Mercurial > libavcodec.hg
annotate dsputil.c @ 6438:e1dd408a7864 libavcodec
move ff_init_scantable() into dsputil
author | aurel |
---|---|
date | Tue, 04 Mar 2008 10:26:56 +0000 |
parents | 5154ab444372 |
children | 2b553c57ec51 |
rev | line source |
---|---|
0 | 1 /* |
2 * DSP utils | |
429 | 3 * Copyright (c) 2000, 2001 Fabrice Bellard. |
1739
07a484280a82
copyright year update of the files i touched and remembered, things look annoyingly unmaintained otherwise
michael
parents:
1729
diff
changeset
|
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> |
0 | 5 * |
5214 | 6 * gmc & q-pel & 32/64 bit based MC by Michael Niedermayer <michaelni@gmx.at> |
7 * | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
8 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
9 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
10 * FFmpeg is free software; you can redistribute it and/or |
429 | 11 * modify it under the terms of the GNU Lesser General Public |
12 * 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:
3807
diff
changeset
|
13 * version 2.1 of the License, or (at your option) any later version. |
0 | 14 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
15 * FFmpeg is distributed in the hope that it will be useful, |
0 | 16 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
429 | 17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
18 * Lesser General Public License for more details. | |
0 | 19 * |
429 | 20 * 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:
3807
diff
changeset
|
21 * 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
|
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
0 | 23 */ |
2967 | 24 |
1106 | 25 /** |
26 * @file dsputil.c | |
27 * DSP utils | |
28 */ | |
2967 | 29 |
0 | 30 #include "avcodec.h" |
31 #include "dsputil.h" | |
936 | 32 #include "mpegvideo.h" |
1092 | 33 #include "simple_idct.h" |
1557 | 34 #include "faandct.h" |
6407 | 35 #include "faanidct.h" |
5277
7b3fcb7c61ce
Avoid linking with h263.c functions when the relevant codecs
aurel
parents:
5256
diff
changeset
|
36 #include "h263.h" |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
37 #include "snow.h" |
676 | 38 |
2522
e25782262d7d
kill warnings patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2448
diff
changeset
|
39 /* snow.c */ |
e25782262d7d
kill warnings patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2448
diff
changeset
|
40 void ff_spatial_dwt(int *buffer, int width, int height, int stride, int type, int decomposition_count); |
e25782262d7d
kill warnings patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2448
diff
changeset
|
41 |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
42 /* vorbis.c */ |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
43 void vorbis_inverse_coupling(float *mag, float *ang, int blocksize); |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
44 |
5737 | 45 /* flacenc.c */ |
46 void ff_flac_compute_autocorr(const int32_t *data, int len, int lag, double *autoc); | |
47 | |
6384 | 48 /* pngdec.c */ |
49 void ff_add_png_paeth_prediction(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp); | |
50 | |
4176 | 51 uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP] = {0, }; |
4179 | 52 uint32_t ff_squareTbl[512] = {0, }; |
0 | 53 |
6387 | 54 // 0x7f7f7f7f or 0x7f7f7f7f7f7f7f7f or whatever, depending on the cpu's native arithmetic size |
55 #define pb_7f (~0UL/255 * 0x7f) | |
56 #define pb_80 (~0UL/255 * 0x80) | |
6385 | 57 |
1064 | 58 const uint8_t ff_zigzag_direct[64] = { |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
59 0, 1, 8, 16, 9, 2, 3, 10, |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
60 17, 24, 32, 25, 18, 11, 4, 5, |
34 | 61 12, 19, 26, 33, 40, 48, 41, 34, |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
62 27, 20, 13, 6, 7, 14, 21, 28, |
34 | 63 35, 42, 49, 56, 57, 50, 43, 36, |
64 29, 22, 15, 23, 30, 37, 44, 51, | |
65 58, 59, 52, 45, 38, 31, 39, 46, | |
66 53, 60, 61, 54, 47, 55, 62, 63 | |
67 }; | |
68 | |
1567 | 69 /* Specific zigzag scan for 248 idct. NOTE that unlike the |
