count_cache[3+8*2 + 2*8*i]= | |
404 h->non_zero_count_cache[0+8*1 + 8*i]= | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
405 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64; |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
406 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
407 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
408 |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
409 if( h->pps.cabac ) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
410 // top_cbp |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
411 if(top_type) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
412 h->top_cbp = h->cbp_table[top_xy]; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
413 } else if(IS_INTRA(mb_type)) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
414 h->top_cbp = 0x1C0; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
415 } else { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
416 h->top_cbp = 0; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
417 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
418 // left_cbp |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
419 if (left_type[0]) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
420 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
421 } else if(IS_INTRA(mb_type)) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
422 h->left_cbp = 0x1C0; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
423 } else { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
424 h->left_cbp = 0; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
425 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
426 if (left_type[0]) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
427 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
428 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
429 if (left_type[1]) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
430 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
431 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
432 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
433 |
1168 | 434 #if 1 |
2623
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
435 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){ |
1168 | 436 int list; |
4533 | 437 for(list=0; list<h->list_count; list++){ |
2623
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
438 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){ |
1168 | 439 /*if(!h->mv_cache_clean[list]){ |
440 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all? | |
441 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t)); | |
442 h->mv_cache_clean[list]= 1; | |
443 }*/ | |
2396 | 444 continue; |
1168 | 445 } |
446 h->mv_cache_clean[list]= 0; | |
2967 | 447 |
3212 | 448 if(USES_LIST(top_type, list)){ |
1168 | 449 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride; |
450 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride; | |
451 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0]; | |
452 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1]; | |
453 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2]; | |
454 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3]; | |
455 h->ref_cache[list][scan8[0] + 0 - 1*8]= | |
456 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0]; | |
457 h->ref_cache[list][scan8[0] + 2 - 1*8]= | |
458 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1]; | |
459 }else{ | |
2967 | 460 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= |
461 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= | |
462 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= | |
1168 | 463 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0; |
464 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101; | |
465 } | |
466 | |
4546 | 467 for(i=0; i<2; i++){ |
468 int cache_idx = scan8[0] - 1 + i*2*8; | |
469 if(USES_LIST(left_type[i], list)){ | |
470 const int b_xy= h->mb2b_xy[left_xy[i]] + 3; | |
471 const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1; | |
472 *(uint32_t*)h->mv_cache[list][cache_idx ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]; | |
473 *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]]; | |
474 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)]; | |
475 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)]; | |
476 }else{ | |
477 *(uint32_t*)h->mv_cache [list][cache_idx ]= | |
478 *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0; | |
479 h->ref_cache[list][cache_idx ]= | |
480 h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
481 } | |
1168 | 482 } |
483 | |
3519 | 484 if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF) |
2449 | 485 continue; |
486 | |
3212 | 487 if(USES_LIST(topleft_type, list)){ |
2623
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
488 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
489 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
490 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy]; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
491 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy]; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
492 }else{ |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
493 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
494 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
495 } |
2967 | 496 |
3212 | 497 if(USES_LIST(topright_type, list)){ |
2623
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
498 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
499 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
500 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy]; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
501 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy]; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
502 }else{ |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
503 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
504 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
505 } |
2967 | 506 |
3519 | 507 if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF) |
3482 | 508 continue; |
2967 | 509 |
510 h->ref_cache[list][scan8[5 ]+1] = | |
511 h->ref_cache[list][scan8[7 ]+1] = | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
512 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else) |
2967 | 513 h->ref_cache[list][scan8[4 ]] = |
1168 | 514 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE; |
515 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]= | |
516 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]= | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
517 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else) |
1168 | 518 *(uint32_t*)h->mv_cache [list][scan8[4 ]]= |
519 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0; | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
520 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
521 if( h->pps.cabac ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
522 /* XXX beurk, Load mvd */ |
3212 | 523 if(USES_LIST(top_type, list)){ |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
524 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
525 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
526 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
527 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
528 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
529 }else{ |
2967 | 530 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]= |
531 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]= | |
532 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]= | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
533 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
534 } |
3212 | 535 if(USES_LIST(left_type[0], list)){ |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
536 const int b_xy= h->mb2b_xy[left_xy[0]] + 3; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
537 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
538 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
539 }else{ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
540 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]= |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
541 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
542 } |
3212 | 543 if(USES_LIST(left_type[1], list)){ |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
544 const int b_xy= h->mb2b_xy[left_xy[1]] + 3; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
545 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
546 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
547 }else{ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
548 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]= |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
549 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
550 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
551 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]= |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
552 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]= |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
553 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else) |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
554 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]= |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
555 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0; |
2396 | 556 |
557 if(h->slice_type == B_TYPE){ | |
558 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1); | |
559 | |
560 if(IS_DIRECT(top_type)){ | |
561 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101; | |
562 }else if(IS_8X8(top_type)){ | |
563 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride; | |
564 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy]; | |
565 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1]; | |
566 }else{ | |
567 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0; | |
568 } | |
2967 | 569 |
3316 | 570 if(IS_DIRECT(left_type[0])) |
571 h->direct_cache[scan8[0] - 1 + 0*8]= 1; | |
572 else if(IS_8X8(left_type[0])) | |
573 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)]; | |
574 else | |
575 h->direct_cache[scan8[0] - 1 + 0*8]= 0; | |
576 | |
577 if(IS_DIRECT(left_type[1])) | |
2396 | 578 h->direct_cache[scan8[0] - 1 + 2*8]= 1; |
3316 | 579 else if(IS_8X8(left_type[1])) |
580 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)]; | |
581 else | |
2396 | 582 h->direct_cache[scan8[0] - 1 + 2*8]= 0; |
3316 | 583 } |
584 } | |
585 | |
586 if(FRAME_MBAFF){ | |
587 #define MAP_MVS\ | |
588 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\ | |
589 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\ | |
590 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\ | |
591 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\ | |
592 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\ | |
593 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\ | |
594 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\ | |
595 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\ | |
596 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\ | |
597 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1]) | |
598 if(MB_FIELD){ | |
599 #define MAP_F2F(idx, mb_type)\ | |
600 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\ | |
601 h->ref_cache[list][idx] <<= 1;\ | |
602 h->mv_cache[list][idx][1] /= 2;\ | |
603 h->mvd_cache[list][idx][1] /= 2;\ | |
2396 | 604 } |
3316 | 605 MAP_MVS |
606 #undef MAP_F2F | |
607 }else{ | |
608 #define MAP_F2F(idx, mb_type)\ | |
609 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\ | |
610 h->ref_cache[list][idx] >>= 1;\ | |
611 h->mv_cache[list][idx][1] <<= 1;\ | |
612 h->mvd_cache[list][idx][1] <<= 1;\ | |
613 } | |
614 MAP_MVS | |
615 #undef MAP_F2F | |
2396 | 616 } |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
617 } |
1168 | 618 } |
619 } | |
620 #endif | |
2755 | 621 |
622 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]); | |
1168 | 623 } |
624 | |
625 static inline void write_back_intra_pred_mode(H264Context *h){ | |
626 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
627 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 628 |
629 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1]; | |
630 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2]; | |
631 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3]; | |
632 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4]; | |
633 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4]; | |
634 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4]; | |
635 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4]; | |
636 } | |
637 | |
638 /** | |
639 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. | |
640 */ | |
641 static inline int check_intra4x4_pred_mode(H264Context *h){ | |
642 MpegEncContext * const s = &h->s; | |
643 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0}; | |
644 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED}; | |
645 int i; | |
2967 | 646 |
1168 | 647 if(!(h->top_samples_available&0x8000)){ |
648 for(i=0; i<4; i++){ | |
649 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ]; | |
650 if(status<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
651 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y); |
1168 | 652 return -1; |
653 } else if(status){ | |
654 h->intra4x4_pred_mode_cache[scan8[0] + i]= status; | |
655 } | |
656 } | |
657 } | |
2967 | 658 |
1168 | 659 if(!(h->left_samples_available&0x8000)){ |
660 for(i=0; i<4; i++){ | |
661 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ]; | |
662 if(status<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
663 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y); |
1168 | 664 return -1; |
665 } else if(status){ | |
666 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status; | |
667 } | |
668 } | |
669 } | |
670 | |
671 return 0; | |
672 } //FIXME cleanup like next | |
673 | |
674 /** | |
675 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. | |
676 */ | |
677 static inline int check_intra_pred_mode(H264Context *h, int mode){ | |
678 MpegEncContext * const s = &h->s; | |
679 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1}; | |
680 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8}; | |
2967 | 681 |
4529 | 682 if(mode > 6U) { |
2392 | 683 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y); |
2163 | 684 return -1; |
2392 | 685 } |
2967 | 686 |
1168 | 687 if(!(h->top_samples_available&0x8000)){ |
688 mode= top[ mode ]; | |
689 if(mode<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
690 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y); |
1168 | 691 return -1; |
692 } | |
693 } | |
2967 | 694 |
1168 | 695 if(!(h->left_samples_available&0x8000)){ |
696 mode= left[ mode ]; | |
697 if(mode<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
698 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y); |
1168 | 699 return -1; |
2967 | 700 } |
1168 | 701 } |
702 | |
703 return mode; | |
704 } | |
705 | |
706 /** | |
707 * gets the predicted intra4x4 prediction mode. | |
708 */ | |
709 static inline int pred_intra_mode(H264Context *h, int n){ | |
710 const int index8= scan8[n]; | |
711 const int left= h->intra4x4_pred_mode_cache[index8 - 1]; | |
712 const int top = h->intra4x4_pred_mode_cache[index8 - 8]; | |
713 const int min= FFMIN(left, top); | |
714 | |
4600 | 715 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min); |
1168 | 716 |
717 if(min<0) return DC_PRED; | |
718 else return min; | |
719 } | |
720 | |
721 static inline void write_back_non_zero_count(H264Context *h){ | |
722 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
723 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
724 |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
725 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
726 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
727 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
728 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4]; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
729 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
730 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
731 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4]; |
2967 | 732 |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
733 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
734 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2]; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
735 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
736 |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
737 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
738 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5]; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
739 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4]; |
3316 | 740 |
741 if(FRAME_MBAFF){ | |
742 // store all luma nnzs, for deblocking | |
743 int v = 0, i; | |
744 for(i=0; i<16; i++) | |
745 v += (!!h->non_zero_count_cache[scan8[i]]) << i; | |
746 *(uint16_t*)&h->non_zero_count[mb_xy][14] = v; | |
747 } | |
1168 | 748 } |
749 | |
750 /** | |
751 * gets the predicted number of non zero coefficients. | |
752 * @param n block index | |
753 */ | |
754 static inline int pred_non_zero_count(H264Context *h, int n){ | |
755 const int index8= scan8[n]; | |
756 const int left= h->non_zero_count_cache[index8 - 1]; | |
757 const int top = h->non_zero_count_cache[index8 - 8]; | |
758 int i= left + top; | |
2967 | 759 |
1168 | 760 if(i<64) i= (i+1)>>1; |
761 | |
4600 | 762 tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31); |
1168 | 763 |
764 return i&31; | |
765 } | |
766 | |
1169 | 767 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){ |
768 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ]; | |
4600 | 769 MpegEncContext *s = &h->s; |
1169 | 770 |
3316 | 771 /* there is no consistent mapping of mvs to neighboring locations that will |
772 * make mbaff happy, so we can't move all this logic to fill_caches */ | |
773 if(FRAME_MBAFF){ | |
3776 | 774 const uint32_t *mb_types = s->current_picture_ptr->mb_type; |
3316 | 775 const int16_t *mv; |
776 *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0; | |
777 *C = h->mv_cache[list][scan8[0]-2]; | |
778 | |
779 if(!MB_FIELD | |
780 && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){ | |
781 int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3); | |
782 if(IS_INTERLACED(mb_types[topright_xy])){ | |
783 #define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\ | |
784 const int x4 = X4, y4 = Y4;\ | |
785 const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\ | |
786 if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\ | |
787 return LIST_NOT_USED;\ | |
788 mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\ | |
789 h->mv_cache[list][scan8[0]-2][0] = mv[0];\ | |
790 h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\ | |
791 return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP; | |
792 | |
793 SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1); | |
794 } | |
795 } | |
796 if(topright_ref == PART_NOT_AVAILABLE | |
797 && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4 | |
798 && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){ | |
799 if(!MB_FIELD | |
800 && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){ | |
801 SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1); | |
802 } | |
803 if(MB_FIELD | |
804 && !IS_INTERLACED(mb_types[h->left_mb_xy[0]]) | |
805 && i >= scan8[0]+8){ | |
806 // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok. | |
807 SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2); | |
808 } | |
809 } | |
810 #undef SET_DIAG_MV | |
811 } | |
812 | |
1169 | 813 if(topright_ref != PART_NOT_AVAILABLE){ |
814 *C= h->mv_cache[list][ i - 8 + part_width ]; | |
815 return topright_ref; | |
816 }else{ | |
4600 | 817 tprintf(s->avctx, "topright MV not available\n"); |
1170 | 818 |
1169 | 819 *C= h->mv_cache[list][ i - 8 - 1 ]; |
820 return h->ref_cache[list][ i - 8 - 1 ]; | |
821 } | |
822 } | |
823 | |
1168 | 824 /** |
825 * gets the predicted MV. | |
826 * @param n the block index | |
827 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4) | |
828 * @param mx the x component of the predicted motion vector | |
829 * @param my the y component of the predicted motion vector | |
830 */ | |
831 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){ | |
832 const int index8= scan8[n]; | |
833 const int top_ref= h->ref_cache[list][ index8 - 8 ]; | |
834 const int left_ref= h->ref_cache[list][ index8 - 1 ]; | |
835 const int16_t * const A= h->mv_cache[list][ index8 - 1 ]; | |
836 const int16_t * const B= h->mv_cache[list][ index8 - 8 ]; | |
1169 | 837 const int16_t * C; |
838 int diagonal_ref, match_count; | |
839 | |
1168 | 840 assert(part_width==1 || part_width==2 || part_width==4); |
1169 | 841 |
1168 | 842 /* mv_cache |
2967 | 843 B . . A T T T T |
1168 | 844 U . . L . . , . |
845 U . . L . . . . | |
846 U . . L . . , . | |
847 . . . L . . . . | |
848 */ | |
1169 | 849 |
850 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width); | |
851 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref); | |
4600 | 852 tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count); |
1169 | 853 if(match_count > 1){ //most common |
854 *mx= mid_pred(A[0], B[0], C[0]); | |
855 *my= mid_pred(A[1], B[1], C[1]); | |
856 }else if(match_count==1){ | |
857 if(left_ref==ref){ | |
858 *mx= A[0]; | |
2967 | 859 *my= A[1]; |
1169 | 860 }else if(top_ref==ref){ |
861 *mx= B[0]; | |
2967 | 862 *my= B[1]; |
1169 | 863 }else{ |
864 *mx= C[0]; | |
2967 | 865 *my= C[1]; |
1169 | 866 } |
867 }else{ | |
868 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){ | |
869 *mx= A[0]; | |
2967 | 870 *my= A[1]; |
1168 | 871 }else{ |
872 *mx= mid_pred(A[0], B[0], C[0]); | |
873 *my= mid_pred(A[1], B[1], C[1]); | |
874 } | |
1169 | 875 } |
2967 | 876 |
4600 | 877 tprintf(h->s.avctx, "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list); |
1168 | 878 } |
879 | |
880 /** | |
881 * gets the directionally predicted 16x8 MV. | |
882 * @param n the block index | |
883 * @param mx the x component of the predicted motion vector | |
884 * @param my the y component of the predicted motion vector | |
885 */ | |
886 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){ | |
887 if(n==0){ | |
888 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ]; | |
889 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ]; | |
890 | |
4600 | 891 tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list); |
2967 | 892 |
1168 | 893 if(top_ref == ref){ |
894 *mx= B[0]; | |
895 *my= B[1]; | |
896 return; | |
897 } | |
898 }else{ | |
899 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ]; | |
900 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ]; | |
2967 | 901 |
4600 | 902 tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list); |
1168 | 903 |
904 if(left_ref == ref){ | |
905 *mx= A[0]; | |
906 *my= A[1]; | |
907 return; | |
908 } | |
909 } | |
910 | |
911 //RARE | |
912 pred_motion(h, n, 4, list, ref, mx, my); | |
913 } | |
914 | |
915 /** | |
916 * gets the directionally predicted 8x16 MV. | |
917 * @param n the block index | |
918 * @param mx the x component of the predicted motion vector | |
919 * @param my the y component of the predicted motion vector | |
920 */ | |
921 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){ | |
922 if(n==0){ | |
923 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ]; | |
924 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ]; | |
2967 | 925 |
4600 | 926 tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list); |
1168 | 927 |
928 if(left_ref == ref){ | |
929 *mx= A[0]; | |
930 *my= A[1]; | |
931 return; | |
932 } | |
933 }else{ | |
1169 | 934 const int16_t * C; |
935 int diagonal_ref; | |
936 | |
937 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2); | |
2967 | 938 |
4600 | 939 tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list); |
1168 | 940 |
2967 | 941 if(diagonal_ref == ref){ |
1168 | 942 *mx= C[0]; |
943 *my= C[1]; | |
944 return; | |
945 } | |
946 } | |
947 | |
948 //RARE | |
949 pred_motion(h, n, 2, list, ref, mx, my); | |
950 } | |
951 | |
952 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){ | |
953 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ]; | |
954 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ]; | |
955 | |
4600 | 956 tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y); |
1168 | 957 |
958 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE | |
959 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0) | |
960 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){ | |
2967 | 961 |
1168 | 962 *mx = *my = 0; |
963 return; | |
964 } | |
2967 | 965 |
1168 | 966 pred_motion(h, 0, 4, 0, 0, mx, my); |
967 | |
968 return; | |
969 } | |
970 | |
2396 | 971 static inline void direct_dist_scale_factor(H264Context * const h){ |
972 const int poc = h->s.current_picture_ptr->poc; | |
973 const int poc1 = h->ref_list[1][0].poc; | |
974 int i; | |
975 for(i=0; i<h->ref_count[0]; i++){ | |
976 int poc0 = h->ref_list[0][i].poc; | |
4594 | 977 int td = av_clip(poc1 - poc0, -128, 127); |
2396 | 978 if(td == 0 /* FIXME || pic0 is a long-term ref */){ |
979 h->dist_scale_factor[i] = 256; | |
980 }else{ | |
4594 | 981 int tb = av_clip(poc - poc0, -128, 127); |
4001 | 982 int tx = (16384 + (FFABS(td) >> 1)) / td; |
4594 | 983 h->dist_scale_factor[i] = av_clip((tb*tx + 32) >> 6, -1024, 1023); |
2396 | 984 } |
985 } | |
3316 | 986 if(FRAME_MBAFF){ |
987 for(i=0; i<h->ref_count[0]; i++){ | |
988 h->dist_scale_factor_field[2*i] = | |
989 h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i]; | |
990 } | |
991 } | |
2396 | 992 } |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
993 static inline void direct_ref_list_init(H264Context * const h){ |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
994 MpegEncContext * const s = &h->s; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
995 Picture * const ref1 = &h->ref_list[1][0]; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
996 Picture * const cur = s->current_picture_ptr; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
997 int list, i, j; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
998 if(cur->pict_type == I_TYPE) |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
999 cur->ref_count[0] = 0; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1000 if(cur->pict_type != B_TYPE) |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1001 cur->ref_count[1] = 0; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1002 for(list=0; list<2; list++){ |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1003 cur->ref_count[list] = h->ref_count[list]; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1004 for(j=0; j<h->ref_count[list]; j++) |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1005 cur->ref_poc[list][j] = h->ref_list[list][j].poc; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1006 } |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1007 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred) |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1008 return; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1009 for(list=0; list<2; list++){ |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1010 for(i=0; i<ref1->ref_count[list]; i++){ |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1011 const int poc = ref1->ref_poc[list][i]; |
3125 | 1012 h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */ |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1013 for(j=0; j<h->ref_count[list]; j++) |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1014 if(h->ref_list[list][j].poc == poc){ |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1015 h->map_col_to_list0[list][i] = j; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1016 break; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1017 } |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1018 } |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1019 } |
3316 | 1020 if(FRAME_MBAFF){ |
1021 for(list=0; list<2; list++){ | |
1022 for(i=0; i<ref1->ref_count[list]; i++){ | |
1023 j = h->map_col_to_list0[list][i]; | |
1024 h->map_col_to_list0_field[list][2*i] = 2*j; | |
1025 h->map_col_to_list0_field[list][2*i+1] = 2*j+1; | |
1026 } | |
1027 } | |
1028 } | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1029 } |
2396 | 1030 |
1031 static inline void pred_direct_motion(H264Context * const h, int *mb_type){ | |
1032 MpegEncContext * const s = &h->s; | |
1033 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride; | |
1034 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride; | |
1035 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; | |
1036 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy]; | |
1037 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy]; | |
2808
932ec96933f5
fixing colocated mv if colocated block is L1 predicted
michael
parents:
2803
diff
changeset
|
1038 const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy]; |
2396 | 1039 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy]; |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1040 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy]; |
2396 | 1041 const int is_b8x8 = IS_8X8(*mb_type); |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
1042 unsigned int sub_mb_type; |
2396 | 1043 int i8, i4; |
1044 | |
3316 | 1045 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM) |
2396 | 1046 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){ |
1047 /* FIXME save sub mb types from previous frames (or derive from MVs) | |
1048 * so we know exactly what block size to use */ | |
1049 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */ | |
2536 | 1050 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1; |
3316 | 1051 }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){ |
2396 | 1052 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ |
1053 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */ | |
1054 }else{ | |
1055 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ | |
2536 | 1056 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1; |
2396 | 1057 } |
1058 if(!is_b8x8) | |
1059 *mb_type |= MB_TYPE_DIRECT2; | |
3316 | 1060 if(MB_FIELD) |
1061 *mb_type |= MB_TYPE_INTERLACED; | |
2396 | 1062 |
4600 | 1063 tprintf(s->avctx, "mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col); |
2967 | 1064 |
2396 | 1065 if(h->direct_spatial_mv_pred){ |
1066 int ref[2]; | |
1067 int mv[2][2]; | |
1068 int list; | |
1069 | |
3316 | 1070 /* FIXME interlacing + spatial direct uses wrong colocated block positions */ |
1071 | |
2396 | 1072 /* ref = min(neighbors) */ |
1073 for(list=0; list<2; list++){ | |
1074 int refa = h->ref_cache[list][scan8[0] - 1]; | |
1075 int refb = h->ref_cache[list][scan8[0] - 8]; | |
1076 int refc = h->ref_cache[list][scan8[0] - 8 + 4]; | |
1077 if(refc == -2) | |
1078 refc = h->ref_cache[list][scan8[0] - 8 - 1]; | |
1079 ref[list] = refa; | |
1080 if(ref[list] < 0 || (refb < ref[list] && refb >= 0)) | |
1081 ref[list] = refb; | |
1082 if(ref[list] < 0 || (refc < ref[list] && refc >= 0)) | |
1083 ref[list] = refc; | |
1084 if(ref[list] < 0) | |
1085 ref[list] = -1; | |
1086 } | |
1087 | |
1088 if(ref[0] < 0 && ref[1] < 0){ | |
1089 ref[0] = ref[1] = 0; | |
1090 mv[0][0] = mv[0][1] = | |
1091 mv[1][0] = mv[1][1] = 0; | |
1092 }else{ | |
1093 for(list=0; list<2; list++){ | |
1094 if(ref[list] >= 0) | |
1095 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]); | |
1096 else | |
1097 mv[list][0] = mv[list][1] = 0; | |
1098 } | |
1099 } | |
1100 | |
1101 if(ref[1] < 0){ | |
1102 *mb_type &= ~MB_TYPE_P0L1; | |
1103 sub_mb_type &= ~MB_TYPE_P0L1; | |
1104 }else if(ref[0] < 0){ | |
1105 *mb_type &= ~MB_TYPE_P0L0; | |
1106 sub_mb_type &= ~MB_TYPE_P0L0; | |
1107 } | |
1108 | |
1109 if(IS_16X16(*mb_type)){ | |
4540 | 1110 int a=0, b=0; |
1111 | |
3001
b52d8ee430f6
fix some potential arithmetic overflows in pred_direct_motion() and
lorenm
parents:
2979
diff
changeset
|
1112 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1); |
b52d8ee430f6
fix some potential arithmetic overflows in pred_direct_motion() and
lorenm
parents:
2979
diff
changeset
|
1113 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1); |
2967 | 1114 if(!IS_INTRA(mb_type_col) |
4001 | 1115 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1) |
1116 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1 | |
2834 | 1117 && (h->x264_build>33 || !h->x264_build)))){ |
2396 | 1118 if(ref[0] > 0) |
4540 | 1119 a= pack16to32(mv[0][0],mv[0][1]); |
2396 | 1120 if(ref[1] > 0) |
4540 | 1121 b= pack16to32(mv[1][0],mv[1][1]); |
2396 | 1122 }else{ |
4540 | 1123 a= pack16to32(mv[0][0],mv[0][1]); |
1124 b= pack16to32(mv[1][0],mv[1][1]); | |
1125 } | |
1126 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4); | |
1127 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4); | |
2396 | 1128 }else{ |
1129 for(i8=0; i8<4; i8++){ | |
1130 const int x8 = i8&1; | |
1131 const int y8 = i8>>1; | |
2967 | 1132 |
2396 | 1133 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) |
1134 continue; | |
1135 h->sub_mb_type[i8] = sub_mb_type; | |
2967 | 1136 |
2396 | 1137 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4); |
1138 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4); | |
3001
b52d8ee430f6
fix some potential arithmetic overflows in pred_direct_motion() and
lorenm
parents:
2979
diff
changeset
|
1139 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1); |
b52d8ee430f6
fix some potential arithmetic overflows in pred_direct_motion() and
lorenm
parents:
2979
diff
changeset
|
1140 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1); |
2967 | 1141 |
2396 | 1142 /* col_zero_flag */ |
2967 | 1143 if(!IS_INTRA(mb_type_col) && ( l1ref0[x8 + y8*h->b8_stride] == 0 |
1144 || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0 | |
2834 | 1145 && (h->x264_build>33 || !h->x264_build)))){ |
2808
932ec96933f5
fixing colocated mv if colocated block is L1 predicted
michael
parents:
2803
diff
changeset
|
1146 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1; |
3002 | 1147 if(IS_SUB_8X8(sub_mb_type)){ |
1148 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride]; | |
4001 | 1149 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){ |
3002 | 1150 if(ref[0] == 0) |
1151 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); | |
1152 if(ref[1] == 0) | |
1153 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); | |
1154 } | |
1155 }else | |
2396 | 1156 for(i4=0; i4<4; i4++){ |
2808
932ec96933f5
fixing colocated mv if colocated block is L1 predicted
michael
parents:
2803
diff
changeset
|
1157 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride]; |
4001 | 1158 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){ |
2396 | 1159 if(ref[0] == 0) |
1160 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0; | |
1161 if(ref[1] == 0) | |
1162 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0; | |
1163 } | |
1164 } | |
1165 } | |
1166 } | |
1167 } | |
1168 }else{ /* direct temporal mv pred */ | |
3316 | 1169 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]}; |
1170 const int *dist_scale_factor = h->dist_scale_factor; | |
1171 | |
1172 if(FRAME_MBAFF){ | |
1173 if(IS_INTERLACED(*mb_type)){ | |
1174 map_col_to_list0[0] = h->map_col_to_list0_field[0]; | |
1175 map_col_to_list0[1] = h->map_col_to_list0_field[1]; | |
1176 dist_scale_factor = h->dist_scale_factor_field; | |
1177 } | |
1178 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){ | |
1179 /* FIXME assumes direct_8x8_inference == 1 */ | |
1180 const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride; | |
1181 int mb_types_col[2]; | |
1182 int y_shift; | |
1183 | |
1184 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1 | |
1185 | (is_b8x8 ? 0 : MB_TYPE_DIRECT2) | |
1186 | (*mb_type & MB_TYPE_INTERLACED); | |
1187 sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16; | |
1188 | |
1189 if(IS_INTERLACED(*mb_type)){ | |
1190 /* frame to field scaling */ | |
1191 mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy]; | |
1192 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride]; | |
1193 if(s->mb_y&1){ | |
1194 l1ref0 -= 2*h->b8_stride; | |
1195 l1ref1 -= 2*h->b8_stride; | |
1196 l1mv0 -= 4*h->b_stride; | |
1197 l1mv1 -= 4*h->b_stride; | |
1198 } | |
1199 y_shift = 0; | |
1200 | |
1201 if( (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA) | |
1202 && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA) | |
1203 && !is_b8x8) | |
1204 *mb_type |= MB_TYPE_16x8; | |
1205 else | |
1206 *mb_type |= MB_TYPE_8x8; | |
1207 }else{ | |
1208 /* field to frame scaling */ | |
1209 /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1) | |
1210 * but in MBAFF, top and bottom POC are equal */ | |
1211 int dy = (s->mb_y&1) ? 1 : 2; | |
1212 mb_types_col[0] = | |
1213 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride]; | |
1214 l1ref0 += dy*h->b8_stride; | |
1215 l1ref1 += dy*h->b8_stride; | |
1216 l1mv0 += 2*dy*h->b_stride; | |
1217 l1mv1 += 2*dy*h->b_stride; | |
1218 y_shift = 2; | |
1219 | |
1220 if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8)) | |
1221 && !is_b8x8) | |
1222 *mb_type |= MB_TYPE_16x16; | |
1223 else | |
1224 *mb_type |= MB_TYPE_8x8; | |
1225 } | |
1226 | |
1227 for(i8=0; i8<4; i8++){ | |
1228 const int x8 = i8&1; | |
1229 const int y8 = i8>>1; | |
1230 int ref0, scale; | |
1231 const int16_t (*l1mv)[2]= l1mv0; | |
1232 | |
1233 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) | |
1234 continue; | |
1235 h->sub_mb_type[i8] = sub_mb_type; | |
1236 | |
1237 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1); | |
1238 if(IS_INTRA(mb_types_col[y8])){ | |
1239 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1); | |
1240 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); | |
1241 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); | |
1242 continue; | |
1243 } | |
1244 | |
1245 ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride]; | |
1246 if(ref0 >= 0) | |
1247 ref0 = map_col_to_list0[0][ref0*2>>y_shift]; | |
1248 else{ | |
1249 ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift]; | |
1250 l1mv= l1mv1; | |
1251 } | |
1252 scale = dist_scale_factor[ref0]; | |
1253 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1); | |
1254 | |
1255 { | |
1256 const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride]; | |
1257 int my_col = (mv_col[1]<<y_shift)/2; | |
1258 int mx = (scale * mv_col[0] + 128) >> 8; | |
1259 int my = (scale * my_col + 128) >> 8; | |
1260 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4); | |
1261 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4); | |
1262 } | |
1263 } | |
1264 return; | |
1265 } | |
1266 } | |
1267 | |
1268 /* one-to-one mv scaling */ | |
1269 | |
2396 | 1270 if(IS_16X16(*mb_type)){ |
4541 | 1271 int ref, mv0, mv1; |
1272 | |
2396 | 1273 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1); |
1274 if(IS_INTRA(mb_type_col)){ | |
4541 | 1275 ref=mv0=mv1=0; |
2396 | 1276 }else{ |
3316 | 1277 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]] |
1278 : map_col_to_list0[1][l1ref1[0]]; | |
1279 const int scale = dist_scale_factor[ref0]; | |
2809
75400dfbe117
fixing colocated mv if colocated block is L1 predicted for the temporal direct case
michael
parents:
2808
diff
changeset
|
1280 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0]; |
2396 | 1281 int mv_l0[2]; |
3316 | 1282 mv_l0[0] = (scale * mv_col[0] + 128) >> 8; |
1283 mv_l0[1] = (scale * mv_col[1] + 128) >> 8; | |
4541 | 1284 ref= ref0; |
1285 mv0= pack16to32(mv_l0[0],mv_l0[1]); | |
1286 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]); | |
1287 } | |
1288 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1); | |
1289 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4); | |
1290 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4); | |
2396 | 1291 }else{ |
1292 for(i8=0; i8<4; i8++){ | |
1293 const int x8 = i8&1; | |
1294 const int y8 = i8>>1; | |
3316 | 1295 int ref0, scale; |
2834 | 1296 const int16_t (*l1mv)[2]= l1mv0; |
2809
75400dfbe117
fixing colocated mv if colocated block is L1 predicted for the temporal direct case
michael
parents:
2808
diff
changeset
|
1297 |
2396 | 1298 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) |
1299 continue; | |
1300 h->sub_mb_type[i8] = sub_mb_type; | |
3316 | 1301 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1); |
2396 | 1302 if(IS_INTRA(mb_type_col)){ |
1303 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1); | |
1304 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); | |
1305 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); | |
1306 continue; | |
1307 } | |
2967 | 1308 |
2396 | 1309 ref0 = l1ref0[x8 + y8*h->b8_stride]; |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1310 if(ref0 >= 0) |
3316 | 1311 ref0 = map_col_to_list0[0][ref0]; |
2809
75400dfbe117
fixing colocated mv if colocated block is L1 predicted for the temporal direct case
michael
parents:
2808
diff
changeset
|
1312 else{ |
3316 | 1313 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]]; |
2809
75400dfbe117
fixing colocated mv if colocated block is L1 predicted for the temporal direct case
michael
parents:
2808
diff
changeset
|
1314 l1mv= l1mv1; |
75400dfbe117
fixing colocated mv if colocated block is L1 predicted for the temporal direct case
michael
parents:
2808
diff
changeset
|
1315 } |
3316 | 1316 scale = dist_scale_factor[ref0]; |
2967 | 1317 |
2396 | 1318 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1); |
3002 | 1319 if(IS_SUB_8X8(sub_mb_type)){ |
1320 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride]; | |
3316 | 1321 int mx = (scale * mv_col[0] + 128) >> 8; |
1322 int my = (scale * mv_col[1] + 128) >> 8; | |
3002 | 1323 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4); |
1324 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4); | |
1325 }else | |
2396 | 1326 for(i4=0; i4<4; i4++){ |
2809
75400dfbe117
fixing colocated mv if colocated block is L1 predicted for the temporal direct case
michael
parents:
2808
diff
changeset
|
1327 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride]; |
2396 | 1328 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]]; |
3316 | 1329 mv_l0[0] = (scale * mv_col[0] + 128) >> 8; |
1330 mv_l0[1] = (scale * mv_col[1] + 128) >> 8; | |
2396 | 1331 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = |
1332 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]); | |
1333 } | |
1334 } | |
1335 } | |
1336 } | |
1337 } | |
1338 | |
1168 | 1339 static inline void write_back_motion(H264Context *h, int mb_type){ |
1340 MpegEncContext * const s = &h->s; | |
1341 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; | |
1342 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride; | |
1343 int list; | |
1344 | |
3219
cf233efbcece
10l in 1.199 (write_back_motion with B-pyramid + spatial)
lorenm
parents:
3215
diff
changeset
|
1345 if(!USES_LIST(mb_type, 0)) |
cf233efbcece
10l in 1.199 (write_back_motion with B-pyramid + spatial)
lorenm
parents:
3215
diff
changeset
|
1346 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1); |
cf233efbcece
10l in 1.199 (write_back_motion with B-pyramid + spatial)
lorenm
parents:
3215
diff
changeset
|
1347 |
4533 | 1348 for(list=0; list<h->list_count; list++){ |
1168 | 1349 int y; |
3212 | 1350 if(!USES_LIST(mb_type, list)) |
2396 | 1351 continue; |
2967 | 1352 |
1168 | 1353 for(y=0; y<4; y++){ |
1354 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y]; | |
1355 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y]; | |
1356 } | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
1357 if( h->pps.cabac ) { |
3927 | 1358 if(IS_SKIP(mb_type)) |
1359 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4); | |
1360 else | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
1361 for(y=0; y<4; y++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
1362 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
1363 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
1364 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
1365 } |
3212 | 1366 |
1367 { | |
3776 | 1368 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy]; |
3212 | 1369 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]]; |
1370 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]]; | |
1371 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]]; | |
1372 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]]; | |
1168 | 1373 } |
1374 } | |
2967 | 1375 |
2396 | 1376 if(h->slice_type == B_TYPE && h->pps.cabac){ |
1377 if(IS_8X8(mb_type)){ | |
3212 | 1378 uint8_t *direct_table = &h->direct_table[b8_xy]; |
1379 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0; | |
1380 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0; | |
1381 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0; | |
2396 | 1382 } |
1383 } | |
1168 | 1384 } |
1385 | |
1386 /** | |
1387 * Decodes a network abstraction layer unit. | |
1388 * @param consumed is the number of bytes used as input | |
1389 * @param length is the length of the array | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
1390 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing? |
2967 | 1391 * @returns decoded bytes, might be src+1 if no escapes |
1168 | 1392 */ |
1393 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){ | |
1394 int i, si, di; | |
1395 uint8_t *dst; | |
1396 | |
2979 | 1397 // src[0]&0x80; //forbidden bit |
1168 | 1398 h->nal_ref_idc= src[0]>>5; |
1399 h->nal_unit_type= src[0]&0x1F; | |
1400 | |
1401 src++; length--; | |
2967 | 1402 #if 0 |
1168 | 1403 for(i=0; i<length; i++) |
1404 printf("%2X ", src[i]); | |
1405 #endif | |
1406 for(i=0; i+1<length; i+=2){ | |
1407 if(src[i]) continue; | |
1408 if(i>0 && src[i-1]==0) i--; | |
1409 if(i+2<length && src[i+1]==0 && src[i+2]<=3){ | |
1410 if(src[i+2]!=3){ | |
1411 /* startcode, so we must be past the end */ | |
1412 length=i; | |
1413 } | |
1414 break; | |
1415 } | |
1416 } | |
1417 | |
1418 if(i>=length-1){ //no escaped 0 | |
1419 *dst_length= length; | |
1420 *consumed= length+1; //+1 for the header | |
2967 | 1421 return src; |
1168 | 1422 } |
1423 | |
1424 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length); | |
1425 dst= h->rbsp_buffer; | |
1426 | |
4362
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
1427 if (dst == NULL){ |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
1428 return NULL; |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
1429 } |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
1430 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
1431 //printf("decoding esc\n"); |
1168 | 1432 si=di=0; |
2967 | 1433 while(si<length){ |
1168 | 1434 //remove escapes (very rare 1:2^22) |
1435 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){ | |
1436 if(src[si+2]==3){ //escape | |
1437 dst[di++]= 0; | |
1438 dst[di++]= 0; | |
1439 si+=3; | |
1957
54411768fa38
h264 nal decoding fix by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1956
diff
changeset
|
1440 continue; |
1168 | 1441 }else //next start code |
1442 break; | |
1443 } | |
1444 | |
1445 dst[di++]= src[si++]; | |
1446 } | |
1447 | |
1448 *dst_length= di; | |
1449 *consumed= si + 1;//+1 for the header | |
1450 //FIXME store exact number of bits in the getbitcontext (its needed for decoding) | |
1451 return dst; | |
1452 } | |
1453 | |
1454 /** | |
1455 * identifies the exact end of the bitstream | |
1456 * @return the length of the trailing, or 0 if damaged | |
1457 */ | |
4600 | 1458 static int decode_rbsp_trailing(H264Context *h, uint8_t *src){ |
1168 | 1459 int v= *src; |
1460 int r; | |
1461 | |
4600 | 1462 tprintf(h->s.avctx, "rbsp trailing %X\n", v); |
1168 | 1463 |
1464 for(r=1; r<9; r++){ | |
1465 if(v&1) return r; | |
1466 v>>=1; | |
1467 } | |
1468 return 0; | |
1469 } | |
1470 | |
1471 /** | |
1472 * idct tranforms the 16 dc values and dequantize them. | |
1473 * @param qp quantization parameter | |
1474 */ | |
2919 | 1475 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){ |
1168 | 1476 #define stride 16 |
1477 int i; | |
1478 int temp[16]; //FIXME check if this is a good idea | |
1479 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride}; | |
1480 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride}; | |
1481 | |
1482 //memset(block, 64, 2*256); | |
1483 //return; | |
1484 for(i=0; i<4; i++){ | |
1485 const int offset= y_offset[i]; | |
1486 const int z0= block[offset+stride*0] + block[offset+stride*4]; | |
1487 const int z1= block[offset+stride*0] - block[offset+stride*4]; | |
1488 const int z2= block[offset+stride*1] - block[offset+stride*5]; | |
1489 const int z3= block[offset+stride*1] + block[offset+stride*5]; | |
1490 | |
1491 temp[4*i+0]= z0+z3; | |
1492 temp[4*i+1]= z1+z2; | |
1493 temp[4*i+2]= z1-z2; | |
1494 temp[4*i+3]= z0-z3; | |
1495 } | |
1496 | |
1497 for(i=0; i<4; i++){ | |
1498 const int offset= x_offset[i]; | |
1499 const int z0= temp[4*0+i] + temp[4*2+i]; | |
1500 const int z1= temp[4*0+i] - temp[4*2+i]; | |
1501 const int z2= temp[4*1+i] - temp[4*3+i]; | |
1502 const int z3= temp[4*1+i] + temp[4*3+i]; | |
1503 | |
2919 | 1504 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual |
1505 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8)); | |
1506 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8)); | |
1507 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8)); | |
1168 | 1508 } |
1509 } | |
1510 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
1511 #if 0 |
1168 | 1512 /** |
1513 * dct tranforms the 16 dc values. | |
1514 * @param qp quantization parameter ??? FIXME | |
1515 */ | |
1516 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){ | |
1517 // const int qmul= dequant_coeff[qp][0]; | |
1518 int i; | |
1519 int temp[16]; //FIXME check if this is a good idea | |
1520 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride}; | |
1521 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride}; | |
1522 | |
1523 for(i=0; i<4; i++){ | |
1524 const int offset= y_offset[i]; | |
1525 const int z0= block[offset+stride*0] + block[offset+stride*4]; | |
1526 const int z1= block[offset+stride*0] - block[offset+stride*4]; | |
1527 const int z2= block[offset+stride*1] - block[offset+stride*5]; | |
1528 const int z3= block[offset+stride*1] + block[offset+stride*5]; | |
1529 | |
1530 temp[4*i+0]= z0+z3; | |
1531 temp[4*i+1]= z1+z2; | |
1532 temp[4*i+2]= z1-z2; | |
1533 temp[4*i+3]= z0-z3; | |
1534 } | |
1535 | |
1536 for(i=0; i<4; i++){ | |
1537 const int offset= x_offset[i]; | |
1538 const int z0= temp[4*0+i] + temp[4*2+i]; | |
1539 const int z1= temp[4*0+i] - temp[4*2+i]; | |
1540 const int z2= temp[4*1+i] - temp[4*3+i]; | |
1541 const int z3= temp[4*1+i] + temp[4*3+i]; | |
1542 | |
1543 block[stride*0 +offset]= (z0 + z3)>>1; | |
1544 block[stride*2 +offset]= (z1 + z2)>>1; | |
1545 block[stride*8 +offset]= (z1 - z2)>>1; | |
1546 block[stride*10+offset]= (z0 - z3)>>1; | |
1547 } | |
1548 } | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
1549 #endif |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
1550 |
1168 | 1551 #undef xStride |
1552 #undef stride | |
1553 | |
2919 | 1554 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){ |
1168 | 1555 const int stride= 16*2; |
1556 const int xStride= 16; | |
1557 int a,b,c,d,e; | |
1558 | |
1559 a= block[stride*0 + xStride*0]; | |
1560 b= block[stride*0 + xStride*1]; | |
1561 c= block[stride*1 + xStride*0]; | |
1562 d= block[stride*1 + xStride*1]; | |
1563 | |
1564 e= a-b; | |
1565 a= a+b; | |
1566 b= c-d; | |
1567 c= c+d; | |
1568 | |
2919 | 1569 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7; |
1570 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7; | |
1571 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7; | |
1572 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7; | |
1168 | 1573 } |
1574 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
1575 #if 0 |
1168 | 1576 static void chroma_dc_dct_c(DCTELEM *block){ |
1577 const int stride= 16*2; | |
1578 const int xStride= 16; | |
1579 int a,b,c,d,e; | |
1580 | |
1581 a= block[stride*0 + xStride*0]; | |
1582 b= block[stride*0 + xStride*1]; | |
1583 c= block[stride*1 + xStride*0]; | |
1584 d= block[stride*1 + xStride*1]; | |
1585 | |
1586 e= a-b; | |
1587 a= a+b; | |
1588 b= c-d; | |
1589 c= c+d; | |
1590 | |
1591 block[stride*0 + xStride*0]= (a+c); | |
1592 block[stride*0 + xStride*1]= (e+b); | |
1593 block[stride*1 + xStride*0]= (a-c); | |
1594 block[stride*1 + xStride*1]= (e-b); | |
1595 } | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
1596 #endif |
1168 | 1597 |
1598 /** | |
1599 * gets the chroma qp. | |
1600 */ | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
1601 static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){ |
2967 | 1602 |
4594 | 1603 return chroma_qp[av_clip(qscale + chroma_qp_index_offset, 0, 51)]; |
1168 | 1604 } |
1605 | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
1606 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close |
4553
1a714d3f0233
cosmetics: Fix a common typo, sepErate --> sepArate.
