Mercurial > libavcodec.hg
annotate h264data.h @ 12197:fbf4d5b1b664 libavcodec
Remove FF_MM_SSE2/3 flags for CPUs where this is generally not faster than
regular MMX code. Examples of this are the Core1 CPU. Instead, set a new flag,
FF_MM_SSE2/3SLOW, which can be checked for particular SSE2/3 functions that
have been checked specifically on such CPUs and are actually faster than
their MMX counterparts.
In addition, use this flag to enable particular VP8 and LPC SSE2 functions
that are faster than their MMX counterparts.
Based on a patch by Loren Merritt <lorenm AT u washington edu>.
author | rbultje |
---|---|
date | Mon, 19 Jul 2010 22:38:23 +0000 |
parents | 7dd2a45249a9 |
children |
rev | line source |
---|---|
1168 | 1 /* |
2 * H26L/H264/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:
3316
diff
changeset
|
5 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3316
diff
changeset
|
6 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3316
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:
3316
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:
3316
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:
3316
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:
2967
diff
changeset
|
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
1168 | 20 */ |
21 | |
22 /** | |
11644
7dd2a45249a9
Remove explicit filename from Doxygen @file commands.
diego
parents:
11145
diff
changeset
|
23 * @file |
2967 | 24 * @brief |
1168 | 25 * H264 / AVC / MPEG4 part10 codec data table |
26 * @author Michael Niedermayer <michaelni@gmx.at> | |
27 */ | |
28 | |
7760 | 29 #ifndef AVCODEC_H264DATA_H |
30 #define AVCODEC_H264DATA_H | |
5163 | 31 |
5162 | 32 #include <stdint.h> |
6763 | 33 #include "libavutil/rational.h" |
5162 | 34 #include "mpegvideo.h" |
8207
2f35f9781c31
Move #defines that are mostly used in h264.c out of h264data.h and into h264.h.
diego
parents:
8189
diff
changeset
|
35 #include "h264.h" |
5162 | 36 |
1168 | 37 |
38 static const uint8_t golomb_to_pict_type[5]= | |
6481 | 39 {FF_P_TYPE, FF_B_TYPE, FF_I_TYPE, FF_SP_TYPE, FF_SI_TYPE}; |
1168 | 40 |
10868
13a84faba50d
Move golomb_to_int*cbp tables back to h264_data.h as svq3.c used them.
michael
parents:
10866
diff
changeset
|
41 static const uint8_t golomb_to_intra4x4_cbp[48]={ |
13a84faba50d
Move golomb_to_int*cbp tables back to h264_data.h as svq3.c used them.
michael
parents:
10866
diff
changeset
|
42 47, 31, 15, 0, 23, 27, 29, 30, 7, 11, 13, 14, 39, 43, 45, 46, |
13a84faba50d
Move golomb_to_int*cbp tables back to h264_data.h as svq3.c used them.
michael
parents:
10866
diff
changeset
|
43 16, 3, 5, 10, 12, 19, 21, 26, 28, 35, 37, 42, 44, 1, 2, 4, |
13a84faba50d
Move golomb_to_int*cbp tables back to h264_data.h as svq3.c used them.
michael
parents:
10866
diff
changeset
|
44 8, 17, 18, 20, 24, 6, 9, 22, 25, 32, 33, 34, 36, 40, 38, 41 |
13a84faba50d
Move golomb_to_int*cbp tables back to h264_data.h as svq3.c used them.
michael
parents:
10866
diff
changeset
|
45 }; |
13a84faba50d
Move golomb_to_int*cbp tables back to h264_data.h as svq3.c used them.
michael
parents:
10866
diff
changeset
|
46 |
13a84faba50d
Move golomb_to_int*cbp tables back to h264_data.h as svq3.c used them.
michael
parents:
10866
diff
changeset
|
47 static const uint8_t golomb_to_inter_cbp[48]={ |
13a84faba50d
Move golomb_to_int*cbp tables back to h264_data.h as svq3.c used them.
michael
parents:
10866
diff
changeset
|
48 0, 16, 1, 2, 4, 8, 32, 3, 5, 10, 12, 15, 47, 7, 11, 13, |
13a84faba50d
Move golomb_to_int*cbp tables back to h264_data.h as svq3.c used them.