70 specification, we interleave the fields */ | |
71 const uint8_t ff_zigzag248_direct[64] = { | |
72 0, 8, 1, 9, 16, 24, 2, 10, | |
73 17, 25, 32, 40, 48, 56, 33, 41, | |
74 18, 26, 3, 11, 4, 12, 19, 27, | |
75 34, 42, 49, 57, 50, 58, 35, 43, | |
76 20, 28, 5, 13, 6, 14, 21, 29, | |
77 36, 44, 51, 59, 52, 60, 37, 45, | |
78 22, 30, 7, 15, 23, 31, 38, 46, | |
79 53, 61, 54, 62, 39, 47, 55, 63, | |
80 }; | |
81 | |
220 | 82 /* not permutated inverse zigzag_direct + 1 for MMX quantizer */ |
4197 | 83 DECLARE_ALIGNED_8(uint16_t, inv_zigzag_direct16[64]) = {0, }; |
220 | 84 |
1064 | 85 const uint8_t ff_alternate_horizontal_scan[64] = { |
2967 | 86 0, 1, 2, 3, 8, 9, 16, 17, |
34 | 87 10, 11, 4, 5, 6, 7, 15, 14, |
2967 | 88 13, 12, 19, 18, 24, 25, 32, 33, |
34 | 89 26, 27, 20, 21, 22, 23, 28, 29, |
2967 | 90 30, 31, 34, 35, 40, 41, 48, 49, |
34 | 91 42, 43, 36, 37, 38, 39, 44, 45, |
2967 | 92 46, 47, 50, 51, 56, 57, 58, 59, |
34 | 93 52, 53, 54, 55, 60, 61, 62, 63, |
94 }; | |
95 | |
1064 | 96 const uint8_t ff_alternate_vertical_scan[64] = { |
2967 | 97 0, 8, 16, 24, 1, 9, 2, 10, |
34 | 98 17, 25, 32, 40, 48, 56, 57, 49, |
2967 | 99 41, 33, 26, 18, 3, 11, 4, 12, |
34 | 100 19, 27, 34, 42, 50, 58, 35, 43, |
2967 | 101 51, 59, 20, 28, 5, 13, 6, 14, |
34 | 102 21, 29, 36, 44, 52, 60, 37, 45, |
2967 | 103 53, 61, 22, 30, 7, 15, 23, 31, |
34 | 104 38, 46, 54, 62, 39, 47, 55, 63, |
105 }; | |
106 | |
220 | 107 /* a*inverse[b]>>32 == a/b for all 0<=a<=65536 && 2<=b<=255 */ |
4174 | 108 const uint32_t ff_inverse[256]={ |
2967 | 109 0, 4294967295U,2147483648U,1431655766, 1073741824, 858993460, 715827883, 613566757, |
110 536870912, 477218589, 429496730, 390451573, 357913942, 330382100, 306783379, 286331154, | |
111 268435456, 252645136, 238609295, 226050911, 214748365, 204522253, 195225787, 186737709, | |
112 178956971, 171798692, 165191050, 159072863, 153391690, 148102321, 143165577, 138547333, | |
113 134217728, 130150525, 126322568, 122713352, 119304648, 116080198, 113025456, 110127367, | |
114 107374183, 104755300, 102261127, 99882961, 97612894, 95443718, 93368855, 91382283, | |
115 89478486, 87652394, 85899346, 84215046, 82595525, 81037119, 79536432, 78090315, | |
116 76695845, 75350304, 74051161, 72796056, 71582789, 70409300, 69273667, 68174085, | |
117 67108864, 66076420, 65075263, 64103990, 63161284, 62245903, 61356676, 60492498, | |
118 59652324, 58835169, 58040099, 57266231, 56512728, 55778797, 55063684, 54366675, | |
119 53687092, 53024288, 52377650, 51746594, 51130564, 50529028, 49941481, 49367441, | |
120 48806447, 48258060, 47721859, 47197443, 46684428, 46182445, 45691142, 45210183, | |
121 44739243, 44278014, 43826197, 43383509, 42949673, 42524429, 42107523, 41698712, | |
122 41297763, 40904451, 40518560, 40139882, 39768216, 39403370, 39045158, 38693400, | |
123 38347923, 38008561, 37675152, 37347542, 37025581, 36709123, 36398028, 36092163, | |
124 35791395, 35495598, 35204650, 34918434, 34636834, 34359739, 34087043, 33818641, | |
125 33554432, 33294321, 33038210, 32786010, 32537632, 32292988, 32051995, 31814573, | |
126 31580642, 31350127, 31122952, 30899046, 30678338, 30460761, 30246249, 30034737, | |
127 29826162, 29620465, 29417585, 29217465, 29020050, 28825284, 28633116, 28443493, | |
128 28256364, 28071682, 27889399, 27709467, 27531842, 27356480, 27183338, 27012373, | |
129 26843546, 26676816, 26512144, 26349493, 26188825, 26030105, 25873297, 25718368, | |
130 25565282, 25414008, 25264514, 25116768, 24970741, 24826401, 24683721, 24542671, | |
131 24403224, 24265352, 24129030, 23994231, 23860930, 23729102, 23598722, 23469767, | |
132 23342214, 23216040, 23091223, 22967740, 22845571, 22724695, 22605092, 22486740, | |
133 22369622, 22253717, 22139007, 22025474, 21913099, 21801865, 21691755, 21582751, | |
134 21474837, 21367997, 21262215, 21157475, 21053762, 20951060, 20849356, 20748635, | |
135 20648882, 20550083, 20452226, 20355296, 20259280, 20164166, 20069941, 19976593, | |
136 19884108, 19792477, 19701685, 19611723, 19522579, 19434242, 19346700, 19259944, | |