diego
parents:
4546
diff
changeset
|
1607 //FIXME check that gcc inlines this (and optimizes intra & separate_dc stuff away) |
1a714d3f0233
cosmetics: Fix a common typo, sepErate --> sepArate.
diego
parents:
4546
diff
changeset
|
1608 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int separate_dc){ |
1168 | 1609 int i; |
1610 const int * const quant_table= quant_coeff[qscale]; | |
1611 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6; | |
1612 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1; | |
1613 const unsigned int threshold2= (threshold1<<1); | |
1614 int last_non_zero; | |
1615 | |
4553
1a714d3f0233
cosmetics: Fix a common typo, sepErate --> sepArate.
diego
parents:
4546
diff
changeset
|
1616 if(separate_dc){ |
1168 | 1617 if(qscale<=18){ |
1618 //avoid overflows | |
1619 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6; | |
1620 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1; | |
1621 const unsigned int dc_threshold2= (dc_threshold1<<1); | |
1622 | |
1623 int level= block[0]*quant_coeff[qscale+18][0]; | |
1624 if(((unsigned)(level+dc_threshold1))>dc_threshold2){ | |
1625 if(level>0){ | |
1626 level= (dc_bias + level)>>(QUANT_SHIFT-2); | |
1627 block[0]= level; | |
1628 }else{ | |
1629 level= (dc_bias - level)>>(QUANT_SHIFT-2); | |
1630 block[0]= -level; | |
1631 } | |
1632 // last_non_zero = i; | |
1633 }else{ | |
1634 block[0]=0; | |
1635 } | |
1636 }else{ | |
1637 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6; | |
1638 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1; | |
1639 const unsigned int dc_threshold2= (dc_threshold1<<1); | |
1640 | |
1641 int level= block[0]*quant_table[0]; | |
1642 if(((unsigned)(level+dc_threshold1))>dc_threshold2){ | |
1643 if(level>0){ | |
1644 level= (dc_bias + level)>>(QUANT_SHIFT+1); | |
1645 block[0]= level; | |
1646 }else{ | |
1647 level= (dc_bias - level)>>(QUANT_SHIFT+1); | |
1648 block[0]= -level; | |
1649 } | |
1650 // last_non_zero = i; | |
1651 }else{ | |
1652 block[0]=0; | |
1653 } | |
1654 } | |
1655 last_non_zero= 0; | |
1656 i=1; | |
1657 }else{ | |
1658 last_non_zero= -1; | |
1659 i=0; | |
1660 } | |
1661 | |
1662 for(; i<16; i++){ | |
1663 const int j= scantable[i]; | |
1664 int level= block[j]*quant_table[j]; | |
1665 | |
1666 // if( bias+level >= (1<<(QMAT_SHIFT - 3)) | |
1667 // || bias-level >= (1<<(QMAT_SHIFT - 3))){ | |
1668 if(((unsigned)(level+threshold1))>threshold2){ | |
1669 if(level>0){ | |
1670 level= (bias + level)>>QUANT_SHIFT; | |
1671 block[j]= level; | |
1672 }else{ | |
1673 level= (bias - level)>>QUANT_SHIFT; | |
1674 block[j]= -level; | |
1675 } | |
1676 last_non_zero = i; | |
1677 }else{ | |
1678 block[j]=0; | |
1679 } | |
1680 } | |
1681 | |
1682 return last_non_zero; | |
1683 } | |
1684 | |
1685 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){ | |
1686 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
1687 ((uint32_t*)(src+0*stride))[0]= a; | |
1688 ((uint32_t*)(src+1*stride))[0]= a; | |
1689 ((uint32_t*)(src+2*stride))[0]= a; | |
1690 ((uint32_t*)(src+3*stride))[0]= a; | |
1691 } | |
1692 | |
1693 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){ | |
1694 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101; | |
1695 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101; | |
1696 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101; | |
1697 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101; | |
1698 } | |
1699 | |
1700 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1701 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] | |
1702 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3; | |
2967 | 1703 |
1704 ((uint32_t*)(src+0*stride))[0]= | |
1705 ((uint32_t*)(src+1*stride))[0]= | |
1706 ((uint32_t*)(src+2*stride))[0]= | |
1707 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1168 | 1708 } |
1709 | |
1710 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1711 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2; | |
2967 | 1712 |
1713 ((uint32_t*)(src+0*stride))[0]= | |
1714 ((uint32_t*)(src+1*stride))[0]= | |
1715 ((uint32_t*)(src+2*stride))[0]= | |
1716 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1168 | 1717 } |
1718 | |
1719 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1720 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2; | |
2967 | 1721 |
1722 ((uint32_t*)(src+0*stride))[0]= | |
1723 ((uint32_t*)(src+1*stride))[0]= | |
1724 ((uint32_t*)(src+2*stride))[0]= | |
1725 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1168 | 1726 } |
1727 | |
1728 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
2967 | 1729 ((uint32_t*)(src+0*stride))[0]= |
1730 ((uint32_t*)(src+1*stride))[0]= | |
1731 ((uint32_t*)(src+2*stride))[0]= | |
1168 | 1732 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U; |
1733 } | |
1734 | |
1735 | |
1736 #define LOAD_TOP_RIGHT_EDGE\ | |
1737 const int t4= topright[0];\ | |
1738 const int t5= topright[1];\ | |
1739 const int t6= topright[2];\ | |
1740 const int t7= topright[3];\ | |
1741 | |
1742 #define LOAD_LEFT_EDGE\ | |
1743 const int l0= src[-1+0*stride];\ | |
1744 const int l1= src[-1+1*stride];\ | |
1745 const int l2= src[-1+2*stride];\ | |
1746 const int l3= src[-1+3*stride];\ | |
1747 | |
1748 #define LOAD_TOP_EDGE\ | |
1749 const int t0= src[ 0-1*stride];\ | |
1750 const int t1= src[ 1-1*stride];\ | |
1751 const int t2= src[ 2-1*stride];\ | |
1752 const int t3= src[ 3-1*stride];\ | |
1753 | |
1754 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){ | |
1755 const int lt= src[-1-1*stride]; | |
1756 LOAD_TOP_EDGE | |
1757 LOAD_LEFT_EDGE | |
1758 | |
2967 | 1759 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; |
1168 | 1760 src[0+2*stride]= |
2967 | 1761 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; |
1168 | 1762 src[0+1*stride]= |
1763 src[1+2*stride]= | |
2967 | 1764 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; |
1168 | 1765 src[0+0*stride]= |
1766 src[1+1*stride]= | |
1767 src[2+2*stride]= | |
2967 | 1768 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; |
1168 | 1769 src[1+0*stride]= |
1770 src[2+1*stride]= | |
1771 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
1772 src[2+0*stride]= | |
1773 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1774 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
1282 | 1775 } |
1168 | 1776 |
1777 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){ | |
2967 | 1778 LOAD_TOP_EDGE |
1779 LOAD_TOP_RIGHT_EDGE | |
1780 // LOAD_LEFT_EDGE | |
1168 | 1781 |
1782 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2; | |
1783 src[1+0*stride]= | |
1784 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2; | |
1785 src[2+0*stride]= | |
1786 src[1+1*stride]= | |
1787 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2; | |
1788 src[3+0*stride]= | |
1789 src[2+1*stride]= | |
1790 src[1+2*stride]= | |
1791 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2; | |
1792 src[3+1*stride]= | |
1793 src[2+2*stride]= | |
1794 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2; | |
1795 src[3+2*stride]= | |
1796 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2; | |
1797 src[3+3*stride]=(t6 + 3*t7 + 2)>>2; | |
1282 | 1798 } |
1168 | 1799 |
1800 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){ | |
1801 const int lt= src[-1-1*stride]; | |
2967 | 1802 LOAD_TOP_EDGE |
1803 LOAD_LEFT_EDGE | |
1168 | 1804 const __attribute__((unused)) int unu= l3; |
1805 | |
1806 src[0+0*stride]= | |
1807 src[1+2*stride]=(lt + t0 + 1)>>1; | |
1808 src[1+0*stride]= | |
1809 src[2+2*stride]=(t0 + t1 + 1)>>1; | |
1810 src[2+0*stride]= | |
1811 src[3+2*stride]=(t1 + t2 + 1)>>1; | |
1812 src[3+0*stride]=(t2 + t3 + 1)>>1; | |
1813 src[0+1*stride]= | |
1814 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2; | |
1815 src[1+1*stride]= | |
1816 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
1817 src[2+1*stride]= | |
1818 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1819 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
1820 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2; | |
1821 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
1282 | 1822 } |
1168 | 1823 |
1824 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){ | |
2967 | 1825 LOAD_TOP_EDGE |
1826 LOAD_TOP_RIGHT_EDGE | |
1168 | 1827 const __attribute__((unused)) int unu= t7; |
1828 | |
1829 src[0+0*stride]=(t0 + t1 + 1)>>1; | |
1830 src[1+0*stride]= | |
1831 src[0+2*stride]=(t1 + t2 + 1)>>1; | |
1832 src[2+0*stride]= | |
1833 src[1+2*stride]=(t2 + t3 + 1)>>1; | |
1834 src[3+0*stride]= | |
1835 src[2+2*stride]=(t3 + t4+ 1)>>1; | |
1836 src[3+2*stride]=(t4 + t5+ 1)>>1; | |
1837 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1838 src[1+1*stride]= | |
1839 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
1840 src[2+1*stride]= | |
1841 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2; | |
1842 src[3+1*stride]= | |
1843 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2; | |
1844 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2; | |
1282 | 1845 } |
1168 | 1846 |
1847 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){ | |
2967 | 1848 LOAD_LEFT_EDGE |
1168 | 1849 |
1850 src[0+0*stride]=(l0 + l1 + 1)>>1; | |
1851 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
1852 src[2+0*stride]= | |
1853 src[0+1*stride]=(l1 + l2 + 1)>>1; | |
1854 src[3+0*stride]= | |
1855 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2; | |
1856 src[2+1*stride]= | |
1857 src[0+2*stride]=(l2 + l3 + 1)>>1; | |
1858 src[3+1*stride]= | |
1859 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2; | |
1860 src[3+2*stride]= | |
1861 src[1+3*stride]= | |
1862 src[0+3*stride]= | |
1863 src[2+2*stride]= | |
1864 src[2+3*stride]= | |
1865 src[3+3*stride]=l3; | |
1282 | 1866 } |
2967 | 1867 |
1168 | 1868 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){ |
1869 const int lt= src[-1-1*stride]; | |
2967 | 1870 LOAD_TOP_EDGE |
1871 LOAD_LEFT_EDGE | |
1168 | 1872 const __attribute__((unused)) int unu= t3; |
1873 | |
1874 src[0+0*stride]= | |
1875 src[2+1*stride]=(lt + l0 + 1)>>1; | |
1876 src[1+0*stride]= | |
1877 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2; | |
1878 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
1879 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1880 src[0+1*stride]= | |
1881 src[2+2*stride]=(l0 + l1 + 1)>>1; | |
1882 src[1+1*stride]= | |
1883 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2; | |
1884 src[0+2*stride]= | |
1885 src[2+3*stride]=(l1 + l2+ 1)>>1; | |
1886 src[1+2*stride]= | |
1887 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
1888 src[0+3*stride]=(l2 + l3 + 1)>>1; | |
1889 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2; | |
1282 | 1890 } |
1168 | 1891 |
4271
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
1892 void ff_pred16x16_vertical_c(uint8_t *src, int stride){ |
1168 | 1893 int i; |
1894 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
1895 const uint32_t b= ((uint32_t*)(src-stride))[1]; | |
1896 const uint32_t c= ((uint32_t*)(src-stride))[2]; | |
1897 const uint32_t d= ((uint32_t*)(src-stride))[3]; | |
2967 | 1898 |
1168 | 1899 for(i=0; i<16; i++){ |
1900 ((uint32_t*)(src+i*stride))[0]= a; | |
1901 ((uint32_t*)(src+i*stride))[1]= b; | |
1902 ((uint32_t*)(src+i*stride))[2]= c; | |
1903 ((uint32_t*)(src+i*stride))[3]= d; | |
1904 } | |
1905 } | |
1906 | |
4271
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
1907 void ff_pred16x16_horizontal_c(uint8_t *src, int stride){ |
1168 | 1908 int i; |
1909 | |
1910 for(i=0; i<16; i++){ | |
1911 ((uint32_t*)(src+i*stride))[0]= | |
1912 ((uint32_t*)(src+i*stride))[1]= | |
1913 ((uint32_t*)(src+i*stride))[2]= | |
1914 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101; | |
1915 } | |
1916 } | |
1917 | |
4271
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
1918 void ff_pred16x16_dc_c(uint8_t *src, int stride){ |
1168 | 1919 int i, dc=0; |
1920 | |
1921 for(i=0;i<16; i++){ | |
1922 dc+= src[-1+i*stride]; | |
1923 } | |
2967 | 1924 |
1168 | 1925 for(i=0;i<16; i++){ |
1926 dc+= src[i-stride]; | |
1927 } | |
1928 | |
1929 dc= 0x01010101*((dc + 16)>>5); | |
1930 | |
1931 for(i=0; i<16; i++){ | |
1932 ((uint32_t*)(src+i*stride))[0]= | |
1933 ((uint32_t*)(src+i*stride))[1]= | |
1934 ((uint32_t*)(src+i*stride))[2]= | |
1935 ((uint32_t*)(src+i*stride))[3]= dc; | |
1936 } | |
1937 } | |
1938 | |
1939 static void pred16x16_left_dc_c(uint8_t *src, int stride){ | |
1940 int i, dc=0; | |
1941 | |
1942 for(i=0;i<16; i++){ | |
1943 dc+= src[-1+i*stride]; | |
1944 } | |
2967 | 1945 |
1168 | 1946 dc= 0x01010101*((dc + 8)>>4); |
1947 | |
1948 for(i=0; i<16; i++){ | |
1949 ((uint32_t*)(src+i*stride))[0]= | |
1950 ((uint32_t*)(src+i*stride))[1]= | |
1951 ((uint32_t*)(src+i*stride))[2]= | |
1952 ((uint32_t*)(src+i*stride))[3]= dc; | |
1953 } | |
1954 } | |
1955 | |
1956 static void pred16x16_top_dc_c(uint8_t *src, int stride){ | |
1957 int i, dc=0; | |
1958 | |
1959 for(i=0;i<16; i++){ | |
1960 dc+= src[i-stride]; | |
1961 } | |
1962 dc= 0x01010101*((dc + 8)>>4); | |
1963 | |
1964 for(i=0; i<16; i++){ | |
1965 ((uint32_t*)(src+i*stride))[0]= | |
1966 ((uint32_t*)(src+i*stride))[1]= | |
1967 ((uint32_t*)(src+i*stride))[2]= | |
1968 ((uint32_t*)(src+i*stride))[3]= dc; | |
1969 } | |
1970 } | |
1971 | |
4271
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
1972 void ff_pred16x16_128_dc_c(uint8_t *src, int stride){ |
1168 | 1973 int i; |
1974 | |
1975 for(i=0; i<16; i++){ | |
1976 ((uint32_t*)(src+i*stride))[0]= | |
1977 ((uint32_t*)(src+i*stride))[1]= | |
1978 ((uint32_t*)(src+i*stride))[2]= | |
1979 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U; | |
1980 } | |
1981 } | |
1982 | |
1234 | 1983 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){ |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1984 int i, j, k; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1985 int a; |
4176 | 1986 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1987 const uint8_t * const src0 = src+7-stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1988 const uint8_t *src1 = src+8*stride-1; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1989 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1990 int H = src0[1] - src0[-1]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1991 int V = src1[0] - src2[ 0]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1992 for(k=2; k<=8; ++k) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1993 src1 += stride; src2 -= stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1994 H += k*(src0[k] - src0[-k]); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1995 V += k*(src1[0] - src2[ 0]); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1996 } |
1234 | 1997 if(svq3){ |
1998 H = ( 5*(H/4) ) / 16; | |
1999 V = ( 5*(V/4) ) / 16; | |
1330
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
2000 |
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
2001 /* required for 100% accuracy */ |
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
2002 i = H; H = V; V = i; |
1234 | 2003 }else{ |
2004 H = ( 5*H+32 ) >> 6; | |
2005 V = ( 5*V+32 ) >> 6; | |
2006 } | |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2007 |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2008 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2009 for(j=16; j>0; --j) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2010 int b = a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2011 a += V; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2012 for(i=-16; i<0; i+=4) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2013 src[16+i] = cm[ (b ) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2014 src[17+i] = cm[ (b+ H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2015 src[18+i] = cm[ (b+2*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2016 src[19+i] = cm[ (b+3*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2017 b += 4*H; |
1168 | 2018 } |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2019 src += stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2020 } |
1168 | 2021 } |
2022 | |
4271
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2023 void ff_pred16x16_plane_c(uint8_t *src, int stride){ |
1234 | 2024 pred16x16_plane_compat_c(src, stride, 0); |
2025 } | |
2026 | |
4271
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2027 void ff_pred8x8_vertical_c(uint8_t *src, int stride){ |
1168 | 2028 int i; |
2029 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
2030 const uint32_t b= ((uint32_t*)(src-stride))[1]; | |
2967 | 2031 |
1168 | 2032 for(i=0; i<8; i++){ |
2033 ((uint32_t*)(src+i*stride))[0]= a; | |
2034 ((uint32_t*)(src+i*stride))[1]= b; | |
2035 } | |
2036 } | |
2037 | |
4271
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2038 void ff_pred8x8_horizontal_c(uint8_t *src, int stride){ |
1168 | 2039 int i; |
2040 | |
2041 for(i=0; i<8; i++){ | |
2042 ((uint32_t*)(src+i*stride))[0]= | |
2043 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101; | |
2044 } | |
2045 } | |
2046 | |
4271
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2047 void ff_pred8x8_128_dc_c(uint8_t *src, int stride){ |
1168 | 2048 int i; |
2049 | |
2755 | 2050 for(i=0; i<8; i++){ |
2967 | 2051 ((uint32_t*)(src+i*stride))[0]= |
1168 | 2052 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U; |
2053 } | |
2054 } | |
2055 | |
2056 static void pred8x8_left_dc_c(uint8_t *src, int stride){ | |
2057 int i; | |
2058 int dc0, dc2; | |
2059 | |
2060 dc0=dc2=0; | |
2061 for(i=0;i<4; i++){ | |
2062 dc0+= src[-1+i*stride]; | |
2063 dc2+= src[-1+(i+4)*stride]; | |
2064 } | |
2065 dc0= 0x01010101*((dc0 + 2)>>2); | |
2066 dc2= 0x01010101*((dc2 + 2)>>2); | |
2067 | |
2068 for(i=0; i<4; i++){ | |
2069 ((uint32_t*)(src+i*stride))[0]= | |
2070 ((uint32_t*)(src+i*stride))[1]= dc0; | |
2071 } | |
2072 for(i=4; i<8; i++){ | |
2073 ((uint32_t*)(src+i*stride))[0]= | |
2074 ((uint32_t*)(src+i*stride))[1]= dc2; | |
2075 } | |
2076 } | |
2077 | |
2078 static void pred8x8_top_dc_c(uint8_t *src, int stride){ | |
2079 int i; | |
2080 int dc0, dc1; | |
2081 | |
2082 dc0=dc1=0; | |
2083 for(i=0;i<4; i++){ | |
2084 dc0+= src[i-stride]; | |
2085 dc1+= src[4+i-stride]; | |
2086 } | |
2087 dc0= 0x01010101*((dc0 + 2)>>2); | |
2088 dc1= 0x01010101*((dc1 + 2)>>2); | |
2089 | |
2090 for(i=0; i<4; i++){ | |
2091 ((uint32_t*)(src+i*stride))[0]= dc0; | |
2092 ((uint32_t*)(src+i*stride))[1]= dc1; | |
2093 } | |
2094 for(i=4; i<8; i++){ | |
2095 ((uint32_t*)(src+i*stride))[0]= dc0; | |
2096 ((uint32_t*)(src+i*stride))[1]= dc1; | |
2097 } | |
2098 } | |
2099 | |
2100 | |
4271
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2101 void ff_pred8x8_dc_c(uint8_t *src, int stride){ |
1168 | 2102 int i; |
2103 int dc0, dc1, dc2, dc3; | |
2104 | |
2105 dc0=dc1=dc2=0; | |
2106 for(i=0;i<4; i++){ | |
2107 dc0+= src[-1+i*stride] + src[i-stride]; | |
2108 dc1+= src[4+i-stride]; | |
2109 dc2+= src[-1+(i+4)*stride]; | |
2110 } | |
2111 dc3= 0x01010101*((dc1 + dc2 + 4)>>3); | |
2112 dc0= 0x01010101*((dc0 + 4)>>3); | |
2113 dc1= 0x01010101*((dc1 + 2)>>2); | |
2114 dc2= 0x01010101*((dc2 + 2)>>2); | |
2115 | |
2116 for(i=0; i<4; i++){ | |
2117 ((uint32_t*)(src+i*stride))[0]= dc0; | |
2118 ((uint32_t*)(src+i*stride))[1]= dc1; | |
2119 } | |
2120 for(i=4; i<8; i++){ | |
2121 ((uint32_t*)(src+i*stride))[0]= dc2; | |
2122 ((uint32_t*)(src+i*stride))[1]= dc3; | |
2123 } | |
2124 } | |
2125 | |
4271
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2126 void ff_pred8x8_plane_c(uint8_t *src, int stride){ |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2127 int j, k; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2128 int a; |
4176 | 2129 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2130 const uint8_t * const src0 = src+3-stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2131 const uint8_t *src1 = src+4*stride-1; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2132 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2133 int H = src0[1] - src0[-1]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2134 int V = src1[0] - src2[ 0]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2135 for(k=2; k<=4; ++k) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2136 src1 += stride; src2 -= stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2137 H += k*(src0[k] - src0[-k]); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2138 V += k*(src1[0] - src2[ 0]); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2139 } |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2140 H = ( 17*H+16 ) >> 5; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2141 V = ( 17*V+16 ) >> 5; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2142 |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2143 a = 16*(src1[0] + src2[8]+1) - 3*(V+H); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2144 for(j=8; j>0; --j) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2145 int b = a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2146 a += V; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2147 src[0] = cm[ (b ) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2148 src[1] = cm[ (b+ H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2149 src[2] = cm[ (b+2*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2150 src[3] = cm[ (b+3*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2151 src[4] = cm[ (b+4*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2152 src[5] = cm[ (b+5*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2153 src[6] = cm[ (b+6*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2154 src[7] = cm[ (b+7*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2155 src += stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2156 } |
1168 | 2157 } |
2158 | |
2755 | 2159 #define SRC(x,y) src[(x)+(y)*stride] |
2160 #define PL(y) \ | |
2161 const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2; | |
2162 #define PREDICT_8x8_LOAD_LEFT \ | |
2163 const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \ | |
2164 + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \ | |
2165 PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \ | |
2834 | 2166 const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2 |
2755 | 2167 |
2168 #define PT(x) \ | |
2169 const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2; | |
2170 #define PREDICT_8x8_LOAD_TOP \ | |
2171 const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \ | |
2172 + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \ | |
2173 PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \ | |
2834 | 2174 const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \ |
2757 | 2175 + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2 |
2755 | 2176 |
2177 #define PTR(x) \ | |
2178 t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2; | |
2179 #define PREDICT_8x8_LOAD_TOPRIGHT \ | |
2180 int t8, t9, t10, t11, t12, t13, t14, t15; \ | |
2181 if(has_topright) { \ | |
2182 PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \ | |
2183 t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \ | |
2184 } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1); | |
2185 | |
2186 #define PREDICT_8x8_LOAD_TOPLEFT \ | |
2757 | 2187 const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2 |
2755 | 2188 |
2189 #define PREDICT_8x8_DC(v) \ | |
2190 int y; \ | |
2191 for( y = 0; y < 8; y++ ) { \ | |
2192 ((uint32_t*)src)[0] = \ | |
2193 ((uint32_t*)src)[1] = v; \ | |
2194 src += stride; \ | |
2195 } | |
2196 | |
2197 static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2198 { | |
2199 PREDICT_8x8_DC(0x80808080); | |
2200 } | |
2201 static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2202 { | |
2203 PREDICT_8x8_LOAD_LEFT; | |
2204 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101; | |
2205 PREDICT_8x8_DC(dc); | |
2206 } | |
2207 static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2208 { | |
2209 PREDICT_8x8_LOAD_TOP; | |
2210 const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101; | |
2211 PREDICT_8x8_DC(dc); | |
2212 } | |
2213 static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2214 { | |
2215 PREDICT_8x8_LOAD_LEFT; | |
2216 PREDICT_8x8_LOAD_TOP; | |
2217 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7 | |
2218 +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101; | |
2219 PREDICT_8x8_DC(dc); | |
2220 } | |
2221 static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2222 { | |
2223 PREDICT_8x8_LOAD_LEFT; | |
2224 #define ROW(y) ((uint32_t*)(src+y*stride))[0] =\ | |
2225 ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y | |
2226 ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7); | |
2227 #undef ROW | |
2228 } | |
2229 static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2230 { | |
2231 int y; | |
2232 PREDICT_8x8_LOAD_TOP; | |
2233 src[0] = t0; | |
2234 src[1] = t1; | |
2235 src[2] = t2; | |
2236 src[3] = t3; | |
2237 src[4] = t4; | |
2238 src[5] = t5; | |
2239 src[6] = t6; | |
2240 src[7] = t7; | |
2241 for( y = 1; y < 8; y++ ) | |
2242 *(uint64_t*)(src+y*stride) = *(uint64_t*)src; | |
2243 } | |
2244 static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2245 { | |
2246 PREDICT_8x8_LOAD_TOP; | |
2247 PREDICT_8x8_LOAD_TOPRIGHT; | |
2248 SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2; | |
2249 SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2; | |
2250 SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2; | |
2251 SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2; | |
2252 SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2; | |
2253 SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2; | |
2254 SRC(0,6)=SRC(1,5)=SRC(2,4)=SRC(3,3)=SRC(4,2)=SRC(5,1)=SRC(6,0)= (t6 + 2*t7 + t8 + 2) >> 2; | |
2255 SRC(0,7)=SRC(1,6)=SRC(2,5)=SRC(3,4)=SRC(4,3)=SRC(5,2)=SRC(6,1)=SRC(7,0)= (t7 + 2*t8 + t9 + 2) >> 2; | |
2256 SRC(1,7)=SRC(2,6)=SRC(3,5)=SRC(4,4)=SRC(5,3)=SRC(6,2)=SRC(7,1)= (t8 + 2*t9 + t10 + 2) >> 2; | |
2257 SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2; | |
2258 SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2; | |
2259 SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2; | |
2260 SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2; | |
2261 SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2; | |
2262 SRC(7,7)= (t14 + 3*t15 + 2) >> 2; | |
2263 } | |
2264 static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2265 { | |
2266 PREDICT_8x8_LOAD_TOP; | |
2267 PREDICT_8x8_LOAD_LEFT; | |
2268 PREDICT_8x8_LOAD_TOPLEFT; | |
2269 SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2; | |
2270 SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2; | |
2271 SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2; | |
2272 SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2; | |
2273 SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2; | |
2274 SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2; | |
2275 SRC(0,1)=SRC(1,2)=SRC(2,3)=SRC(3,4)=SRC(4,5)=SRC(5,6)=SRC(6,7)= (l1 + 2*l0 + lt + 2) >> 2; | |
2276 SRC(0,0)=SRC(1,1)=SRC(2,2)=SRC(3,3)=SRC(4,4)=SRC(5,5)=SRC(6,6)=SRC(7,7)= (l0 + 2*lt + t0 + 2) >> 2; | |
2277 SRC(1,0)=SRC(2,1)=SRC(3,2)=SRC(4,3)=SRC(5,4)=SRC(6,5)=SRC(7,6)= (lt + 2*t0 + t1 + 2) >> 2; | |
2278 SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2; | |
2279 SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2; | |
2280 SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2; | |
2281 SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2; | |
2282 SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2; | |
2283 SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2; | |
2967 | 2284 |
2755 | 2285 } |
2286 static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2287 { | |
2288 PREDICT_8x8_LOAD_TOP; | |
2289 PREDICT_8x8_LOAD_LEFT; | |
2290 PREDICT_8x8_LOAD_TOPLEFT; | |
2291 SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2; | |
2292 SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2; | |
2293 SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2; | |
2294 SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2; | |
2295 SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2; | |
2296 SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2; | |
2297 SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2; | |
2298 SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1; | |
2299 SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2; | |
2300 SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1; | |
2301 SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2; | |
2302 SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1; | |
2303 SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2; | |
2304 SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1; | |
2305 SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2; | |
2306 SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1; | |
2307 SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2; | |
2308 SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1; | |
2309 SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2; | |
2310 SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1; | |
2311 SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2; | |
2312 SRC(7,0)= (t6 + t7 + 1) >> 1; | |
2313 } | |
2314 static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2315 { | |
2316 PREDICT_8x8_LOAD_TOP; | |
2317 PREDICT_8x8_LOAD_LEFT; | |
2318 PREDICT_8x8_LOAD_TOPLEFT; | |
2319 SRC(0,7)= (l6 + l7 + 1) >> 1; | |
2320 SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2; | |
2321 SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1; | |
2322 SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2; | |
2323 SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1; | |
2324 SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2; | |
2325 SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1; | |
2326 SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2; | |
2327 SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1; | |
2328 SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2; | |
2329 SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1; | |
2330 SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2; | |
2331 SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1; | |
2332 SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2; | |
2333 SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1; | |
2334 SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2; | |
2335 SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2; | |
2336 SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2; | |
2337 SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2; | |
2338 SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2; | |
2339 SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2; | |
2340 SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2; | |
2341 } | |
2342 static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2343 { | |
2344 PREDICT_8x8_LOAD_TOP; | |
2345 PREDICT_8x8_LOAD_TOPRIGHT; | |
2346 SRC(0,0)= (t0 + t1 + 1) >> 1; | |
2347 SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2; | |
2348 SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1; | |
2349 SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2; | |
2350 SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1; | |
2351 SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2; | |
2352 SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1; | |
2353 SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2; | |
2354 SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1; | |
2355 SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2; | |
2356 SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1; | |
2357 SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2; | |
2358 SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1; | |
2359 SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2; | |
2360 SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1; | |
2361 SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2; | |
2362 SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1; | |
2363 SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2; | |
2364 SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1; | |
2365 SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2; | |
2366 SRC(7,6)= (t10 + t11 + 1) >> 1; | |
2367 SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2; | |
2368 } | |
2369 static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2370 { | |
2371 PREDICT_8x8_LOAD_LEFT; | |
2372 SRC(0,0)= (l0 + l1 + 1) >> 1; | |
2373 SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2; | |
2374 SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1; | |
2375 SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2; | |
2376 SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1; | |
2377 SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2; | |
2378 SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1; | |
2379 SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2; | |
2380 SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1; | |
2381 SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2; | |
2382 SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1; | |
2383 SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2; | |
2384 SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1; | |
2385 SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2; | |
2386 SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)= | |
2387 SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)= | |
2388 SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)= | |
2389 SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7; | |
2390 } | |
2391 #undef PREDICT_8x8_LOAD_LEFT | |
2392 #undef PREDICT_8x8_LOAD_TOP | |
2393 #undef PREDICT_8x8_LOAD_TOPLEFT | |
2394 #undef PREDICT_8x8_LOAD_TOPRIGHT | |
2395 #undef PREDICT_8x8_DC | |
2396 #undef PTR | |
2397 #undef PT | |
2398 #undef PL | |
2399 #undef SRC | |
2400 | |
1168 | 2401 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list, |
2402 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
2403 int src_x_offset, int src_y_offset, | |
2404 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){ | |
2405 MpegEncContext * const s = &h->s; | |
2406 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8; | |
3316 | 2407 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8; |
1168 | 2408 const int luma_xy= (mx&3) + ((my&3)<<2); |
3316 | 2409 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize; |
2410 uint8_t * src_cb, * src_cr; | |
2411 int extra_width= h->emu_edge_width; | |
2412 int extra_height= h->emu_edge_height; | |
1168 | 2413 int emu=0; |
2414 const int full_mx= mx>>2; | |
2415 const int full_my= my>>2; | |
2861
3b920e274b26
fix emulated edge motion compensation when resolution%16!=0
lorenm
parents:
2846
diff
changeset
|
2416 const int pic_width = 16*s->mb_width; |
3316 | 2417 const int pic_height = 16*s->mb_height >> MB_MBAFF; |
2967 | 2418 |
4390 | 2419 if(!pic->data[0]) //FIXME this is unacceptable, some senseable error concealment must be done for missing reference frames |
3125 | 2420 return; |
2967 | 2421 |
1168 | 2422 if(mx&7) extra_width -= 3; |
2423 if(my&7) extra_height -= 3; | |
2967 | 2424 |
2425 if( full_mx < 0-extra_width | |
2426 || full_my < 0-extra_height | |
2427 || full_mx + 16/*FIXME*/ > pic_width + extra_width | |
2861
3b920e274b26
fix emulated edge motion compensation when resolution%16!=0
lorenm
parents:
2846
diff
changeset
|
2428 || full_my + 16/*FIXME*/ > pic_height + extra_height){ |
3316 | 2429 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height); |
2430 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize; | |
1168 | 2431 emu=1; |
2432 } | |
2967 | 2433 |
3316 | 2434 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps? |
1168 | 2435 if(!square){ |
3316 | 2436 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize); |
1168 | 2437 } |
2967 | 2438 |
1168 | 2439 if(s->flags&CODEC_FLAG_GRAY) return; |
2967 | 2440 |
3316 | 2441 if(MB_MBAFF){ |
2442 // chroma offset when predicting from a field of opposite parity | |
2443 my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1)); | |
2444 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1); | |
2445 } | |
2446 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize; | |
2447 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize; | |
1168 | 2448 |
2449 if(emu){ | |
3316 | 2450 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1); |
2451 src_cb= s->edge_emu_buffer; | |
2452 } | |
2453 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7); | |
2454 | |
2455 if(emu){ | |
2456 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1); | |
1168 | 2457 src_cr= s->edge_emu_buffer; |
2458 } | |
3316 | 2459 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7); |
1168 | 2460 } |
2461 | |
2415 | 2462 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta, |
1168 | 2463 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, |
2464 int x_offset, int y_offset, | |
2465 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put, | |
2466 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg, | |
2467 int list0, int list1){ | |
2468 MpegEncContext * const s = &h->s; | |
2469 qpel_mc_func *qpix_op= qpix_put; | |
2470 h264_chroma_mc_func chroma_op= chroma_put; | |
2967 | 2471 |
3316 | 2472 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize; |
2473 dest_cb += x_offset + y_offset*h->mb_uvlinesize; | |
2474 dest_cr += x_offset + y_offset*h->mb_uvlinesize; | |
1168 | 2475 x_offset += 8*s->mb_x; |
3316 | 2476 y_offset += 8*(s->mb_y >> MB_MBAFF); |
2967 | 2477 |
1168 | 2478 if(list0){ |
1169 | 2479 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ]; |
1168 | 2480 mc_dir_part(h, ref, n, square, chroma_height, delta, 0, |
2481 dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2482 qpix_op, chroma_op); | |
2483 | |
2484 qpix_op= qpix_avg; | |
2485 chroma_op= chroma_avg; | |
2486 } | |
2487 | |
2488 if(list1){ | |
1169 | 2489 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ]; |
1168 | 2490 mc_dir_part(h, ref, n, square, chroma_height, delta, 1, |
2491 dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2492 qpix_op, chroma_op); | |
2493 } | |
2494 } | |
2495 | |
2415 | 2496 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta, |
2497 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
2498 int x_offset, int y_offset, | |
2499 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put, | |
2500 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op, | |
2501 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg, | |
2502 int list0, int list1){ | |
2503 MpegEncContext * const s = &h->s; | |
2504 | |
3316 | 2505 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize; |
2506 dest_cb += x_offset + y_offset*h->mb_uvlinesize; | |
2507 dest_cr += x_offset + y_offset*h->mb_uvlinesize; | |
2415 | 2508 x_offset += 8*s->mb_x; |
3316 | 2509 y_offset += 8*(s->mb_y >> MB_MBAFF); |
2967 | 2510 |
2415 | 2511 if(list0 && list1){ |
2512 /* don't optimize for luma-only case, since B-frames usually | |
2513 * use implicit weights => chroma too. */ | |
2514 uint8_t *tmp_cb = s->obmc_scratchpad; | |
3316 | 2515 uint8_t *tmp_cr = s->obmc_scratchpad + 8; |
2516 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize; | |
2415 | 2517 int refn0 = h->ref_cache[0][ scan8[n] ]; |
2518 int refn1 = h->ref_cache[1][ scan8[n] ]; | |
2519 | |
2520 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0, | |
2521 dest_y, dest_cb, dest_cr, | |
2522 x_offset, y_offset, qpix_put, chroma_put); | |
2523 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1, | |
2524 tmp_y, tmp_cb, tmp_cr, | |
2525 x_offset, y_offset, qpix_put, chroma_put); | |
2526 | |
2527 if(h->use_weight == 2){ | |
2528 int weight0 = h->implicit_weight[refn0][refn1]; | |
2529 int weight1 = 64 - weight0; | |
3316 | 2530 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0); |
2531 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0); | |
2532 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0); | |
2415 | 2533 }else{ |
3316 | 2534 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom, |
2967 | 2535 h->luma_weight[0][refn0], h->luma_weight[1][refn1], |
3029 | 2536 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]); |
3316 | 2537 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom, |
2967 | 2538 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0], |
3029 | 2539 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]); |
3316 | 2540 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom, |
2967 | 2541 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1], |
3029 | 2542 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]); |
2415 | 2543 } |
2544 }else{ | |
2545 int list = list1 ? 1 : 0; | |
2546 int refn = h->ref_cache[list][ scan8[n] ]; | |
2547 Picture *ref= &h->ref_list[list][refn]; | |
2548 mc_dir_part(h, ref, n, square, chroma_height, delta, list, | |
2549 dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2550 qpix_put, chroma_put); | |
2551 | |
3316 | 2552 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom, |
2415 | 2553 h->luma_weight[list][refn], h->luma_offset[list][refn]); |
2554 if(h->use_weight_chroma){ | |
3316 | 2555 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom, |
2415 | 2556 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]); |
3316 | 2557 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom, |
2415 | 2558 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]); |
2559 } | |
2560 } | |
2561 } | |
2562 | |
2563 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta, | |
2564 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
2565 int x_offset, int y_offset, | |
2566 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put, | |
2567 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg, | |
2967 | 2568 h264_weight_func *weight_op, h264_biweight_func *weight_avg, |
2415 | 2569 int list0, int list1){ |
2570 if((h->use_weight==2 && list0 && list1 | |
2571 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32)) | |
2572 || h->use_weight==1) | |
2573 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr, | |
2574 x_offset, y_offset, qpix_put, chroma_put, | |
2575 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1); | |
2576 else | |
2577 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr, | |
2578 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1); | |
2579 } | |
2580 | |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2581 static inline void prefetch_motion(H264Context *h, int list){ |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2582 /* fetch pixels for estimated mv 4 macroblocks ahead |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2583 * optimized for 64byte cache lines */ |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2584 MpegEncContext * const s = &h->s; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2585 const int refn = h->ref_cache[list][scan8[0]]; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2586 if(refn >= 0){ |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2587 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2588 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2589 uint8_t **src= h->ref_list[list][refn].data; |
3316 | 2590 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64; |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2591 s->dsp.prefetch(src[0]+off, s->linesize, 4); |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2592 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2593 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2); |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2594 } |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2595 } |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2596 |
1168 | 2597 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, |
2598 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put), | |
2415 | 2599 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg), |
2600 h264_weight_func *weight_op, h264_biweight_func *weight_avg){ | |
1168 | 2601 MpegEncContext * const s = &h->s; |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
2602 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 2603 const int mb_type= s->current_picture.mb_type[mb_xy]; |
2967 | 2604 |
1168 | 2605 assert(IS_INTER(mb_type)); |
2967 | 2606 |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2607 prefetch_motion(h, 0); |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2608 |
1168 | 2609 if(IS_16X16(mb_type)){ |
2610 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0, | |
2611 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0], | |
2415 | 2612 &weight_op[0], &weight_avg[0], |
1168 | 2613 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); |
2614 }else if(IS_16X8(mb_type)){ | |
2615 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0, | |
2616 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0], | |
2415 | 2617 &weight_op[1], &weight_avg[1], |
1168 | 2618 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); |
2619 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4, | |
2620 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0], | |
2415 | 2621 &weight_op[1], &weight_avg[1], |
1168 | 2622 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1)); |
2623 }else if(IS_8X16(mb_type)){ | |
3316 | 2624 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0, |
1168 | 2625 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], |
2415 | 2626 &weight_op[2], &weight_avg[2], |
1168 | 2627 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); |
3316 | 2628 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0, |
1168 | 2629 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], |
2415 | 2630 &weight_op[2], &weight_avg[2], |
1168 | 2631 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1)); |
2632 }else{ | |
2633 int i; | |
2967 | 2634 |
1168 | 2635 assert(IS_8X8(mb_type)); |
2636 | |
2637 for(i=0; i<4; i++){ | |
2638 const int sub_mb_type= h->sub_mb_type[i]; | |
2639 const int n= 4*i; | |
2640 int x_offset= (i&1)<<2; | |
2641 int y_offset= (i&2)<<1; | |
2642 | |
2643 if(IS_SUB_8X8(sub_mb_type)){ | |
2644 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2645 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], | |
2415 | 2646 &weight_op[3], &weight_avg[3], |
1168 | 2647 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); |
2648 }else if(IS_SUB_8X4(sub_mb_type)){ | |
2649 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2650 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1], | |
2415 | 2651 &weight_op[4], &weight_avg[4], |
1168 | 2652 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); |
2653 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2, | |
2654 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1], | |
2415 | 2655 &weight_op[4], &weight_avg[4], |
1168 | 2656 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); |
2657 }else if(IS_SUB_4X8(sub_mb_type)){ | |
3316 | 2658 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset, |
1168 | 2659 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], |
2415 | 2660 &weight_op[5], &weight_avg[5], |
1168 | 2661 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); |
3316 | 2662 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset, |
1168 | 2663 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], |
2415 | 2664 &weight_op[5], &weight_avg[5], |
1168 | 2665 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); |
2666 }else{ | |
2667 int j; | |
2668 assert(IS_SUB_4X4(sub_mb_type)); | |
2669 for(j=0; j<4; j++){ | |
2670 int sub_x_offset= x_offset + 2*(j&1); | |
2671 int sub_y_offset= y_offset + (j&2); | |
2672 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset, | |
2673 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], | |
2415 | 2674 &weight_op[6], &weight_avg[6], |
1168 | 2675 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); |
2676 } | |
2677 } | |
2678 } | |
2679 } | |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2680 |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
2681 prefetch_motion(h, 1); |
1168 | 2682 } |
2683 | |
4783
011fb289e3b0
changes some function declarations from () to (void) as per ansi c.