michael
parents:
10866
diff
changeset
|
49 14, 6, 9, 31, 35, 37, 42, 44, 33, 34, 36, 40, 39, 43, 45, 46, |
13a84faba50d
Move golomb_to_int*cbp tables back to h264_data.h as svq3.c used them.
michael
parents:
10866
diff
changeset
|
50 17, 18, 20, 24, 19, 21, 26, 28, 23, 27, 29, 30, 22, 25, 38, 41 |
13a84faba50d
Move golomb_to_int*cbp tables back to h264_data.h as svq3.c used them.
michael
parents:
10866
diff
changeset
|
51 }; |
13a84faba50d
Move golomb_to_int*cbp tables back to h264_data.h as svq3.c used them.
michael
parents:
10866
diff
changeset
|
52 |
1168 | 53 static const uint8_t zigzag_scan[16]={ |
2967 | 54 0+0*4, 1+0*4, 0+1*4, 0+2*4, |
55 1+1*4, 2+0*4, 3+0*4, 2+1*4, | |
56 1+2*4, 0+3*4, 1+3*4, 2+2*4, | |
57 3+1*4, 3+2*4, 2+3*4, 3+3*4, | |
1168 | 58 }; |
59 | |
60 static const uint8_t field_scan[16]={ | |
2967 | 61 0+0*4, 0+1*4, 1+0*4, 0+2*4, |
1168 | 62 0+3*4, 1+1*4, 1+2*4, 1+3*4, |
2967 | 63 2+0*4, 2+1*4, 2+2*4, 2+3*4, |
1168 | 64 3+0*4, 3+1*4, 3+2*4, 3+3*4, |
65 }; | |
66 | |
67 static const uint8_t luma_dc_zigzag_scan[16]={ | |
68 0*16 + 0*64, 1*16 + 0*64, 2*16 + 0*64, 0*16 + 2*64, | |
69 3*16 + 0*64, 0*16 + 1*64, 1*16 + 1*64, 2*16 + 1*64, | |
70 1*16 + 2*64, 2*16 + 2*64, 3*16 + 2*64, 0*16 + 3*64, | |
71 3*16 + 1*64, 1*16 + 3*64, 2*16 + 3*64, 3*16 + 3*64, | |
72 }; | |
73 | |
74 static const uint8_t luma_dc_field_scan[16]={ | |
2967 | 75 0*16 + 0*64, 2*16 + 0*64, 1*16 + 0*64, 0*16 + 2*64, |
76 2*16 + 2*64, 3*16 + 0*64, 1*16 + 2*64, 3*16 + 2*64, | |
77 0*16 + 1*64, 2*16 + 1*64, 0*16 + 3*64, 2*16 + 3*64, | |
1168 | 78 1*16 + 1*64, 3*16 + 1*64, 1*16 + 3*64, 3*16 + 3*64, |
79 }; | |
80 | |
81 static const uint8_t chroma_dc_scan[4]={ | |
2967 | 82 (0+0*2)*16, (1+0*2)*16, |
1168 | 83 (0+1*2)*16, (1+1*2)*16, //FIXME |
84 }; | |
85 | |
2755 | 86 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)] |
87 static const uint8_t zigzag_scan8x8_cavlc[64]={ | |
88 0+0*8, 1+1*8, 1+2*8, 2+2*8, | |
89 4+1*8, 0+5*8, 3+3*8, 7+0*8, | |
90 3+4*8, 1+7*8, 5+3*8, 6+3*8, | |
91 2+7*8, 6+4*8, 5+6*8, 7+5*8, | |
92 1+0*8, 2+0*8, 0+3*8, 3+1*8, | |
93 3+2*8, 0+6*8, 4+2*8, 6+1*8, | |
94 2+5*8, 2+6*8, 6+2*8, 5+4*8, | |
95 3+7*8, 7+3*8, 4+7*8, 7+6*8, | |
96 0+1*8, 3+0*8, 0+4*8, 4+0*8, | |
97 2+3*8, 1+5*8, 5+1*8, 5+2*8, | |
98 1+6*8, 3+5*8, 7+1*8, 4+5*8, | |