137 19173962, 19088744, 19004281, 18920561, 18837576, 18755316, 18673771, 18592933, | |
138 18512791, 18433337, 18354562, 18276457, 18199014, 18122225, 18046082, 17970575, | |
139 17895698, 17821442, 17747799, 17674763, 17602325, 17530479, 17459217, 17388532, | |
220 | 140 17318417, 17248865, 17179870, 17111424, 17043522, 16976156, 16909321, 16843010, |
141 }; | |
142 | |
1092 | 143 /* Input permutation for the simple_idct_mmx */ |
144 static const uint8_t simple_mmx_permutation[64]={ | |
2979 | 145 0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, |
146 0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, | |
147 0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, | |
148 0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, | |
149 0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, | |
150 0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, | |
151 0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, | |
152 0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F, | |
1092 | 153 }; |
154 | |
6438 | 155 void ff_init_scantable(uint8_t *permutation, ScanTable *st, const uint8_t *src_scantable){ |
156 int i; | |
157 int end; | |
158 | |
159 st->scantable= src_scantable; | |
160 | |
161 for(i=0; i<64; i++){ | |
162 int j; | |
163 j = src_scantable[i]; | |
164 st->permutated[i] = permutation[j]; | |
165 #ifdef ARCH_POWERPC | |
166 st->inverse[j] = i; | |
167 #endif | |
168 } | |
169 | |
170 end=-1; | |
171 for(i=0; i<64; i++){ | |
172 int j; | |
173 j = st->permutated[i]; | |
174 if(j>end) end=j; | |
175 st->raster_end[i]= end; | |
176 } | |
177 } | |
178 | |
1064 | 179 static int pix_sum_c(uint8_t * pix, int line_size) |
612 | 180 { |
181 int s, i, j; | |
182 | |
183 s = 0; | |
184 for (i = 0; i < 16; i++) { | |
2979 | 185 for (j = 0; j < 16; j += 8) { |
186 s += pix[0]; | |
187 s += pix[1]; | |
188 s += pix[2]; | |
189 s += pix[3]; | |
190 s += pix[4]; | |
191 s += pix[5]; | |
192 s += pix[6]; | |
193 s += pix[7]; | |
194 pix += 8; | |
195 } | |
196 pix += line_size - 16; | |
612 | 197 } |
198 return s; | |
199 } | |
200 | |
1064 | 201 static int pix_norm1_c(uint8_t * pix, int line_size) |
612 | 202 { |
203 int s, i, j; | |
4179 | 204 uint32_t *sq = ff_squareTbl + 256; |
612 | 205 |
206 s = 0; | |
207 for (i = 0; i < 16; i++) { | |
2979 | 208 for (j = 0; j < 16; j += 8) { |
997
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
209 #if 0 |
2979 | 210 s += sq[pix[0]]; |
211 s += sq[pix[1]]; | |
212 s += sq[pix[2]]; | |
213 s += sq[pix[3]]; | |
214 s += sq[pix[4]]; | |
215 s += sq[pix[5]]; | |
216 s += sq[pix[6]]; | |
217 s += sq[pix[7]]; | |
997
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
218 #else |
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
219 #if LONG_MAX > 2147483647 |
2979 | 220 register uint64_t x=*(uint64_t*)pix; |
221 s += sq[x&0xff]; | |
222 s += sq[(x>>8)&0xff]; | |
223 s += sq[(x>>16)&0xff]; | |
224 s += sq[(x>>24)&0xff]; | |
997
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
225 s += sq[(x>>32)&0xff]; |
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
226 s += sq[(x>>40)&0xff]; |
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
227 s += sq[(x>>48)&0xff]; |
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
228 s += sq[(x>>56)&0xff]; |
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
229 #else |
2979 | 230 register uint32_t x=*(uint32_t*)pix; |
231 s += sq[x&0xff]; | |
232 s += sq[(x>>8)&0xff]; | |
233 s += sq[(x>>16)&0xff]; | |
234 s += sq[(x>>24)&0xff]; | |
997
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
235 x=*(uint32_t*)(pix+4); |
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
236 s += sq[x&0xff]; |
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
237 s += sq[(x>>8)&0xff]; |
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
238 s += sq[(x>>16)&0xff]; |
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
239 s += sq[(x>>24)&0xff]; |
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
240 #endif |
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