gpoirier
parents:
4745
diff
changeset
|
2684 static void decode_init_vlc(void){ |
1168 | 2685 static int done = 0; |
2686 | |
2687 if (!done) { | |
2688 int i; | |
2689 done = 1; | |
2690 | |
2967 | 2691 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, |
1168 | 2692 &chroma_dc_coeff_token_len [0], 1, 1, |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2365
diff
changeset
|
2693 &chroma_dc_coeff_token_bits[0], 1, 1, 1); |
1168 | 2694 |
2695 for(i=0; i<4; i++){ | |
2967 | 2696 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, |
1168 | 2697 &coeff_token_len [i][0], 1, 1, |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2365
diff
changeset
|
2698 &coeff_token_bits[i][0], 1, 1, 1); |
1168 | 2699 } |
2700 | |
2701 for(i=0; i<3; i++){ | |
2702 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4, | |
2703 &chroma_dc_total_zeros_len [i][0], 1, 1, | |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2365
diff
changeset
|
2704 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1); |
1168 | 2705 } |
2706 for(i=0; i<15; i++){ | |
2967 | 2707 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, |
1168 | 2708 &total_zeros_len [i][0], 1, 1, |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2365
diff
changeset
|
2709 &total_zeros_bits[i][0], 1, 1, 1); |
1168 | 2710 } |
2711 | |
2712 for(i=0; i<6; i++){ | |
2967 | 2713 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, |
1168 | 2714 &run_len [i][0], 1, 1, |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2365
diff
changeset
|
2715 &run_bits[i][0], 1, 1, 1); |
1168 | 2716 } |
2967 | 2717 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, |
1168 | 2718 &run_len [6][0], 1, 1, |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2365
diff
changeset
|
2719 &run_bits[6][0], 1, 1, 1); |
1168 | 2720 } |
2721 } | |
2722 | |
2723 /** | |
2724 * Sets the intra prediction function pointers. | |
2725 */ | |
2726 static void init_pred_ptrs(H264Context *h){ | |
2727 // MpegEncContext * const s = &h->s; | |
2728 | |
2729 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c; | |
2730 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c; | |
2731 h->pred4x4[DC_PRED ]= pred4x4_dc_c; | |
2732 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c; | |
2733 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c; | |
2734 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c; | |
2735 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c; | |
2736 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c; | |
2737 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c; | |
2738 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c; | |
2739 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c; | |
2740 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c; | |
2741 | |
2755 | 2742 h->pred8x8l[VERT_PRED ]= pred8x8l_vertical_c; |
2743 h->pred8x8l[HOR_PRED ]= pred8x8l_horizontal_c; | |
2744 h->pred8x8l[DC_PRED ]= pred8x8l_dc_c; | |
2745 h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c; | |
2746 h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c; | |
2747 h->pred8x8l[VERT_RIGHT_PRED ]= pred8x8l_vertical_right_c; | |
2748 h->pred8x8l[HOR_DOWN_PRED ]= pred8x8l_horizontal_down_c; | |
2749 h->pred8x8l[VERT_LEFT_PRED ]= pred8x8l_vertical_left_c; | |
2750 h->pred8x8l[HOR_UP_PRED ]= pred8x8l_horizontal_up_c; | |
2751 h->pred8x8l[LEFT_DC_PRED ]= pred8x8l_left_dc_c; | |
2752 h->pred8x8l[TOP_DC_PRED ]= pred8x8l_top_dc_c; | |
2753 h->pred8x8l[DC_128_PRED ]= pred8x8l_128_dc_c; | |
2754 | |
4271
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2755 h->pred8x8[DC_PRED8x8 ]= ff_pred8x8_dc_c; |
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2756 h->pred8x8[VERT_PRED8x8 ]= ff_pred8x8_vertical_c; |
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2757 h->pred8x8[HOR_PRED8x8 ]= ff_pred8x8_horizontal_c; |
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2758 h->pred8x8[PLANE_PRED8x8 ]= ff_pred8x8_plane_c; |
1168 | 2759 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c; |
2760 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c; | |
4271
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2761 h->pred8x8[DC_128_PRED8x8 ]= ff_pred8x8_128_dc_c; |
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2762 |
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2763 h->pred16x16[DC_PRED8x8 ]= ff_pred16x16_dc_c; |
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2764 h->pred16x16[VERT_PRED8x8 ]= ff_pred16x16_vertical_c; |
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2765 h->pred16x16[HOR_PRED8x8 ]= ff_pred16x16_horizontal_c; |
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2766 h->pred16x16[PLANE_PRED8x8 ]= ff_pred16x16_plane_c; |
1168 | 2767 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c; |
2768 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c; | |
4271
8cb3d28d7e40
Export symbols of several H.264 intra prediction functions
takis
parents:
4270
diff
changeset
|
2769 h->pred16x16[DC_128_PRED8x8 ]= ff_pred16x16_128_dc_c; |
1168 | 2770 } |
2771 | |
2772 static void free_tables(H264Context *h){ | |
2773 av_freep(&h->intra4x4_pred_mode); | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
2774 av_freep(&h->chroma_pred_mode_table); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
2775 av_freep(&h->cbp_table); |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
2776 av_freep(&h->mvd_table[0]); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
2777 av_freep(&h->mvd_table[1]); |
2396 | 2778 av_freep(&h->direct_table); |
1168 | 2779 av_freep(&h->non_zero_count); |
2780 av_freep(&h->slice_table_base); | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
2781 av_freep(&h->top_borders[1]); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
2782 av_freep(&h->top_borders[0]); |
1168 | 2783 h->slice_table= NULL; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
2784 |
1168 | 2785 av_freep(&h->mb2b_xy); |
2786 av_freep(&h->mb2b8_xy); | |
2415 | 2787 |
2788 av_freep(&h->s.obmc_scratchpad); | |
1168 | 2789 } |
2790 | |
2919 | 2791 static void init_dequant8_coeff_table(H264Context *h){ |
2792 int i,q,x; | |
3174 | 2793 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly |
2919 | 2794 h->dequant8_coeff[0] = h->dequant8_buffer[0]; |
2795 h->dequant8_coeff[1] = h->dequant8_buffer[1]; | |
2796 | |
2797 for(i=0; i<2; i++ ){ | |
2798 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){ | |
2799 h->dequant8_coeff[1] = h->dequant8_buffer[0]; | |
2800 break; | |
2801 } | |
2802 | |
2803 for(q=0; q<52; q++){ | |
4277
113f3b395bac
Making rem6 and div6 globally visible and thus adding prefixes.
takis
parents:
4276
diff
changeset
|
2804 int shift = ff_div6[q]; |
113f3b395bac
Making rem6 and div6 globally visible and thus adding prefixes.
takis
parents:
4276
diff
changeset
|
2805 int idx = ff_rem6[q]; |
2919 | 2806 for(x=0; x<64; x++) |
3174 | 2807 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] = |
2808 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] * | |
2809 h->pps.scaling_matrix8[i][x]) << shift; | |
2919 | 2810 } |
2811 } | |
2812 } | |
2813 | |
2814 static void init_dequant4_coeff_table(H264Context *h){ | |
2815 int i,j,q,x; | |
3005 | 2816 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly |
2919 | 2817 for(i=0; i<6; i++ ){ |
2818 h->dequant4_coeff[i] = h->dequant4_buffer[i]; | |
2819 for(j=0; j<i; j++){ | |
2820 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){ | |
2821 h->dequant4_coeff[i] = h->dequant4_buffer[j]; | |
2822 break; | |
2823 } | |
2824 } | |
2825 if(j<i) | |
2826 continue; | |
2827 | |
2828 for(q=0; q<52; q++){ | |
4277
113f3b395bac
Making rem6 and div6 globally visible and thus adding prefixes.
takis
parents:
4276
diff
changeset
|
2829 int shift = ff_div6[q] + 2; |
113f3b395bac
Making rem6 and div6 globally visible and thus adding prefixes.
takis
parents:
4276
diff
changeset
|
2830 int idx = ff_rem6[q]; |
2919 | 2831 for(x=0; x<16; x++) |
3005 | 2832 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] = |
2833 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] * | |
2919 | 2834 h->pps.scaling_matrix4[i][x]) << shift; |
2835 } | |
2836 } | |
2837 } | |
2838 | |
2839 static void init_dequant_tables(H264Context *h){ | |
2840 int i,x; | |
2841 init_dequant4_coeff_table(h); | |
2842 if(h->pps.transform_8x8_mode) | |
2843 init_dequant8_coeff_table(h); | |
2844 if(h->sps.transform_bypass){ | |
2845 for(i=0; i<6; i++) | |
2846 for(x=0; x<16; x++) | |
2847 h->dequant4_coeff[i][0][x] = 1<<6; | |
2848 if(h->pps.transform_8x8_mode) | |
2849 for(i=0; i<2; i++) | |
2850 for(x=0; x<64; x++) | |
2851 h->dequant8_coeff[i][0][x] = 1<<6; | |
2852 } | |
2853 } | |
2854 | |
2855 | |
1168 | 2856 /** |
2857 * allocates tables. | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
2858 * needs width/height |
1168 | 2859 */ |
2860 static int alloc_tables(H264Context *h){ | |
2861 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
2862 const int big_mb_num= s->mb_stride * (s->mb_height+1); |
2919 | 2863 int x,y; |
1168 | 2864 |
2865 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t)) | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
2866 |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2867 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t)) |
3316 | 2868 CHECKED_ALLOCZ(h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(uint8_t)) |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
2869 CHECKED_ALLOCZ(h->top_borders[0] , s->mb_width * (16+8+8) * sizeof(uint8_t)) |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
2870 CHECKED_ALLOCZ(h->top_borders[1] , s->mb_width * (16+8+8) * sizeof(uint8_t)) |
2336 | 2871 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t)) |
1168 | 2872 |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
2873 if( h->pps.cabac ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
2874 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t)) |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
2875 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t)); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
2876 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t)); |
2396 | 2877 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t)); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
2878 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
2879 |
3316 | 2880 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(uint8_t)); |
2881 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1; | |
1168 | 2882 |
2641
c337f851d0f9
fix storage of motion vectors for frames with more than 4096 macroblocks.
lorenm
parents:
2640
diff
changeset
|
2883 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t)); |
c337f851d0f9
fix storage of motion vectors for frames with more than 4096 macroblocks.
lorenm
parents:
2640
diff
changeset
|
2884 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t)); |
1168 | 2885 for(y=0; y<s->mb_height; y++){ |
2886 for(x=0; x<s->mb_width; x++){ | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
2887 const int mb_xy= x + y*s->mb_stride; |
1168 | 2888 const int b_xy = 4*x + 4*y*h->b_stride; |
2889 const int b8_xy= 2*x + 2*y*h->b8_stride; | |
2967 | 2890 |
1168 | 2891 h->mb2b_xy [mb_xy]= b_xy; |
2892 h->mb2b8_xy[mb_xy]= b8_xy; | |
2893 } | |
2894 } | |
2415 | 2895 |
2417 | 2896 s->obmc_scratchpad = NULL; |
2897 | |
2920 | 2898 if(!h->dequant4_coeff[0]) |
2899 init_dequant_tables(h); | |
2900 | |
1168 | 2901 return 0; |
2902 fail: | |
2903 free_tables(h); | |
2904 return -1; | |
2905 } | |
2906 | |
2907 static void common_init(H264Context *h){ | |
2908 MpegEncContext * const s = &h->s; | |
2909 | |
2910 s->width = s->avctx->width; | |
2911 s->height = s->avctx->height; | |
2912 s->codec_id= s->avctx->codec->id; | |
2967 | 2913 |
1168 | 2914 init_pred_ptrs(h); |
2915 | |
2919 | 2916 h->dequant_coeff_pps= -1; |
1698 | 2917 s->unrestricted_mv=1; |
1168 | 2918 s->decode=1; //FIXME |
2920 | 2919 |
2920 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t)); | |
2921 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t)); | |
1168 | 2922 } |
2923 | |
2924 static int decode_init(AVCodecContext *avctx){ | |
2925 H264Context *h= avctx->priv_data; | |
2926 MpegEncContext * const s = &h->s; | |
2927 | |
1892 | 2928 MPV_decode_defaults(s); |
2967 | 2929 |
1168 | 2930 s->avctx = avctx; |
2931 common_init(h); | |
2932 | |
2933 s->out_format = FMT_H264; | |
2934 s->workaround_bugs= avctx->workaround_bugs; | |
2935 | |
2936 // set defaults | |
2937 // s->decode_mb= ff_h263_decode_mb; | |
2938 s->low_delay= 1; | |
2939 avctx->pix_fmt= PIX_FMT_YUV420P; | |
2940 | |
4164
171e768324cd
Remove the H264Context parameter from decode_init_vlc() as it is not being used.
takis
parents:
4118
diff
changeset
|
2941 decode_init_vlc(); |
2967 | 2942 |
2547
c5781912ad8a
improved detection of "AVC1" style H.264 patch by (M«©ns Rullg«©rd <mru inprovide com>)
michael
parents:
2538
diff
changeset
|
2943 if(avctx->extradata_size > 0 && avctx->extradata && |
c5781912ad8a
improved detection of "AVC1" style H.264 patch by (M«©ns Rullg«©rd <mru inprovide com>)
michael
parents:
2538
diff
changeset
|
2944 *(char *)avctx->extradata == 1){ |
2227 | 2945 h->is_avc = 1; |
2946 h->got_avcC = 0; | |
2547
c5781912ad8a
improved detection of "AVC1" style H.264 patch by (M«©ns Rullg«©rd <mru inprovide com>)
michael
parents:
2538
diff
changeset
|
2947 } else { |
c5781912ad8a
improved detection of "AVC1" style H.264 patch by (M«©ns Rullg«©rd <mru inprovide com>)
michael
parents:
2538
diff
changeset
|
2948 h->is_avc = 0; |
2227 | 2949 } |
2950 | |
1168 | 2951 return 0; |
2952 } | |
2953 | |
2935 | 2954 static int frame_start(H264Context *h){ |
1168 | 2955 MpegEncContext * const s = &h->s; |
2956 int i; | |
2957 | |
2935 | 2958 if(MPV_frame_start(s, s->avctx) < 0) |
2959 return -1; | |
1168 | 2960 ff_er_frame_start(s); |
2961 | |
2962 assert(s->linesize && s->uvlinesize); | |
2963 | |
2964 for(i=0; i<16; i++){ | |
2965 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3); | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
2966 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3); |
1168 | 2967 } |
2968 for(i=0; i<4; i++){ | |
2969 h->block_offset[16+i]= | |
2970 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3); | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
2971 h->block_offset[24+16+i]= |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
2972 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3); |
1168 | 2973 } |
2974 | |
2416
06b7d678b582
10l: scratchpad could be allocated before its size was known.
lorenm
parents:
2415
diff
changeset
|
2975 /* can't be in alloc_tables because linesize isn't known there. |
06b7d678b582
10l: scratchpad could be allocated before its size was known.
lorenm
parents:
2415
diff
changeset
|
2976 * FIXME: redo bipred weight to not require extra buffer? */ |
06b7d678b582
10l: scratchpad could be allocated before its size was known.
lorenm
parents:
2415
diff
changeset
|
2977 if(!s->obmc_scratchpad) |
3316 | 2978 s->obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize); |
2979 | |
2980 /* some macroblocks will be accessed before they're available */ | |
2981 if(FRAME_MBAFF) | |
2982 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t)); | |
2416
06b7d678b582
10l: scratchpad could be allocated before its size was known.
lorenm
parents:
2415
diff
changeset
|
2983 |
1168 | 2984 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1; |
2935 | 2985 return 0; |
1168 | 2986 } |
2987 | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2988 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2989 MpegEncContext * const s = &h->s; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2990 int i; |
2967 | 2991 |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2992 src_y -= linesize; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2993 src_cb -= uvlinesize; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2994 src_cr -= uvlinesize; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2995 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
2996 // There are two lines saved, the line above the the top macroblock of a pair, |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
2997 // and the line above the bottom macroblock |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
2998 h->left_border[0]= h->top_borders[0][s->mb_x][15]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2999 for(i=1; i<17; i++){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3000 h->left_border[i]= src_y[15+i* linesize]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3001 } |
2967 | 3002 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3003 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3004 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize); |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3005 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3006 if(!(s->flags&CODEC_FLAG_GRAY)){ |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3007 h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3008 h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3009 for(i=1; i<9; i++){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3010 h->left_border[i+17 ]= src_cb[7+i*uvlinesize]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3011 h->left_border[i+17+9]= src_cr[7+i*uvlinesize]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3012 } |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3013 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3014 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize); |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3015 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3016 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3017 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3018 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3019 MpegEncContext * const s = &h->s; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3020 int temp8, i; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3021 uint64_t temp64; |
2200
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3022 int deblock_left = (s->mb_x > 0); |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3023 int deblock_top = (s->mb_y > 0); |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3024 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3025 src_y -= linesize + 1; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3026 src_cb -= uvlinesize + 1; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3027 src_cr -= uvlinesize + 1; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3028 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3029 #define XCHG(a,b,t,xchg)\ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3030 t= a;\ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3031 if(xchg)\ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3032 a= b;\ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3033 b= t; |
2200
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3034 |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3035 if(deblock_left){ |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3036 for(i = !deblock_top; i<17; i++){ |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3037 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg); |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3038 } |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3039 } |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3040 |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3041 if(deblock_top){ |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3042 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3043 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1); |
2803 | 3044 if(s->mb_x+1 < s->mb_width){ |
2755 | 3045 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1); |
3046 } | |
2200
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3047 } |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3048 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3049 if(!(s->flags&CODEC_FLAG_GRAY)){ |
2200
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3050 if(deblock_left){ |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3051 for(i = !deblock_top; i<9; i++){ |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3052 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg); |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3053 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg); |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3054 } |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3055 } |
2200
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3056 if(deblock_top){ |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3057 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3058 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3059 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3060 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3061 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3062 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3063 static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3064 MpegEncContext * const s = &h->s; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3065 int i; |
2967 | 3066 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3067 src_y -= 2 * linesize; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3068 src_cb -= 2 * uvlinesize; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3069 src_cr -= 2 * uvlinesize; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3070 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
3071 // There are two lines saved, the line above the the top macroblock of a pair, |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3072 // and the line above the bottom macroblock |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3073 h->left_border[0]= h->top_borders[0][s->mb_x][15]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3074 h->left_border[1]= h->top_borders[1][s->mb_x][15]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3075 for(i=2; i<34; i++){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3076 h->left_border[i]= src_y[15+i* linesize]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3077 } |
2967 | 3078 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3079 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3080 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3081 *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3082 *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3083 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3084 if(!(s->flags&CODEC_FLAG_GRAY)){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3085 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3086 h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3087 h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3088 h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3089 for(i=2; i<18; i++){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3090 h->left_border[i+34 ]= src_cb[7+i*uvlinesize]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3091 h->left_border[i+34+18]= src_cr[7+i*uvlinesize]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3092 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3093 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3094 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3095 *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3096 *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3097 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3098 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3099 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3100 static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3101 MpegEncContext * const s = &h->s; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3102 int temp8, i; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3103 uint64_t temp64; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3104 int deblock_left = (s->mb_x > 0); |
3316 | 3105 int deblock_top = (s->mb_y > 1); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3106 |
4600 | 3107 tprintf(s->avctx, "xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3108 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3109 src_y -= 2 * linesize + 1; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3110 src_cb -= 2 * uvlinesize + 1; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3111 src_cr -= 2 * uvlinesize + 1; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3112 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3113 #define XCHG(a,b,t,xchg)\ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3114 t= a;\ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3115 if(xchg)\ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3116 a= b;\ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3117 b= t; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3118 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3119 if(deblock_left){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3120 for(i = (!deblock_top)<<1; i<34; i++){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3121 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3122 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3123 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3124 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3125 if(deblock_top){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3126 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3127 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3128 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3129 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1); |
3316 | 3130 if(s->mb_x+1 < s->mb_width){ |
3131 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1); | |
3132 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1); | |
3133 } | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3134 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3135 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3136 if(!(s->flags&CODEC_FLAG_GRAY)){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3137 if(deblock_left){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3138 for(i = (!deblock_top) << 1; i<18; i++){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3139 XCHG(h->left_border[i+34 ], src_cb[i*uvlinesize], temp8, xchg); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3140 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3141 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3142 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3143 if(deblock_top){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3144 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3145 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3146 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3147 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1); |
2200
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3148 } |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3149 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3150 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3151 |
4908
777f250df232
Fix multiple "¡Æinline/static¡Ç is not at beginning of declaration" warnings.
diego
parents:
4896
diff
changeset
|
3152 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){ |
1168 | 3153 MpegEncContext * const s = &h->s; |
3154 const int mb_x= s->mb_x; | |
3155 const int mb_y= s->mb_y; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
3156 const int mb_xy= mb_x + mb_y*s->mb_stride; |
1168 | 3157 const int mb_type= s->current_picture.mb_type[mb_xy]; |
3158 uint8_t *dest_y, *dest_cb, *dest_cr; | |
3159 int linesize, uvlinesize /*dct_offset*/; | |
3160 int i; | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3161 int *block_offset = &h->block_offset[0]; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3162 const unsigned int bottom = mb_y & 1; |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3163 const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass), is_h264 = (simple || s->codec_id == CODEC_ID_H264); |
2763 | 3164 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride); |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3165 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride); |
1168 | 3166 |
3167 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16; | |
3168 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; | |
3169 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; | |
3170 | |
4000 | 3171 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4); |
3172 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2); | |
3173 | |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3174 if (!simple && MB_FIELD) { |
3316 | 3175 linesize = h->mb_linesize = s->linesize * 2; |
3176 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2; | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3177 block_offset = &h->block_offset[24]; |
1168 | 3178 if(mb_y&1){ //FIXME move out of this func? |
3179 dest_y -= s->linesize*15; | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3180 dest_cb-= s->uvlinesize*7; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3181 dest_cr-= s->uvlinesize*7; |
1168 | 3182 } |
3316 | 3183 if(FRAME_MBAFF) { |
3184 int list; | |
4533 | 3185 for(list=0; list<h->list_count; list++){ |
3316 | 3186 if(!USES_LIST(mb_type, list)) |
3187 continue; | |
3188 if(IS_16X16(mb_type)){ | |
3189 int8_t *ref = &h->ref_cache[list][scan8[0]]; | |
3190 fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1); | |
3191 }else{ | |
3192 for(i=0; i<16; i+=4){ | |
3193 //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ? | |
3194 int ref = h->ref_cache[list][scan8[i]]; | |
3195 if(ref >= 0) | |
3196 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1); | |
3197 } | |
3198 } | |
3199 } | |
3200 } | |
1168 | 3201 } else { |
3316 | 3202 linesize = h->mb_linesize = s->linesize; |
3203 uvlinesize = h->mb_uvlinesize = s->uvlinesize; | |
1168 | 3204 // dct_offset = s->linesize * 16; |
3205 } | |
2967 | 3206 |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3207 if(transform_bypass){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3208 idct_dc_add = |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3209 idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3210 }else if(IS_8x8DCT(mb_type)){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3211 idct_dc_add = s->dsp.h264_idct8_dc_add; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3212 idct_add = s->dsp.h264_idct8_add; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3213 }else{ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3214 idct_dc_add = s->dsp.h264_idct_dc_add; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3215 idct_add = s->dsp.h264_idct_add; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3216 } |
1168 | 3217 |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3218 if(!simple && FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type) |
3316 | 3219 && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){ |
3220 int mbt_y = mb_y&~1; | |
3221 uint8_t *top_y = s->current_picture.data[0] + (mbt_y * 16* s->linesize ) + mb_x * 16; | |
3222 uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8; | |
3223 uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8; | |
3224 xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1); | |
3225 } | |
3226 | |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3227 if (!simple && IS_INTRA_PCM(mb_type)) { |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3228 unsigned int x, y; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3229 |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3230 // The pixels are stored in h->mb array in the same order as levels, |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3231 // copy them in output in the correct order. |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3232 for(i=0; i<16; i++) { |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3233 for (y=0; y<4; y++) { |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3234 for (x=0; x<4; x++) { |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3235 *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x]; |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3236 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3237 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3238 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3239 for(i=16; i<16+4; i++) { |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3240 for (y=0; y<4; y++) { |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3241 for (x=0; x<4; x++) { |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3242 *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x]; |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3243 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3244 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3245 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3246 for(i=20; i<20+4; i++) { |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3247 for (y=0; y<4; y++) { |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3248 for (x=0; x<4; x++) { |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3249 *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x]; |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3250 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3251 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3252 } |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3253 } else { |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3254 if(IS_INTRA(mb_type)){ |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3255 if(h->deblocking_filter && (simple || !FRAME_MBAFF)) |
3316 | 3256 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1); |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3257 |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3258 if(simple || !(s->flags&CODEC_FLAG_GRAY)){ |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3259 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize); |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3260 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize); |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3261 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3262 |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3263 if(IS_INTRA4x4(mb_type)){ |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3264 if(simple || !s->encoding){ |
2755 | 3265 if(IS_8x8DCT(mb_type)){ |
3266 for(i=0; i<16; i+=4){ | |
3267 uint8_t * const ptr= dest_y + block_offset[i]; | |
3268 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ]; | |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3269 const int nnz = h->non_zero_count_cache[ scan8[i] ]; |
2755 | 3270 h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000, |
4546 | 3271 (h->topright_samples_available<<i)&0x4000, linesize); |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3272 if(nnz){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3273 if(nnz == 1 && h->mb[i*16]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3274 idct_dc_add(ptr, h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3275 else |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3276 idct_add(ptr, h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3277 } |
2755 | 3278 } |
3279 }else | |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3280 for(i=0; i<16; i++){ |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3281 uint8_t * const ptr= dest_y + block_offset[i]; |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3282 uint8_t *topright; |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3283 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ]; |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3284 int nnz, tr; |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3285 |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3286 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3287 const int topright_avail= (h->topright_samples_available<<i)&0x8000; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3288 assert(mb_y || linesize <= block_offset[i]); |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3289 if(!topright_avail){ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3290 tr= ptr[3 - linesize]*0x01010101; |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3291 topright= (uint8_t*) &tr; |
2967 | 3292 }else |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3293 topright= ptr + 4 - linesize; |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3294 }else |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3295 topright= NULL; |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3296 |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3297 h->pred4x4[ dir ](ptr, topright, linesize); |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3298 nnz = h->non_zero_count_cache[ scan8[i] ]; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3299 if(nnz){ |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3300 if(is_h264){ |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3301 if(nnz == 1 && h->mb[i*16]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3302 idct_dc_add(ptr, h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3303 else |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3304 idct_add(ptr, h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3305 }else |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3306 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0); |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3307 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3308 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3309 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3310 }else{ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3311 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize); |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3312 if(is_h264){ |
2763 | 3313 if(!transform_bypass) |
2919 | 3314 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]); |
2763 | 3315 }else |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3316 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale); |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3317 } |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3318 if(h->deblocking_filter && (simple || !FRAME_MBAFF)) |
3316 | 3319 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0); |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3320 }else if(is_h264){ |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3321 hl_motion(h, dest_y, dest_cb, dest_cr, |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
3322 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab, |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
3323 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab, |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3324 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab); |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3325 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3326 |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3327 |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3328 if(!IS_INTRA4x4(mb_type)){ |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3329 if(is_h264){ |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3330 if(IS_INTRA16x16(mb_type)){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3331 for(i=0; i<16; i++){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3332 if(h->non_zero_count_cache[ scan8[i] ]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3333 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3334 else if(h->mb[i*16]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3335 idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3336 } |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3337 }else{ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3338 const int di = IS_8x8DCT(mb_type) ? 4 : 1; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3339 for(i=0; i<16; i+=di){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3340 int nnz = h->non_zero_count_cache[ scan8[i] ]; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3341 if(nnz){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3342 if(nnz==1 && h->mb[i*16]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3343 idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3344 else |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3345 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3346 } |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3347 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3348 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3349 }else{ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3350 for(i=0; i<16; i++){ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3351 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3352 uint8_t * const ptr= dest_y + block_offset[i]; |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3353 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0); |
1234 | 3354 } |
1168 | 3355 } |
3356 } | |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3357 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3358 |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3359 if(simple || !(s->flags&CODEC_FLAG_GRAY)){ |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3360 uint8_t *dest[2] = {dest_cb, dest_cr}; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3361 if(transform_bypass){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3362 idct_add = idct_dc_add = s->dsp.add_pixels4; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3363 }else{ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3364 idct_add = s->dsp.h264_idct_add; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3365 idct_dc_add = s->dsp.h264_idct_dc_add; |
2919 | 3366 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp][0]); |
3367 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp][0]); | |
2763 | 3368 } |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3369 if(is_h264){ |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3370 for(i=16; i<16+8; i++){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3371 if(h->non_zero_count_cache[ scan8[i] ]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3372 idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3373 else if(h->mb[i*16]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3374 idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize); |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3375 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3376 }else{ |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3377 for(i=16; i<16+8; i++){ |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3378 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3379 uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i]; |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3380 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2); |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3381 } |
1250 | 3382 } |
1168 | 3383 } |
3384 } | |
3385 } | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3386 if(h->deblocking_filter) { |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3387 if (!simple && FRAME_MBAFF) { |
3316 | 3388 //FIXME try deblocking one mb at a time? |
3389 // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3390 const int mb_y = s->mb_y - 1; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3391 uint8_t *pair_dest_y, *pair_dest_cb, *pair_dest_cr; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3392 const int mb_xy= mb_x + mb_y*s->mb_stride; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3393 const int mb_type_top = s->current_picture.mb_type[mb_xy]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3394 const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3395 if (!bottom) return; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3396 pair_dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3397 pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3398 pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3399 |
3316 | 3400 if(IS_INTRA(mb_type_top | mb_type_bottom)) |
3401 xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0); | |
3402 | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3403 backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize); |
3316 | 3404 // deblock a pair |
2967 | 3405 // top |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3406 s->mb_y--; |
4600 | 3407 tprintf(h->s.avctx, "call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y); |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
3408 fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb |
3316 | 3409 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy]); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3410 filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3411 // bottom |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3412 s->mb_y++; |
4600 | 3413 tprintf(h->s.avctx, "call mbaff filter_mb\n"); |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
3414 fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb |
3316 | 3415 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy+s->mb_stride]); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3416 filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3417 } else { |
4600 | 3418 tprintf(h->s.avctx, "call filter_mb\n"); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3419 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize); |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