99 4+6*8, 7+4*8, 5+7*8, 6+7*8, | |
100 0+2*8, 2+1*8, 1+3*8, 5+0*8, | |
101 1+4*8, 2+4*8, 6+0*8, 4+3*8, | |
102 0+7*8, 4+4*8, 7+2*8, 3+6*8, | |
103 5+5*8, 6+5*8, 6+6*8, 7+7*8, | |
104 }; | |
105 | |
3316 | 106 static const uint8_t field_scan8x8[64]={ |
107 0+0*8, 0+1*8, 0+2*8, 1+0*8, | |
108 1+1*8, 0+3*8, 0+4*8, 1+2*8, | |
109 2+0*8, 1+3*8, 0+5*8, 0+6*8, | |
110 0+7*8, 1+4*8, 2+1*8, 3+0*8, | |
111 2+2*8, 1+5*8, 1+6*8, 1+7*8, | |
112 2+3*8, 3+1*8, 4+0*8, 3+2*8, | |
113 2+4*8, 2+5*8, 2+6*8, 2+7*8, | |
114 3+3*8, 4+1*8, 5+0*8, 4+2*8, | |
115 3+4*8, 3+5*8, 3+6*8, 3+7*8, | |
116 4+3*8, 5+1*8, 6+0*8, 5+2*8, | |
117 4+4*8, 4+5*8, 4+6*8, 4+7*8, | |
118 5+3*8, 6+1*8, 6+2*8, 5+4*8, | |
119 5+5*8, 5+6*8, 5+7*8, 6+3*8, | |
120 7+0*8, 7+1*8, 6+4*8, 6+5*8, | |
121 6+6*8, 6+7*8, 7+2*8, 7+3*8, | |
122 7+4*8, 7+5*8, 7+6*8, 7+7*8, | |
123 }; | |
124 | |
125 static const uint8_t field_scan8x8_cavlc[64]={ | |
126 0+0*8, 1+1*8, 2+0*8, 0+7*8, | |
127 2+2*8, 2+3*8, 2+4*8, 3+3*8, | |
128 3+4*8, 4+3*8, 4+4*8, 5+3*8, | |
129 5+5*8, 7+0*8, 6+6*8, 7+4*8, | |
130 0+1*8, 0+3*8, 1+3*8, 1+4*8, | |
131 1+5*8, 3+1*8, 2+5*8, 4+1*8, | |
132 3+5*8, 5+1*8, 4+5*8, 6+1*8, | |
133 5+6*8, 7+1*8, 6+7*8, 7+5*8, | |
134 0+2*8, 0+4*8, 0+5*8, 2+1*8, | |
135 1+6*8, 4+0*8, 2+6*8, 5+0*8, | |
136 3+6*8, 6+0*8, 4+6*8, 6+2*8, | |
137 5+7*8, 6+4*8, 7+2*8, 7+6*8, | |
138 1+0*8, 1+2*8, 0+6*8, 3+0*8, | |
139 1+7*8, 3+2*8, 2+7*8, 4+2*8, | |
140 3+7*8, 5+2*8, 4+7*8, 5+4*8, | |
141 6+3*8, 6+5*8, 7+3*8, 7+7*8, | |
142 }; | |
143 | |
1168 | 144 typedef struct IMbInfo{ |
145 uint16_t type; | |
146 uint8_t pred_mode; | |
147 uint8_t cbp; | |
148 } IMbInfo; | |
149 | |
150 static const IMbInfo i_mb_type_info[26]={ | |
151 {MB_TYPE_INTRA4x4 , -1, -1}, | |
152 {MB_TYPE_INTRA16x16, 2, 0}, | |
153 {MB_TYPE_INTRA16x16, 1, 0}, | |
154 {MB_TYPE_INTRA16x16, 0, 0}, | |
155 {MB_TYPE_INTRA16x16, 3, 0}, | |
156 {MB_TYPE_INTRA16x16, 2, 16}, | |
157 {MB_TYPE_INTRA16x16, 1, 16}, | |
158 {MB_TYPE_INTRA16x16, 0, 16}, | |
159 {MB_TYPE_INTRA16x16, 3, 16}, | |
160 {MB_TYPE_INTRA16x16, 2, 32}, | |
161 {MB_TYPE_INTRA16x16, 1, 32}, | |
162 {MB_TYPE_INTRA16x16, 0, 32}, | |
163 {MB_TYPE_INTRA16x16, 3, 32}, | |
164 {MB_TYPE_INTRA16x16, 2, 15+0}, | |
165 {MB_TYPE_INTRA16x16, 1, 15+0}, | |
166 {MB_TYPE_INTRA16x16, 0, 15+0}, | |