241 #endif |
2979 | 242 pix += 8; |
243 } | |
244 pix += line_size - 16; | |
612 | 245 } |
246 return s; | |
247 } | |
248 | |
6241 | 249 static void bswap_buf(uint32_t *dst, const uint32_t *src, int w){ |
1273 | 250 int i; |
2967 | 251 |
1273 | 252 for(i=0; i+8<=w; i+=8){ |
253 dst[i+0]= bswap_32(src[i+0]); | |
254 dst[i+1]= bswap_32(src[i+1]); | |
255 dst[i+2]= bswap_32(src[i+2]); | |
256 dst[i+3]= bswap_32(src[i+3]); | |
257 dst[i+4]= bswap_32(src[i+4]); | |
258 dst[i+5]= bswap_32(src[i+5]); | |
259 dst[i+6]= bswap_32(src[i+6]); | |
260 dst[i+7]= bswap_32(src[i+7]); | |
261 } | |
262 for(;i<w; i++){ | |
263 dst[i+0]= bswap_32(src[i+0]); | |
264 } | |
265 } | |
612 | 266 |
2184 | 267 static int sse4_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
268 { | |
269 int s, i; | |
4179 | 270 uint32_t *sq = ff_squareTbl + 256; |
2184 | 271 |
272 s = 0; | |
273 for (i = 0; i < h; i++) { | |
274 s += sq[pix1[0] - pix2[0]]; | |
275 s += sq[pix1[1] - pix2[1]]; | |
276 s += sq[pix1[2] - pix2[2]]; | |
277 s += sq[pix1[3] - pix2[3]]; | |
278 pix1 += line_size; | |
279 pix2 += line_size; | |
280 } | |
281 return s; | |
282 } | |
283 | |
1708 | 284 static int sse8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
936 | 285 { |
286 int s, i; | |
4179 | 287 uint32_t *sq = ff_squareTbl + 256; |
936 | 288 |
289 s = 0; | |
1708 | 290 for (i = 0; i < h; i++) { |
936 | 291 s += sq[pix1[0] - pix2[0]]; |
292 s += sq[pix1[1] - pix2[1]]; | |
293 s += sq[pix1[2] - pix2[2]]; | |
294 s += sq[pix1[3] - pix2[3]]; | |
295 s += sq[pix1[4] - pix2[4]]; | |
296 s += sq[pix1[5] - pix2[5]]; | |
297 s += sq[pix1[6] - pix2[6]]; | |
298 s += sq[pix1[7] - pix2[7]]; | |
299 pix1 += line_size; | |
300 pix2 += line_size; | |
301 } | |
302 return s; | |
303 } | |
304 | |
1708 | 305 static int sse16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
884 | 306 { |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
307 int s, i; |
4179 | 308 uint32_t *sq = ff_squareTbl + 256; |
884 | 309 |
310 s = 0; | |
1708 | 311 for (i = 0; i < h; i++) { |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
312 s += sq[pix1[ 0] - pix2[ 0]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
313 s += sq[pix1[ 1] - pix2[ 1]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
314 s += sq[pix1[ 2] - pix2[ 2]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
315 s += sq[pix1[ 3] - pix2[ 3]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
316 s += sq[pix1[ 4] - pix2[ 4]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
317 s += sq[pix1[ 5] - pix2[ 5]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
318 s += sq[pix1[ 6] - pix2[ 6]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
319 s += sq[pix1[ 7] - pix2[ 7]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
320 s += sq[pix1[ 8] - pix2[ 8]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
321 s += sq[pix1[ 9] - pix2[ 9]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
322 s += sq[pix1[10] - pix2[10]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
323 s += sq[pix1[11] - pix2[11]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
324 s += sq[pix1[12] - pix2[12]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
325 s += sq[pix1[13] - pix2[13]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
326 s += sq[pix1[14] - pix2[14]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
327 s += sq[pix1[15] - pix2[15]]; |
997
4dfe15ae0078
sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents:
996
diff
changeset
|
328 |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
329 pix1 += line_size; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
330 pix2 += line_size; |
884 | 331 } |
332 return s; | |
333 } | |
334 | |
2184 | 335 |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
336 #ifdef CONFIG_SNOW_ENCODER //dwt is in snow.c |
2184 | 337 static inline int w_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int w, int h, int type){ |
338 int s, i, j; | |
339 const int dec_count= w==8 ? 3 : 4; | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