3420 fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
3421 filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3422 } |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3423 } |
1168 | 3424 } |
3425 | |
3426 /** | |
4587
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3427 * Process a macroblock; this case avoids checks for expensive uncommon cases. |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3428 */ |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3429 static void hl_decode_mb_simple(H264Context *h){ |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3430 hl_decode_mb_internal(h, 1); |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3431 } |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3432 |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3433 /** |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3434 * Process a macroblock; this handles edge cases, such as interlacing. |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3435 */ |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3436 static void av_noinline hl_decode_mb_complex(H264Context *h){ |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3437 hl_decode_mb_internal(h, 0); |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3438 } |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3439 |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3440 static void hl_decode_mb(H264Context *h){ |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3441 MpegEncContext * const s = &h->s; |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3442 const int mb_x= s->mb_x; |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3443 const int mb_y= s->mb_y; |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3444 const int mb_xy= mb_x + mb_y*s->mb_stride; |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3445 const int mb_type= s->current_picture.mb_type[mb_xy]; |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3446 int is_complex = FRAME_MBAFF || MB_FIELD || IS_INTRA_PCM(mb_type) || s->codec_id != CODEC_ID_H264 || (s->flags&CODEC_FLAG_GRAY) || s->encoding; |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3447 |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3448 if(!s->decode) |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3449 return; |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3450 |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3451 if (is_complex) |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3452 hl_decode_mb_complex(h); |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3453 else hl_decode_mb_simple(h); |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3454 } |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3455 |
644687c58928
h264 optimization: common case hl_decode_mb patch by (Alexander Strange astrange ithinksw com)
michael
parents:
4556
diff
changeset
|
3456 /** |
1168 | 3457 * fills the default_ref_list. |
3458 */ | |
3459 static int fill_default_ref_list(H264Context *h){ | |
3460 MpegEncContext * const s = &h->s; | |
3461 int i; | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3462 int smallest_poc_greater_than_current = -1; |
2582 | 3463 Picture sorted_short_ref[32]; |
2967 | 3464 |
1168 | 3465 if(h->slice_type==B_TYPE){ |
3466 int out_i; | |
2713 | 3467 int limit= INT_MIN; |
1168 | 3468 |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3469 /* sort frame according to poc in B slice */ |
1168 | 3470 for(out_i=0; out_i<h->short_ref_count; out_i++){ |
2713 | 3471 int best_i=INT_MIN; |
2255
507690ff49a2
assertion when playing AVC/H.264 streams fix by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2254
diff
changeset
|
3472 int best_poc=INT_MAX; |
1168 | 3473 |
3474 for(i=0; i<h->short_ref_count; i++){ | |
3475 const int poc= h->short_ref[i]->poc; | |
3476 if(poc > limit && poc < best_poc){ | |
3477 best_poc= poc; | |
3478 best_i= i; | |
3479 } | |
3480 } | |
2967 | 3481 |
2713 | 3482 assert(best_i != INT_MIN); |
2967 | 3483 |
1168 | 3484 limit= best_poc; |
3485 sorted_short_ref[out_i]= *h->short_ref[best_i]; | |
4600 | 3486 tprintf(h->s.avctx, "sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num); |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3487 if (-1 == smallest_poc_greater_than_current) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3488 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3489 smallest_poc_greater_than_current = out_i; |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3490 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3491 } |
1168 | 3492 } |
3493 } | |
3494 | |
3495 if(s->picture_structure == PICT_FRAME){ | |
3496 if(h->slice_type==B_TYPE){ | |
3497 int list; | |
4600 | 3498 tprintf(h->s.avctx, "current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current); |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3499 |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3500 // find the largest poc |
1168 | 3501 for(list=0; list<2; list++){ |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3502 int index = 0; |
2447 | 3503 int j= -99; |
3504 int step= list ? -1 : 1; | |
3505 | |
3506 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) { | |
3507 while(j<0 || j>= h->short_ref_count){ | |
2680
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
3508 if(j != -99 && step == (list ? -1 : 1)) |
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
3509 return -1; |
2447 | 3510 step = -step; |
3511 j= smallest_poc_greater_than_current + (step>>1); | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3512 } |
2447 | 3513 if(sorted_short_ref[j].reference != 3) continue; |
3514 h->default_ref_list[list][index ]= sorted_short_ref[j]; | |
3515 h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num; | |
1168 | 3516 } |
3517 | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3518 for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3519 if(h->long_ref[i] == NULL) continue; |
1168 | 3520 if(h->long_ref[i]->reference != 3) continue; |
3521 | |
3522 h->default_ref_list[ list ][index ]= *h->long_ref[i]; | |
3523 h->default_ref_list[ list ][index++].pic_id= i;; | |
3524 } | |
2967 | 3525 |
2447 | 3526 if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){ |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3527 // swap the two first elements of L1 when |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3528 // L0 and L1 are identical |
1168 | 3529 Picture temp= h->default_ref_list[1][0]; |
3530 h->default_ref_list[1][0] = h->default_ref_list[1][1]; | |
2673 | 3531 h->default_ref_list[1][1] = temp; |
1168 | 3532 } |
3533 | |
3534 if(index < h->ref_count[ list ]) | |
3535 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index)); | |
3536 } | |
3537 }else{ | |
3538 int index=0; | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
3539 for(i=0; i<h->short_ref_count; i++){ |
1168 | 3540 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit |
3541 h->default_ref_list[0][index ]= *h->short_ref[i]; | |
3542 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num; | |
3543 } | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
3544 for(i = 0; i < 16; i++){ |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3545 if(h->long_ref[i] == NULL) continue; |
1168 | 3546 if(h->long_ref[i]->reference != 3) continue; |
3547 h->default_ref_list[0][index ]= *h->long_ref[i]; | |
3548 h->default_ref_list[0][index++].pic_id= i;; | |
3549 } | |
3550 if(index < h->ref_count[0]) | |
3551 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index)); | |
3552 } | |
3553 }else{ //FIELD | |
3554 if(h->slice_type==B_TYPE){ | |
3555 }else{ | |
3556 //FIXME second field balh | |
3557 } | |
3558 } | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3559 #ifdef TRACE |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3560 for (i=0; i<h->ref_count[0]; i++) { |
4600 | 3561 tprintf(h->s.avctx, "List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]); |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3562 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3563 if(h->slice_type==B_TYPE){ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3564 for (i=0; i<h->ref_count[1]; i++) { |
4600 | 3565 tprintf(h->s.avctx, "List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[0][i].data[0]); |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3566 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3567 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3568 #endif |
1168 | 3569 return 0; |
3570 } | |
3571 | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3572 static void print_short_term(H264Context *h); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3573 static void print_long_term(H264Context *h); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3574 |
1168 | 3575 static int decode_ref_pic_list_reordering(H264Context *h){ |
3576 MpegEncContext * const s = &h->s; | |
2680
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
3577 int list, index; |
2967 | 3578 |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3579 print_short_term(h); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3580 print_long_term(h); |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
3581 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func |
2967 | 3582 |
4533 | 3583 for(list=0; list<h->list_count; list++){ |
1168 | 3584 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]); |
3585 | |
3586 if(get_bits1(&s->gb)){ | |
3587 int pred= h->curr_pic_num; | |
3588 | |
3589 for(index=0; ; index++){ | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
3590 unsigned int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
3591 unsigned int pic_id; |
1168 | 3592 int i; |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
3593 Picture *ref = NULL; |
2967 | 3594 |
3595 if(reordering_of_pic_nums_idc==3) | |
2284
6d26e105f68f
h.264 ref list reordering bugfix patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2272
diff
changeset
|
3596 break; |
2967 | 3597 |
1168 | 3598 if(index >= h->ref_count[list]){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3599 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n"); |
1168 | 3600 return -1; |
3601 } | |
2967 | 3602 |
1168 | 3603 if(reordering_of_pic_nums_idc<3){ |
3604 if(reordering_of_pic_nums_idc<2){ | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
3605 const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1; |
1168 | 3606 |
3607 if(abs_diff_pic_num >= h->max_pic_num){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3608 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n"); |
1168 | 3609 return -1; |
3610 } | |
3611 | |
3612 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num; | |
3613 else pred+= abs_diff_pic_num; | |
3614 pred &= h->max_pic_num - 1; | |
2967 | 3615 |
2683 | 3616 for(i= h->short_ref_count-1; i>=0; i--){ |
3617 ref = h->short_ref[i]; | |
3618 assert(ref->reference == 3); | |
3619 assert(!ref->long_ref); | |
3620 if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer | |
1168 | 3621 break; |
3622 } | |
2683 | 3623 if(i>=0) |
3624 ref->pic_id= ref->frame_num; | |
1168 | 3625 }else{ |
3626 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
3627 if(pic_id>31){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
3628 av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n"); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
3629 return -1; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
3630 } |
2683 | 3631 ref = h->long_ref[pic_id]; |
4362
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
3632 if(ref){ |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
3633 ref->pic_id= pic_id; |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
3634 assert(ref->reference == 3); |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
3635 assert(ref->long_ref); |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
3636 i=0; |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
3637 }else{ |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
3638 i=-1; |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
3639 } |
1168 | 3640 } |
3641 | |
2485
2844b8cf4e11
H.264 multiplce instance in reference list patch by (Loic <lll+ffmpeg m4x org )
michael
parents:
2484
diff
changeset
|
3642 if (i < 0) { |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3643 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n"); |
1168 | 3644 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME |
2683 | 3645 } else { |
3646 for(i=index; i+1<h->ref_count[list]; i++){ | |
3647 if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id) | |
3648 break; | |
2650
d321fbf0b2ce
reverse 1 hunk from 1.96->1.97, fixes decoding of (MR1_BT_A.h264 and MR1_MW_A.264)
michael
parents:
2649
diff
changeset
|
3649 } |
d321fbf0b2ce
reverse 1 hunk from 1.96->1.97, fixes decoding of (MR1_BT_A.h264 and MR1_MW_A.264)
michael
parents:
2649
diff
changeset
|
3650 for(; i > index; i--){ |
d321fbf0b2ce
reverse 1 hunk from 1.96->1.97, fixes decoding of (MR1_BT_A.h264 and MR1_MW_A.264)
michael
parents:
2649
diff
changeset
|
3651 h->ref_list[list][i]= h->ref_list[list][i-1]; |
d321fbf0b2ce
reverse 1 hunk from 1.96->1.97, fixes decoding of (MR1_BT_A.h264 and MR1_MW_A.264)
michael
parents:
2649
diff
changeset
|
3652 } |
2683 | 3653 h->ref_list[list][index]= *ref; |
1168 | 3654 } |
2284
6d26e105f68f
h.264 ref list reordering bugfix patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2272
diff
changeset
|
3655 }else{ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3656 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n"); |
1168 | 3657 return -1; |
3658 } | |
3659 } | |
3660 } | |
4533 | 3661 } |
3662 for(list=0; list<h->list_count; list++){ | |
2680
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
3663 for(index= 0; index < h->ref_count[list]; index++){ |
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
3664 if(!h->ref_list[list][index].data[0]) |
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
3665 h->ref_list[list][index]= s->current_picture; |
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
3666 } |
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
3667 } |
2967 | 3668 |
2396 | 3669 if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred) |
3670 direct_dist_scale_factor(h); | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
3671 direct_ref_list_init(h); |
2967 | 3672 return 0; |
1168 | 3673 } |
3674 | |
3379
69901769c811
fill_mbaff_ref_list is missing a return statement, its return value
mru
parents:
3341
diff
changeset
|
3675 static void fill_mbaff_ref_list(H264Context *h){ |
3316 | 3676 int list, i, j; |
4533 | 3677 for(list=0; list<2; list++){ //FIXME try list_count |
3316 | 3678 for(i=0; i<h->ref_count[list]; i++){ |
3679 Picture *frame = &h->ref_list[list][i]; | |
3680 Picture *field = &h->ref_list[list][16+2*i]; | |
3681 field[0] = *frame; | |
3682 for(j=0; j<3; j++) | |
3683 field[0].linesize[j] <<= 1; | |
3684 field[1] = field[0]; | |
3685 for(j=0; j<3; j++) | |
3686 field[1].data[j] += frame->linesize[j]; | |
3687 | |
3688 h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i]; | |
3689 h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i]; | |
3690 for(j=0; j<2; j++){ | |
3691 h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j]; | |
3692 h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j]; | |
3693 } | |
3694 } | |
3695 } | |
3696 for(j=0; j<h->ref_count[1]; j++){ | |
3697 for(i=0; i<h->ref_count[0]; i++) | |
3698 h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i]; | |
3699 memcpy(h->implicit_weight[16+2*j], h->implicit_weight[j], sizeof(*h->implicit_weight)); | |
3700 memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight)); | |
3701 } | |
3702 } | |
3703 | |
1168 | 3704 static int pred_weight_table(H264Context *h){ |
3705 MpegEncContext * const s = &h->s; | |
3706 int list, i; | |
2415 | 3707 int luma_def, chroma_def; |
2967 | 3708 |
2415 | 3709 h->use_weight= 0; |
3710 h->use_weight_chroma= 0; | |
1168 | 3711 h->luma_log2_weight_denom= get_ue_golomb(&s->gb); |
3712 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb); | |
2415 | 3713 luma_def = 1<<h->luma_log2_weight_denom; |
3714 chroma_def = 1<<h->chroma_log2_weight_denom; | |
1168 | 3715 |
3716 for(list=0; list<2; list++){ | |
3717 for(i=0; i<h->ref_count[list]; i++){ | |
3718 int luma_weight_flag, chroma_weight_flag; | |
2967 | 3719 |
1168 | 3720 luma_weight_flag= get_bits1(&s->gb); |
3721 if(luma_weight_flag){ | |
3722 h->luma_weight[list][i]= get_se_golomb(&s->gb); | |
3723 h->luma_offset[list][i]= get_se_golomb(&s->gb); | |
2415 | 3724 if( h->luma_weight[list][i] != luma_def |
3725 || h->luma_offset[list][i] != 0) | |
3726 h->use_weight= 1; | |
3727 }else{ | |
3728 h->luma_weight[list][i]= luma_def; | |
3729 h->luma_offset[list][i]= 0; | |
1168 | 3730 } |
3731 | |
3732 chroma_weight_flag= get_bits1(&s->gb); | |
3733 if(chroma_weight_flag){ | |
3734 int j; | |
3735 for(j=0; j<2; j++){ | |
3736 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb); | |
3737 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb); | |
2415 | 3738 if( h->chroma_weight[list][i][j] != chroma_def |
3739 || h->chroma_offset[list][i][j] != 0) | |
3740 h->use_weight_chroma= 1; | |
3741 } | |
3742 }else{ | |
3743 int j; | |
3744 for(j=0; j<2; j++){ | |
3745 h->chroma_weight[list][i][j]= chroma_def; | |
3746 h->chroma_offset[list][i][j]= 0; | |
1168 | 3747 } |
3748 } | |
3749 } | |
3750 if(h->slice_type != B_TYPE) break; | |
3751 } | |
2415 | 3752 h->use_weight= h->use_weight || h->use_weight_chroma; |
1168 | 3753 return 0; |
3754 } | |
3755 | |
2415 | 3756 static void implicit_weight_table(H264Context *h){ |
3757 MpegEncContext * const s = &h->s; | |
3758 int ref0, ref1; | |
3759 int cur_poc = s->current_picture_ptr->poc; | |
3760 | |
3761 if( h->ref_count[0] == 1 && h->ref_count[1] == 1 | |
3762 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){ | |
3763 h->use_weight= 0; | |
3764 h->use_weight_chroma= 0; | |
3765 return; | |
3766 } | |
3767 | |
3768 h->use_weight= 2; | |
3769 h->use_weight_chroma= 2; | |
3770 h->luma_log2_weight_denom= 5; | |
3771 h->chroma_log2_weight_denom= 5; | |
3772 | |
3773 for(ref0=0; ref0 < h->ref_count[0]; ref0++){ | |
3774 int poc0 = h->ref_list[0][ref0].poc; | |
3775 for(ref1=0; ref1 < h->ref_count[1]; ref1++){ | |
2519 | 3776 int poc1 = h->ref_list[1][ref1].poc; |
4594 | 3777 int td = av_clip(poc1 - poc0, -128, 127); |
2415 | 3778 if(td){ |
4594 | 3779 int tb = av_clip(cur_poc - poc0, -128, 127); |
4001 | 3780 int tx = (16384 + (FFABS(td) >> 1)) / td; |
4594 | 3781 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2; |
2415 | 3782 if(dist_scale_factor < -64 || dist_scale_factor > 128) |
3783 h->implicit_weight[ref0][ref1] = 32; | |
3784 else | |
3785 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor; | |
3786 }else | |
3787 h->implicit_weight[ref0][ref1] = 32; | |
3788 } | |
3789 } | |
3790 } | |
3791 | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3792 static inline void unreference_pic(H264Context *h, Picture *pic){ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3793 int i; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3794 pic->reference=0; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3795 if(pic == h->delayed_output_pic) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3796 pic->reference=1; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3797 else{ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3798 for(i = 0; h->delayed_pic[i]; i++) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3799 if(pic == h->delayed_pic[i]){ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3800 pic->reference=1; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3801 break; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3802 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3803 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3804 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3805 |
1168 | 3806 /** |
2392 | 3807 * instantaneous decoder refresh. |
1168 | 3808 */ |
3809 static void idr(H264Context *h){ | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3810 int i; |
1168 | 3811 |
2484
a3188eb4266c
correct long term picture management patch by (Loic <lll+ffmpeg m4x org>)
michael
parents:
2471
diff
changeset
|
3812 for(i=0; i<16; i++){ |
a3188eb4266c
correct long term picture management patch by (Loic <lll+ffmpeg m4x org>)
michael
parents:
2471
diff
changeset
|
3813 if (h->long_ref[i] != NULL) { |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3814 unreference_pic(h, h->long_ref[i]); |
2484
a3188eb4266c
correct long term picture management patch by (Loic <lll+ffmpeg m4x org>)
michael
parents:
2471
diff
changeset
|
3815 h->long_ref[i]= NULL; |
a3188eb4266c
correct long term picture management patch by (Loic <lll+ffmpeg m4x org>)
michael
parents:
2471
diff
changeset
|
3816 } |
1168 | 3817 } |
3818 h->long_ref_count=0; | |
3819 | |
3820 for(i=0; i<h->short_ref_count; i++){ | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3821 unreference_pic(h, h->short_ref[i]); |
1168 | 3822 h->short_ref[i]= NULL; |
3823 } | |
3824 h->short_ref_count=0; | |
3825 } | |
3826 | |
2640 | 3827 /* forget old pics after a seek */ |
3828 static void flush_dpb(AVCodecContext *avctx){ | |
3829 H264Context *h= avctx->priv_data; | |
3830 int i; | |
3144 | 3831 for(i=0; i<16; i++) { |
3832 if(h->delayed_pic[i]) | |
3833 h->delayed_pic[i]->reference= 0; | |
2640 | 3834 h->delayed_pic[i]= NULL; |
3144 | 3835 } |
3836 if(h->delayed_output_pic) | |
3837 h->delayed_output_pic->reference= 0; | |
2640 | 3838 h->delayed_output_pic= NULL; |
3839 idr(h); | |
2751
3247049dfc7a
fix segfault if flush is called before the first frame
mru
parents:
2748
diff
changeset
|
3840 if(h->s.current_picture_ptr) |
3247049dfc7a
fix segfault if flush is called before the first frame
mru
parents:
2748
diff
changeset
|
3841 h->s.current_picture_ptr->reference= 0; |
2640 | 3842 } |
3843 | |
1168 | 3844 /** |
3845 * | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
3846 * @return the removed picture or NULL if an error occurs |
1168 | 3847 */ |
3848 static Picture * remove_short(H264Context *h, int frame_num){ | |
1169 | 3849 MpegEncContext * const s = &h->s; |
1168 | 3850 int i; |
2967 | 3851 |
1169 | 3852 if(s->avctx->debug&FF_DEBUG_MMCO) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3853 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count); |
2967 | 3854 |
1168 | 3855 for(i=0; i<h->short_ref_count; i++){ |
3856 Picture *pic= h->short_ref[i]; | |
1169 | 3857 if(s->avctx->debug&FF_DEBUG_MMCO) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3858 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic); |
1168 | 3859 if(pic->frame_num == frame_num){ |
3860 h->short_ref[i]= NULL; | |
3861 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*)); | |
3862 h->short_ref_count--; | |
3863 return pic; | |
3864 } | |
3865 } | |
3866 return NULL; | |
3867 } | |
3868 | |
3869 /** | |
3870 * | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
3871 * @return the removed picture or NULL if an error occurs |
1168 | 3872 */ |
3873 static Picture * remove_long(H264Context *h, int i){ | |
3874 Picture *pic; | |
3875 | |
3876 pic= h->long_ref[i]; | |
3877 h->long_ref[i]= NULL; | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3878 if(pic) h->long_ref_count--; |
1168 | 3879 |
3880 return pic; | |
3881 } | |
3882 | |
3883 /** | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3884 * print short term list |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3885 */ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3886 static void print_short_term(H264Context *h) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3887 uint32_t i; |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3888 if(h->s.avctx->debug&FF_DEBUG_MMCO) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3889 av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n"); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3890 for(i=0; i<h->short_ref_count; i++){ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3891 Picture *pic= h->short_ref[i]; |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3892 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3893 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3894 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3895 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3896 |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3897 /** |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3898 * print long term list |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3899 */ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3900 static void print_long_term(H264Context *h) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3901 uint32_t i; |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3902 if(h->s.avctx->debug&FF_DEBUG_MMCO) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3903 av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n"); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3904 for(i = 0; i < 16; i++){ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3905 Picture *pic= h->long_ref[i]; |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3906 if (pic) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3907 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3908 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3909 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3910 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3911 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3912 |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3913 /** |
1168 | 3914 * Executes the reference picture marking (memory management control operations). |
3915 */ | |
3916 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){ | |
3917 MpegEncContext * const s = &h->s; | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3918 int i, j; |
1168 | 3919 int current_is_long=0; |
3920 Picture *pic; | |
2967 | 3921 |
1168 | 3922 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3923 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n"); |
2967 | 3924 |
1168 | 3925 for(i=0; i<mmco_count; i++){ |
3926 if(s->avctx->debug&FF_DEBUG_MMCO) | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3927 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index); |
1168 | 3928 |
3929 switch(mmco[i].opcode){ | |
3930 case MMCO_SHORT2UNUSED: | |
3931 pic= remove_short(h, mmco[i].short_frame_num); | |
2766
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
3932 if(pic) |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
3933 unreference_pic(h, pic); |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
3934 else if(s->avctx->debug&FF_DEBUG_MMCO) |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
3935 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_short() failure\n"); |
1168 | 3936 break; |
3937 case MMCO_SHORT2LONG: | |
3938 pic= remove_long(h, mmco[i].long_index); | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3939 if(pic) unreference_pic(h, pic); |
2967 | 3940 |
1168 | 3941 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num); |
4362
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
3942 if (h->long_ref[ mmco[i].long_index ]){ |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
3943 h->long_ref[ mmco[i].long_index ]->long_ref=1; |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
3944 h->long_ref_count++; |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
3945 } |
1168 | 3946 break; |
3947 case MMCO_LONG2UNUSED: | |
3948 pic= remove_long(h, mmco[i].long_index); | |
2766
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
3949 if(pic) |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
3950 unreference_pic(h, pic); |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
3951 else if(s->avctx->debug&FF_DEBUG_MMCO) |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
3952 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_long() failure\n"); |
1168 | 3953 break; |
3954 case MMCO_LONG: | |
3955 pic= remove_long(h, mmco[i].long_index); | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3956 if(pic) unreference_pic(h, pic); |
2967 | 3957 |
1168 | 3958 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr; |
3959 h->long_ref[ mmco[i].long_index ]->long_ref=1; | |
3960 h->long_ref_count++; | |
2967 | 3961 |
1168 | 3962 current_is_long=1; |
3963 break; | |
3964 case MMCO_SET_MAX_LONG: | |
3965 assert(mmco[i].long_index <= 16); | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3966 // just remove the long term which index is greater than new max |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3967 for(j = mmco[i].long_index; j<16; j++){ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3968 pic = remove_long(h, j); |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3969 if (pic) unreference_pic(h, pic); |
1168 | 3970 } |
3971 break; | |
3972 case MMCO_RESET: | |
3973 while(h->short_ref_count){ | |
3974 pic= remove_short(h, h->short_ref[0]->frame_num); | |
4362
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
3975 if(pic) unreference_pic(h, pic); |
1168 | 3976 } |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3977 for(j = 0; j < 16; j++) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3978 pic= remove_long(h, j); |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3979 if(pic) unreference_pic(h, pic); |
1168 | 3980 } |
3981 break; | |
3982 default: assert(0); | |
3983 } | |
3984 } | |
2967 | 3985 |
1168 | 3986 if(!current_is_long){ |
3987 pic= remove_short(h, s->current_picture_ptr->frame_num); | |
3988 if(pic){ | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
3989 unreference_pic(h, pic); |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3990 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n"); |
1168 | 3991 } |
2967 | 3992 |
1168 | 3993 if(h->short_ref_count) |
1169 | 3994 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*)); |
3995 | |
3996 h->short_ref[0]= s->current_picture_ptr; | |
1168 | 3997 h->short_ref[0]->long_ref=0; |
3998 h->short_ref_count++; | |
3999 } | |
2967 | 4000 |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4001 print_short_term(h); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4002 print_long_term(h); |
2967 | 4003 return 0; |
1168 | 4004 } |
4005 | |
4006 static int decode_ref_pic_marking(H264Context *h){ | |
4007 MpegEncContext * const s = &h->s; | |
4008 int i; | |
2967 | 4009 |
1168 | 4010 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields |
4011 s->broken_link= get_bits1(&s->gb) -1; | |
4012 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx | |
4013 if(h->mmco[0].long_index == -1) | |
4014 h->mmco_index= 0; | |
4015 else{ | |
4016 h->mmco[0].opcode= MMCO_LONG; | |
4017 h->mmco_index= 1; | |
2967 | 4018 } |
1168 | 4019 }else{ |
4020 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag | |
2967 | 4021 for(i= 0; i<MAX_MMCO_COUNT; i++) { |
1168 | 4022 MMCOOpcode opcode= get_ue_golomb(&s->gb);; |
4023 | |
4024 h->mmco[i].opcode= opcode; | |
4025 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){ | |
4026 h->mmco[i].short_frame_num= (h->frame_num - get_ue_golomb(&s->gb) - 1) & ((1<<h->sps.log2_max_frame_num)-1); //FIXME fields | |
4027 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){ | |
2846
40765c51a7a9
Compilation fixes part 1 patch by (Arvind R. and Burkhard Plaum, plaum, ipf uni-stuttgart de)
michael
parents:
2844
diff
changeset
|
4028 av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco); |
1168 | 4029 return -1; |
4030 }*/ | |
4031 } | |
4032 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){ | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4033 unsigned int long_index= get_ue_golomb(&s->gb); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4034 if(/*h->mmco[i].long_index >= h->long_ref_count || h->long_ref[ h->mmco[i].long_index ] == NULL*/ long_index >= 16){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4035 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode); |
1168 | 4036 return -1; |
4037 } | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4038 h->mmco[i].long_index= long_index; |
1168 | 4039 } |
2967 | 4040 |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4041 if(opcode > (unsigned)MMCO_LONG){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4042 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode); |
1168 | 4043 return -1; |
4044 } | |
2255
507690ff49a2
assertion when playing AVC/H.264 streams fix by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2254
diff
changeset
|
4045 if(opcode == MMCO_END) |
507690ff49a2
assertion when playing AVC/H.264 streams fix by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2254
diff
changeset
|
4046 break; |
1168 | 4047 } |
4048 h->mmco_index= i; | |
4049 }else{ | |
4050 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count); | |
4051 | |
4052 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields | |
4053 h->mmco[0].opcode= MMCO_SHORT2UNUSED; | |
4054 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num; | |
4055 h->mmco_index= 1; | |
4056 }else | |
4057 h->mmco_index= 0; | |
4058 } | |
4059 } | |
2967 | 4060 |
4061 return 0; | |
1168 | 4062 } |
4063 | |
4064 static int init_poc(H264Context *h){ | |
4065 MpegEncContext * const s = &h->s; | |
4066 const int max_frame_num= 1<<h->sps.log2_max_frame_num; | |
4067 int field_poc[2]; | |
4068 | |
4069 if(h->nal_unit_type == NAL_IDR_SLICE){ | |
4070 h->frame_num_offset= 0; | |
4071 }else{ | |
4072 if(h->frame_num < h->prev_frame_num) | |
4073 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num; | |
4074 else | |
4075 h->frame_num_offset= h->prev_frame_num_offset; | |
4076 } | |
4077 | |
4078 if(h->sps.poc_type==0){ | |
4079 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb; | |
4080 | |
2649
b2e6c5e9240b
fix decoding of conformace streams AUD_MW_E.264 and BA1_FT_C.264 again (there where last correctly decoded with h264.c rev 1.96)
michael
parents:
2648
diff
changeset
|
4081 if(h->nal_unit_type == NAL_IDR_SLICE){ |
b2e6c5e9240b
fix decoding of conformace streams AUD_MW_E.264 and BA1_FT_C.264 again (there where last correctly decoded with h264.c rev 1.96)
michael
parents:
2648
diff
changeset
|
4082 h->prev_poc_msb= |
b2e6c5e9240b
fix decoding of conformace streams AUD_MW_E.264 and BA1_FT_C.264 again (there where last correctly decoded with h264.c rev 1.96)
michael
parents:
2648
diff
changeset
|
4083 h->prev_poc_lsb= 0; |
b2e6c5e9240b
fix decoding of conformace streams AUD_MW_E.264 and BA1_FT_C.264 again (there where last correctly decoded with h264.c rev 1.96)
michael
parents:
2648
diff
changeset
|
4084 } |
b2e6c5e9240b
fix decoding of conformace streams AUD_MW_E.264 and BA1_FT_C.264 again (there where last correctly decoded with h264.c rev 1.96)
michael
parents:
2648
diff
changeset
|
4085 |
1168 | 4086 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2) |
4087 h->poc_msb = h->prev_poc_msb + max_poc_lsb; | |
4088 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2) | |
4089 h->poc_msb = h->prev_poc_msb - max_poc_lsb; | |
4090 else | |
4091 h->poc_msb = h->prev_poc_msb; | |
4092 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb); | |
2967 | 4093 field_poc[0] = |
1168 | 4094 field_poc[1] = h->poc_msb + h->poc_lsb; |
2967 | 4095 if(s->picture_structure == PICT_FRAME) |
1168 | 4096 field_poc[1] += h->delta_poc_bottom; |
4097 }else if(h->sps.poc_type==1){ | |
4098 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc; | |
4099 int i; | |
4100 | |
4101 if(h->sps.poc_cycle_length != 0) | |
4102 abs_frame_num = h->frame_num_offset + h->frame_num; | |
4103 else | |
4104 abs_frame_num = 0; | |
4105 | |
4106 if(h->nal_ref_idc==0 && abs_frame_num > 0) | |
4107 abs_frame_num--; | |
2967 | 4108 |
1168 | 4109 expected_delta_per_poc_cycle = 0; |
4110 for(i=0; i < h->sps.poc_cycle_length; i++) | |
4111 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse | |
4112 | |
4113 if(abs_frame_num > 0){ | |
4114 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length; | |
4115 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length; | |
4116 | |
4117 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle; | |
4118 for(i = 0; i <= frame_num_in_poc_cycle; i++) | |
4119 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ]; | |
4120 } else | |
4121 expectedpoc = 0; | |
4122 | |
2967 | 4123 if(h->nal_ref_idc == 0) |
1168 | 4124 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic; |
2967 | 4125 |
1168 | 4126 field_poc[0] = expectedpoc + h->delta_poc[0]; |
4127 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field; | |
4128 | |
4129 if(s->picture_structure == PICT_FRAME) | |
4130 field_poc[1] += h->delta_poc[1]; | |
4131 }else{ | |
4132 int poc; | |
4133 if(h->nal_unit_type == NAL_IDR_SLICE){ | |
4134 poc= 0; | |
4135 }else{ | |
4136 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num); | |
4137 else poc= 2*(h->frame_num_offset + h->frame_num) - 1; | |
4138 } | |
4139 field_poc[0]= poc; | |
4140 field_poc[1]= poc; | |
4141 } | |
2967 | 4142 |
1168 | 4143 if(s->picture_structure != PICT_BOTTOM_FIELD) |
4144 s->current_picture_ptr->field_poc[0]= field_poc[0]; | |
4145 if(s->picture_structure != PICT_TOP_FIELD) | |
4146 s->current_picture_ptr->field_poc[1]= field_poc[1]; | |
4147 if(s->picture_structure == PICT_FRAME) // FIXME field pix? | |
4148 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]); | |
4149 | |
4150 return 0; | |
4151 } | |
4152 | |
4153 /** | |
4154 * decodes a slice header. | |
4155 * this will allso call MPV_common_init() and frame_start() as needed | |
4156 */ | |
4157 static int decode_slice_header(H264Context *h){ | |
4158 MpegEncContext * const s = &h->s; | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4159 unsigned int first_mb_in_slice; |
4362
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
4160 unsigned int pps_id; |
1168 | 4161 int num_ref_idx_active_override_flag; |
4162 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE}; | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4163 unsigned int slice_type, tmp; |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4164 int default_ref_list_done = 0; |
1168 | 4165 |
4166 s->current_picture.reference= h->nal_ref_idc != 0; | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
4167 s->dropable= h->nal_ref_idc == 0; |
1168 | 4168 |
4169 first_mb_in_slice= get_ue_golomb(&s->gb); | |
4170 | |
4616
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
4171 if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){ |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
4172 h->slice_num = 0; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
4173 s->current_picture_ptr= NULL; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
4174 } |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
4175 |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4176 slice_type= get_ue_golomb(&s->gb); |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4177 if(slice_type > 9){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4178 av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y); |
2392 | 4179 return -1; |
1168 | 4180 } |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4181 if(slice_type > 4){ |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4182 slice_type -= 5; |
1168 | 4183 h->slice_type_fixed=1; |
4184 }else | |
4185 h->slice_type_fixed=0; | |
2967 | 4186 |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4187 slice_type= slice_type_map[ slice_type ]; |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4188 if (slice_type == I_TYPE |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4189 || (h->slice_num != 0 && slice_type == h->slice_type) ) { |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4190 default_ref_list_done = 1; |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4191 } |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4192 h->slice_type= slice_type; |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4193 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
4194 s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though |
2967 | 4195 |
1168 | 4196 pps_id= get_ue_golomb(&s->gb); |
4362
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
4197 if(pps_id>=MAX_PPS_COUNT){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4198 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n"); |
1168 | 4199 return -1; |
4200 } | |
4201 h->pps= h->pps_buffer[pps_id]; | |
1174 | 4202 if(h->pps.slice_group_count == 0){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4203 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n"); |
1174 | 4204 return -1; |
4205 } | |
4206 | |
1168 | 4207 h->sps= h->sps_buffer[ h->pps.sps_id ]; |
1174 | 4208 if(h->sps.log2_max_frame_num == 0){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4209 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n"); |
1174 | 4210 return -1; |
4211 } | |
2919 | 4212 |
4363 | 4213 if(h->dequant_coeff_pps != pps_id){ |
4214 h->dequant_coeff_pps = pps_id; | |
2919 | 4215 init_dequant_tables(h); |
4216 } | |
2967 | 4217 |
1168 | 4218 s->mb_width= h->sps.mb_width; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4219 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag); |
2967 | 4220 |
3178 | 4221 h->b_stride= s->mb_width*4; |
4222 h->b8_stride= s->mb_width*2; | |
1168 | 4223 |
1371 | 4224 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right ); |
1168 | 4225 if(h->sps.frame_mbs_only_flag) |
1371 | 4226 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom); |
1168 | 4227 else |
1371 | 4228 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck |
2967 | 4229 |
4230 if (s->context_initialized | |
1548 | 4231 && ( s->width != s->avctx->width || s->height != s->avctx->height)) { |
1168 | 4232 free_tables(h); |
4233 MPV_common_end(s); | |
4234 } | |
4235 if (!s->context_initialized) { | |
4236 if (MPV_common_init(s) < 0) | |
4237 return -1; | |
2967 | 4238 |
2758
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4239 if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4240 memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t)); |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4241 memcpy(h-> field_scan, field_scan, 16*sizeof(uint8_t)); |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4242 }else{ |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4243 int i; |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4244 for(i=0; i<16; i++){ |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4245 #define T(x) (x>>2) | ((x<<2) & 0xF) |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4246 h->zigzag_scan[i] = T(zigzag_scan[i]); |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4247 h-> field_scan[i] = T( field_scan[i]); |
3174 | 4248 #undef T |
4249 } | |
4250 } | |
4251 if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){ | |
3316 | 4252 memcpy(h->zigzag_scan8x8, zigzag_scan8x8, 64*sizeof(uint8_t)); |
3174 | 4253 memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t)); |
3316 | 4254 memcpy(h->field_scan8x8, field_scan8x8, 64*sizeof(uint8_t)); |
4255 memcpy(h->field_scan8x8_cavlc, field_scan8x8_cavlc, 64*sizeof(uint8_t)); | |
3174 | 4256 }else{ |
4257 int i; | |
4258 for(i=0; i<64; i++){ | |
4259 #define T(x) (x>>3) | ((x&7)<<3) | |
3316 | 4260 h->zigzag_scan8x8[i] = T(zigzag_scan8x8[i]); |
3174 | 4261 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]); |
3316 | 4262 h->field_scan8x8[i] = T(field_scan8x8[i]); |
4263 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]); | |
3174 | 4264 #undef T |
2758
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4265 } |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4266 } |
2763 | 4267 if(h->sps.transform_bypass){ //FIXME same ugly |
3316 | 4268 h->zigzag_scan_q0 = zigzag_scan; |
4269 h->zigzag_scan8x8_q0 = zigzag_scan8x8; | |
3174 | 4270 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc; |
3316 | 4271 h->field_scan_q0 = field_scan; |
4272 h->field_scan8x8_q0 = field_scan8x8; | |
4273 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc; | |
2763 | 4274 }else{ |
3316 | 4275 h->zigzag_scan_q0 = h->zigzag_scan; |
4276 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8; | |
3174 | 4277 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc; |
3316 | 4278 h->field_scan_q0 = h->field_scan; |
4279 h->field_scan8x8_q0 = h->field_scan8x8; | |
4280 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc; | |
2763 | 4281 } |
1168 | 4282 |
4283 alloc_tables(h); | |
4284 | |
4285 s->avctx->width = s->width; | |
4286 s->avctx->height = s->height; | |
1548 | 4287 s->avctx->sample_aspect_ratio= h->sps.sar; |
2440 | 4288 if(!s->avctx->sample_aspect_ratio.den) |
4289 s->avctx->sample_aspect_ratio.den = 1; | |
2174
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
4290 |
2648 | 4291 if(h->sps.timing_info_present_flag){ |
3052 | 4292 s->avctx->time_base= (AVRational){h->sps.num_units_in_tick * 2, h->sps.time_scale}; |
3063
f02d0b59279c
Remove all stray tabs and trailing whitespace, this time for good.
diego
parents:
3052
diff
changeset
|
4293 if(h->x264_build > 0 && h->x264_build < 44) |
f02d0b59279c
Remove all stray tabs and trailing whitespace, this time for good.
diego
parents:
3052
diff
changeset
|
4294 s->avctx->time_base.den *= 2; |
f02d0b59279c
Remove all stray tabs and trailing whitespace, this time for good.
diego
parents:
3052
diff
changeset
|
4295 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den, |
f02d0b59279c
Remove all stray tabs and trailing whitespace, this time for good.