167 {MB_TYPE_INTRA16x16, 3, 15+0}, | |
168 {MB_TYPE_INTRA16x16, 2, 15+16}, | |
169 {MB_TYPE_INTRA16x16, 1, 15+16}, | |
170 {MB_TYPE_INTRA16x16, 0, 15+16}, | |
171 {MB_TYPE_INTRA16x16, 3, 15+16}, | |
172 {MB_TYPE_INTRA16x16, 2, 15+32}, | |
173 {MB_TYPE_INTRA16x16, 1, 15+32}, | |
174 {MB_TYPE_INTRA16x16, 0, 15+32}, | |
175 {MB_TYPE_INTRA16x16, 3, 15+32}, | |
176 {MB_TYPE_INTRA_PCM , -1, -1}, | |
177 }; | |
178 | |
179 typedef struct PMbInfo{ | |
180 uint16_t type; | |
181 uint8_t partition_count; | |
182 } PMbInfo; | |
183 | |
184 static const PMbInfo p_mb_type_info[5]={ | |
185 {MB_TYPE_16x16|MB_TYPE_P0L0 , 1}, | |
186 {MB_TYPE_16x8 |MB_TYPE_P0L0|MB_TYPE_P1L0, 2}, | |
187 {MB_TYPE_8x16 |MB_TYPE_P0L0|MB_TYPE_P1L0, 2}, | |
2535 | 188 {MB_TYPE_8x8 |MB_TYPE_P0L0|MB_TYPE_P1L0, 4}, |
189 {MB_TYPE_8x8 |MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_REF0, 4}, | |
1168 | 190 }; |
191 | |
192 static const PMbInfo p_sub_mb_type_info[4]={ | |
193 {MB_TYPE_16x16|MB_TYPE_P0L0 , 1}, | |
194 {MB_TYPE_16x8 |MB_TYPE_P0L0 , 2}, | |
195 {MB_TYPE_8x16 |MB_TYPE_P0L0 , 2}, | |
196 {MB_TYPE_8x8 |MB_TYPE_P0L0 , 4}, | |
197 }; | |
198 | |
199 static const PMbInfo b_mb_type_info[23]={ | |
11145
8805efb4c3fd
Move setting MB_TYPE_L0L1 for direct MBs up, this is simpler.
michael
parents:
10869
diff
changeset
|
200 {MB_TYPE_DIRECT2|MB_TYPE_L0L1 , 1, }, |
1168 | 201 {MB_TYPE_16x16|MB_TYPE_P0L0 , 1, }, |
202 {MB_TYPE_16x16 |MB_TYPE_P0L1 , 1, }, | |
203 {MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1 , 1, }, | |
204 {MB_TYPE_16x8 |MB_TYPE_P0L0 |MB_TYPE_P1L0 , 2, }, | |
205 {MB_TYPE_8x16 |MB_TYPE_P0L0 |MB_TYPE_P1L0 , 2, }, | |
206 {MB_TYPE_16x8 |MB_TYPE_P0L1 |MB_TYPE_P1L1, 2, }, | |
207 {MB_TYPE_8x16 |MB_TYPE_P0L1 |MB_TYPE_P1L1, 2, }, | |
208 {MB_TYPE_16x8 |MB_TYPE_P0L0 |MB_TYPE_P1L1, 2, }, | |
209 {MB_TYPE_8x16 |MB_TYPE_P0L0 |MB_TYPE_P1L1, 2, }, | |
210 {MB_TYPE_16x8 |MB_TYPE_P0L1|MB_TYPE_P1L0 , 2, }, | |
211 {MB_TYPE_8x16 |MB_TYPE_P0L1|MB_TYPE_P1L0 , 2, }, | |
212 {MB_TYPE_16x8 |MB_TYPE_P0L0 |MB_TYPE_P1L0|MB_TYPE_P1L1, 2, }, | |
213 {MB_TYPE_8x16 |MB_TYPE_P0L0 |MB_TYPE_P1L0|MB_TYPE_P1L1, 2, }, | |
214 {MB_TYPE_16x8 |MB_TYPE_P0L1|MB_TYPE_P1L0|MB_TYPE_P1L1, 2, }, | |
215 {MB_TYPE_8x16 |MB_TYPE_P0L1|MB_TYPE_P1L0|MB_TYPE_P1L1, 2, }, | |