diego
parents:
3052
diff
changeset
|
4296 s->avctx->time_base.num, s->avctx->time_base.den, 1<<30); |
2174
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
4297 } |
1168 | 4298 } |
4299 | |
2392 | 4300 if(h->slice_num == 0){ |
2935 | 4301 if(frame_start(h) < 0) |
4302 return -1; | |
1168 | 4303 } |
4304 | |
1169 | 4305 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup |
1168 | 4306 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num); |
4307 | |
3316 | 4308 h->mb_mbaff = 0; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
4309 h->mb_aff_frame = 0; |
1168 | 4310 if(h->sps.frame_mbs_only_flag){ |
4311 s->picture_structure= PICT_FRAME; | |
4312 }else{ | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
4313 if(get_bits1(&s->gb)) { //field_pic_flag |
1168 | 4314 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag |
3316 | 4315 av_log(h->s.avctx, AV_LOG_ERROR, "PAFF interlacing is not implemented\n"); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
4316 } else { |
1168 | 4317 s->picture_structure= PICT_FRAME; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
4318 h->mb_aff_frame = h->sps.mb_aff; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4319 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4320 } |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4321 assert(s->mb_num == s->mb_width * s->mb_height); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4322 if(first_mb_in_slice << h->mb_aff_frame >= s->mb_num || |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4323 first_mb_in_slice >= s->mb_num){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4324 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n"); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4325 return -1; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4326 } |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4327 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width; |
3316 | 4328 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << h->mb_aff_frame; |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4329 assert(s->mb_y < s->mb_height); |
2967 | 4330 |
1168 | 4331 if(s->picture_structure==PICT_FRAME){ |
4332 h->curr_pic_num= h->frame_num; | |
4333 h->max_pic_num= 1<< h->sps.log2_max_frame_num; | |
4334 }else{ | |
4335 h->curr_pic_num= 2*h->frame_num; | |
4336 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1); | |
4337 } | |
2967 | 4338 |
1168 | 4339 if(h->nal_unit_type == NAL_IDR_SLICE){ |
1453 | 4340 get_ue_golomb(&s->gb); /* idr_pic_id */ |
1168 | 4341 } |
2967 | 4342 |
1168 | 4343 if(h->sps.poc_type==0){ |
4344 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb); | |
2967 | 4345 |
1168 | 4346 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){ |
4347 h->delta_poc_bottom= get_se_golomb(&s->gb); | |
4348 } | |
4349 } | |
2967 | 4350 |
1168 | 4351 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){ |
4352 h->delta_poc[0]= get_se_golomb(&s->gb); | |
2967 | 4353 |
1168 | 4354 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME) |
4355 h->delta_poc[1]= get_se_golomb(&s->gb); | |
4356 } | |
2967 | 4357 |
1168 | 4358 init_poc(h); |
2967 | 4359 |
1168 | 4360 if(h->pps.redundant_pic_cnt_present){ |
4361 h->redundant_pic_count= get_ue_golomb(&s->gb); | |
4362 } | |
4363 | |
4364 //set defaults, might be overriden a few line later | |
4365 h->ref_count[0]= h->pps.ref_count[0]; | |
4366 h->ref_count[1]= h->pps.ref_count[1]; | |
4367 | |
4368 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){ | |
4369 if(h->slice_type == B_TYPE){ | |
4370 h->direct_spatial_mv_pred= get_bits1(&s->gb); | |
3316 | 4371 if(h->sps.mb_aff && h->direct_spatial_mv_pred) |
4372 av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF + spatial direct mode is not implemented\n"); | |
1168 | 4373 } |
4374 num_ref_idx_active_override_flag= get_bits1(&s->gb); | |
2967 | 4375 |
1168 | 4376 if(num_ref_idx_active_override_flag){ |
4377 h->ref_count[0]= get_ue_golomb(&s->gb) + 1; | |
4378 if(h->slice_type==B_TYPE) | |
4379 h->ref_count[1]= get_ue_golomb(&s->gb) + 1; | |
4380 | |
4532 | 4381 if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4382 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n"); |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4383 h->ref_count[0]= h->ref_count[1]= 1; |
1168 | 4384 return -1; |
4385 } | |
4386 } | |
4532 | 4387 if(h->slice_type == B_TYPE) |
4388 h->list_count= 2; | |
4389 else | |
4390 h->list_count= 1; | |
4391 }else | |
4392 h->list_count= 0; | |
1168 | 4393 |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4394 if(!default_ref_list_done){ |
1168 | 4395 fill_default_ref_list(h); |
4396 } | |
4397 | |
2766
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
4398 if(decode_ref_pic_list_reordering(h) < 0) |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
4399 return -1; |
1168 | 4400 |
2967 | 4401 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) |
1168 | 4402 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) ) |
4403 pred_weight_table(h); | |
2415 | 4404 else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE) |
4405 implicit_weight_table(h); | |
4406 else | |
4407 h->use_weight = 0; | |
2967 | 4408 |
2580
b4f6f89ec2f6
The cvs version 1.103 of h264.c brokes 13 conformance streams, this
michael
parents:
2561
diff
changeset
|
4409 if(s->current_picture.reference) |
1168 | 4410 decode_ref_pic_marking(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
4411 |
3316 | 4412 if(FRAME_MBAFF) |
4413 fill_mbaff_ref_list(h); | |
4414 | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4415 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac ){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4416 tmp = get_ue_golomb(&s->gb); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4417 if(tmp > 2){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4418 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n"); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4419 return -1; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4420 } |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4421 h->cabac_init_idc= tmp; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4422 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
4423 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
4424 h->last_qscale_diff = 0; |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4425 tmp = h->pps.init_qp + get_se_golomb(&s->gb); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4426 if(tmp>51){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4427 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp); |
1898 | 4428 return -1; |
4429 } | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4430 s->qscale= tmp; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
4431 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale); |
1168 | 4432 //FIXME qscale / qp ... stuff |
4433 if(h->slice_type == SP_TYPE){ | |
1453 | 4434 get_bits1(&s->gb); /* sp_for_switch_flag */ |
1168 | 4435 } |
4436 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){ | |
1453 | 4437 get_se_golomb(&s->gb); /* slice_qs_delta */ |
1168 | 4438 } |
4439 | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4440 h->deblocking_filter = 1; |
1898 | 4441 h->slice_alpha_c0_offset = 0; |
4442 h->slice_beta_offset = 0; | |
1168 | 4443 if( h->pps.deblocking_filter_parameters_present ) { |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4444 tmp= get_ue_golomb(&s->gb); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4445 if(tmp > 2){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4446 av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4447 return -1; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4448 } |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4449 h->deblocking_filter= tmp; |
2967 | 4450 if(h->deblocking_filter < 2) |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4451 h->deblocking_filter^= 1; // 1<->0 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4452 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4453 if( h->deblocking_filter ) { |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4454 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4455 h->slice_beta_offset = get_se_golomb(&s->gb) << 1; |
1168 | 4456 } |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4457 } |
2792 | 4458 if( s->avctx->skip_loop_filter >= AVDISCARD_ALL |
4459 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type != I_TYPE) | |
4460 ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR && h->slice_type == B_TYPE) | |
4461 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0)) | |
4462 h->deblocking_filter= 0; | |
1168 | 4463 |
4464 #if 0 //FMO | |
4465 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5) | |
4466 slice_group_change_cycle= get_bits(&s->gb, ?); | |
4467 #endif | |
4468 | |
2392 | 4469 h->slice_num++; |
4470 | |
3316 | 4471 h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; |
4472 h->emu_edge_height= FRAME_MBAFF ? 0 : h->emu_edge_width; | |
4473 | |
1168 | 4474 if(s->avctx->debug&FF_DEBUG_PICT_INFO){ |
4362
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
4475 av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s\n", |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4476 h->slice_num, |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4477 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"), |
2967 | 4478 first_mb_in_slice, |
1264 | 4479 av_get_pict_type_char(h->slice_type), |
1168 | 4480 pps_id, h->frame_num, |
4481 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1], | |
4482 h->ref_count[0], h->ref_count[1], | |
4483 s->qscale, | |
2583 | 4484 h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2, |
2415 | 4485 h->use_weight, |
4486 h->use_weight==1 && h->use_weight_chroma ? "c" : "" | |
1168 | 4487 ); |
4488 } | |
4489 | |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4490 if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !s->current_picture.reference){ |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4491 s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4492 s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4493 }else{ |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4494 s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4495 s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4496 } |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4497 |
1168 | 4498 return 0; |
4499 } | |
4500 | |
4501 /** | |
4502 * | |
4503 */ | |
4504 static inline int get_level_prefix(GetBitContext *gb){ | |
4505 unsigned int buf; | |
4506 int log; | |
2967 | 4507 |
1168 | 4508 OPEN_READER(re, gb); |
4509 UPDATE_CACHE(re, gb); | |
4510 buf=GET_CACHE(re, gb); | |
2967 | 4511 |
1168 | 4512 log= 32 - av_log2(buf); |
4513 #ifdef TRACE | |
4514 print_bin(buf>>(32-log), log); | |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2255
diff
changeset
|
4515 av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__); |
1168 | 4516 #endif |
4517 | |
4518 LAST_SKIP_BITS(re, gb, log); | |
4519 CLOSE_READER(re, gb); | |
4520 | |
4521 return log-1; | |
4522 } | |
4523 | |
2755 | 4524 static inline int get_dct8x8_allowed(H264Context *h){ |
4525 int i; | |
4526 for(i=0; i<4; i++){ | |
4527 if(!IS_SUB_8X8(h->sub_mb_type[i]) | |
2781
899a2ea0907c
parenthesing to avoid compiler errors in the future - compared asm outputs, behaviour didnt changed
alex
parents:
2774
diff
changeset
|
4528 || (!h->sps.direct_8x8_inference_flag && IS_DIRECT(h->sub_mb_type[i]))) |
2755 | 4529 return 0; |
4530 } | |
4531 return 1; | |
4532 } | |
4533 | |
1168 | 4534 /** |
4535 * decodes a residual block. | |
4536 * @param n block index | |
4537 * @param scantable scantable | |
4538 * @param max_coeff number of coefficients in the block | |
4539 * @return <0 if an error occured | |
4540 */ | |
2919 | 4541 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){ |
1168 | 4542 MpegEncContext * const s = &h->s; |
4543 static const int coeff_token_table_index[17]= {0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3}; | |
2897 | 4544 int level[16]; |
4545 int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before; | |
1168 | 4546 |
4547 //FIXME put trailing_onex into the context | |
4548 | |
4549 if(n == CHROMA_DC_BLOCK_INDEX){ | |
4550 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1); | |
4551 total_coeff= coeff_token>>2; | |
2967 | 4552 }else{ |
1168 | 4553 if(n == LUMA_DC_BLOCK_INDEX){ |
4554 total_coeff= pred_non_zero_count(h, 0); | |
4555 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2); | |
4556 total_coeff= coeff_token>>2; | |
4557 }else{ | |
4558 total_coeff= pred_non_zero_count(h, n); | |
4559 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2); | |
4560 total_coeff= coeff_token>>2; | |
4561 h->non_zero_count_cache[ scan8[n] ]= total_coeff; | |
4562 } | |
4563 } | |
4564 | |
4565 //FIXME set last_non_zero? | |
4566 | |
4567 if(total_coeff==0) | |
4568 return 0; | |
4391
20944c70a3c0
fix segfault with darkkben.free.fr/corrupted_h264.mp4
michael
parents:
4390
diff
changeset
|
4569 if(total_coeff > (unsigned)max_coeff) { |
20944c70a3c0
fix segfault with darkkben.free.fr/corrupted_h264.mp4
michael
parents:
4390
diff
changeset
|
4570 av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff); |
4270 | 4571 return -1; |
4572 } | |
2967 | 4573 |
1168 | 4574 trailing_ones= coeff_token&3; |
4600 | 4575 tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff); |
1168 | 4576 assert(total_coeff<=16); |
2967 | 4577 |
1168 | 4578 for(i=0; i<trailing_ones; i++){ |
4579 level[i]= 1 - 2*get_bits1(gb); | |
4580 } | |
4581 | |
2897 | 4582 if(i<total_coeff) { |
1168 | 4583 int level_code, mask; |
2897 | 4584 int suffix_length = total_coeff > 10 && trailing_ones < 3; |
4585 int prefix= get_level_prefix(gb); | |
4586 | |
4587 //first coefficient has suffix_length equal to 0 or 1 | |
1168 | 4588 if(prefix<14){ //FIXME try to build a large unified VLC table for all this |
4589 if(suffix_length) | |
4590 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part | |
4591 else | |
4592 level_code= (prefix<<suffix_length); //part | |
4593 }else if(prefix==14){ | |
4594 if(suffix_length) | |
4595 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part | |
4596 else | |
4597 level_code= prefix + get_bits(gb, 4); //part | |
4598 }else if(prefix==15){ | |
4599 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
4600 if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense |
1168 | 4601 }else{ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4602 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y); |
1168 | 4603 return -1; |
4604 } | |
4605 | |
2897 | 4606 if(trailing_ones < 3) level_code += 2; |
4607 | |
4608 suffix_length = 1; | |
4609 if(level_code > 5) | |
4610 suffix_length++; | |
1168 | 4611 mask= -(level_code&1); |
4612 level[i]= (((2+level_code)>>1) ^ mask) - mask; | |
2897 | 4613 i++; |
4614 | |
4615 //remaining coefficients have suffix_length > 0 | |
4616 for(;i<total_coeff;i++) { | |
4617 static const int suffix_limit[7] = {0,5,11,23,47,95,INT_MAX }; | |
4618 prefix = get_level_prefix(gb); | |
4619 if(prefix<15){ | |
4620 level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length); | |
4621 }else if(prefix==15){ | |
4622 level_code = (prefix<<suffix_length) + get_bits(gb, 12); | |
4623 }else{ | |
4624 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y); | |
4625 return -1; | |
4626 } | |
4627 mask= -(level_code&1); | |
4628 level[i]= (((2+level_code)>>1) ^ mask) - mask; | |
4629 if(level_code > suffix_limit[suffix_length]) | |
4630 suffix_length++; | |
4631 } | |
1168 | 4632 } |
4633 | |
4634 if(total_coeff == max_coeff) | |
4635 zeros_left=0; | |
4636 else{ | |
4637 if(n == CHROMA_DC_BLOCK_INDEX) | |
4638 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1); | |
4639 else | |
4640 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1); | |
4641 } | |
2897 | 4642 |
4643 coeff_num = zeros_left + total_coeff - 1; | |
4644 j = scantable[coeff_num]; | |
4645 if(n > 24){ | |
4646 block[j] = level[0]; | |
4647 for(i=1;i<total_coeff;i++) { | |
4648 if(zeros_left <= 0) | |
4649 run_before = 0; | |
4650 else if(zeros_left < 7){ | |
4651 run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1); | |
4652 }else{ | |
4653 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); | |
4654 } | |
4655 zeros_left -= run_before; | |
4656 coeff_num -= 1 + run_before; | |
4657 j= scantable[ coeff_num ]; | |
4658 | |
4659 block[j]= level[i]; | |
4660 } | |
4661 }else{ | |
2919 | 4662 block[j] = (level[0] * qmul[j] + 32)>>6; |
2897 | 4663 for(i=1;i<total_coeff;i++) { |
4664 if(zeros_left <= 0) | |
4665 run_before = 0; | |
4666 else if(zeros_left < 7){ | |
4667 run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1); | |
4668 }else{ | |
4669 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); | |
4670 } | |
4671 zeros_left -= run_before; | |
4672 coeff_num -= 1 + run_before; | |
4673 j= scantable[ coeff_num ]; | |
4674 | |
2919 | 4675 block[j]= (level[i] * qmul[j] + 32)>>6; |
2897 | 4676 } |
1168 | 4677 } |
4678 | |
4679 if(zeros_left<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4680 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y); |
1168 | 4681 return -1; |
4682 } | |
2897 | 4683 |
1168 | 4684 return 0; |
4685 } | |
4686 | |
3316 | 4687 static void predict_field_decoding_flag(H264Context *h){ |
4688 MpegEncContext * const s = &h->s; | |
4689 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; | |
4690 int mb_type = (h->slice_table[mb_xy-1] == h->slice_num) | |
4691 ? s->current_picture.mb_type[mb_xy-1] | |
4692 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num) | |
4693 ? s->current_picture.mb_type[mb_xy-s->mb_stride] | |
4694 : 0; | |
4695 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0; | |
4696 } | |
4697 | |
1168 | 4698 /** |
2396 | 4699 * decodes a P_SKIP or B_SKIP macroblock |
4700 */ | |
4701 static void decode_mb_skip(H264Context *h){ | |
4702 MpegEncContext * const s = &h->s; | |
4703 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; | |
2674
58fb8e1efcae
10l (mb_type uninitalized and then changed and overwritten)
michael
parents:
2673
diff
changeset
|
4704 int mb_type=0; |
2967 | 4705 |
2396 | 4706 memset(h->non_zero_count[mb_xy], 0, 16); |
4707 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui | |
4708 | |
3316 | 4709 if(MB_FIELD) |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
4710 mb_type|= MB_TYPE_INTERLACED; |
2674
58fb8e1efcae
10l (mb_type uninitalized and then changed and overwritten)
michael
parents:
2673
diff
changeset
|
4711 |
2396 | 4712 if( h->slice_type == B_TYPE ) |
4713 { | |
4714 // just for fill_caches. pred_direct_motion will set the real mb_type | |
2674
58fb8e1efcae
10l (mb_type uninitalized and then changed and overwritten)
michael
parents:
2673
diff
changeset
|
4715 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP; |
2396 | 4716 |
2449 | 4717 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ... |
2396 | 4718 pred_direct_motion(h, &mb_type); |
3927 | 4719 mb_type|= MB_TYPE_SKIP; |
2396 | 4720 } |
4721 else | |
4722 { | |
4723 int mx, my; | |
2674
58fb8e1efcae
10l (mb_type uninitalized and then changed and overwritten)
michael
parents:
2673
diff
changeset
|
4724 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP; |
2396 | 4725 |
2449 | 4726 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ... |
2396 | 4727 pred_pskip_motion(h, &mx, &my); |
4728 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1); | |
4729 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4); | |
4730 } | |
4731 | |
4732 write_back_motion(h, mb_type); | |
3927 | 4733 s->current_picture.mb_type[mb_xy]= mb_type; |
2396 | 4734 s->current_picture.qscale_table[mb_xy]= s->qscale; |
4735 h->slice_table[ mb_xy ]= h->slice_num; | |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2623
diff
changeset
|
4736 h->prev_mb_skipped= 1; |
2396 | 4737 } |
4738 | |
4739 /** | |
1168 | 4740 * decodes a macroblock |
4741 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed | |
4742 */ | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
4743 static int decode_mb_cavlc(H264Context *h){ |
1168 | 4744 MpegEncContext * const s = &h->s; |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
4745 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4746 int partition_count; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4747 unsigned int mb_type, cbp; |
2755 | 4748 int dct8x8_allowed= h->pps.transform_8x8_mode; |
1168 | 4749 |
2967 | 4750 s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong? |
1168 | 4751 |
4600 | 4752 tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y); |
1453 | 4753 cbp = 0; /* avoid warning. FIXME: find a solution without slowing |
4754 down the code */ | |
1168 | 4755 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){ |
4756 if(s->mb_skip_run==-1) | |
4757 s->mb_skip_run= get_ue_golomb(&s->gb); | |
2967 | 4758 |
1168 | 4759 if (s->mb_skip_run--) { |
3316 | 4760 if(FRAME_MBAFF && (s->mb_y&1) == 0){ |
4761 if(s->mb_skip_run==0) | |
4762 h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb); | |
4763 else | |
4764 predict_field_decoding_flag(h); | |
4765 } | |
2396 | 4766 decode_mb_skip(h); |
1168 | 4767 return 0; |
4768 } | |
4769 } | |
3316 | 4770 if(FRAME_MBAFF){ |
4771 if( (s->mb_y&1) == 0 ) | |
4772 h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb); | |
1168 | 4773 }else |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4774 h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME); |
2967 | 4775 |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2623
diff
changeset
|
4776 h->prev_mb_skipped= 0; |
2967 | 4777 |
1168 | 4778 mb_type= get_ue_golomb(&s->gb); |
4779 if(h->slice_type == B_TYPE){ | |
4780 if(mb_type < 23){ | |
4781 partition_count= b_mb_type_info[mb_type].partition_count; | |
4782 mb_type= b_mb_type_info[mb_type].type; | |
4783 }else{ | |
4784 mb_type -= 23; | |
4785 goto decode_intra_mb; | |
4786 } | |
4787 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){ | |
4788 if(mb_type < 5){ | |
4789 partition_count= p_mb_type_info[mb_type].partition_count; | |
4790 mb_type= p_mb_type_info[mb_type].type; | |
4791 }else{ | |
4792 mb_type -= 5; | |
4793 goto decode_intra_mb; | |
4794 } | |
4795 }else{ | |
4796 assert(h->slice_type == I_TYPE); | |
4797 decode_intra_mb: | |
4798 if(mb_type > 25){ | |
3954 | 4799 av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice too large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y); |
1168 | 4800 return -1; |
4801 } | |
4802 partition_count=0; | |
4803 cbp= i_mb_type_info[mb_type].cbp; | |
4804 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode; | |
4805 mb_type= i_mb_type_info[mb_type].type; | |
4806 } | |
4807 | |
3316 | 4808 if(MB_FIELD) |
1168 | 4809 mb_type |= MB_TYPE_INTERLACED; |
4810 | |
4811 h->slice_table[ mb_xy ]= h->slice_num; | |
2967 | 4812 |
1168 | 4813 if(IS_INTRA_PCM(mb_type)){ |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
4814 unsigned int x, y; |
2967 | 4815 |
1168 | 4816 // we assume these blocks are very rare so we dont optimize it |
4817 align_get_bits(&s->gb); | |
2967 | 4818 |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
4819 // The pixels are stored in the same order as levels in h->mb array. |
1168 | 4820 for(y=0; y<16; y++){ |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
4821 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3); |
1168 | 4822 for(x=0; x<16; x++){ |
4600 | 4823 tprintf(s->avctx, "LUMA ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8)); |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
4824 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= get_bits(&s->gb, 8); |
1168 | 4825 } |
4826 } | |
4827 for(y=0; y<8; y++){ | |
4828 const int index= 256 + 4*(y&3) + 32*(y>>2); | |
4829 for(x=0; x<8; x++){ | |
4600 | 4830 tprintf(s->avctx, "CHROMA U ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8)); |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
4831 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8); |
1168 | 4832 } |
4833 } | |
4834 for(y=0; y<8; y++){ | |
4835 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2); | |
4836 for(x=0; x<8; x++){ | |
4600 | 4837 tprintf(s->avctx, "CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8)); |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
4838 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8); |
1168 | 4839 } |
4840 } | |
2967 | 4841 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
4842 // In deblocking, the quantizer is 0 |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
4843 s->current_picture.qscale_table[mb_xy]= 0; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
4844 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0); |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
4845 // All coeffs are present |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4846 memset(h->non_zero_count[mb_xy], 16, 16); |
2967 | 4847 |
2755 | 4848 s->current_picture.mb_type[mb_xy]= mb_type; |
1168 | 4849 return 0; |
4850 } | |
2967 | 4851 |
3316 | 4852 if(MB_MBAFF){ |
4853 h->ref_count[0] <<= 1; | |
4854 h->ref_count[1] <<= 1; | |
4855 } | |
4856 | |
2449 | 4857 fill_caches(h, mb_type, 0); |
1168 | 4858 |
4859 //mb_pred | |
4860 if(IS_INTRA(mb_type)){ | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4861 int pred_mode; |
1168 | 4862 // init_top_left_availability(h); |
4863 if(IS_INTRA4x4(mb_type)){ | |
4864 int i; | |
2755 | 4865 int di = 1; |
4866 if(dct8x8_allowed && get_bits1(&s->gb)){ | |
4867 mb_type |= MB_TYPE_8x8DCT; | |
4868 di = 4; | |
4869 } | |
1168 | 4870 |
4871 // fill_intra4x4_pred_table(h); | |
2755 | 4872 for(i=0; i<16; i+=di){ |
3409 | 4873 int mode= pred_intra_mode(h, i); |
4874 | |
4875 if(!get_bits1(&s->gb)){ | |
1168 | 4876 const int rem_mode= get_bits(&s->gb, 3); |
3409 | 4877 mode = rem_mode + (rem_mode >= mode); |
1168 | 4878 } |
2967 | 4879 |
2755 | 4880 if(di==4) |
4881 fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 ); | |
4882 else | |
4883 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode; | |
1168 | 4884 } |
4885 write_back_intra_pred_mode(h); | |
4886 if( check_intra4x4_pred_mode(h) < 0) | |
4887 return -1; | |
4888 }else{ | |
4889 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode); | |
4890 if(h->intra16x16_pred_mode < 0) | |
4891 return -1; | |
4892 } | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4893 |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4894 pred_mode= check_intra_pred_mode(h, get_ue_golomb(&s->gb)); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4895 if(pred_mode < 0) |
1168 | 4896 return -1; |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4897 h->chroma_pred_mode= pred_mode; |
1168 | 4898 }else if(partition_count==4){ |
4899 int i, j, sub_partition_count[4], list, ref[2][4]; | |
2967 | 4900 |
1168 | 4901 if(h->slice_type == B_TYPE){ |
4902 for(i=0; i<4; i++){ | |
4903 h->sub_mb_type[i]= get_ue_golomb(&s->gb); | |
4904 if(h->sub_mb_type[i] >=13){ | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4905 av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y); |
1168 | 4906 return -1; |
4907 } | |
4908 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; | |
4909 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type; | |
4910 } | |
2396 | 4911 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1]) |
3003 | 4912 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) { |
2396 | 4913 pred_direct_motion(h, &mb_type); |
3003 | 4914 h->ref_cache[0][scan8[4]] = |
4915 h->ref_cache[1][scan8[4]] = | |
4916 h->ref_cache[0][scan8[12]] = | |
4917 h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE; | |
4918 } | |
1168 | 4919 }else{ |
4920 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ? | |
4921 for(i=0; i<4; i++){ | |
4922 h->sub_mb_type[i]= get_ue_golomb(&s->gb); | |
4923 if(h->sub_mb_type[i] >=4){ | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4924 av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y); |
1168 | 4925 return -1; |
4926 } | |
4927 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; | |
4928 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type; | |
4929 } | |
4930 } | |
2967 | 4931 |
4533 | 4932 for(list=0; list<h->list_count; list++){ |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
4933 int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list]; |
1168 | 4934 for(i=0; i<4; i++){ |
2396 | 4935 if(IS_DIRECT(h->sub_mb_type[i])) continue; |
4936 if(IS_DIR(h->sub_mb_type[i], 0, list)){ | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4937 unsigned int tmp = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip? |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4938 if(tmp>=ref_count){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4939 av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", tmp); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4940 return -1; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4941 } |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
4942 ref[list][i]= tmp; |
1168 | 4943 }else{ |
4944 //FIXME | |
4945 ref[list][i] = -1; | |
4946 } | |
4947 } | |
4948 } | |
2967 | 4949 |
2755 | 4950 if(dct8x8_allowed) |
4951 dct8x8_allowed = get_dct8x8_allowed(h); | |
2967 | 4952 |
4533 | 4953 for(list=0; list<h->list_count; list++){ |
1168 | 4954 for(i=0; i<4; i++){ |
3003 | 4955 if(IS_DIRECT(h->sub_mb_type[i])) { |
4956 h->ref_cache[list][ scan8[4*i] ] = h->ref_cache[list][ scan8[4*i]+1 ]; | |
4957 continue; | |
4958 } | |
1168 | 4959 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]= |
4960 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i]; | |
4961 | |
2396 | 4962 if(IS_DIR(h->sub_mb_type[i], 0, list)){ |
1168 | 4963 const int sub_mb_type= h->sub_mb_type[i]; |
4964 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1; | |
4965 for(j=0; j<sub_partition_count[i]; j++){ | |
4966 int mx, my; | |
4967 const int index= 4*i + block_width*j; | |
4968 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ]; | |
4969 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my); | |
4970 mx += get_se_golomb(&s->gb); | |
4971 my += get_se_golomb(&s->gb); | |
4600 | 4972 tprintf(s->avctx, "final mv:%d %d\n", mx, my); |
1170 | 4973 |
1168 | 4974 if(IS_SUB_8X8(sub_mb_type)){ |
4530 | 4975 mv_cache[ 1 ][0]= |
1168 | 4976 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx; |
4530 | 4977 mv_cache[ 1 ][1]= |
1168 | 4978 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my; |
4979 }else if(IS_SUB_8X4(sub_mb_type)){ | |
4530 | 4980 mv_cache[ 1 ][0]= mx; |
4981 mv_cache[ 1 ][1]= my; | |
1168 | 4982 }else if(IS_SUB_4X8(sub_mb_type)){ |
4530 | 4983 mv_cache[ 8 ][0]= mx; |
4984 mv_cache[ 8 ][1]= my; | |
1168 | 4985 } |
4530 | 4986 mv_cache[ 0 ][0]= mx; |
4987 mv_cache[ 0 ][1]= my; | |
1168 | 4988 } |
4989 }else{ | |
4990 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0]; | |
4991 p[0] = p[1]= | |
4992 p[8] = p[9]= 0; | |
4993 } | |
4994 } | |
4995 } | |
2396 | 4996 }else if(IS_DIRECT(mb_type)){ |
4997 pred_direct_motion(h, &mb_type); | |
2755 | 4998 dct8x8_allowed &= h->sps.direct_8x8_inference_flag; |
2396 | 4999 }else{ |
1168 | 5000 int list, mx, my, i; |
5001 //FIXME we should set ref_idx_l? to 0 if we use that later ... | |
5002 if(IS_16X16(mb_type)){ | |
4534 | 5003 for(list=0; list<h->list_count; list++){ |
5004 unsigned int val; | |
1168 | 5005 if(IS_DIR(mb_type, 0, list)){ |
4534 | 5006 val= get_te0_golomb(&s->gb, h->ref_count[list]); |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5007 if(val >= h->ref_count[list]){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5008 av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5009 return -1; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5010 } |
2523 | 5011 }else |
4534 | 5012 val= LIST_NOT_USED&0xFF; |
5013 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1); | |
5014 } | |
5015 for(list=0; list<h->list_count; list++){ | |
5016 unsigned int val; | |
1168 | 5017 if(IS_DIR(mb_type, 0, list)){ |
5018 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my); | |
5019 mx += get_se_golomb(&s->gb); | |
5020 my += get_se_golomb(&s->gb); | |
4600 | 5021 tprintf(s->avctx, "final mv:%d %d\n", mx, my); |
1170 | 5022 |
4534 | 5023 val= pack16to32(mx,my); |
2523 | 5024 }else |
4534 | 5025 val=0; |
5026 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, val, 4); | |
1168 | 5027 } |
5028 } | |
5029 else if(IS_16X8(mb_type)){ | |
4534 | 5030 for(list=0; list<h->list_count; list++){ |
1168 | 5031 for(i=0; i<2; i++){ |
4534 | 5032 unsigned int val; |
1168 | 5033 if(IS_DIR(mb_type, i, list)){ |
4534 | 5034 val= get_te0_golomb(&s->gb, h->ref_count[list]); |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5035 if(val >= h->ref_count[list]){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5036 av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5037 return -1; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5038 } |
2523 | 5039 }else |
4534 | 5040 val= LIST_NOT_USED&0xFF; |
5041 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1); | |
1168 | 5042 } |
4534 | 5043 } |
5044 for(list=0; list<h->list_count; list++){ | |
1168 | 5045 for(i=0; i<2; i++){ |
4534 | 5046 unsigned int val; |
1168 | 5047 if(IS_DIR(mb_type, i, list)){ |
5048 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my); | |
5049 mx += get_se_golomb(&s->gb); | |
5050 my += get_se_golomb(&s->gb); | |
4600 | 5051 tprintf(s->avctx, "final mv:%d %d\n", mx, my); |
1170 | 5052 |
4534 | 5053 val= pack16to32(mx,my); |
2396 | 5054 }else |
4534 | 5055 val=0; |
5056 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 4); | |
1168 | 5057 } |
5058 } | |
5059 }else{ | |
5060 assert(IS_8X16(mb_type)); | |
4534 | 5061 for(list=0; list<h->list_count; list++){ |
1168 | 5062 for(i=0; i<2; i++){ |
4534 | 5063 unsigned int val; |
1168 | 5064 if(IS_DIR(mb_type, i, list)){ //FIXME optimize |
4534 | 5065 val= get_te0_golomb(&s->gb, h->ref_count[list]); |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5066 if(val >= h->ref_count[list]){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5067 av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5068 return -1; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5069 } |
2523 | 5070 }else |
4534 | 5071 val= LIST_NOT_USED&0xFF; |
5072 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1); | |
1168 | 5073 } |
4534 | 5074 } |
5075 for(list=0; list<h->list_count; list++){ | |
1168 | 5076 for(i=0; i<2; i++){ |
4534 | 5077 unsigned int val; |
1168 | 5078 if(IS_DIR(mb_type, i, list)){ |
5079 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my); | |
5080 mx += get_se_golomb(&s->gb); | |
5081 my += get_se_golomb(&s->gb); | |
4600 | 5082 tprintf(s->avctx, "final mv:%d %d\n", mx, my); |
1170 | 5083 |
4534 | 5084 val= pack16to32(mx,my); |
2396 | 5085 }else |
4534 | 5086 val=0; |
5087 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 4); | |
1168 | 5088 } |
5089 } | |
5090 } | |
5091 } | |
2967 | 5092 |
1168 | 5093 if(IS_INTER(mb_type)) |
5094 write_back_motion(h, mb_type); | |
2967 | 5095 |
1168 | 5096 if(!IS_INTRA16x16(mb_type)){ |
5097 cbp= get_ue_golomb(&s->gb); | |
5098 if(cbp > 47){ | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5099 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, s->mb_x, s->mb_y); |
1168 | 5100 return -1; |
5101 } | |
2967 | 5102 |
1168 | 5103 if(IS_INTRA4x4(mb_type)) |
5104 cbp= golomb_to_intra4x4_cbp[cbp]; | |
5105 else | |
5106 cbp= golomb_to_inter_cbp[cbp]; | |
5107 } | |
3651 | 5108 h->cbp = cbp; |
1168 | 5109 |
2755 | 5110 if(dct8x8_allowed && (cbp&15) && !IS_INTRA(mb_type)){ |
5111 if(get_bits1(&s->gb)) | |
5112 mb_type |= MB_TYPE_8x8DCT; | |
5113 } | |
5114 s->current_picture.mb_type[mb_xy]= mb_type; | |
5115 | |
1168 | 5116 if(cbp || IS_INTRA16x16(mb_type)){ |
5117 int i8x8, i4x4, chroma_idx; | |
5118 int chroma_qp, dquant; | |
5119 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr; | |
3174 | 5120 const uint8_t *scan, *scan8x8, *dc_scan; |
2967 | 5121 |
1168 | 5122 // fill_non_zero_count_cache(h); |
5123 | |
5124 if(IS_INTERLACED(mb_type)){ | |
3316 | 5125 scan8x8= s->qscale ? h->field_scan8x8_cavlc : h->field_scan8x8_cavlc_q0; |
2763 | 5126 scan= s->qscale ? h->field_scan : h->field_scan_q0; |
1168 | 5127 dc_scan= luma_dc_field_scan; |
5128 }else{ | |
3316 | 5129 scan8x8= s->qscale ? h->zigzag_scan8x8_cavlc : h->zigzag_scan8x8_cavlc_q0; |
2763 | 5130 scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0; |
1168 | 5131 dc_scan= luma_dc_zigzag_scan; |
5132 } | |
5133 | |
5134 dquant= get_se_golomb(&s->gb); | |
5135 | |
5136 if( dquant > 25 || dquant < -26 ){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
5137 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y); |
1168 | 5138 return -1; |
5139 } | |
2967 | 5140 |
1168 | 5141 s->qscale += dquant; |
5142 if(((unsigned)s->qscale) > 51){ | |
5143 if(s->qscale<0) s->qscale+= 52; | |
5144 else s->qscale-= 52; | |
5145 } | |
2967 | 5146 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
5147 h->chroma_qp= chroma_qp= get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale); |
1168 | 5148 if(IS_INTRA16x16(mb_type)){ |
2919 | 5149 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, h->dequant4_coeff[0][s->qscale], 16) < 0){ |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
5150 return -1; //FIXME continue if partitioned and other return -1 too |
1168 | 5151 } |
5152 | |
5153 assert((cbp&15) == 0 || (cbp&15) == 15); | |
5154 | |
5155 if(cbp&15){ | |
5156 for(i8x8=0; i8x8<4; i8x8++){ | |
5157 for(i4x4=0; i4x4<4; i4x4++){ | |
5158 const int index= i4x4 + 4*i8x8; | |
2919 | 5159 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 ){ |
1168 | 5160 return -1; |
5161 } | |
5162 } | |
5163 } | |
5164 }else{ | |
1636 | 5165 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1); |
1168 | 5166 } |
5167 }else{ | |
5168 for(i8x8=0; i8x8<4; i8x8++){ | |
5169 if(cbp & (1<<i8x8)){ | |
2755 | 5170 if(IS_8x8DCT(mb_type)){ |
5171 DCTELEM *buf = &h->mb[64*i8x8]; | |
2757 | 5172 uint8_t *nnz; |
2755 | 5173 for(i4x4=0; i4x4<4; i4x4++){ |
3174 | 5174 if( decode_residual(h, gb, buf, i4x4+4*i8x8, scan8x8+16*i4x4, |
2919 | 5175 h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 16) <0 ) |
2755 | 5176 return -1; |
5177 } | |
2757 | 5178 nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ]; |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
5179 nnz[0] += nnz[1] + nnz[8] + nnz[9]; |
2755 | 5180 }else{ |
5181 for(i4x4=0; i4x4<4; i4x4++){ | |
5182 const int index= i4x4 + 4*i8x8; | |
2967 | 5183 |
2919 | 5184 if( decode_residual(h, gb, h->mb + 16*index, index, scan, h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale], 16) <0 ){ |
2755 | 5185 return -1; |
5186 } | |
1168 | 5187 } |
5188 } | |
5189 }else{ | |
5190 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ]; | |
5191 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0; | |
5192 } | |
5193 } | |
5194 } | |
2967 | 5195 |
1168 | 5196 if(cbp&0x30){ |
5197 for(chroma_idx=0; chroma_idx<2; chroma_idx++) | |
2919 | 5198 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, NULL, 4) < 0){ |
1168 | 5199 return -1; |
5200 } | |
5201 } | |
5202 | |
5203 if(cbp&0x20){ | |
5204 for(chroma_idx=0; chroma_idx<2; chroma_idx++){ | |
5205 for(i4x4=0; i4x4<4; i4x4++){ | |
5206 const int index= 16 + 4*chroma_idx + i4x4; | |
2919 | 5207 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, h->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][chroma_qp], 15) < 0){ |
1168 | 5208 return -1; |
5209 } | |
5210 } | |
5211 } | |
5212 }else{ | |
5213 uint8_t * const nnz= &h->non_zero_count_cache[0]; | |
5214 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] = | |
5215 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0; | |
5216 } | |
5217 }else{ | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
5218 uint8_t * const nnz= &h->non_zero_count_cache[0]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
5219 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
5220 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] = |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
5221 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0; |
1168 | 5222 } |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
5223 s->current_picture.qscale_table[mb_xy]= s->qscale; |
1168 | 5224 write_back_non_zero_count(h); |
5225 | |
3316 | 5226 if(MB_MBAFF){ |
5227 h->ref_count[0] >>= 1; | |
5228 h->ref_count[1] >>= 1; | |
5229 } | |
5230 | |
1168 | 5231 return 0; |
5232 } | |
5233 | |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5234 static int decode_cabac_field_decoding_flag(H264Context *h) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5235 MpegEncContext * const s = &h->s; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5236 const int mb_x = s->mb_x; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5237 const int mb_y = s->mb_y & ~1; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5238 const int mba_xy = mb_x - 1 + mb_y *s->mb_stride; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5239 const int mbb_xy = mb_x + (mb_y-2)*s->mb_stride; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5240 |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5241 unsigned int ctx = 0; |
2967 | 5242 |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5243 if( h->slice_table[mba_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) ) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5244 ctx += 1; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5245 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5246 if( h->slice_table[mbb_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) ) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5247 ctx += 1; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5248 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5249 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5250 return get_cabac_noinline( &h->cabac, &h->cabac_state[70 + ctx] ); |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5251 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5252 |
2312 | 5253 static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) { |
5254 uint8_t *state= &h->cabac_state[ctx_base]; | |
5255 int mb_type; | |
2967 | 5256 |
2312 | 5257 if(intra_slice){ |
5258 MpegEncContext * const s = &h->s; | |
2597
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5259 const int mba_xy = h->left_mb_xy[0]; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5260 const int mbb_xy = h->top_mb_xy; |
2312 | 5261 int ctx=0; |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5262 if( h->slice_table[mba_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mba_xy] ) ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5263 ctx++; |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5264 if( h->slice_table[mbb_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mbb_xy] ) ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5265 ctx++; |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5266 if( get_cabac_noinline( &h->cabac, &state[ctx] ) == 0 ) |
2312 | 5267 return 0; /* I4x4 */ |
5268 state += 2; | |
5269 }else{ | |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5270 if( get_cabac_noinline( &h->cabac, &state[0] ) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5271 return 0; /* I4x4 */ |
2312 | 5272 } |
5273 | |
5274 if( get_cabac_terminate( &h->cabac ) ) | |
5275 return 25; /* PCM */ | |
5276 | |
5277 mb_type = 1; /* I16x16 */ | |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5278 mb_type += 12 * get_cabac_noinline( &h->cabac, &state[1] ); /* cbp_luma != 0 */ |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5279 if( get_cabac_noinline( &h->cabac, &state[2] ) ) /* cbp_chroma */ |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5280 mb_type += 4 + 4 * get_cabac_noinline( &h->cabac, &state[2+intra_slice] ); |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5281 mb_type += 2 * get_cabac_noinline( &h->cabac, &state[3+intra_slice] ); |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5282 mb_type += 1 * get_cabac_noinline( &h->cabac, &state[3+2*intra_slice] ); |
2312 | 5283 return mb_type; |
5284 } | |
5285 | |
5286 static int decode_cabac_mb_type( H264Context *h ) { | |
5287 MpegEncContext * const s = &h->s; | |
5288 | |
5289 if( h->slice_type == I_TYPE ) { | |
5290 return decode_cabac_intra_mb_type(h, 3, 1); | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5291 } else if( h->slice_type == P_TYPE ) { |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5292 if( get_cabac_noinline( &h->cabac, &h->cabac_state[14] ) == 0 ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5293 /* P-type */ |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5294 if( get_cabac_noinline( &h->cabac, &h->cabac_state[15] ) == 0 ) { |
3138 | 5295 /* P_L0_D16x16, P_8x8 */ |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5296 return 3 * get_cabac_noinline( &h->cabac, &h->cabac_state[16] ); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5297 } else { |
3138 | 5298 /* P_L0_D8x16, P_L0_D16x8 */ |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5299 return 2 - get_cabac_noinline( &h->cabac, &h->cabac_state[17] ); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5300 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5301 } else { |
2312 | 5302 return decode_cabac_intra_mb_type(h, 17, 0) + 5; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5303 } |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5304 } else if( h->slice_type == B_TYPE ) { |
2597
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5305 const int mba_xy = h->left_mb_xy[0]; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5306 const int mbb_xy = h->top_mb_xy; |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5307 int ctx = 0; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5308 int bits; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5309 |
3138 | 5310 if( h->slice_table[mba_xy] == h->slice_num && !IS_DIRECT( s->current_picture.mb_type[mba_xy] ) ) |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5311 ctx++; |
3138 | 5312 if( h->slice_table[mbb_xy] == h->slice_num && !IS_DIRECT( s->current_picture.mb_type[mbb_xy] ) ) |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5313 ctx++; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5314 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5315 if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+ctx] ) ) |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5316 return 0; /* B_Direct_16x16 */ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5317 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5318 if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+3] ) ) { |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5319 return 1 + get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */ |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5320 } |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5321 |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5322 bits = get_cabac_noinline( &h->cabac, &h->cabac_state[27+4] ) << 3; |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5323 bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ) << 2; |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5324 bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ) << 1; |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5325 bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ); |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5326 if( bits < 8 ) |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5327 return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5328 else if( bits == 13 ) { |
2312 | 5329 return decode_cabac_intra_mb_type(h, 32, 0) + 23; |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5330 } else if( bits == 14 ) |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5331 return 11; /* B_L1_L0_8x16 */ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5332 else if( bits == 15 ) |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5333 return 22; /* B_8x8 */ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5334 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5335 bits= ( bits<<1 ) | get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ); |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5336 return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */ |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5337 } else { |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5338 /* TODO SI/SP frames? */ |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5339 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5340 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5341 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5342 |
3316 | 5343 static int decode_cabac_mb_skip( H264Context *h, int mb_x, int mb_y ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5344 MpegEncContext * const s = &h->s; |
3316 | 5345 int mba_xy, mbb_xy; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5346 int ctx = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5347 |
3316 | 5348 if(FRAME_MBAFF){ //FIXME merge with the stuff in fill_caches? |
5349 int mb_xy = mb_x + (mb_y&~1)*s->mb_stride; | |
5350 mba_xy = mb_xy - 1; | |
5351 if( (mb_y&1) | |
5352 && h->slice_table[mba_xy] == h->slice_num | |
5353 && MB_FIELD == !!IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) ) | |
5354 mba_xy += s->mb_stride; | |
5355 if( MB_FIELD ){ | |
5356 mbb_xy = mb_xy - s->mb_stride; | |
5357 if( !(mb_y&1) | |
5358 && h->slice_table[mbb_xy] == h->slice_num | |
5359 && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) ) | |
5360 mbb_xy -= s->mb_stride; | |
5361 }else | |
5362 mbb_xy = mb_x + (mb_y-1)*s->mb_stride; | |
5363 }else{ | |
5364 int mb_xy = mb_x + mb_y*s->mb_stride; | |
5365 mba_xy = mb_xy - 1; | |
5366 mbb_xy = mb_xy - s->mb_stride; | |
5367 } | |
5368 | |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5369 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] )) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5370 ctx++; |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5371 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] )) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5372 ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5373 |
3138 | 5374 if( h->slice_type == B_TYPE ) |
5375 ctx += 13; | |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5376 return get_cabac_noinline( &h->cabac, &h->cabac_state[11+ctx] ); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5377 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5378 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5379 static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5380 int mode = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5381 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5382 if( get_cabac( &h->cabac, &h->cabac_state[68] ) ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5383 return pred_mode; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5384 |
2928 | 5385 mode += 1 * get_cabac( &h->cabac, &h->cabac_state[69] ); |
5386 mode += 2 * get_cabac( &h->cabac, &h->cabac_state[69] ); | |
5387 mode += 4 * get_cabac( &h->cabac, &h->cabac_state[69] ); | |
5388 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5389 if( mode >= pred_mode ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5390 return mode + 1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5391 else |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5392 return mode; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5393 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5394 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5395 static int decode_cabac_mb_chroma_pre_mode( H264Context *h) { |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5396 const int mba_xy = h->left_mb_xy[0]; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5397 const int mbb_xy = h->top_mb_xy; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5398 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5399 int ctx = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5400 |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
5401 /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */ |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5402 if( h->slice_table[mba_xy] == h->slice_num && h->chroma_pred_mode_table[mba_xy] != 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5403 ctx++; |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
5404 |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5405 if( h->slice_table[mbb_xy] == h->slice_num && h->chroma_pred_mode_table[mbb_xy] != 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5406 ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5407 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5408 if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+ctx] ) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5409 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5410 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5411 if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+3] ) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5412 return 1; |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5413 if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+3] ) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5414 return 2; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5415 else |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5416 return 3; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5417 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5418 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5419 static const uint8_t block_idx_x[16] = { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5420 0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5421 }; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5422 static const uint8_t block_idx_y[16] = { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5423 0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5424 }; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5425 static const uint8_t block_idx_xy[4][4] = { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5426 { 0, 2, 8, 10}, |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5427 { 1, 3, 9, 11}, |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5428 { 4, 6, 12, 14}, |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5429 { 5, 7, 13, 15} |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5430 }; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5431 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5432 static int decode_cabac_mb_cbp_luma( H264Context *h) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5433 int cbp = 0; |
3138 | 5434 int cbp_b = -1; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5435 int i8x8; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5436 |
3138 | 5437 if( h->slice_table[h->top_mb_xy] == h->slice_num ) { |
5438 cbp_b = h->top_cbp; | |
4600 | 5439 tprintf(h->s.avctx, "cbp_b = top_cbp = %x\n", cbp_b); |
3138 | 5440 } |
5441 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5442 for( i8x8 = 0; i8x8 < 4; i8x8++ ) { |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5443 int cbp_a = -1; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5444 int x, y; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5445 int ctx = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5446 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5447 x = block_idx_x[4*i8x8]; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5448 y = block_idx_y[4*i8x8]; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5449 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5450 if( x > 0 ) |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5451 cbp_a = cbp; |
3138 | 5452 else if( h->slice_table[h->left_mb_xy[0]] == h->slice_num ) { |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5453 cbp_a = h->left_cbp; |
4600 | 5454 tprintf(h->s.avctx, "cbp_a = left_cbp = %x\n", cbp_a); |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5455 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5456 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5457 if( y > 0 ) |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5458 cbp_b = cbp; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5459 |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
5460 /* No need to test for skip as we put 0 for skip block */ |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
5461 /* No need to test for IPCM as we put 1 for IPCM block */ |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5462 if( cbp_a >= 0 ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5463 int i8x8a = block_idx_xy[(x-1)&0x03][y]/4; |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5464 if( ((cbp_a >> i8x8a)&0x01) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5465 ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5466 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5467 |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5468 if( cbp_b >= 0 ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5469 int i8x8b = block_idx_xy[x][(y-1)&0x03]/4; |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5470 if( ((cbp_b >> i8x8b)&0x01) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5471 ctx += 2; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5472 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5473 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5474 if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5475 cbp |= 1 << i8x8; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5476 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5477 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5478 return cbp; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5479 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5480 static int decode_cabac_mb_cbp_chroma( H264Context *h) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5481 int ctx; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5482 int cbp_a, cbp_b; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5483 |
2314 | 5484 cbp_a = (h->left_cbp>>4)&0x03; |
5485 cbp_b = (h-> top_cbp>>4)&0x03; | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5486 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5487 ctx = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5488 if( cbp_a > 0 ) ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5489 if( cbp_b > 0 ) ctx += 2; |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5490 if( get_cabac_noinline( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5491 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5492 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5493 ctx = 4; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5494 if( cbp_a == 2 ) ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5495 if( cbp_b == 2 ) ctx += 2; |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5496 return 1 + get_cabac_noinline( &h->cabac, &h->cabac_state[77 + ctx] ); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5497 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5498 static int decode_cabac_mb_dqp( H264Context *h) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5499 MpegEncContext * const s = &h->s; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5500 int mbn_xy; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5501 int ctx = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5502 int val = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5503 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5504 if( s->mb_x > 0 ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5505 mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5506 else |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5507 mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5508 |
3316 | 5509 if( h->last_qscale_diff != 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5510 ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5511 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5512 while( get_cabac_noinline( &h->cabac, &h->cabac_state[60 + ctx] ) ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5513 if( ctx < 2 ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5514 ctx = 2; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5515 else |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5516 ctx = 3; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5517 val++; |
3128 | 5518 if(val > 102) //prevent infinite loop |
2844
5f20ab245501
prevent assert failures and infinite loops with broken streams
michael
parents:
2834
diff
changeset
|
5519 return INT_MIN; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5520 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5521 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5522 if( val&0x01 ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5523 return (val + 1)/2; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5524 else |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5525 return -(val + 1)/2; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5526 } |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5527 static int decode_cabac_p_mb_sub_type( H264Context *h ) { |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5528 if( get_cabac( &h->cabac, &h->cabac_state[21] ) ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5529 return 0; /* 8x8 */ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5530 if( !get_cabac( &h->cabac, &h->cabac_state[22] ) ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5531 return 1; /* 8x4 */ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5532 if( get_cabac( &h->cabac, &h->cabac_state[23] ) ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5533 return 2; /* 4x8 */ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5534 return 3; /* 4x4 */ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5535 } |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5536 static int decode_cabac_b_mb_sub_type( H264Context *h ) { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5537 int type; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5538 if( !get_cabac( &h->cabac, &h->cabac_state[36] ) ) |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5539 return 0; /* B_Direct_8x8 */ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5540 if( !get_cabac( &h->cabac, &h->cabac_state[37] ) ) |
2311
cdbb2f30e08b
small typo patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
2310
diff
changeset
|
5541 return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */ |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5542 type = 3; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5543 if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5544 if( get_cabac( &h->cabac, &h->cabac_state[39] ) ) |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5545 return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5546 type += 4; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5547 } |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5548 type += 2*get_cabac( &h->cabac, &h->cabac_state[39] ); |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5549 type += get_cabac( &h->cabac, &h->cabac_state[39] ); |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5550 return type; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5551 } |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5552 |
2755 | 5553 static inline int decode_cabac_mb_transform_size( H264Context *h ) { |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5554 return get_cabac_noinline( &h->cabac, &h->cabac_state[399 + h->neighbor_transform_size] ); |
2755 | 5555 } |
5556 | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5557 static int decode_cabac_mb_ref( H264Context *h, int list, int n ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5558 int refa = h->ref_cache[list][scan8[n] - 1]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5559 int refb = h->ref_cache[list][scan8[n] - 8]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5560 int ref = 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5561 int ctx = 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5562 |
2396 | 5563 if( h->slice_type == B_TYPE) { |
5564 if( refa > 0 && !h->direct_cache[scan8[n] - 1] ) | |
5565 ctx++; | |
5566 if( refb > 0 && !h->direct_cache[scan8[n] - 8] ) | |
5567 ctx += 2; | |
5568 } else { | |
5569 if( refa > 0 ) | |
5570 ctx++; | |
5571 if( refb > 0 ) | |
5572 ctx += 2; | |
5573 } | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5574 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5575 while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5576 ref++; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5577 if( ctx < 4 ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5578 ctx = 4; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5579 else |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5580 ctx = 5; |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5581 if(ref >= 32 /*h->ref_list[list]*/){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5582 av_log(h->s.avctx, AV_LOG_ERROR, "overflow in decode_cabac_mb_ref\n"); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5583 return 0; //FIXME we should return -1 and check the return everywhere |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5584 } |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5585 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5586 return ref; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5587 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5588 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5589 static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5590 int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) + |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5591 abs( h->mvd_cache[list][scan8[n] - 8][l] ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5592 int ctxbase = (l == 0) ? 40 : 47; |
2317 | 5593 int ctx, mvd; |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5594 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5595 if( amvd < 3 ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5596 ctx = 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5597 else if( amvd > 32 ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5598 ctx = 2; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5599 else |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5600 ctx = 1; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5601 |
2317 | 5602 if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx])) |
5603 return 0; | |
5604 | |
5605 mvd= 1; | |
5606 ctx= 3; | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5607 while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5608 mvd++; |
2317 | 5609 if( ctx < 6 ) |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5610 ctx++; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5611 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5612 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5613 if( mvd >= 9 ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5614 int k = 3; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5615 while( get_cabac_bypass( &h->cabac ) ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5616 mvd += 1 << k; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5617 k++; |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5618 if(k>24){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5619 av_log(h->s.avctx, AV_LOG_ERROR, "overflow in decode_cabac_mb_mvd\n"); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5620 return INT_MIN; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
5621 } |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5622 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5623 while( k-- ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5624 if( get_cabac_bypass( &h->cabac ) ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5625 mvd += 1 << k; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5626 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5627 } |
4042 | 5628 return get_cabac_bypass_sign( &h->cabac, -mvd ); |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5629 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5630 |
4908
777f250df232
Fix multiple "¡Æinline/static¡Ç is not at beginning of declaration" warnings.
diego
parents:
4896
diff
changeset
|
5631 static inline int get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) { |
2314 | 5632 int nza, nzb; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5633 int ctx = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5634 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5635 if( cat == 0 ) { |
2314 | 5636 nza = h->left_cbp&0x100; |
5637 nzb = h-> top_cbp&0x100; | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5638 } else if( cat == 1 || cat == 2 ) { |
2314 | 5639 nza = h->non_zero_count_cache[scan8[idx] - 1]; |
5640 nzb = h->non_zero_count_cache[scan8[idx] - 8]; | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5641 } else if( cat == 3 ) { |
2314 | 5642 nza = (h->left_cbp>>(6+idx))&0x01; |
5643 nzb = (h-> top_cbp>>(6+idx))&0x01; | |
5644 } else { | |
5645 assert(cat == 4); | |
5646 nza = h->non_zero_count_cache[scan8[16+idx] - 1]; | |
5647 nzb = h->non_zero_count_cache[scan8[16+idx] - 8]; | |
5648 } | |
5649 | |
5650 if( nza > 0 ) | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5651 ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5652 |
2314 | 5653 if( nzb > 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5654 ctx += 2; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5655 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5656 return ctx + 4 * cat; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5657 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5658 |
4051 | 5659 static const __attribute((used)) uint8_t last_coeff_flag_offset_8x8[63] = { |
5660 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
5661 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
5662 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, | |
5663 5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8 | |
5664 }; | |
5665 | |
3028 | 5666 static int decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5667 const int mb_xy = h->s.mb_x + h->s.mb_y*h->s.mb_stride; |
3316 | 5668 static const int significant_coeff_flag_offset[2][6] = { |
5669 { 105+0, 105+15, 105+29, 105+44, 105+47, 402 }, | |
5670 { 277+0, 277+15, 277+29, 277+44, 277+47, 436 } | |
5671 }; | |
5672 static const int last_coeff_flag_offset[2][6] = { | |
5673 { 166+0, 166+15, 166+29, 166+44, 166+47, 417 }, | |
5674 { 338+0, 338+15, 338+29, 338+44, 338+47, 451 } | |
5675 }; | |
5676 static const int coeff_abs_level_m1_offset[6] = { | |
5677 227+0, 227+10, 227+20, 227+30, 227+39, 426 | |
5678 }; | |
4007 | 5679 static const uint8_t significant_coeff_flag_offset_8x8[2][63] = { |
3316 | 5680 { 0, 1, 2, 3, 4, 5, 5, 4, 4, 3, 3, 4, 4, 4, 5, 5, |
2755 | 5681 4, 4, 4, 4, 3, 3, 6, 7, 7, 7, 8, 9,10, 9, 8, 7, |
5682 7, 6,11,12,13,11, 6, 7, 8, 9,14,10, 9, 8, 6,11, | |
3316 | 5683 12,13,11, 6, 9,14,10, 9,11,12,13,11,14,10,12 }, |
5684 { 0, 1, 1, 2, 2, 3, 3, 4, 5, 6, 7, 7, 7, 8, 4, 5, | |
5685 6, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,11,12,11, | |
5686 9, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,13,13, 9, | |
5687 9,10,10, 8,13,13, 9, 9,10,10,14,14,14,14,14 } | |
2755 | 5688 }; |
5689 | |
5690 int index[64]; | |
2313 | 5691 |
4041
0113ceb0f953
1 variable less in decode_cabac_residual() no speed change (but less vars means less things gcc can load and store redundantly)
michael
parents:
4040
diff
changeset
|
5692 int last; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5693 int coeff_count = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5694 |
2316 | 5695 int abslevel1 = 1; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5696 int abslevelgt1 = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5697 |
2898 | 5698 uint8_t *significant_coeff_ctx_base; |
5699 uint8_t *last_coeff_ctx_base; | |
5700 uint8_t *abs_level_m1_ctx_base; | |
2755 | 5701 |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5702 #ifndef ARCH_X86 |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5703 #define CABAC_ON_STACK |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5704 #endif |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5705 #ifdef CABAC_ON_STACK |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5706 #define CC &cc |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5707 CABACContext cc; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5708 cc.range = h->cabac.range; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5709 cc.low = h->cabac.low; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5710 cc.bytestream= h->cabac.bytestream; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5711 #else |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5712 #define CC &h->cabac |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5713 #endif |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5714 |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5715 |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5716 /* cat: 0-> DC 16x16 n = 0 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5717 * 1-> AC 16x16 n = luma4x4idx |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5718 * 2-> Luma4x4 n = luma4x4idx |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5719 * 3-> DC Chroma n = iCbCr |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5720 * 4-> AC Chroma n = 4 * iCbCr + chroma4x4idx |
2755 | 5721 * 5-> Luma8x8 n = 4 * luma8x8idx |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5722 */ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5723 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5724 /* read coded block flag */ |
2898 | 5725 if( cat != 5 ) { |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5726 if( get_cabac( CC, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) { |
2755 | 5727 if( cat == 1 || cat == 2 ) |
5728 h->non_zero_count_cache[scan8[n]] = 0; | |
5729 else if( cat == 4 ) | |
5730 h->non_zero_count_cache[scan8[16+n]] = 0; | |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5731 #ifdef CABAC_ON_STACK |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5732 h->cabac.range = cc.range ; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5733 h->cabac.low = cc.low ; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5734 h->cabac.bytestream= cc.bytestream; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5735 #endif |
2755 | 5736 return 0; |
5737 } | |
2898 | 5738 } |
5739 | |
5740 significant_coeff_ctx_base = h->cabac_state | |
3316 | 5741 + significant_coeff_flag_offset[MB_FIELD][cat]; |
2898 | 5742 last_coeff_ctx_base = h->cabac_state |
3316 | 5743 + last_coeff_flag_offset[MB_FIELD][cat]; |
2898 | 5744 abs_level_m1_ctx_base = h->cabac_state |
5745 + coeff_abs_level_m1_offset[cat]; | |
5746 | |
5747 if( cat == 5 ) { | |
5748 #define DECODE_SIGNIFICANCE( coefs, sig_off, last_off ) \ | |
5749 for(last= 0; last < coefs; last++) { \ | |
5750 uint8_t *sig_ctx = significant_coeff_ctx_base + sig_off; \ | |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5751 if( get_cabac( CC, sig_ctx )) { \ |
2898 | 5752 uint8_t *last_ctx = last_coeff_ctx_base + last_off; \ |
5753 index[coeff_count++] = last; \ | |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5754 if( get_cabac( CC, last_ctx ) ) { \ |
2898 | 5755 last= max_coeff; \ |
5756 break; \ | |
5757 } \ | |
5758 } \ | |
4037
53be304c7f54
x86 asm version of the decode significance loop (not 8x8) of decode_residual() 5% faster decode_residual() on P3
michael
parents:
4016
diff
changeset
|
5759 }\ |
53be304c7f54
x86 asm version of the decode significance loop (not 8x8) of decode_residual() 5% faster decode_residual() on P3
michael
parents:
4016
diff
changeset
|
5760 if( last == max_coeff -1 ) {\ |
53be304c7f54
x86 asm version of the decode significance loop (not 8x8) of decode_residual() 5% faster decode_residual() on P3
michael
parents:
4016
diff
changeset
|
5761 index[coeff_count++] = last;\ |
2898 | 5762 } |
4009 | 5763 const uint8_t *sig_off = significant_coeff_flag_offset_8x8[MB_FIELD]; |
4882
8131ccb4ea72
Mark code parts that cannot work on AMD64 due to broken relocations as such.
diego
parents:
4881
diff
changeset
|
5764 #if defined(ARCH_X86) && defined(CONFIG_7REGS) && defined(HAVE_EBX_AVAILABLE) && !defined(BROKEN_RELOCATIONS) |
4051 | 5765 coeff_count= decode_significance_8x8_x86(CC, significant_coeff_ctx_base, index, sig_off); |
5766 } else { | |
5767 coeff_count= decode_significance_x86(CC, max_coeff, significant_coeff_ctx_base, index); | |
5768 #else | |
3316 | 5769 DECODE_SIGNIFICANCE( 63, sig_off[last], last_coeff_flag_offset_8x8[last] ); |
2898 | 5770 } else { |
5771 DECODE_SIGNIFICANCE( max_coeff - 1, last, last ); | |
4037
53be304c7f54
x86 asm version of the decode significance loop (not 8x8) of decode_residual() 5% faster decode_residual() on P3
michael
parents:
4016
diff
changeset
|
5772 #endif |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5773 } |
2316 | 5774 assert(coeff_count > 0); |
5775 | |
5776 if( cat == 0 ) | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5777 h->cbp_table[mb_xy] |= 0x100; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5778 else if( cat == 1 || cat == 2 ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5779 h->non_zero_count_cache[scan8[n]] = coeff_count; |
2316 | 5780 else if( cat == 3 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5781 h->cbp_table[mb_xy] |= 0x40 << n; |
2755 | 5782 else if( cat == 4 ) |
5783 h->non_zero_count_cache[scan8[16+n]] = coeff_count; | |
2316 | 5784 else { |
2755 | 5785 assert( cat == 5 ); |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
5786 fill_rectangle(&h->non_zero_count_cache[scan8[n]], 2, 2, 8, coeff_count, 1); |
2316 | 5787 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5788 |
4041
0113ceb0f953
1 variable less in decode_cabac_residual() no speed change (but less vars means less things gcc can load and store redundantly)
michael
parents:
4040
diff
changeset
|
5789 for( coeff_count--; coeff_count >= 0; coeff_count-- ) { |
2898 | 5790 uint8_t *ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 )) + abs_level_m1_ctx_base; |
4041
0113ceb0f953
1 variable less in decode_cabac_residual() no speed change (but less vars means less things gcc can load and store redundantly)
michael
parents:
4040
diff
changeset
|
5791 int j= scantable[index[coeff_count]]; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5792 |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5793 if( get_cabac( CC, ctx ) == 0 ) { |
2898 | 5794 if( !qmul ) { |
4040 | 5795 block[j] = get_cabac_bypass_sign( CC, -1); |
2316 | 5796 }else{ |
4040 | 5797 block[j] = (get_cabac_bypass_sign( CC, -qmul[j]) + 32) >> 6;; |
2316 | 5798 } |
2967 | 5799 |
2313 | 5800 abslevel1++; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5801 } else { |
2313 | 5802 int coeff_abs = 2; |
2898 | 5803 ctx = 5 + FFMIN( 4, abslevelgt1 ) + abs_level_m1_ctx_base; |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5804 while( coeff_abs < 15 && get_cabac( CC, ctx ) ) { |
2313 | 5805 coeff_abs++; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5806 } |
2313 | 5807 |
5808 if( coeff_abs >= 15 ) { | |
5809 int j = 0; | |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5810 while( get_cabac_bypass( CC ) ) { |
2313 | 5811 j++; |
5812 } | |
2967 | 5813 |
4016 | 5814 coeff_abs=1; |
2313 | 5815 while( j-- ) { |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5816 coeff_abs += coeff_abs + get_cabac_bypass( CC ); |
2313 | 5817 } |
4016 | 5818 coeff_abs+= 14; |
2313 | 5819 } |
5820 | |
2898 | 5821 if( !qmul ) { |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5822 if( get_cabac_bypass( CC ) ) block[j] = -coeff_abs; |
2316 | 5823 else block[j] = coeff_abs; |
5824 }else{ | |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5825 if( get_cabac_bypass( CC ) ) block[j] = (-coeff_abs * qmul[j] + 32) >> 6; |
2919 | 5826 else block[j] = ( coeff_abs * qmul[j] + 32) >> 6; |
2316 | 5827 } |
2967 | 5828 |
2313 | 5829 abslevelgt1++; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5830 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5831 } |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5832 #ifdef CABAC_ON_STACK |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5833 h->cabac.range = cc.range ; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5834 h->cabac.low = cc.low ; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5835 h->cabac.bytestream= cc.bytestream; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
5836 #endif |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5837 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5838 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5839 |
4908
777f250df232
Fix multiple "¡Æinline/static¡Ç is not at beginning of declaration" warnings.
diego
parents:
4896
diff
changeset
|
5840 static inline void compute_mb_neighbors(H264Context *h) |
2597
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5841 { |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5842 MpegEncContext * const s = &h->s; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5843 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5844 h->top_mb_xy = mb_xy - s->mb_stride; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5845 h->left_mb_xy[0] = mb_xy - 1; |
3316 | 5846 if(FRAME_MBAFF){ |
2597
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5847 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5848 const int top_pair_xy = pair_xy - s->mb_stride; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5849 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]); |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5850 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]); |
3316 | 5851 const int curr_mb_frame_flag = !MB_FIELD; |
2597
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5852 const int bottom = (s->mb_y & 1); |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5853 if (bottom |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5854 ? !curr_mb_frame_flag // bottom macroblock |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5855 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5856 ) { |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5857 h->top_mb_xy -= s->mb_stride; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5858 } |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5859 if (left_mb_frame_flag != curr_mb_frame_flag) { |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5860 h->left_mb_xy[0] = pair_xy - 1; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5861 } |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5862 } |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5863 return; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5864 } |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5865 |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5866 /** |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5867 * decodes a macroblock |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5868 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5869 */ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5870 static int decode_mb_cabac(H264Context *h) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5871 MpegEncContext * const s = &h->s; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5872 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5873 int mb_type, partition_count, cbp = 0; |
2755 | 5874 int dct8x8_allowed= h->pps.transform_8x8_mode; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5875 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
5876 s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5877 |
4600 | 5878 tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5879 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) { |
3316 | 5880 int skip; |
5881 /* a skipped mb needs the aff flag from the following mb */ | |
5882 if( FRAME_MBAFF && s->mb_x==0 && (s->mb_y&1)==0 ) | |
5883 predict_field_decoding_flag(h); | |
5884 if( FRAME_MBAFF && (s->mb_y&1)==1 && h->prev_mb_skipped ) | |
5885 skip = h->next_mb_skipped; | |
5886 else | |
5887 skip = decode_cabac_mb_skip( h, s->mb_x, s->mb_y ); | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5888 /* read skip flags */ |
3316 | 5889 if( skip ) { |
5890 if( FRAME_MBAFF && (s->mb_y&1)==0 ){ | |
5891 s->current_picture.mb_type[mb_xy] = MB_TYPE_SKIP; | |
5892 h->next_mb_skipped = decode_cabac_mb_skip( h, s->mb_x, s->mb_y+1 ); | |
5893 if(h->next_mb_skipped) | |
5894 predict_field_decoding_flag(h); | |
5895 else | |
5896 h->mb_mbaff = h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h); | |
5897 } | |
5898 | |
2396 | 5899 decode_mb_skip(h); |
5900 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5901 h->cbp_table[mb_xy] = 0; |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
5902 h->chroma_pred_mode_table[mb_xy] = 0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5903 h->last_qscale_diff = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5904 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5905 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5906 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5907 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5908 } |
3316 | 5909 if(FRAME_MBAFF){ |
5910 if( (s->mb_y&1) == 0 ) | |
5911 h->mb_mbaff = | |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5912 h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h); |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5913 }else |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5914 h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME); |
5357b214eda0
CABAC support for MBAFF I frames patch by (Lo«Ác Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5915 |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2623
diff
changeset
|
5916 h->prev_mb_skipped = 0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5917 |
3077 | 5918 compute_mb_neighbors(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5919 if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5920 av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" ); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5921 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5922 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5923 |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5924 if( h->slice_type == B_TYPE ) { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5925 if( mb_type < 23 ){ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5926 partition_count= b_mb_type_info[mb_type].partition_count; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5927 mb_type= b_mb_type_info[mb_type].type; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5928 }else{ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5929 mb_type -= 23; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5930 goto decode_intra_mb; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5931 } |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5932 } else if( h->slice_type == P_TYPE ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5933 if( mb_type < 5) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5934 partition_count= p_mb_type_info[mb_type].partition_count; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5935 mb_type= p_mb_type_info[mb_type].type; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5936 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5937 mb_type -= 5; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5938 goto decode_intra_mb; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5939 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5940 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5941 assert(h->slice_type == I_TYPE); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5942 decode_intra_mb: |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5943 partition_count = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5944 cbp= i_mb_type_info[mb_type].cbp; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5945 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5946 mb_type= i_mb_type_info[mb_type].type; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5947 } |
3316 | 5948 if(MB_FIELD) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5949 mb_type |= MB_TYPE_INTERLACED; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5950 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5951 h->slice_table[ mb_xy ]= h->slice_num; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5952 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5953 if(IS_INTRA_PCM(mb_type)) { |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5954 const uint8_t *ptr; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5955 unsigned int x, y; |
2967 | 5956 |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5957 // We assume these blocks are very rare so we dont optimize it. |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5958 // FIXME The two following lines get the bitstream position in the cabac |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5959 // decode, I think it should be done by a function in cabac.h (or cabac.c). |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5960 ptr= h->cabac.bytestream; |
4344 | 5961 if(h->cabac.low&0x1) ptr--; |
5962 if(CABAC_BITS==16){ | |
5963 if(h->cabac.low&0x1FF) ptr--; | |
5964 } | |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5965 |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5966 // The pixels are stored in the same order as levels in h->mb array. |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5967 for(y=0; y<16; y++){ |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5968 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5969 for(x=0; x<16; x++){ |
4600 | 5970 tprintf(s->avctx, "LUMA ICPM LEVEL (%3d)\n", *ptr); |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5971 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= *ptr++; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5972 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5973 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5974 for(y=0; y<8; y++){ |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5975 const int index= 256 + 4*(y&3) + 32*(y>>2); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5976 for(x=0; x<8; x++){ |
4600 | 5977 tprintf(s->avctx, "CHROMA U ICPM LEVEL (%3d)\n", *ptr); |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5978 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5979 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5980 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5981 for(y=0; y<8; y++){ |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5982 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5983 for(x=0; x<8; x++){ |
4600 | 5984 tprintf(s->avctx, "CHROMA V ICPM LEVEL (%3d)\n", *ptr); |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5985 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5986 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5987 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5988 |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5989 ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5990 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
5991 // All blocks are present |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5992 h->cbp_table[mb_xy] = 0x1ef; |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
5993 h->chroma_pred_mode_table[mb_xy] = 0; |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
5994 // In deblocking, the quantizer is 0 |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5995 s->current_picture.qscale_table[mb_xy]= 0; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
5996 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0); |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
5997 // All coeffs are present |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5998 memset(h->non_zero_count[mb_xy], 16, 16); |
2755 | 5999 s->current_picture.mb_type[mb_xy]= mb_type; |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6000 return 0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6001 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6002 |
3316 | 6003 if(MB_MBAFF){ |
6004 h->ref_count[0] <<= 1; | |
6005 h->ref_count[1] <<= 1; | |
6006 } | |
6007 | |
2449 | 6008 fill_caches(h, mb_type, 0); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6009 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6010 if( IS_INTRA( mb_type ) ) { |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
6011 int i, pred_mode; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6012 if( IS_INTRA4x4( mb_type ) ) { |
2755 | 6013 if( dct8x8_allowed && decode_cabac_mb_transform_size( h ) ) { |
6014 mb_type |= MB_TYPE_8x8DCT; | |
6015 for( i = 0; i < 16; i+=4 ) { | |
6016 int pred = pred_intra_mode( h, i ); | |
6017 int mode = decode_cabac_mb_intra4x4_pred_mode( h, pred ); | |
6018 fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 ); | |
6019 } | |
6020 } else { | |
6021 for( i = 0; i < 16; i++ ) { | |
6022 int pred = pred_intra_mode( h, i ); | |
6023 h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred ); | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6024 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6025 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] ); |
2755 | 6026 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6027 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6028 write_back_intra_pred_mode(h); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6029 if( check_intra4x4_pred_mode(h) < 0 ) return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6030 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6031 h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode ); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6032 if( h->intra16x16_pred_mode < 0 ) return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6033 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6034 h->chroma_pred_mode_table[mb_xy] = |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
6035 pred_mode = decode_cabac_mb_chroma_pre_mode( h ); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
6036 |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
6037 pred_mode= check_intra_pred_mode( h, pred_mode ); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
6038 if( pred_mode < 0 ) return -1; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
6039 h->chroma_pred_mode= pred_mode; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6040 } else if( partition_count == 4 ) { |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6041 int i, j, sub_partition_count[4], list, ref[2][4]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6042 |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6043 if( h->slice_type == B_TYPE ) { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6044 for( i = 0; i < 4; i++ ) { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6045 h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h ); |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6046 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6047 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6048 } |
3482 | 6049 if( IS_DIRECT(h->sub_mb_type[0] | h->sub_mb_type[1] | |
6050 h->sub_mb_type[2] | h->sub_mb_type[3]) ) { | |
2396 | 6051 pred_direct_motion(h, &mb_type); |
6052 if( h->ref_count[0] > 1 || h->ref_count[1] > 1 ) { | |
6053 for( i = 0; i < 4; i++ ) | |
6054 if( IS_DIRECT(h->sub_mb_type[i]) ) | |
6055 fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, 1, 1 ); | |
6056 } | |
6057 } | |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6058 } else { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6059 for( i = 0; i < 4; i++ ) { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6060 h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h ); |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6061 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6062 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6063 } |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6064 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6065 |
4533 | 6066 for( list = 0; list < h->list_count; list++ ) { |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6067 for( i = 0; i < 4; i++ ) { |
2396 | 6068 if(IS_DIRECT(h->sub_mb_type[i])) continue; |
6069 if(IS_DIR(h->sub_mb_type[i], 0, list)){ | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6070 if( h->ref_count[list] > 1 ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6071 ref[list][i] = decode_cabac_mb_ref( h, list, 4*i ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6072 else |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6073 ref[list][i] = 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6074 } else { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6075 ref[list][i] = -1; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6076 } |
2110 | 6077 h->ref_cache[list][ scan8[4*i]+1 ]= |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6078 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6079 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6080 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6081 |
2755 | 6082 if(dct8x8_allowed) |
6083 dct8x8_allowed = get_dct8x8_allowed(h); | |
6084 | |
4533 | 6085 for(list=0; list<h->list_count; list++){ |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6086 for(i=0; i<4; i++){ |
2396 | 6087 if(IS_DIRECT(h->sub_mb_type[i])){ |
6088 fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 4); | |
6089 continue; | |
6090 } | |
2110 | 6091 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]; |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6092 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6093 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6094 const int sub_mb_type= h->sub_mb_type[i]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6095 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6096 for(j=0; j<sub_partition_count[i]; j++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6097 int mpx, mpy; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6098 int mx, my; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6099 const int index= 4*i + block_width*j; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6100 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6101 int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6102 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6103 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6104 mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6105 my = mpy + decode_cabac_mb_mvd( h, list, index, 1 ); |
4600 | 6106 tprintf(s->avctx, "final mv:%d %d\n", mx, my); |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6107 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6108 if(IS_SUB_8X8(sub_mb_type)){ |
4530 | 6109 mv_cache[ 1 ][0]= |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6110 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx; |
4530 | 6111 mv_cache[ 1 ][1]= |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6112 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6113 |
4530 | 6114 mvd_cache[ 1 ][0]= |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6115 mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx; |
4530 | 6116 mvd_cache[ 1 ][1]= |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6117 mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6118 }else if(IS_SUB_8X4(sub_mb_type)){ |
4530 | 6119 mv_cache[ 1 ][0]= mx; |
6120 mv_cache[ 1 ][1]= my; | |
6121 | |
6122 mvd_cache[ 1 ][0]= mx - mpx; | |
6123 mvd_cache[ 1 ][1]= my - mpy; | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6124 }else if(IS_SUB_4X8(sub_mb_type)){ |
4530 | 6125 mv_cache[ 8 ][0]= mx; |
6126 mv_cache[ 8 ][1]= my; | |
6127 | |
6128 mvd_cache[ 8 ][0]= mx - mpx; | |
6129 mvd_cache[ 8 ][1]= my - mpy; | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6130 } |
4530 | 6131 mv_cache[ 0 ][0]= mx; |
6132 mv_cache[ 0 ][1]= my; | |
6133 | |
6134 mvd_cache[ 0 ][0]= mx - mpx; | |
6135 mvd_cache[ 0 ][1]= my - mpy; | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6136 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6137 }else{ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6138 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6139 uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6140 p[0] = p[1] = p[8] = p[9] = 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6141 pd[0]= pd[1]= pd[8]= pd[9]= 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6142 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6143 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6144 } |
2396 | 6145 } else if( IS_DIRECT(mb_type) ) { |
6146 pred_direct_motion(h, &mb_type); | |
6147 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4); | |
6148 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4); | |
2755 | 6149 dct8x8_allowed &= h->sps.direct_8x8_inference_flag; |
2396 | 6150 } else { |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6151 int list, mx, my, i, mpx, mpy; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6152 if(IS_16X16(mb_type)){ |
4534 | 6153 for(list=0; list<h->list_count; list++){ |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6154 if(IS_DIR(mb_type, 0, list)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6155 const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6156 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1); |
2523 | 6157 }else |
4534 | 6158 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1); //FIXME factorize and the other fill_rect below too |
6159 } | |
6160 for(list=0; list<h->list_count; list++){ | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6161 if(IS_DIR(mb_type, 0, list)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6162 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6163 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6164 mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6165 my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 ); |
4600 | 6166 tprintf(s->avctx, "final mv:%d %d\n", mx, my); |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6167 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6168 fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6169 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4); |
2523 | 6170 }else |
6171 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4); | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6172 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6173 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6174 else if(IS_16X8(mb_type)){ |
4533 | 6175 for(list=0; list<h->list_count; list++){ |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6176 for(i=0; i<2; i++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6177 if(IS_DIR(mb_type, i, list)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6178 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6179 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1); |
2396 | 6180 }else |
6181 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1); | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6182 } |
4533 | 6183 } |
6184 for(list=0; list<h->list_count; list++){ | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6185 for(i=0; i<2; i++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6186 if(IS_DIR(mb_type, i, list)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6187 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6188 mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6189 my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 ); |
4600 | 6190 tprintf(s->avctx, "final mv:%d %d\n", mx, my); |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6191 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6192 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6193 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4); |
2523 | 6194 }else{ |
2396 | 6195 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4); |
6196 fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4); | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6197 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6198 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6199 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6200 }else{ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6201 assert(IS_8X16(mb_type)); |
4533 | 6202 for(list=0; list<h->list_count; list++){ |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6203 for(i=0; i<2; i++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6204 if(IS_DIR(mb_type, i, list)){ //FIXME optimize |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6205 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6206 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1); |
2396 | 6207 }else |
6208 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1); | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6209 } |
4533 | 6210 } |
6211 for(list=0; list<h->list_count; list++){ | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6212 for(i=0; i<2; i++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6213 if(IS_DIR(mb_type, i, list)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6214 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6215 mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6216 my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6217 |
4600 | 6218 tprintf(s->avctx, "final mv:%d %d\n", mx, my); |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6219 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6220 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4); |
2523 | 6221 }else{ |
2396 | 6222 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4); |
6223 fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4); | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6224 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6225 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6226 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6227 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6228 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6229 |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
6230 if( IS_INTER( mb_type ) ) { |
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
6231 h->chroma_pred_mode_table[mb_xy] = 0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6232 write_back_motion( h, mb_type ); |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
6233 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6234 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6235 if( !IS_INTRA16x16( mb_type ) ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6236 cbp = decode_cabac_mb_cbp_luma( h ); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6237 cbp |= decode_cabac_mb_cbp_chroma( h ) << 4; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6238 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6239 |
3651 | 6240 h->cbp_table[mb_xy] = h->cbp = cbp; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6241 |
2755 | 6242 if( dct8x8_allowed && (cbp&15) && !IS_INTRA( mb_type ) ) { |
6243 if( decode_cabac_mb_transform_size( h ) ) | |
6244 mb_type |= MB_TYPE_8x8DCT; | |
6245 } | |
6246 s->current_picture.mb_type[mb_xy]= mb_type; | |
6247 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6248 if( cbp || IS_INTRA16x16( mb_type ) ) { |
3174 | 6249 const uint8_t *scan, *scan8x8, *dc_scan; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6250 int dqp; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6251 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6252 if(IS_INTERLACED(mb_type)){ |
3316 | 6253 scan8x8= s->qscale ? h->field_scan8x8 : h->field_scan8x8_q0; |
2763 | 6254 scan= s->qscale ? h->field_scan : h->field_scan_q0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6255 dc_scan= luma_dc_field_scan; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6256 }else{ |
3316 | 6257 scan8x8= s->qscale ? h->zigzag_scan8x8 : h->zigzag_scan8x8_q0; |
2763 | 6258 scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6259 dc_scan= luma_dc_zigzag_scan; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6260 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6261 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6262 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h ); |
2883
a95e55679dfa
check return value of decode_cabac_mb_dqp, it is INT_MIN on error.
reimar
parents:
2867
diff
changeset
|
6263 if( dqp == INT_MIN ){ |
a95e55679dfa
check return value of decode_cabac_mb_dqp, it is INT_MIN on error.
reimar
parents:
2867
diff
changeset
|
6264 av_log(h->s.avctx, AV_LOG_ERROR, "cabac decode of qscale diff failed at %d %d\n", s->mb_x, s->mb_y); |
a95e55679dfa
check return value of decode_cabac_mb_dqp, it is INT_MIN on error.
reimar
parents:
2867
diff
changeset
|
6265 return -1; |
a95e55679dfa
check return value of decode_cabac_mb_dqp, it is INT_MIN on error.
reimar
parents:
2867
diff
changeset
|
6266 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6267 s->qscale += dqp; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6268 if(((unsigned)s->qscale) > 51){ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6269 if(s->qscale<0) s->qscale+= 52; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6270 else s->qscale-= 52; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6271 } |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6272 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6273 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6274 if( IS_INTRA16x16( mb_type ) ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6275 int i; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6276 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" ); |
2898 | 6277 if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, NULL, 16) < 0) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6278 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6279 if( cbp&15 ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6280 for( i = 0; i < 16; i++ ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6281 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i ); |
2919 | 6282 if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6283 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6284 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6285 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6286 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6287 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6288 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6289 int i8x8, i4x4; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6290 for( i8x8 = 0; i8x8 < 4; i8x8++ ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6291 if( cbp & (1<<i8x8) ) { |
2755 | 6292 if( IS_8x8DCT(mb_type) ) { |
6293 if( decode_cabac_residual(h, h->mb + 64*i8x8, 5, 4*i8x8, | |
3174 | 6294 scan8x8, h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 64) < 0 ) |
2755 | 6295 return -1; |
6296 } else | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6297 for( i4x4 = 0; i4x4 < 4; i4x4++ ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6298 const int index = 4*i8x8 + i4x4; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6299 //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index ); |
3972
f5f1c9af095d
move outcommented START/STOP_TIMER to a hopefully better place for benchmarking ...
michael
parents:
3955
diff
changeset
|
6300 //START_TIMER |
2919 | 6301 if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale], 16) < 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6302 return -1; |
3972
f5f1c9af095d
move outcommented START/STOP_TIMER to a hopefully better place for benchmarking ...