216 {MB_TYPE_16x8 |MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_P1L0 , 2, }, | |
217 {MB_TYPE_8x16 |MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_P1L0 , 2, }, | |
218 {MB_TYPE_16x8 |MB_TYPE_P0L0|MB_TYPE_P0L1 |MB_TYPE_P1L1, 2, }, | |
219 {MB_TYPE_8x16 |MB_TYPE_P0L0|MB_TYPE_P0L1 |MB_TYPE_P1L1, 2, }, | |
220 {MB_TYPE_16x8 |MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_P1L0|MB_TYPE_P1L1, 2, }, | |
221 {MB_TYPE_8x16 |MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_P1L0|MB_TYPE_P1L1, 2, }, | |
2535 | 222 {MB_TYPE_8x8 |MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_P1L0|MB_TYPE_P1L1, 4, }, |
1168 | 223 }; |
224 | |
225 static const PMbInfo b_sub_mb_type_info[13]={ | |
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:
1168
diff
changeset
|
226 {MB_TYPE_DIRECT2 , 1, }, |
1168 | 227 {MB_TYPE_16x16|MB_TYPE_P0L0 , 1, }, |
228 {MB_TYPE_16x16 |MB_TYPE_P0L1 , 1, }, | |
229 {MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1 , 1, }, | |
230 {MB_TYPE_16x8 |MB_TYPE_P0L0 |MB_TYPE_P1L0 , 2, }, | |
231 {MB_TYPE_8x16 |MB_TYPE_P0L0 |MB_TYPE_P1L0 , 2, }, | |
232 {MB_TYPE_16x8 |MB_TYPE_P0L1 |MB_TYPE_P1L1, 2, }, | |
233 {MB_TYPE_8x16 |MB_TYPE_P0L1 |MB_TYPE_P1L1, 2, }, | |
234 {MB_TYPE_16x8 |MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_P1L0|MB_TYPE_P1L1, 2, }, | |
235 {MB_TYPE_8x16 |MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_P1L0|MB_TYPE_P1L1, 2, }, | |
236 {MB_TYPE_8x8 |MB_TYPE_P0L0 |MB_TYPE_P1L0 , 4, }, | |
237 {MB_TYPE_8x8 |MB_TYPE_P0L1 |MB_TYPE_P1L1, 4, }, | |
238 {MB_TYPE_8x8 |MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_P1L0|MB_TYPE_P1L1, 4, }, | |
239 }; | |
240 | |
5578 | 241 static const uint8_t dequant4_coeff_init[6][3]={ |
2919 | 242 {10,13,16}, |
243 {11,14,18}, | |
244 {13,16,20}, | |
245 {14,18,23}, | |
246 {16,20,25}, | |
247 {18,23,29}, | |
1168 | 248 }; |
249 | |
5578 | 250 static const uint8_t dequant8_coeff_init_scan[16] = { |
2755 | 251 0,3,4,3, 3,1,5,1, 4,5,2,5, 3,1,5,1 |
252 }; | |
5578 | 253 static const uint8_t dequant8_coeff_init[6][6]={ |
2755 | 254 {20,18,32,19,25,24}, |
255 {22,19,35,21,28,26}, | |
256 {26,23,42,24,33,31}, | |
257 {28,25,45,26,35,33}, | |
258 {32,28,51,30,40,38}, | |
259 {36,32,58,34,46,43}, | |
260 }; | |
261 | |
7760 | 262 #endif /* AVCODEC_H264DATA_H */ |