michael
parents:
3955
diff
changeset
|
6303 //STOP_TIMER("decode_residual") |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6304 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6305 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6306 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ]; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6307 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6308 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6309 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6310 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6311 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6312 if( cbp&0x30 ){ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6313 int c; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6314 for( c = 0; c < 2; c++ ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6315 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c ); |
2898 | 6316 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, NULL, 4) < 0) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6317 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6318 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6319 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6320 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6321 if( cbp&0x20 ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6322 int c, i; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6323 for( c = 0; c < 2; c++ ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6324 for( i = 0; i < 4; i++ ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6325 const int index = 16 + 4 * c + i; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6326 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 ); |
2919 | 6327 if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->dequant4_coeff[c+1+(IS_INTRA( mb_type ) ? 0:3)][h->chroma_qp], 15) < 0) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6328 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6329 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6330 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6331 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6332 uint8_t * const nnz= &h->non_zero_count_cache[0]; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6333 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] = |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6334 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6335 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6336 } else { |
2315 | 6337 uint8_t * const nnz= &h->non_zero_count_cache[0]; |
6338 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1); | |
6339 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] = | |
6340 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0; | |
3316 | 6341 h->last_qscale_diff = 0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6342 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6343 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6344 s->current_picture.qscale_table[mb_xy]= s->qscale; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6345 write_back_non_zero_count(h); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6346 |
3316 | 6347 if(MB_MBAFF){ |
6348 h->ref_count[0] >>= 1; | |
6349 h->ref_count[1] >>= 1; | |
6350 } | |
6351 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6352 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6353 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6354 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6355 |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6356 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) { |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6357 int i, d; |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6358 const int index_a = qp + h->slice_alpha_c0_offset; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6359 const int alpha = (alpha_table+52)[index_a]; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6360 const int beta = (beta_table+52)[qp + h->slice_beta_offset]; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6361 |
2633 | 6362 if( bS[0] < 4 ) { |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2683
diff
changeset
|
6363 int8_t tc[4]; |
2633 | 6364 for(i=0; i<4; i++) |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6365 tc[i] = bS[i] ? (tc0_table+52)[index_a][bS[i] - 1] : -1; |
2633 | 6366 h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc); |
6367 } else { | |
6368 /* 16px edge length, because bS=4 is triggered by being at | |
6369 * the edge of an intra MB, so all 4 bS are the same */ | |
6370 for( d = 0; d < 16; d++ ) { | |
1898 | 6371 const int p0 = pix[-1]; |
6372 const int p1 = pix[-2]; | |
6373 const int p2 = pix[-3]; | |
6374 | |
6375 const int q0 = pix[0]; | |
6376 const int q1 = pix[1]; | |
6377 const int q2 = pix[2]; | |
6378 | |
4001 | 6379 if( FFABS( p0 - q0 ) < alpha && |
6380 FFABS( p1 - p0 ) < beta && | |
6381 FFABS( q1 - q0 ) < beta ) { | |
6382 | |
6383 if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){ | |
6384 if( FFABS( p2 - p0 ) < beta) | |
1898 | 6385 { |
6386 const int p3 = pix[-4]; | |
6387 /* p0', p1', p2' */ | |
6388 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3; | |
6389 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2; | |
6390 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3; | |
6391 } else { | |
6392 /* p0' */ | |
6393 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
6394 } | |
4001 | 6395 if( FFABS( q2 - q0 ) < beta) |
1898 | 6396 { |
6397 const int q3 = pix[3]; | |
6398 /* q0', q1', q2' */ | |
6399 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3; | |
6400 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2; | |
6401 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3; | |
6402 } else { | |
6403 /* q0' */ | |
6404 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
6405 } | |
6406 }else{ | |
6407 /* p0', q0' */ | |
6408 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
6409 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
6410 } | |
4600 | 6411 tprintf(h->s.avctx, "filter_mb_edgev i:%d d:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, p2, p1, p0, q0, q1, q2, pix[-2], pix[-1], pix[0], pix[1]); |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6412 } |
1898 | 6413 pix += stride; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6414 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6415 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6416 } |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6417 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) { |
2834 | 6418 int i; |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6419 const int index_a = qp + h->slice_alpha_c0_offset; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6420 const int alpha = (alpha_table+52)[index_a]; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6421 const int beta = (beta_table+52)[qp + h->slice_beta_offset]; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6422 |
2633 | 6423 if( bS[0] < 4 ) { |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2683
diff
changeset
|
6424 int8_t tc[4]; |
2633 | 6425 for(i=0; i<4; i++) |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6426 tc[i] = bS[i] ? (tc0_table+52)[index_a][bS[i] - 1] + 1 : 0; |
2633 | 6427 h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc); |
6428 } else { | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2683
diff
changeset
|
6429 h->s.dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta); |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6430 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6431 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6432 |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6433 static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[8], int qp[2] ) { |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6434 int i; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6435 for( i = 0; i < 16; i++, pix += stride) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6436 int index_a; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6437 int alpha; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6438 int beta; |
2967 | 6439 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6440 int qp_index; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6441 int bS_index = (i >> 1); |
3316 | 6442 if (!MB_FIELD) { |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6443 bS_index &= ~1; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6444 bS_index |= (i & 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6445 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6446 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6447 if( bS[bS_index] == 0 ) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6448 continue; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6449 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6450 |
3316 | 6451 qp_index = MB_FIELD ? (i >> 3) : (i & 1); |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6452 index_a = qp[qp_index] + h->slice_alpha_c0_offset; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6453 alpha = (alpha_table+52)[index_a]; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6454 beta = (beta_table+52)[qp[qp_index] + h->slice_beta_offset]; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6455 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6456 if( bS[bS_index] < 4 ) { |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6457 const int tc0 = (tc0_table+52)[index_a][bS[bS_index] - 1]; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6458 const int p0 = pix[-1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6459 const int p1 = pix[-2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6460 const int p2 = pix[-3]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6461 const int q0 = pix[0]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6462 const int q1 = pix[1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6463 const int q2 = pix[2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6464 |
4001 | 6465 if( FFABS( p0 - q0 ) < alpha && |
6466 FFABS( p1 - p0 ) < beta && | |
6467 FFABS( q1 - q0 ) < beta ) { | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6468 int tc = tc0; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6469 int i_delta; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6470 |
4001 | 6471 if( FFABS( p2 - p0 ) < beta ) { |
4594 | 6472 pix[-2] = p1 + av_clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 ); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6473 tc++; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6474 } |
4001 | 6475 if( FFABS( q2 - q0 ) < beta ) { |
4594 | 6476 pix[1] = q1 + av_clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 ); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6477 tc++; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6478 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6479 |
4594 | 6480 i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
6481 pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */ | |
6482 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */ | |
4600 | 6483 tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6484 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6485 }else{ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6486 const int p0 = pix[-1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6487 const int p1 = pix[-2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6488 const int p2 = pix[-3]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6489 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6490 const int q0 = pix[0]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6491 const int q1 = pix[1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6492 const int q2 = pix[2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6493 |
4001 | 6494 if( FFABS( p0 - q0 ) < alpha && |
6495 FFABS( p1 - p0 ) < beta && | |
6496 FFABS( q1 - q0 ) < beta ) { | |
6497 | |
6498 if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){ | |
6499 if( FFABS( p2 - p0 ) < beta) | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6500 { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6501 const int p3 = pix[-4]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6502 /* p0', p1', p2' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6503 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6504 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6505 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6506 } else { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6507 /* p0' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6508 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6509 } |
4001 | 6510 if( FFABS( q2 - q0 ) < beta) |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6511 { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6512 const int q3 = pix[3]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6513 /* q0', q1', q2' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6514 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6515 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6516 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6517 } else { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6518 /* q0' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6519 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6520 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6521 }else{ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6522 /* p0', q0' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6523 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6524 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6525 } |
4600 | 6526 tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, p2, p1, p0, q0, q1, q2, pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6527 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6528 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6529 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6530 } |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6531 static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[8], int qp[2] ) { |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6532 int i; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6533 for( i = 0; i < 8; i++, pix += stride) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6534 int index_a; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6535 int alpha; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6536 int beta; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6537 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6538 int qp_index; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6539 int bS_index = i; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6540 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6541 if( bS[bS_index] == 0 ) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6542 continue; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6543 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6544 |
3316 | 6545 qp_index = MB_FIELD ? (i >> 2) : (i & 1); |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6546 index_a = qp[qp_index] + h->slice_alpha_c0_offset; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6547 alpha = (alpha_table+52)[index_a]; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6548 beta = (beta_table+52)[qp[qp_index] + h->slice_beta_offset]; |
3316 | 6549 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6550 if( bS[bS_index] < 4 ) { |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6551 const int tc = (tc0_table+52)[index_a][bS[bS_index] - 1] + 1; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6552 const int p0 = pix[-1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6553 const int p1 = pix[-2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6554 const int q0 = pix[0]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6555 const int q1 = pix[1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6556 |
4001 | 6557 if( FFABS( p0 - q0 ) < alpha && |
6558 FFABS( p1 - p0 ) < beta && | |
6559 FFABS( q1 - q0 ) < beta ) { | |
4594 | 6560 const int i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
6561 | |
6562 pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */ | |
6563 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */ | |
4600 | 6564 tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6565 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6566 }else{ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6567 const int p0 = pix[-1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6568 const int p1 = pix[-2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6569 const int q0 = pix[0]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6570 const int q1 = pix[1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6571 |
4001 | 6572 if( FFABS( p0 - q0 ) < alpha && |
6573 FFABS( p1 - p0 ) < beta && | |
6574 FFABS( q1 - q0 ) < beta ) { | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6575 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6576 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6577 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */ |
4600 | 6578 tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, pix[-3], p1, p0, q0, q1, pix[2], pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6579 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6580 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6581 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6582 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6583 |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6584 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) { |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6585 int i, d; |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6586 const int index_a = qp + h->slice_alpha_c0_offset; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6587 const int alpha = (alpha_table+52)[index_a]; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6588 const int beta = (beta_table+52)[qp + h->slice_beta_offset]; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6589 const int pix_next = stride; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6590 |
2633 | 6591 if( bS[0] < 4 ) { |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2683
diff
changeset
|
6592 int8_t tc[4]; |
2633 | 6593 for(i=0; i<4; i++) |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6594 tc[i] = bS[i] ? (tc0_table+52)[index_a][bS[i] - 1] : -1; |
2633 | 6595 h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc); |
6596 } else { | |
6597 /* 16px edge length, see filter_mb_edgev */ | |
6598 for( d = 0; d < 16; d++ ) { | |
1898 | 6599 const int p0 = pix[-1*pix_next]; |
6600 const int p1 = pix[-2*pix_next]; | |
6601 const int p2 = pix[-3*pix_next]; | |
6602 const int q0 = pix[0]; | |
6603 const int q1 = pix[1*pix_next]; | |
6604 const int q2 = pix[2*pix_next]; | |
6605 | |
4001 | 6606 if( FFABS( p0 - q0 ) < alpha && |
6607 FFABS( p1 - p0 ) < beta && | |
6608 FFABS( q1 - q0 ) < beta ) { | |
1898 | 6609 |
6610 const int p3 = pix[-4*pix_next]; | |
6611 const int q3 = pix[ 3*pix_next]; | |
6612 | |
4001 | 6613 if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){ |
6614 if( FFABS( p2 - p0 ) < beta) { | |
1898 | 6615 /* p0', p1', p2' */ |
6616 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3; | |
6617 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2; | |
6618 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3; | |
6619 } else { | |
6620 /* p0' */ | |
6621 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
6622 } | |
4001 | 6623 if( FFABS( q2 - q0 ) < beta) { |
1898 | 6624 /* q0', q1', q2' */ |
6625 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3; | |
6626 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2; | |
6627 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3; | |
6628 } else { | |
6629 /* q0' */ | |
6630 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
6631 } | |
6632 }else{ | |
6633 /* p0', q0' */ | |
6634 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
6635 pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
6636 } | |
4600 | 6637 tprintf(h->s.avctx, "filter_mb_edgeh i:%d d:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, qp, index_a, alpha, beta, bS[i], p2, p1, p0, q0, q1, q2, pix[-2*pix_next], pix[-pix_next], pix[0], pix[pix_next]); |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6638 } |
1898 | 6639 pix++; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6640 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6641 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6642 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6643 |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6644 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) { |
2834 | 6645 int i; |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6646 const int index_a = qp + h->slice_alpha_c0_offset; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6647 const int alpha = (alpha_table+52)[index_a]; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6648 const int beta = (beta_table+52)[qp + h->slice_beta_offset]; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6649 |
2633 | 6650 if( bS[0] < 4 ) { |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2683
diff
changeset
|
6651 int8_t tc[4]; |
2633 | 6652 for(i=0; i<4; i++) |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6653 tc[i] = bS[i] ? (tc0_table+52)[index_a][bS[i] - 1] + 1 : 0; |
2633 | 6654 h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc); |
6655 } else { | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2683
diff
changeset
|
6656 h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta); |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6657 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6658 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6659 |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6660 static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6661 MpegEncContext * const s = &h->s; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6662 int mb_xy, mb_type; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6663 int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6664 |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6665 if(mb_x==0 || mb_y==0 || !s->dsp.h264_loop_filter_strength) { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6666 filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6667 return; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6668 } |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6669 assert(!FRAME_MBAFF); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6670 |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6671 mb_xy = mb_x + mb_y*s->mb_stride; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6672 mb_type = s->current_picture.mb_type[mb_xy]; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6673 qp = s->current_picture.qscale_table[mb_xy]; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6674 qp0 = s->current_picture.qscale_table[mb_xy-1]; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6675 qp1 = s->current_picture.qscale_table[h->top_mb_xy]; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6676 qpc = get_chroma_qp( h->pps.chroma_qp_index_offset, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6677 qpc0 = get_chroma_qp( h->pps.chroma_qp_index_offset, qp0 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6678 qpc1 = get_chroma_qp( h->pps.chroma_qp_index_offset, qp1 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6679 qp0 = (qp + qp0 + 1) >> 1; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6680 qp1 = (qp + qp1 + 1) >> 1; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6681 qpc0 = (qpc + qpc0 + 1) >> 1; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6682 qpc1 = (qpc + qpc1 + 1) >> 1; |
3651 | 6683 qp_thresh = 15 - h->slice_alpha_c0_offset; |
6684 if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh && | |
6685 qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh) | |
6686 return; | |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6687 |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6688 if( IS_INTRA(mb_type) ) { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6689 int16_t bS4[4] = {4,4,4,4}; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6690 int16_t bS3[4] = {3,3,3,3}; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6691 if( IS_8x8DCT(mb_type) ) { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6692 filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6693 filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6694 filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bS4, qp1 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6695 filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6696 } else { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6697 filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6698 filter_mb_edgev( h, &img_y[4*1], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6699 filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6700 filter_mb_edgev( h, &img_y[4*3], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6701 filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bS4, qp1 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6702 filter_mb_edgeh( h, &img_y[4*1*linesize], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6703 filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6704 filter_mb_edgeh( h, &img_y[4*3*linesize], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6705 } |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6706 filter_mb_edgecv( h, &img_cb[2*0], uvlinesize, bS4, qpc0 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6707 filter_mb_edgecv( h, &img_cb[2*2], uvlinesize, bS3, qpc ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6708 filter_mb_edgecv( h, &img_cr[2*0], uvlinesize, bS4, qpc0 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6709 filter_mb_edgecv( h, &img_cr[2*2], uvlinesize, bS3, qpc ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6710 filter_mb_edgech( h, &img_cb[2*0*uvlinesize], uvlinesize, bS4, qpc1 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6711 filter_mb_edgech( h, &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6712 filter_mb_edgech( h, &img_cr[2*0*uvlinesize], uvlinesize, bS4, qpc1 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6713 filter_mb_edgech( h, &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6714 return; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6715 } else { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6716 DECLARE_ALIGNED_8(int16_t, bS[2][4][4]); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6717 uint64_t (*bSv)[4] = (uint64_t(*)[4])bS; |
3651 | 6718 int edges; |
6719 if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) { | |
6720 edges = 4; | |
6721 bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL; | |
6722 } else { | |
6723 int mask_edge1 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : | |
6724 (mb_type & MB_TYPE_16x8) ? 1 : 0; | |
6725 int mask_edge0 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) | |
6726 && (s->current_picture.mb_type[mb_xy-1] & (MB_TYPE_16x16 | MB_TYPE_8x16)) | |
6727 ? 3 : 0; | |
6728 int step = IS_8x8DCT(mb_type) ? 2 : 1; | |
6729 edges = (mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4; | |
6730 s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache, | |
6731 (h->slice_type == B_TYPE), edges, step, mask_edge0, mask_edge1 ); | |
6732 } | |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6733 if( IS_INTRA(s->current_picture.mb_type[mb_xy-1]) ) |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6734 bSv[0][0] = 0x0004000400040004ULL; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6735 if( IS_INTRA(s->current_picture.mb_type[h->top_mb_xy]) ) |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6736 bSv[1][0] = 0x0004000400040004ULL; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6737 |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6738 #define FILTER(hv,dir,edge)\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6739 if(bSv[dir][edge]) {\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6740 filter_mb_edge##hv( h, &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir );\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6741 if(!(edge&1)) {\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6742 filter_mb_edgec##hv( h, &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6743 filter_mb_edgec##hv( h, &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6744 }\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6745 } |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6746 if( edges == 1 ) { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6747 FILTER(v,0,0); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6748 FILTER(h,1,0); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6749 } else if( IS_8x8DCT(mb_type) ) { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6750 FILTER(v,0,0); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6751 FILTER(v,0,2); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6752 FILTER(h,1,0); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6753 FILTER(h,1,2); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6754 } else { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6755 FILTER(v,0,0); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6756 FILTER(v,0,1); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6757 FILTER(v,0,2); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6758 FILTER(v,0,3); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6759 FILTER(h,1,0); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6760 FILTER(h,1,1); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6761 FILTER(h,1,2); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6762 FILTER(h,1,3); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6763 } |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6764 #undef FILTER |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6765 } |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6766 } |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6767 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6768 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) { |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6769 MpegEncContext * const s = &h->s; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6770 const int mb_xy= mb_x + mb_y*s->mb_stride; |
3316 | 6771 const int mb_type = s->current_picture.mb_type[mb_xy]; |
6772 const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4; | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6773 int first_vertical_edge_done = 0; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6774 int dir; |
2629
202cd69d8d2e
fix ref comparison in B-frame deblocker ('unused' and 'unavailable' were treated as different)
lorenm
parents:
2628
diff
changeset
|
6775 /* FIXME: A given frame may occupy more than one position in |
202cd69d8d2e
fix ref comparison in B-frame deblocker ('unused' and 'unavailable' were treated as different)
lorenm
parents:
2628
diff
changeset
|
6776 * the reference list. So ref2frm should be populated with |
202cd69d8d2e
fix ref comparison in B-frame deblocker ('unused' and 'unavailable' were treated as different)
lorenm
parents:
2628
diff
changeset
|
6777 * frame numbers, not indices. */ |
3316 | 6778 static const int ref2frm[34] = {-1,-1,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15, |
6779 16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31}; | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6780 |
3101 | 6781 //for sufficiently low qp, filtering wouldn't do anything |
6782 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp | |
3316 | 6783 if(!FRAME_MBAFF){ |
3101 | 6784 int qp_thresh = 15 - h->slice_alpha_c0_offset - FFMAX(0, h->pps.chroma_qp_index_offset); |
6785 int qp = s->current_picture.qscale_table[mb_xy]; | |
6786 if(qp <= qp_thresh | |
6787 && (mb_x == 0 || ((qp + s->current_picture.qscale_table[mb_xy-1] + 1)>>1) <= qp_thresh) | |
6788 && (mb_y == 0 || ((qp + s->current_picture.qscale_table[h->top_mb_xy] + 1)>>1) <= qp_thresh)){ | |
6789 return; | |
6790 } | |
6791 } | |
6792 | |
3316 | 6793 if (FRAME_MBAFF |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6794 // left mb is in picture |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6795 && h->slice_table[mb_xy-1] != 255 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6796 // and current and left pair do not have the same interlaced type |
3316 | 6797 && (IS_INTERLACED(mb_type) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1])) |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6798 // and left mb is in the same slice if deblocking_filter == 2 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6799 && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6800 /* First vertical edge is different in MBAFF frames |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
6801 * There are 8 different bS to compute and 2 different Qp |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6802 */ |
3316 | 6803 const int pair_xy = mb_x + (mb_y&~1)*s->mb_stride; |
6804 const int left_mb_xy[2] = { pair_xy-1, pair_xy-1+s->mb_stride }; | |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6805 int16_t bS[8]; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6806 int qp[2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6807 int chroma_qp[2]; |
3316 | 6808 int mb_qp, mbn0_qp, mbn1_qp; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6809 int i; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6810 first_vertical_edge_done = 1; |
3316 | 6811 |
6812 if( IS_INTRA(mb_type) ) | |
6813 bS[0] = bS[1] = bS[2] = bS[3] = bS[4] = bS[5] = bS[6] = bS[7] = 4; | |
6814 else { | |
6815 for( i = 0; i < 8; i++ ) { | |
6816 int mbn_xy = MB_FIELD ? left_mb_xy[i>>2] : left_mb_xy[i&1]; | |
6817 | |
6818 if( IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) | |
6819 bS[i] = 4; | |
6820 else if( h->non_zero_count_cache[12+8*(i>>1)] != 0 || | |
6821 /* FIXME: with 8x8dct + cavlc, should check cbp instead of nnz */ | |
6822 h->non_zero_count[mbn_xy][MB_FIELD ? i&3 : (i>>2)+(mb_y&1)*2] ) | |
6823 bS[i] = 2; | |
6824 else | |
6825 bS[i] = 1; | |
6826 } | |
6827 } | |
6828 | |
6829 mb_qp = s->current_picture.qscale_table[mb_xy]; | |
6830 mbn0_qp = s->current_picture.qscale_table[left_mb_xy[0]]; | |
6831 mbn1_qp = s->current_picture.qscale_table[left_mb_xy[1]]; | |
6832 qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1; | |
6833 chroma_qp[0] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, mb_qp ) + | |
6834 get_chroma_qp( h->pps.chroma_qp_index_offset, mbn0_qp ) + 1 ) >> 1; | |
6835 qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1; | |
6836 chroma_qp[1] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, mb_qp ) + | |
6837 get_chroma_qp( h->pps.chroma_qp_index_offset, mbn1_qp ) + 1 ) >> 1; | |
6838 | |
6839 /* Filter edge */ | |
4600 | 6840 tprintf(s->avctx, "filter mb:%d/%d MBAFF, QPy:%d/%d, QPc:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], chroma_qp[0], chroma_qp[1], linesize, uvlinesize); |
6841 { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); } | |
3316 | 6842 filter_mb_mbaff_edgev ( h, &img_y [0], linesize, bS, qp ); |
6843 filter_mb_mbaff_edgecv( h, &img_cb[0], uvlinesize, bS, chroma_qp ); | |
6844 filter_mb_mbaff_edgecv( h, &img_cr[0], uvlinesize, bS, chroma_qp ); | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6845 } |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6846 /* dir : 0 -> vertical edge, 1 -> horizontal edge */ |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6847 for( dir = 0; dir < 2; dir++ ) |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6848 { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6849 int edge; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6850 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy; |
2924 | 6851 const int mbm_type = s->current_picture.mb_type[mbm_xy]; |
2454 | 6852 int start = h->slice_table[mbm_xy] == 255 ? 1 : 0; |
6853 | |
3101 | 6854 const int edges = (mb_type & (MB_TYPE_16x16|MB_TYPE_SKIP)) |
6855 == (MB_TYPE_16x16|MB_TYPE_SKIP) ? 1 : 4; | |
2926 | 6856 // how often to recheck mv-based bS when iterating between edges |
6857 const int mask_edge = (mb_type & (MB_TYPE_16x16 | (MB_TYPE_16x8 << dir))) ? 3 : | |
6858 (mb_type & (MB_TYPE_8x16 >> dir)) ? 1 : 0; | |
6859 // how often to recheck mv-based bS when iterating along each edge | |
6860 const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)); | |
6861 | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6862 if (first_vertical_edge_done) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6863 start = 1; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6864 first_vertical_edge_done = 0; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6865 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6866 |
2454 | 6867 if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy]) |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6868 start = 1; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6869 |
3316 | 6870 if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0) && start == 0 |
6871 && !IS_INTERLACED(mb_type) | |
6872 && IS_INTERLACED(mbm_type) | |
6873 ) { | |
6874 // This is a special case in the norm where the filtering must | |
6875 // be done twice (one each of the field) even if we are in a | |
6876 // frame macroblock. | |
6877 // | |
6878 static const int nnz_idx[4] = {4,5,6,3}; | |
6879 unsigned int tmp_linesize = 2 * linesize; | |
6880 unsigned int tmp_uvlinesize = 2 * uvlinesize; | |
6881 int mbn_xy = mb_xy - 2 * s->mb_stride; | |
6882 int qp, chroma_qp; | |
6883 int i, j; | |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6884 int16_t bS[4]; |
3316 | 6885 |
6886 for(j=0; j<2; j++, mbn_xy += s->mb_stride){ | |
6887 if( IS_INTRA(mb_type) || | |
6888 IS_INTRA(s->current_picture.mb_type[mbn_xy]) ) { | |
6889 bS[0] = bS[1] = bS[2] = bS[3] = 3; | |
6890 } else { | |
6891 const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy]; | |
6892 for( i = 0; i < 4; i++ ) { | |
6893 if( h->non_zero_count_cache[scan8[0]+i] != 0 || | |
6894 mbn_nnz[nnz_idx[i]] != 0 ) | |
6895 bS[i] = 2; | |
6896 else | |
6897 bS[i] = 1; | |
6898 } | |
6899 } | |
6900 // Do not use s->qscale as luma quantizer because it has not the same | |
6901 // value in IPCM macroblocks. | |
6902 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1; | |
4600 | 6903 tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize); |
6904 { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); } | |
3316 | 6905 filter_mb_edgeh( h, &img_y[j*linesize], tmp_linesize, bS, qp ); |
6906 chroma_qp = ( h->chroma_qp + | |
6907 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1; | |
6908 filter_mb_edgech( h, &img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp ); | |
6909 filter_mb_edgech( h, &img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp ); | |
6910 } | |
6911 | |
6912 start = 1; | |
6913 } | |
6914 | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6915 /* Calculate bS */ |
2924 | 6916 for( edge = start; edge < edges; edge++ ) { |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
6917 /* mbn_xy: neighbor macroblock */ |
2924 | 6918 const int mbn_xy = edge > 0 ? mb_xy : mbm_xy; |
6919 const int mbn_type = s->current_picture.mb_type[mbn_xy]; | |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6920 int16_t bS[4]; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6921 int qp; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6922 |
2924 | 6923 if( (edge&1) && IS_8x8DCT(mb_type) ) |
2755 | 6924 continue; |
6925 | |
2924 | 6926 if( IS_INTRA(mb_type) || |
6927 IS_INTRA(mbn_type) ) { | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6928 int value; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6929 if (edge == 0) { |
2924 | 6930 if ( (!IS_INTERLACED(mb_type) && !IS_INTERLACED(mbm_type)) |
3316 | 6931 || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0)) |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6932 ) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6933 value = 4; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6934 } else { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6935 value = 3; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6936 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6937 } else { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6938 value = 3; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6939 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6940 bS[0] = bS[1] = bS[2] = bS[3] = value; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6941 } else { |
2924 | 6942 int i, l; |
6943 int mv_done; | |
6944 | |
6945 if( edge & mask_edge ) { | |
6946 bS[0] = bS[1] = bS[2] = bS[3] = 0; | |
6947 mv_done = 1; | |
6948 } | |
3316 | 6949 else if( FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbn_type)) { |
6950 bS[0] = bS[1] = bS[2] = bS[3] = 1; | |
6951 mv_done = 1; | |
6952 } | |
2924 | 6953 else if( mask_par0 && (edge || (mbn_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) { |
6954 int b_idx= 8 + 4 + edge * (dir ? 8:1); | |
6955 int bn_idx= b_idx - (dir ? 8:1); | |
6956 int v = 0; | |
6957 for( l = 0; !v && l < 1 + (h->slice_type == B_TYPE); l++ ) { | |
6958 v |= ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] || | |
4001 | 6959 FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 || |
6960 FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit; | |
2924 | 6961 } |
6962 bS[0] = bS[1] = bS[2] = bS[3] = v; | |
6963 mv_done = 1; | |
6964 } | |
6965 else | |
6966 mv_done = 0; | |
6967 | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6968 for( i = 0; i < 4; i++ ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6969 int x = dir == 0 ? edge : i; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6970 int y = dir == 0 ? i : edge; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
6971 int b_idx= 8 + 4 + x + 8*y; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
6972 int bn_idx= b_idx - (dir ? 8:1); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
6973 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
6974 if( h->non_zero_count_cache[b_idx] != 0 || |
2449 | 6975 h->non_zero_count_cache[bn_idx] != 0 ) { |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6976 bS[i] = 2; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6977 } |
2924 | 6978 else if(!mv_done) |
2523 | 6979 { |
6980 bS[i] = 0; | |
6981 for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) { | |
2629
202cd69d8d2e
fix ref comparison in B-frame deblocker ('unused' and 'unavailable' were treated as different)
lorenm
parents:
2628
diff
changeset
|
6982 if( ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] || |
4001 | 6983 FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 || |
6984 FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit ) { | |
2523 | 6985 bS[i] = 1; |
6986 break; | |
6987 } | |
6988 } | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6989 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6990 } |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
6991 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
6992 if(bS[0]+bS[1]+bS[2]+bS[3] == 0) |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
6993 continue; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6994 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6995 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6996 /* Filter edge */ |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
6997 // Do not use s->qscale as luma quantizer because it has not the same |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6998 // value in IPCM macroblocks. |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6999 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1; |
4600 | 7000 //tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp, s->current_picture.qscale_table[mbn_xy]); |
7001 tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize); | |
7002 { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); } | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7003 if( dir == 0 ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7004 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp ); |
1898 | 7005 if( (edge&1) == 0 ) { |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
7006 int chroma_qp = ( h->chroma_qp + |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7007 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7008 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7009 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7010 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7011 } else { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7012 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp ); |
1898 | 7013 if( (edge&1) == 0 ) { |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
7014 int chroma_qp = ( h->chroma_qp + |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7015 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7016 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7017 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7018 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7019 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7020 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7021 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7022 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7023 |
1168 | 7024 static int decode_slice(H264Context *h){ |
7025 MpegEncContext * const s = &h->s; | |
7026 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F; | |
7027 | |
7028 s->mb_skip_run= -1; | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7029 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7030 if( h->pps.cabac ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7031 int i; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7032 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7033 /* realign */ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7034 align_get_bits( &s->gb ); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7035 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7036 /* init cabac */ |
3993
8b7c59b7af01
make state transition tables global as they are constant and the code is slightly faster that way
michael
parents:
3991
diff
changeset
|
7037 ff_init_cabac_states( &h->cabac); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7038 ff_init_cabac_decoder( &h->cabac, |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7039 s->gb.buffer + get_bits_count(&s->gb)/8, |
2116 | 7040 ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7041 /* calculate pre-state */ |
2755 | 7042 for( i= 0; i < 460; i++ ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7043 int pre; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7044 if( h->slice_type == I_TYPE ) |
4594 | 7045 pre = av_clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 ); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7046 else |
4594 | 7047 pre = av_clip( ((cabac_context_init_PB[h->cabac_init_idc][i][0] * s->qscale) >>4 ) + cabac_context_init_PB[h->cabac_init_idc][i][1], 1, 126 ); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7048 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7049 if( pre <= 63 ) |
3976
27e90123b346
reverse remainder of the failed attempt to optimize *state=c->mps_state[s]
michael
parents:
3972
diff
changeset
|
7050 h->cabac_state[i] = 2 * ( 63 - pre ) + 0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7051 else |
3976
27e90123b346
reverse remainder of the failed attempt to optimize *state=c->mps_state[s]
michael
parents:
3972
diff
changeset
|
7052 h->cabac_state[i] = 2 * ( pre - 64 ) + 1; |
1168 | 7053 } |
7054 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7055 for(;;){ |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
7056 //START_TIMER |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7057 int ret = decode_mb_cabac(h); |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7058 int eos; |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
7059 //STOP_TIMER("decode_mb_cabac") |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7060 |
2163 | 7061 if(ret>=0) hl_decode_mb(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7062 |
3316 | 7063 if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ? |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7064 s->mb_y++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7065 |
2163 | 7066 if(ret>=0) ret = decode_mb_cabac(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7067 |
2844
5f20ab245501
prevent assert failures and infinite loops with broken streams
michael
parents:
2834
diff
changeset
|
7068 if(ret>=0) hl_decode_mb(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7069 s->mb_y--; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7070 } |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7071 eos = get_cabac_terminate( &h->cabac ); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7072 |
3948
3edbf131ee44
refill cabac variables in 16bit steps, 3% faster get_cabac()
michael
parents:
3947
diff
changeset
|
7073 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) { |
3297
9ae9653d706c
decoders should (IMHO) not duplicate frames, the application can do this if it wants
michael
parents:
3291
diff
changeset
|
7074 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%d)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7075 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7076 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7077 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7078 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7079 if( ++s->mb_x >= s->mb_width ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7080 s->mb_x = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7081 ff_draw_horiz_band(s, 16*s->mb_y, 16); |
2392 | 7082 ++s->mb_y; |
3316 | 7083 if(FRAME_MBAFF) { |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7084 ++s->mb_y; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7085 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7086 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7087 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7088 if( eos || s->mb_y >= s->mb_height ) { |
4600 | 7089 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7090 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7091 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7092 } |
1168 | 7093 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7094 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7095 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7096 for(;;){ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7097 int ret = decode_mb_cavlc(h); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7098 |
2163 | 7099 if(ret>=0) hl_decode_mb(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7100 |
3316 | 7101 if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ? |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7102 s->mb_y++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7103 ret = decode_mb_cavlc(h); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7104 |
2163 | 7105 if(ret>=0) hl_decode_mb(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7106 s->mb_y--; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7107 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7108 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7109 if(ret<0){ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7110 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7111 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7112 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7113 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7114 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7115 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7116 if(++s->mb_x >= s->mb_width){ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7117 s->mb_x=0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7118 ff_draw_horiz_band(s, 16*s->mb_y, 16); |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7119 ++s->mb_y; |
3316 | 7120 if(FRAME_MBAFF) { |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7121 ++s->mb_y; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7122 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7123 if(s->mb_y >= s->mb_height){ |
4600 | 7124 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7125 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7126 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7127 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7128 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7129 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7130 }else{ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7131 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7132 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7133 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7134 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7135 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7136 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7137 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7138 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){ |
4600 | 7139 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7140 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){ |
1168 | 7141 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); |
7142 | |
7143 return 0; | |
7144 }else{ | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7145 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); |
1168 | 7146 |
7147 return -1; | |
7148 } | |
7149 } | |
7150 } | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7151 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7152 |
1168 | 7153 #if 0 |
7154 for(;s->mb_y < s->mb_height; s->mb_y++){ | |
7155 for(;s->mb_x < s->mb_width; s->mb_x++){ | |
7156 int ret= decode_mb(h); | |
2967 | 7157 |
1168 | 7158 hl_decode_mb(h); |
7159 | |
7160 if(ret<0){ | |
3177 | 7161 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y); |
1168 | 7162 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); |
7163 | |
7164 return -1; | |
7165 } | |
2967 | 7166 |
1168 | 7167 if(++s->mb_x >= s->mb_width){ |
7168 s->mb_x=0; | |
7169 if(++s->mb_y >= s->mb_height){ | |
7170 if(get_bits_count(s->gb) == s->gb.size_in_bits){ | |
7171 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); | |
7172 | |
7173 return 0; | |
7174 }else{ | |
7175 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); | |
7176 | |
7177 return -1; | |
7178 } | |
7179 } | |
7180 } | |
2967 | 7181 |
1168 | 7182 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){ |
7183 if(get_bits_count(s->gb) == s->gb.size_in_bits){ | |
7184 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); | |
7185 | |
7186 return 0; | |
7187 }else{ | |
7188 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); | |
7189 | |
7190 return -1; | |
7191 } | |
7192 } | |
7193 } | |
7194 s->mb_x=0; | |
7195 ff_draw_horiz_band(s, 16*s->mb_y, 16); | |
7196 } | |
7197 #endif | |
7198 return -1; //not reached | |
7199 } | |
7200 | |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7201 static int decode_unregistered_user_data(H264Context *h, int size){ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7202 MpegEncContext * const s = &h->s; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7203 uint8_t user_data[16+256]; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7204 int e, build, i; |
2967 | 7205 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7206 if(size<16) |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7207 return -1; |
2967 | 7208 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7209 for(i=0; i<sizeof(user_data)-1 && i<size; i++){ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7210 user_data[i]= get_bits(&s->gb, 8); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7211 } |
2967 | 7212 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7213 user_data[i]= 0; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7214 e= sscanf(user_data+16, "x264 - core %d"/*%s - H.264/MPEG-4 AVC codec - Copyleft 2005 - http://www.videolan.org/x264.html*/, &build); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7215 if(e==1 && build>=0) |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7216 h->x264_build= build; |
2967 | 7217 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7218 if(s->avctx->debug & FF_DEBUG_BUGS) |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7219 av_log(s->avctx, AV_LOG_DEBUG, "user data:\"%s\"\n", user_data+16); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7220 |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7221 for(; i<size; i++) |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7222 skip_bits(&s->gb, 8); |
2967 | 7223 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7224 return 0; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7225 } |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7226 |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7227 static int decode_sei(H264Context *h){ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7228 MpegEncContext * const s = &h->s; |
2967 | 7229 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7230 while(get_bits_count(&s->gb) + 16 < s->gb.size_in_bits){ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7231 int size, type; |
2967 | 7232 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7233 type=0; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7234 do{ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7235 type+= show_bits(&s->gb, 8); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7236 }while(get_bits(&s->gb, 8) == 255); |
2967 | 7237 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7238 size=0; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7239 do{ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7240 size+= show_bits(&s->gb, 8); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7241 }while(get_bits(&s->gb, 8) == 255); |
2967 | 7242 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7243 switch(type){ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7244 case 5: |
3318 | 7245 if(decode_unregistered_user_data(h, size) < 0) |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7246 return -1; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7247 break; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7248 default: |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7249 skip_bits(&s->gb, 8*size); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7250 } |
2967 | 7251 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7252 //FIXME check bits here |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7253 align_get_bits(&s->gb); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7254 } |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7255 |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7256 return 0; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7257 } |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7258 |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7259 static inline void decode_hrd_parameters(H264Context *h, SPS *sps){ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7260 MpegEncContext * const s = &h->s; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7261 int cpb_count, i; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7262 cpb_count = get_ue_golomb(&s->gb) + 1; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7263 get_bits(&s->gb, 4); /* bit_rate_scale */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7264 get_bits(&s->gb, 4); /* cpb_size_scale */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7265 for(i=0; i<cpb_count; i++){ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7266 get_ue_golomb(&s->gb); /* bit_rate_value_minus1 */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7267 get_ue_golomb(&s->gb); /* cpb_size_value_minus1 */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7268 get_bits1(&s->gb); /* cbr_flag */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7269 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7270 get_bits(&s->gb, 5); /* initial_cpb_removal_delay_length_minus1 */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7271 get_bits(&s->gb, 5); /* cpb_removal_delay_length_minus1 */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7272 get_bits(&s->gb, 5); /* dpb_output_delay_length_minus1 */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7273 get_bits(&s->gb, 5); /* time_offset_length */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7274 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7275 |
1168 | 7276 static inline int decode_vui_parameters(H264Context *h, SPS *sps){ |
7277 MpegEncContext * const s = &h->s; | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7278 int aspect_ratio_info_present_flag; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7279 unsigned int aspect_ratio_idc; |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7280 int nal_hrd_parameters_present_flag, vcl_hrd_parameters_present_flag; |
1168 | 7281 |
7282 aspect_ratio_info_present_flag= get_bits1(&s->gb); | |
2967 | 7283 |
1168 | 7284 if( aspect_ratio_info_present_flag ) { |
7285 aspect_ratio_idc= get_bits(&s->gb, 8); | |
7286 if( aspect_ratio_idc == EXTENDED_SAR ) { | |
1548 | 7287 sps->sar.num= get_bits(&s->gb, 16); |
7288 sps->sar.den= get_bits(&s->gb, 16); | |
3196
15157293beea
fixing coverity warnig CID: 253 (reading over the end of the aspect_ratio array for illegal aspect ratios)
michael
parents:
3178
diff
changeset
|
7289 }else if(aspect_ratio_idc < 14){ |
1548 | 7290 sps->sar= pixel_aspect[aspect_ratio_idc]; |
1168 | 7291 }else{ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
7292 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n"); |
1168 | 7293 return -1; |
7294 } | |
7295 }else{ | |
2967 | 7296 sps->sar.num= |
1548 | 7297 sps->sar.den= 0; |
1168 | 7298 } |
7299 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height); | |
2174
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7300 |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7301 if(get_bits1(&s->gb)){ /* overscan_info_present_flag */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7302 get_bits1(&s->gb); /* overscan_appropriate_flag */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7303 } |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7304 |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7305 if(get_bits1(&s->gb)){ /* video_signal_type_present_flag */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7306 get_bits(&s->gb, 3); /* video_format */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7307 get_bits1(&s->gb); /* video_full_range_flag */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7308 if(get_bits1(&s->gb)){ /* colour_description_present_flag */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7309 get_bits(&s->gb, 8); /* colour_primaries */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7310 get_bits(&s->gb, 8); /* transfer_characteristics */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7311 get_bits(&s->gb, 8); /* matrix_coefficients */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7312 } |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7313 } |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7314 |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7315 if(get_bits1(&s->gb)){ /* chroma_location_info_present_flag */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7316 get_ue_golomb(&s->gb); /* chroma_sample_location_type_top_field */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7317 get_ue_golomb(&s->gb); /* chroma_sample_location_type_bottom_field */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7318 } |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7319 |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7320 sps->timing_info_present_flag = get_bits1(&s->gb); |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7321 if(sps->timing_info_present_flag){ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7322 sps->num_units_in_tick = get_bits_long(&s->gb, 32); |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7323 sps->time_scale = get_bits_long(&s->gb, 32); |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7324 sps->fixed_frame_rate_flag = get_bits1(&s->gb); |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7325 } |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (M«©ns Rullg«©rd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7326 |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7327 nal_hrd_parameters_present_flag = get_bits1(&s->gb); |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7328 if(nal_hrd_parameters_present_flag) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7329 decode_hrd_parameters(h, sps); |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7330 vcl_hrd_parameters_present_flag = get_bits1(&s->gb); |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7331 if(vcl_hrd_parameters_present_flag) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7332 decode_hrd_parameters(h, sps); |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7333 if(nal_hrd_parameters_present_flag || vcl_hrd_parameters_present_flag) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7334 get_bits1(&s->gb); /* low_delay_hrd_flag */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7335 get_bits1(&s->gb); /* pic_struct_present_flag */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7336 |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7337 sps->bitstream_restriction_flag = get_bits1(&s->gb); |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7338 if(sps->bitstream_restriction_flag){ |
4389 | 7339 unsigned int num_reorder_frames; |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7340 get_bits1(&s->gb); /* motion_vectors_over_pic_boundaries_flag */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7341 get_ue_golomb(&s->gb); /* max_bytes_per_pic_denom */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7342 get_ue_golomb(&s->gb); /* max_bits_per_mb_denom */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7343 get_ue_golomb(&s->gb); /* log2_max_mv_length_horizontal */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7344 get_ue_golomb(&s->gb); /* log2_max_mv_length_vertical */ |
4389 | 7345 num_reorder_frames= get_ue_golomb(&s->gb); |
7346 get_ue_golomb(&s->gb); /*max_dec_frame_buffering*/ | |
7347 | |
7348 if(num_reorder_frames > 16 /*max_dec_frame_buffering || max_dec_frame_buffering > 16*/){ | |
7349 av_log(h->s.avctx, AV_LOG_ERROR, "illegal num_reorder_frames %d\n", num_reorder_frames); | |
7350 return -1; | |
7351 } | |
7352 | |
7353 sps->num_reorder_frames= num_reorder_frames; | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7354 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7355 |
1168 | 7356 return 0; |
7357 } | |
7358 | |
3004 | 7359 static void decode_scaling_list(H264Context *h, uint8_t *factors, int size, |
7360 const uint8_t *jvt_list, const uint8_t *fallback_list){ | |
2919 | 7361 MpegEncContext * const s = &h->s; |
7362 int i, last = 8, next = 8; | |
7363 const uint8_t *scan = size == 16 ? zigzag_scan : zigzag_scan8x8; | |
3004 | 7364 if(!get_bits1(&s->gb)) /* matrix not written, we use the predicted one */ |
7365 memcpy(factors, fallback_list, size*sizeof(uint8_t)); | |
2919 | 7366 else |
7367 for(i=0;i<size;i++){ | |
7368 if(next) | |
7369 next = (last + get_se_golomb(&s->gb)) & 0xff; | |
3004 | 7370 if(!i && !next){ /* matrix not written, we use the preset one */ |
7371 memcpy(factors, jvt_list, size*sizeof(uint8_t)); | |
2919 | 7372 break; |
7373 } | |
7374 last = factors[scan[i]] = next ? next : last; | |
7375 } | |
7376 } | |
7377 | |
7378 static void decode_scaling_matrices(H264Context *h, SPS *sps, PPS *pps, int is_sps, | |
7379 uint8_t (*scaling_matrix4)[16], uint8_t (*scaling_matrix8)[64]){ | |
7380 MpegEncContext * const s = &h->s; | |
7381 int fallback_sps = !is_sps && sps->scaling_matrix_present; | |
7382 const uint8_t *fallback[4] = { | |
7383 fallback_sps ? sps->scaling_matrix4[0] : default_scaling4[0], | |
7384 fallback_sps ? sps->scaling_matrix4[3] : default_scaling4[1], | |
7385 fallback_sps ? sps->scaling_matrix8[0] : default_scaling8[0], | |
7386 fallback_sps ? sps->scaling_matrix8[1] : default_scaling8[1] | |
7387 }; | |
7388 if(get_bits1(&s->gb)){ | |
7389 sps->scaling_matrix_present |= is_sps; | |
3004 | 7390 decode_scaling_list(h,scaling_matrix4[0],16,default_scaling4[0],fallback[0]); // Intra, Y |
7391 decode_scaling_list(h,scaling_matrix4[1],16,default_scaling4[0],scaling_matrix4[0]); // Intra, Cr | |
7392 decode_scaling_list(h,scaling_matrix4[2],16,default_scaling4[0],scaling_matrix4[1]); // Intra, Cb | |
7393 decode_scaling_list(h,scaling_matrix4[3],16,default_scaling4[1],fallback[1]); // Inter, Y | |
7394 decode_scaling_list(h,scaling_matrix4[4],16,default_scaling4[1],scaling_matrix4[3]); // Inter, Cr | |
7395 decode_scaling_list(h,scaling_matrix4[5],16,default_scaling4[1],scaling_matrix4[4]); // Inter, Cb | |
2919 | 7396 if(is_sps || pps->transform_8x8_mode){ |
3004 | 7397 decode_scaling_list(h,scaling_matrix8[0],64,default_scaling8[0],fallback[2]); // Intra, Y |
7398 decode_scaling_list(h,scaling_matrix8[1],64,default_scaling8[1],fallback[3]); // Inter, Y | |
2919 | 7399 } |
7400 } else if(fallback_sps) { | |
7401 memcpy(scaling_matrix4, sps->scaling_matrix4, 6*16*sizeof(uint8_t)); | |
7402 memcpy(scaling_matrix8, sps->scaling_matrix8, 2*64*sizeof(uint8_t)); | |
7403 } | |
7404 } | |
7405 | |
1168 | 7406 static inline int decode_seq_parameter_set(H264Context *h){ |
7407 MpegEncContext * const s = &h->s; | |
1371 | 7408 int profile_idc, level_idc; |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7409 unsigned int sps_id, tmp, mb_width, mb_height; |
4362
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7410 int i; |
1168 | 7411 SPS *sps; |
2967 | 7412 |
1168 | 7413 profile_idc= get_bits(&s->gb, 8); |
1371 | 7414 get_bits1(&s->gb); //constraint_set0_flag |
7415 get_bits1(&s->gb); //constraint_set1_flag | |
7416 get_bits1(&s->gb); //constraint_set2_flag | |
2312 | 7417 get_bits1(&s->gb); //constraint_set3_flag |
7418 get_bits(&s->gb, 4); // reserved | |
1168 | 7419 level_idc= get_bits(&s->gb, 8); |
7420 sps_id= get_ue_golomb(&s->gb); | |
2967 | 7421 |
4362
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7422 if (sps_id >= MAX_SPS_COUNT){ |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7423 // ok it has gone out of hand, someone is sending us bad stuff. |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7424 av_log(h->s.avctx, AV_LOG_ERROR, "illegal sps_id (%d)\n", sps_id); |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7425 return -1; |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7426 } |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7427 |
1168 | 7428 sps= &h->sps_buffer[ sps_id ]; |
7429 sps->profile_idc= profile_idc; | |
7430 sps->level_idc= level_idc; | |
2312 | 7431 |
2755 | 7432 if(sps->profile_idc >= 100){ //high profile |
7433 if(get_ue_golomb(&s->gb) == 3) //chroma_format_idc | |
7434 get_bits1(&s->gb); //residual_color_transform_flag | |
7435 get_ue_golomb(&s->gb); //bit_depth_luma_minus8 | |
7436 get_ue_golomb(&s->gb); //bit_depth_chroma_minus8 | |
2763 | 7437 sps->transform_bypass = get_bits1(&s->gb); |
2919 | 7438 decode_scaling_matrices(h, sps, NULL, 1, sps->scaling_matrix4, sps->scaling_matrix8); |
7439 }else | |
7440 sps->scaling_matrix_present = 0; | |
2755 | 7441 |
1168 | 7442 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4; |
7443 sps->poc_type= get_ue_golomb(&s->gb); | |
2967 | 7444 |
1168 | 7445 if(sps->poc_type == 0){ //FIXME #define |
7446 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4; | |
7447 } else if(sps->poc_type == 1){//FIXME #define | |
7448 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb); | |
7449 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb); | |
7450 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb); | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7451 tmp= get_ue_golomb(&s->gb); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7452 |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7453 if(tmp >= sizeof(sps->offset_for_ref_frame) / sizeof(sps->offset_for_ref_frame[0])){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7454 av_log(h->s.avctx, AV_LOG_ERROR, "poc_cycle_length overflow %u\n", tmp); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7455 return -1; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7456 } |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7457 sps->poc_cycle_length= tmp; |
2967 | 7458 |
1168 | 7459 for(i=0; i<sps->poc_cycle_length; i++) |
7460 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb); | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7461 }else if(sps->poc_type != 2){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
7462 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type); |
1168 | 7463 return -1; |
7464 } | |
7465 | |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7466 tmp= get_ue_golomb(&s->gb); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7467 if(tmp > MAX_PICTURE_COUNT-2){ |
2254
0dfe4e32b19c
H.264 max reference pictures fix by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2227
diff
changeset
|
7468 av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n"); |
0dfe4e32b19c
H.264 max reference pictures fix by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2227
diff
changeset
|
7469 } |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7470 sps->ref_frame_count= tmp; |
1371 | 7471 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb); |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7472 mb_width= get_ue_golomb(&s->gb) + 1; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7473 mb_height= get_ue_golomb(&s->gb) + 1; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7474 if(mb_width >= INT_MAX/16 || mb_height >= INT_MAX/16 || |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7475 avcodec_check_dimensions(NULL, 16*mb_width, 16*mb_height)){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7476 av_log(h->s.avctx, AV_LOG_ERROR, "mb_width/height overflow\n"); |
2422 | 7477 return -1; |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7478 } |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7479 sps->mb_width = mb_width; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7480 sps->mb_height= mb_height; |
2422 | 7481 |
1168 | 7482 sps->frame_mbs_only_flag= get_bits1(&s->gb); |
7483 if(!sps->frame_mbs_only_flag) | |
7484 sps->mb_aff= get_bits1(&s->gb); | |
7485 else | |
7486 sps->mb_aff= 0; | |
7487 | |
7488 sps->direct_8x8_inference_flag= get_bits1(&s->gb); | |
7489 | |
3316 | 7490 #ifndef ALLOW_INTERLACE |
7491 if(sps->mb_aff) | |
3954 | 7492 av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF support not included; enable it at compile-time.\n"); |
3316 | 7493 #endif |
7494 if(!sps->direct_8x8_inference_flag && sps->mb_aff) | |
7495 av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF + !direct_8x8_inference is not implemented\n"); | |
7496 | |
1371 | 7497 sps->crop= get_bits1(&s->gb); |
7498 if(sps->crop){ | |
7499 sps->crop_left = get_ue_golomb(&s->gb); | |
7500 sps->crop_right = get_ue_golomb(&s->gb); | |
7501 sps->crop_top = get_ue_golomb(&s->gb); | |
7502 sps->crop_bottom= get_ue_golomb(&s->gb); | |
7503 if(sps->crop_left || sps->crop_top){ | |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2623
diff
changeset
|
7504 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completely supported, this could look slightly wrong ...\n"); |
1371 | 7505 } |
7506 }else{ | |
2967 | 7507 sps->crop_left = |
7508 sps->crop_right = | |
7509 sps->crop_top = | |
1371 | 7510 sps->crop_bottom= 0; |
7511 } | |
7512 | |
1168 | 7513 sps->vui_parameters_present_flag= get_bits1(&s->gb); |
7514 if( sps->vui_parameters_present_flag ) | |
7515 decode_vui_parameters(h, sps); | |
2967 | 7516 |
1168 | 7517 if(s->avctx->debug&FF_DEBUG_PICT_INFO){ |
4362
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7518 av_log(h->s.avctx, AV_LOG_DEBUG, "sps:%u profile:%d/%d poc:%d ref:%d %dx%d %s %s crop:%d/%d/%d/%d %s\n", |
1168 | 7519 sps_id, sps->profile_idc, sps->level_idc, |
7520 sps->poc_type, | |
7521 sps->ref_frame_count, | |
7522 sps->mb_width, sps->mb_height, | |
7523 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"), | |
7524 sps->direct_8x8_inference_flag ? "8B8" : "", | |
2967 | 7525 sps->crop_left, sps->crop_right, |
7526 sps->crop_top, sps->crop_bottom, | |
1168 | 7527 sps->vui_parameters_present_flag ? "VUI" : "" |
7528 ); | |
7529 } | |
7530 return 0; | |
7531 } | |
7532 | |
2755 | 7533 static inline int decode_picture_parameter_set(H264Context *h, int bit_length){ |
1168 | 7534 MpegEncContext * const s = &h->s; |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7535 unsigned int tmp, pps_id= get_ue_golomb(&s->gb); |
4362
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7536 PPS *pps; |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7537 |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7538 if(pps_id>=MAX_PPS_COUNT){ |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7539 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n"); |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7540 return -1; |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7541 } |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7542 pps = &h->pps_buffer[pps_id]; |
2967 | 7543 |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7544 tmp= get_ue_golomb(&s->gb); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7545 if(tmp>=MAX_SPS_COUNT){ |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7546 av_log(h->s.avctx, AV_LOG_ERROR, "sps_id out of range\n"); |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7547 return -1; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7548 } |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7549 pps->sps_id= tmp; |
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7550 |
1168 | 7551 pps->cabac= get_bits1(&s->gb); |
7552 pps->pic_order_present= get_bits1(&s->gb); | |
7553 pps->slice_group_count= get_ue_golomb(&s->gb) + 1; | |
7554 if(pps->slice_group_count > 1 ){ | |
7555 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb); | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
7556 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n"); |
1168 | 7557 switch(pps->mb_slice_group_map_type){ |
7558 case 0: | |
7559 #if 0 | |
7560 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | | | |
7561 | run_length[ i ] |1 |ue(v) | | |
7562 #endif | |
7563 break; | |
7564 case 2: | |
7565 #if 0 | |
7566 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | | | |
7567 |{ | | | | |
7568 | top_left_mb[ i ] |1 |ue(v) | | |
7569 | bottom_right_mb[ i ] |1 |ue(v) | | |
7570 | } | | | | |
7571 #endif | |
7572 break; | |
7573 case 3: | |
7574 case 4: | |
7575 case 5: | |
7576 #if 0 | |
7577 | slice_group_change_direction_flag |1 |u(1) | | |
7578 | slice_group_change_rate_minus1 |1 |ue(v) | | |
7579 #endif | |
7580 break; | |
7581 case 6: | |
7582 #if 0 | |
7583 | slice_group_id_cnt_minus1 |1 |ue(v) | | |
7584 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | | | |
7585 |) | | | | |
7586 | slice_group_id[ i ] |1 |u(v) | | |
7587 #endif | |
1214 | 7588 break; |
1168 | 7589 } |
7590 } | |
7591 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1; | |
7592 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1; | |
4531 | 7593 if(pps->ref_count[0]-1 > 32-1 || pps->ref_count[1]-1 > 32-1){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
7594 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n"); |
4365
9cebff821565
checking bitstream values and other related changes
michael
parents:
4364
diff
changeset
|
7595 pps->ref_count[0]= pps->ref_count[1]= 1; |
1168 | 7596 return -1; |
7597 } | |
2967 | 7598 |
1168 | 7599 pps->weighted_pred= get_bits1(&s->gb); |
7600 pps->weighted_bipred_idc= get_bits(&s->gb, 2); | |
7601 pps->init_qp= get_se_golomb(&s->gb) + 26; | |
7602 pps->init_qs= get_se_golomb(&s->gb) + 26; | |
7603 pps->chroma_qp_index_offset= get_se_golomb(&s->gb); | |
7604 pps->deblocking_filter_parameters_present= get_bits1(&s->gb); | |
7605 pps->constrained_intra_pred= get_bits1(&s->gb); | |
7606 pps->redundant_pic_cnt_present = get_bits1(&s->gb); | |
2967 | 7607 |
3291
454de57e45cf
reinit quant matrices if pps is overwritten by another of the same pps_id.
lorenm
parents:
3285
diff
changeset
|
7608 pps->transform_8x8_mode= 0; |
454de57e45cf
reinit quant matrices if pps is overwritten by another of the same pps_id.
lorenm
parents:
3285
diff
changeset
|
7609 h->dequant_coeff_pps= -1; //contents of sps/pps can change even if id doesn't, so reinit |
2919 | 7610 memset(pps->scaling_matrix4, 16, 6*16*sizeof(uint8_t)); |
7611 memset(pps->scaling_matrix8, 16, 2*64*sizeof(uint8_t)); | |
2755 | 7612 |
7613 if(get_bits_count(&s->gb) < bit_length){ | |
7614 pps->transform_8x8_mode= get_bits1(&s->gb); | |
2919 | 7615 decode_scaling_matrices(h, &h->sps_buffer[pps->sps_id], pps, 0, pps->scaling_matrix4, pps->scaling_matrix8); |
2755 | 7616 get_se_golomb(&s->gb); //second_chroma_qp_index_offset |
7617 } | |
2967 | 7618 |
1168 | 7619 if(s->avctx->debug&FF_DEBUG_PICT_INFO){ |
4362
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7620 av_log(h->s.avctx, AV_LOG_DEBUG, "pps:%u sps:%u %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s %s\n", |
1168 | 7621 pps_id, pps->sps_id, |
7622 pps->cabac ? "CABAC" : "CAVLC", | |
7623 pps->slice_group_count, | |
7624 pps->ref_count[0], pps->ref_count[1], | |
7625 pps->weighted_pred ? "weighted" : "", | |
7626 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset, | |
7627 pps->deblocking_filter_parameters_present ? "LPAR" : "", | |
7628 pps->constrained_intra_pred ? "CONSTR" : "", | |
2755 | 7629 pps->redundant_pic_cnt_present ? "REDU" : "", |
7630 pps->transform_8x8_mode ? "8x8DCT" : "" | |
1168 | 7631 ); |
7632 } | |
2967 | 7633 |
1168 | 7634 return 0; |
7635 } | |
7636 | |
7637 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){ | |
7638 MpegEncContext * const s = &h->s; | |
7639 AVCodecContext * const avctx= s->avctx; | |
7640 int buf_index=0; | |
1322 | 7641 #if 0 |
1168 | 7642 int i; |
2867
7c7edddaa835
dont be so picky with .mp4 ... fixes Mr&MrsSmith.mp4
michael
parents:
2861
diff
changeset
|
7643 for(i=0; i<50; i++){ |
7c7edddaa835
dont be so picky with .mp4 ... fixes Mr&MrsSmith.mp4
michael
parents:
2861
diff
changeset
|
7644 av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]); |
1168 | 7645 } |
7646 #endif | |
4616
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7647 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){ |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7648 h->slice_num = 0; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7649 s->current_picture_ptr= NULL; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7650 } |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7651 |
1168 | 7652 for(;;){ |
7653 int consumed; | |
7654 int dst_length; | |
7655 int bit_length; | |
7656 uint8_t *ptr; | |
2227 | 7657 int i, nalsize = 0; |
2967 | 7658 |
2227 | 7659 if(h->is_avc) { |
7660 if(buf_index >= buf_size) break; | |
7661 nalsize = 0; | |
7662 for(i = 0; i < h->nal_length_size; i++) | |
7663 nalsize = (nalsize << 8) | buf[buf_index++]; | |
4886
7eba8b456a3f
prevent going out of the buffer if the nal size does not fit in the buffer.
gpoirier
parents:
4882
diff
changeset
|
7664 if(nalsize <= 1 || (nalsize+buf_index > buf_size)){ |
3126 | 7665 if(nalsize == 1){ |
7666 buf_index++; | |
7667 continue; | |
7668 }else{ | |
7669 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize); | |
7670 break; | |
7671 } | |
7672 } | |
2227 | 7673 } else { |
1168 | 7674 // start code prefix search |
7675 for(; buf_index + 3 < buf_size; buf_index++){ | |
4795 | 7676 // This should always succeed in the first iteration. |
1168 | 7677 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1) |
7678 break; | |
7679 } | |
2967 | 7680 |
1168 | 7681 if(buf_index+3 >= buf_size) break; |
2967 | 7682 |
1168 | 7683 buf_index+=3; |
2967 | 7684 } |
7685 | |
2401
46898a9fd6dc
Fix avc1 if there is nore than one nal per mov frame
rtognimp
parents:
2396
diff
changeset
|
7686 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, h->is_avc ? nalsize : buf_size - buf_index); |
4828
58f1856608f6
scenario: A properly coded frame is followed by an end of sequence NAL unit,
diego
parents:
4795
diff
changeset
|
7687 if (ptr==NULL || dst_length < 0){ |
4362
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7688 return -1; |
0271b214458b
harden h264 decoding to prevent some crashes when input data is corrupted.
gpoirier
parents:
4354
diff
changeset
|
7689 } |
4829
e3161603a048
Remove a NAL unit's trailing zero bytes even when dst_length is 1.
diego
parents:
4828
diff
changeset
|
7690 while(ptr[dst_length - 1] == 0 && dst_length > 0) |
3341 | 7691 dst_length--; |
4830
5a523aee7379
When dst_length == 0 bit_length has to be 0, too, but the current code still
diego
parents:
4829
diff
changeset
|
7692 bit_length= !dst_length ? 0 : (8*dst_length - decode_rbsp_trailing(h, ptr + dst_length - 1)); |
1168 | 7693 |
7694 if(s->avctx->debug&FF_DEBUG_STARTCODE){ | |
2402
f9d4e1eddbc5
- correct several errors on the deblocking accross slice boundaries.
michael
parents:
2401
diff
changeset
|
7695 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", h->nal_unit_type, buf_index, buf_size, dst_length); |
1168 | 7696 } |
2967 | 7697 |
2227 | 7698 if (h->is_avc && (nalsize != consumed)) |
7699 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize); | |
7700 | |
1168 | 7701 buf_index += consumed; |
7702 | |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7703 if( (s->hurry_up == 1 && h->nal_ref_idc == 0) //FIXME dont discard SEI id |
2792 | 7704 ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)) |
1168 | 7705 continue; |
2967 | 7706 |
1168 | 7707 switch(h->nal_unit_type){ |
7708 case NAL_IDR_SLICE: | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
7709 idr(h); //FIXME ensure we don't loose some frames if there is reordering |
1168 | 7710 case NAL_SLICE: |
7711 init_get_bits(&s->gb, ptr, bit_length); | |
7712 h->intra_gb_ptr= | |
7713 h->inter_gb_ptr= &s->gb; | |
7714 s->data_partitioning = 0; | |
2967 | 7715 |
2771 | 7716 if(decode_slice_header(h) < 0){ |
7717 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n"); | |
7718 break; | |
7719 } | |
3161
e3e94632c6e9
keyframe flag was wrong if SEIs followed the keyframe
lorenm
parents:
3145
diff
changeset
|
7720 s->current_picture_ptr->key_frame= (h->nal_unit_type == NAL_IDR_SLICE); |
2967 | 7721 if(h->redundant_pic_count==0 && s->hurry_up < 5 |
2793 | 7722 && (avctx->skip_frame < AVDISCARD_NONREF || h->nal_ref_idc) |
7723 && (avctx->skip_frame < AVDISCARD_BIDIR || h->slice_type!=B_TYPE) | |
7724 && (avctx->skip_frame < AVDISCARD_NONKEY || h->slice_type==I_TYPE) | |
7725 && avctx->skip_frame < AVDISCARD_ALL) | |
1168 | 7726 decode_slice(h); |
7727 break; | |
7728 case NAL_DPA: | |
7729 init_get_bits(&s->gb, ptr, bit_length); | |
7730 h->intra_gb_ptr= | |
7731 h->inter_gb_ptr= NULL; | |
7732 s->data_partitioning = 1; | |
2967 | 7733 |
2771 | 7734 if(decode_slice_header(h) < 0){ |
7735 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n"); | |
7736 } | |
1168 | 7737 break; |
7738 case NAL_DPB: | |
7739 init_get_bits(&h->intra_gb, ptr, bit_length); | |
7740 h->intra_gb_ptr= &h->intra_gb; | |
7741 break; | |
7742 case NAL_DPC: | |
7743 init_get_bits(&h->inter_gb, ptr, bit_length); | |
7744 h->inter_gb_ptr= &h->inter_gb; | |
1174 | 7745 |
2967 | 7746 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning |
4354
d18568fb0187
Avoid calling decode_slice when context_initialized is not set.
reimar
parents:
4344
diff
changeset
|
7747 && s->context_initialized |
2793 | 7748 && s->hurry_up < 5 |
7749 && (avctx->skip_frame < AVDISCARD_NONREF || h->nal_ref_idc) | |
7750 && (avctx->skip_frame < AVDISCARD_BIDIR || h->slice_type!=B_TYPE) | |
7751 && (avctx->skip_frame < AVDISCARD_NONKEY || h->slice_type==I_TYPE) | |
7752 && avctx->skip_frame < AVDISCARD_ALL) | |
1168 | 7753 decode_slice(h); |
7754 break; | |
7755 case NAL_SEI: | |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7756 init_get_bits(&s->gb, ptr, bit_length); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7757 decode_sei(h); |
1168 | 7758 break; |
7759 case NAL_SPS: | |
7760 init_get_bits(&s->gb, ptr, bit_length); | |
7761 decode_seq_parameter_set(h); | |
2967 | 7762 |
1168 | 7763 if(s->flags& CODEC_FLAG_LOW_DELAY) |
7764 s->low_delay=1; | |
2967 | 7765 |
2538 | 7766 if(avctx->has_b_frames < 2) |
7767 avctx->has_b_frames= !s->low_delay; | |
1168 | 7768 break; |
7769 case NAL_PPS: | |
7770 init_get_bits(&s->gb, ptr, bit_length); | |
2967 | 7771 |
2755 | 7772 decode_picture_parameter_set(h, bit_length); |
1168 | 7773 |
7774 break; | |
2960 | 7775 case NAL_AUD: |
7776 case NAL_END_SEQUENCE: | |
7777 case NAL_END_STREAM: | |
7778 case NAL_FILLER_DATA: | |
7779 case NAL_SPS_EXT: | |
7780 case NAL_AUXILIARY_SLICE: | |
1168 | 7781 break; |
2979 | 7782 default: |
7783 av_log(avctx, AV_LOG_ERROR, "Unknown NAL code: %d\n", h->nal_unit_type); | |
2967 | 7784 } |
7785 } | |
7786 | |
1168 | 7787 return buf_index; |
7788 } | |
7789 | |
7790 /** | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
7791 * returns the number of bytes consumed for building the current frame |
1168 | 7792 */ |
7793 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){ | |
7794 if(s->flags&CODEC_FLAG_TRUNCATED){ | |
7795 pos -= s->parse_context.last_index; | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
7796 if(pos<0) pos=0; // FIXME remove (unneeded?) |
2967 | 7797 |
1168 | 7798 return pos; |
7799 }else{ | |
7800 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...) | |
7801 if(pos+10>buf_size) pos=buf_size; // oops ;) | |
7802 | |
7803 return pos; | |
7804 } | |
7805 } | |
7806 | |
2967 | 7807 static int decode_frame(AVCodecContext *avctx, |
1168 | 7808 void *data, int *data_size, |
7809 uint8_t *buf, int buf_size) | |
7810 { | |
7811 H264Context *h = avctx->priv_data; | |
7812 MpegEncContext *s = &h->s; | |
2967 | 7813 AVFrame *pict = data; |
1168 | 7814 int buf_index; |
2967 | 7815 |
1168 | 7816 s->flags= avctx->flags; |
1754
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1706
diff
changeset
|
7817 s->flags2= avctx->flags2; |
1168 | 7818 |
7819 /* no supplementary picture */ | |
7820 if (buf_size == 0) { | |
4556 | 7821 Picture *out; |
7822 int i, out_idx; | |
7823 | |
7824 //FIXME factorize this with the output code below | |
7825 out = h->delayed_pic[0]; | |
7826 out_idx = 0; | |
7827 for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame; i++) | |
7828 if(h->delayed_pic[i]->poc < out->poc){ | |
7829 out = h->delayed_pic[i]; | |
7830 out_idx = i; | |
7831 } | |
7832 | |
7833 for(i=out_idx; h->delayed_pic[i]; i++) | |
7834 h->delayed_pic[i] = h->delayed_pic[i+1]; | |
7835 | |
7836 if(out){ | |
7837 *data_size = sizeof(AVFrame); | |
7838 *pict= *(AVFrame*)out; | |
7839 } | |
7840 | |
1168 | 7841 return 0; |
7842 } | |
2967 | 7843 |
1168 | 7844 if(s->flags&CODEC_FLAG_TRUNCATED){ |
4975 | 7845 int next= ff_h264_find_frame_end(h, buf, buf_size); |
2967 | 7846 |
4931
0d1cc37d9430
make some parser parameters const to avoid casting const to non-const
aurel
parents:
4908
diff
changeset
|
7847 if( ff_combine_frame(&s->parse_context, next, (const uint8_t **)&buf, &buf_size) < 0 ) |
1168 | 7848 return buf_size; |
7849 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index); | |
7850 } | |
7851 | |
2227 | 7852 if(h->is_avc && !h->got_avcC) { |
7853 int i, cnt, nalsize; | |
7854 unsigned char *p = avctx->extradata; | |
7855 if(avctx->extradata_size < 7) { | |
7856 av_log(avctx, AV_LOG_ERROR, "avcC too short\n"); | |
7857 return -1; | |
7858 } | |
7859 if(*p != 1) { | |
7860 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p); | |
7861 return -1; | |
7862 } | |
7863 /* sps and pps in the avcC always have length coded with 2 bytes, | |
7864 so put a fake nal_length_size = 2 while parsing them */ | |
7865 h->nal_length_size = 2; | |
7866 // Decode sps from avcC | |
7867 cnt = *(p+5) & 0x1f; // Number of sps | |
7868 p += 6; | |
7869 for (i = 0; i < cnt; i++) { | |
4364 | 7870 nalsize = AV_RB16(p) + 2; |
2867
7c7edddaa835
dont be so picky with .mp4 ... fixes Mr&MrsSmith.mp4
michael
parents:
2861
diff
changeset
|
7871 if(decode_nal_units(h, p, nalsize) < 0) { |
2227 | 7872 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i); |
7873 return -1; | |
7874 } | |
7875 p += nalsize; | |
2967 | 7876 } |
2227 | 7877 // Decode pps from avcC |
7878 cnt = *(p++); // Number of pps | |
7879 for (i = 0; i < cnt; i++) { | |
4364 | 7880 nalsize = AV_RB16(p) + 2; |
2227 | 7881 if(decode_nal_units(h, p, nalsize) != nalsize) { |
7882 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i); | |
7883 return -1; | |
7884 } | |
7885 p += nalsize; | |
2967 | 7886 } |
2227 | 7887 // Now store right nal length size, that will be use to parse all other nals |
7888 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1; | |
7889 // Do not reparse avcC | |
7890 h->got_avcC = 1; | |
7891 } | |
7892 | |
4507
b80c704183e9
always decode extradata when of non-avc stream (like RTSP)
gpoirier
parents:
4391
diff
changeset
|
7893 if(avctx->frame_number==0 && !h->is_avc && s->avctx->extradata_size){ |
2967 | 7894 if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0) |
1168 | 7895 return -1; |
7896 } | |
7897 | |
7898 buf_index=decode_nal_units(h, buf, buf_size); | |
2967 | 7899 if(buf_index < 0) |
1168 | 7900 return -1; |
7901 | |
4620 | 7902 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){ |
7903 av_log(avctx, AV_LOG_ERROR, "no frame!\n"); | |
7904 return -1; | |
7905 } | |
7906 | |
4616
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7907 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){ |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7908 Picture *out = s->current_picture_ptr; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7909 Picture *cur = s->current_picture_ptr; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7910 Picture *prev = h->delayed_output_pic; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7911 int i, pics, cross_idr, out_of_order, out_idx; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7912 |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7913 s->mb_y= 0; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7914 |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7915 s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7916 s->current_picture_ptr->pict_type= s->pict_type; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7917 |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7918 h->prev_frame_num_offset= h->frame_num_offset; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7919 h->prev_frame_num= h->frame_num; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7920 if(s->current_picture_ptr->reference){ |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7921 h->prev_poc_msb= h->poc_msb; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7922 h->prev_poc_lsb= h->poc_lsb; |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7923 } |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7924 if(s->current_picture_ptr->reference) |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7925 execute_ref_pic_marking(h, h->mmco, h->mmco_index); |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7926 |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7927 ff_er_frame_end(s); |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7928 |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7929 MPV_frame_end(s); |
8036b117ae26
support feeding individual NAL units to the decoder instead of just complete frames
michael
parents:
4600
diff
changeset
|
7930 |
2967 | 7931 //FIXME do something with unavailable reference frames |
7932 | |
2561 | 7933 #if 0 //decode order |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7934 *data_size = sizeof(AVFrame); |
2561 | 7935 #else |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
7936 /* Sort B-frames into display order */ |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7937 |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7938 if(h->sps.bitstream_restriction_flag |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7939 && s->avctx->has_b_frames < h->sps.num_reorder_frames){ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7940 s->avctx->has_b_frames = h->sps.num_reorder_frames; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7941 s->low_delay = 0; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7942 } |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
7943 |
3298
9637da0a9c1b
cosmetics (initalize variables where they are used)
michael
parents:
3297
diff
changeset
|
7944 pics = 0; |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
7945 while(h->delayed_pic[pics]) pics++; |
4388 | 7946 |
7947 assert(pics+1 < sizeof(h->delayed_pic) / sizeof(h->delayed_pic[0])); | |
7948 | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
7949 h->delayed_pic[pics++] = cur; |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7950 if(cur->reference == 0) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7951 cur->reference = 1; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7952 |
3298
9637da0a9c1b
cosmetics (initalize variables where they are used)
michael
parents:
3297
diff
changeset
|
7953 cross_idr = 0; |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7954 for(i=0; h->delayed_pic[i]; i++) |
2649
b2e6c5e9240b
fix decoding of conformace streams AUD_MW_E.264 and BA1_FT_C.264 again (there where last correctly decoded with h264.c rev 1.96)
michael
parents:
2648
diff
changeset
|
7955 if(h->delayed_pic[i]->key_frame || h->delayed_pic[i]->poc==0) |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7956 cross_idr = 1; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7957 |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
7958 out = h->delayed_pic[0]; |
3298
9637da0a9c1b
cosmetics (initalize variables where they are used)
michael
parents:
3297
diff
changeset
|
7959 out_idx = 0; |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7960 for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame; i++) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7961 if(h->delayed_pic[i]->poc < out->poc){ |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
7962 out = h->delayed_pic[i]; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
7963 out_idx = i; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
7964 } |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7965 |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7966 out_of_order = !cross_idr && prev && out->poc < prev->poc; |
3125 | 7967 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames) |
7968 { } | |
7969 else if(prev && pics <= s->avctx->has_b_frames) | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7970 out = prev; |
2923 | 7971 else if((out_of_order && pics-1 == s->avctx->has_b_frames && pics < 15) |
2967 | 7972 || (s->low_delay && |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7973 ((!cross_idr && prev && out->poc > prev->poc + 2) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7974 || cur->pict_type == B_TYPE))) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7975 { |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7976 s->low_delay = 0; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7977 s->avctx->has_b_frames++; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7978 out = prev; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7979 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7980 else if(out_of_order) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7981 out = prev; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7982 |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7983 if(out_of_order || pics > s->avctx->has_b_frames){ |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
7984 for(i=out_idx; h->delayed_pic[i]; i++) |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
7985 h->delayed_pic[i] = h->delayed_pic[i+1]; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
7986 } |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
7987 |
3297
9ae9653d706c
decoders should (IMHO) not duplicate frames, the application can do this if it wants
michael
parents:
3291
diff
changeset
|
7988 if(prev == out) |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7989 *data_size = 0; |
2561 | 7990 else |
7991 *data_size = sizeof(AVFrame); | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7992 if(prev && prev != out && prev->reference == 1) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7993 prev->reference = 0; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7994 h->delayed_output_pic = out; |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
7995 #endif |
2409 | 7996 |
2937 | 7997 if(out) |
7998 *pict= *(AVFrame*)out; | |
7999 else | |
8000 av_log(avctx, AV_LOG_DEBUG, "no picture\n"); | |
8001 } | |
8002 | |
8003 assert(pict->data[0] || !*data_size); | |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1698
diff
changeset
|
8004 ff_print_debug_info(s, pict); |
1168 | 8005 //printf("out %d\n", (int)pict->data[0]); |
8006 #if 0 //? | |
8007 | |
8008 /* Return the Picture timestamp as the frame number */ | |
8009 /* we substract 1 because it is added on utils.c */ | |
8010 avctx->frame_number = s->picture_number - 1; | |
8011 #endif | |
8012 return get_consumed_bytes(s, buf_index, buf_size); | |
8013 } | |
8014 #if 0 | |
8015 static inline void fill_mb_avail(H264Context *h){ | |
8016 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
8017 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 8018 |
8019 if(s->mb_y){ | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
8020 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
8021 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
8022 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num; |
1168 | 8023 }else{ |
8024 h->mb_avail[0]= | |
8025 h->mb_avail[1]= | |
8026 h->mb_avail[2]= 0; | |
8027 } | |
8028 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num; | |
8029 h->mb_avail[4]= 1; //FIXME move out | |
8030 h->mb_avail[5]= 0; //FIXME move out | |
8031 } | |
8032 #endif | |
8033 | |
8034 #if 0 //selftest | |
8035 #define COUNT 8000 | |
8036 #define SIZE (COUNT*40) | |
8037 int main(){ | |
8038 int i; | |
8039 uint8_t temp[SIZE]; | |
8040 PutBitContext pb; | |
8041 GetBitContext gb; | |
8042 // int int_temp[10000]; | |
8043 DSPContext dsp; | |
8044 AVCodecContext avctx; | |
2967 | 8045 |
1168 | 8046 dsputil_init(&dsp, &avctx); |
8047 | |
1522
79dddc5cd990
removed the obsolete and unused parameters of init_put_bits
alex
parents:
1453
diff
changeset
|
8048 init_put_bits(&pb, temp, SIZE); |
1168 | 8049 printf("testing unsigned exp golomb\n"); |
8050 for(i=0; i<COUNT; i++){ | |
8051 START_TIMER | |
8052 set_ue_golomb(&pb, i); | |
8053 STOP_TIMER("set_ue_golomb"); | |
8054 } | |
8055 flush_put_bits(&pb); | |
2967 | 8056 |
1168 | 8057 init_get_bits(&gb, temp, 8*SIZE); |
8058 for(i=0; i<COUNT; i++){ | |
8059 int j, s; | |
2967 | 8060 |
1168 | 8061 s= show_bits(&gb, 24); |
2967 | 8062 |
1168 | 8063 START_TIMER |
8064 j= get_ue_golomb(&gb); | |
8065 if(j != i){ | |
8066 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s); | |
8067 // return -1; | |
8068 } | |
8069 STOP_TIMER("get_ue_golomb"); | |
8070 } | |
2967 | 8071 |
8072 | |
1524 | 8073 init_put_bits(&pb, temp, SIZE); |
1168 | 8074 printf("testing signed exp golomb\n"); |
8075 for(i=0; i<COUNT; i++){ | |
8076 START_TIMER | |
8077 set_se_golomb(&pb, i - COUNT/2); | |
8078 STOP_TIMER("set_se_golomb"); | |
8079 } | |
8080 flush_put_bits(&pb); | |
2967 | 8081 |
1168 | 8082 init_get_bits(&gb, temp, 8*SIZE); |
8083 for(i=0; i<COUNT; i++){ | |
8084 int j, s; | |
2967 | 8085 |
1168 | 8086 s= show_bits(&gb, 24); |
2967 | 8087 |
1168 | 8088 START_TIMER |
8089 j= get_se_golomb(&gb); | |
8090 if(j != i - COUNT/2){ | |
8091 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s); | |
8092 // return -1; | |
8093 } | |
8094 STOP_TIMER("get_se_golomb"); | |
8095 } | |
8096 | |
8097 printf("testing 4x4 (I)DCT\n"); | |
2967 | 8098 |
1168 | 8099 DCTELEM block[16]; |
8100 uint8_t src[16], ref[16]; | |
8101 uint64_t error= 0, max_error=0; | |
8102 | |
8103 for(i=0; i<COUNT; i++){ | |
8104 int j; | |
8105 // printf("%d %d %d\n", r1, r2, (r2-r1)*16); | |
8106 for(j=0; j<16; j++){ | |
8107 ref[j]= random()%255; | |
8108 src[j]= random()%255; | |
8109 } | |
8110 | |
8111 h264_diff_dct_c(block, src, ref, 4); | |
2967 | 8112 |
1168 | 8113 //normalize |
8114 for(j=0; j<16; j++){ | |
8115 // printf("%d ", block[j]); | |
8116 block[j]= block[j]*4; | |
8117 if(j&1) block[j]= (block[j]*4 + 2)/5; | |
8118 if(j&4) block[j]= (block[j]*4 + 2)/5; | |
8119 } | |
8120 // printf("\n"); | |
2967 | 8121 |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2255
diff
changeset
|
8122 s->dsp.h264_idct_add(ref, block, 4); |
1168 | 8123 /* for(j=0; j<16; j++){ |
8124 printf("%d ", ref[j]); | |
8125 } | |
8126 printf("\n");*/ | |
2967 | 8127 |
1168 | 8128 for(j=0; j<16; j++){ |
4001 | 8129 int diff= FFABS(src[j] - ref[j]); |
2967 | 8130 |
1168 | 8131 error+= diff*diff; |
8132 max_error= FFMAX(max_error, diff); | |
8133 } | |
8134 } | |
8135 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error ); | |
8136 #if 0 | |
8137 printf("testing quantizer\n"); | |
8138 for(qp=0; qp<52; qp++){ | |
8139 for(i=0; i<16; i++) | |
8140 src1_block[i]= src2_block[i]= random()%255; | |
2967 | 8141 |
1168 | 8142 } |
8143 #endif | |
8144 printf("Testing NAL layer\n"); | |
2967 | 8145 |
1168 | 8146 uint8_t bitstream[COUNT]; |
8147 uint8_t nal[COUNT*2]; | |
8148 H264Context h; | |
8149 memset(&h, 0, sizeof(H264Context)); | |
2967 | 8150 |
1168 | 8151 for(i=0; i<COUNT; i++){ |
8152 int zeros= i; | |
8153 int nal_length; | |
8154 int consumed; | |
8155 int out_length; | |
8156 uint8_t *out; | |
8157 int j; | |
2967 | 8158 |
1168 | 8159 for(j=0; j<COUNT; j++){ |
8160 bitstream[j]= (random() % 255) + 1; | |
8161 } | |
2967 | 8162 |
1168 | 8163 for(j=0; j<zeros; j++){ |
8164 int pos= random() % COUNT; | |
8165 while(bitstream[pos] == 0){ | |
8166 pos++; | |
8167 pos %= COUNT; | |
8168 } | |
8169 bitstream[pos]=0; | |
8170 } | |
2967 | 8171 |
1168 | 8172 START_TIMER |
2967 | 8173 |
1168 | 8174 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2); |
8175 if(nal_length<0){ | |
8176 printf("encoding failed\n"); | |
8177 return -1; | |
8178 } | |
2967 | 8179 |
1168 | 8180 out= decode_nal(&h, nal, &out_length, &consumed, nal_length); |
8181 | |
8182 STOP_TIMER("NAL") | |
2967 | 8183 |
1168 | 8184 if(out_length != COUNT){ |
8185 printf("incorrect length %d %d\n", out_length, COUNT); | |
8186 return -1; | |
8187 } | |
2967 | 8188 |
1168 | 8189 if(consumed != nal_length){ |
8190 printf("incorrect consumed length %d %d\n", nal_length, consumed); | |
8191 return -1; | |
8192 } | |
2967 | 8193 |
1168 | 8194 if(memcmp(bitstream, out, COUNT)){ |
8195 printf("missmatch\n"); | |
8196 return -1; | |
8197 } | |
8198 } | |
2967 | 8199 |
1168 | 8200 printf("Testing RBSP\n"); |
2967 | 8201 |
8202 | |
1168 | 8203 return 0; |
8204 } | |
8205 #endif | |
8206 | |
8207 | |
8208 static int decode_end(AVCodecContext *avctx) | |
8209 { | |
8210 H264Context *h = avctx->priv_data; | |
8211 MpegEncContext *s = &h->s; | |
2967 | 8212 |
2891
4c6eb826e9cb
Just noticed there is a memory leak in h264.c with the usage of rbsp_buffer.
michael
parents:
2883
diff
changeset
|
8213 av_freep(&h->rbsp_buffer); |
1168 | 8214 free_tables(h); //FIXME cleanup init stuff perhaps |
8215 MPV_common_end(s); | |
8216 | |
8217 // memset(h, 0, sizeof(H264Context)); | |
2967 | 8218 |
1168 | 8219 return 0; |
8220 } | |
8221 | |
8222 | |
8223 AVCodec h264_decoder = { | |
8224 "h264", | |
8225 CODEC_TYPE_VIDEO, | |
8226 CODEC_ID_H264, | |
8227 sizeof(H264Context), | |
8228 decode_init, | |
8229 NULL, | |
8230 decode_end, | |
8231 decode_frame, | |
2453 | 8232 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED | CODEC_CAP_DELAY, |
2640 | 8233 .flush= flush_dpb, |
1168 | 8234 }; |
8235 | |
1234 | 8236 #include "svq3.c" |