340 int tmp[32*32]; |
2184 | 341 int level, ori; |
2967 | 342 static const int scale[2][2][4][4]={ |
2184 | 343 { |
344 { | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
345 // 9/7 8x8 dec=3 |
2184 | 346 {268, 239, 239, 213}, |
347 { 0, 224, 224, 152}, | |
348 { 0, 135, 135, 110}, | |
349 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
350 // 9/7 16x16 or 32x32 dec=4 |
2184 | 351 {344, 310, 310, 280}, |
352 { 0, 320, 320, 228}, | |
353 { 0, 175, 175, 136}, | |
354 { 0, 129, 129, 102}, | |
355 } | |
356 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
357 { |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
358 // 5/3 8x8 dec=3 |
2184 | 359 {275, 245, 245, 218}, |
360 { 0, 230, 230, 156}, | |
361 { 0, 138, 138, 113}, | |
362 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
363 // 5/3 16x16 or 32x32 dec=4 |
2184 | 364 {352, 317, 317, 286}, |
365 { 0, 328, 328, 233}, | |
366 { 0, 180, 180, 140}, | |
367 { 0, 132, 132, 105}, | |
368 } | |
369 } | |
370 }; | |
371 | |
372 for (i = 0; i < h; i++) { | |
373 for (j = 0; j < w; j+=4) { | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
374 tmp[32*i+j+0] = (pix1[j+0] - pix2[j+0])<<4; |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
375 tmp[32*i+j+1] = (pix1[j+1] - pix2[j+1])<<4; |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
376 tmp[32*i+j+2] = (pix1[j+2] - pix2[j+2])<<4; |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
377 tmp[32*i+j+3] = (pix1[j+3] - pix2[j+3])<<4; |
2184 | 378 } |
379 pix1 += line_size; | |
380 pix2 += line_size; | |
381 } | |
2639 | 382 |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
383 ff_spatial_dwt(tmp, w, h, 32, type, dec_count); |
2184 | 384 |
385 s=0; | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
386 assert(w==h); |
2184 | 387 for(level=0; level<dec_count; level++){ |
388 for(ori= level ? 1 : 0; ori<4; ori++){ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
389 int size= w>>(dec_count-level); |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
390 int sx= (ori&1) ? size : 0; |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
391 int stride= 32<<(dec_count-level); |
2184 | 392 int sy= (ori&2) ? stride>>1 : 0; |
2967 | 393 |
2184 | 394 for(i=0; i<size; i++){ |
395 for(j=0; j<size; j++){ | |
396 int v= tmp[sx + sy + i*stride + j] * scale[type][dec_count-3][level][ori]; | |
4001 | 397 s += FFABS(v); |
2184 | 398 } |
399 } | |
400 } | |
401 } | |
2967 | 402 assert(s>=0); |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
403 return s>>9; |
2184 | 404 } |
405 | |
406 static int w53_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
407 return w_c(v, pix1, pix2, line_size, 8, h, 1); | |
408 } | |
409 | |
410 static int w97_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
411 return w_c(v, pix1, pix2, line_size, 8, h, 0); | |
412 } | |
413 | |
414 static int w53_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
415 return w_c(v, pix1, pix2, line_size, 16, h, 1); | |
416 } | |
417 | |
418 static int w97_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
419 return w_c(v, pix1, pix2, line_size, 16, h, 0); | |
420 } | |
421 | |
4197 | 422 int w53_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
423 return w_c(v, pix1, pix2, line_size, 32, h, 1); |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
424 } |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
425 |
4197 | 426 int w97_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
427 return w_c(v, pix1, pix2, line_size, 32, h, 0); |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
428 } |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
429 #endif |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
430 |
6437 | 431 /* draw the edges of width 'w' of an image of size width, height */ |
432 //FIXME check that this is ok for mpeg4 interlaced | |
433 static void draw_edges_c(uint8_t *buf, int wrap, int width, int height, int w) | |
434 { | |
435 uint8_t *ptr, *last_line; | |
436 int i; | |
437 | |
438 last_line = buf + (height - 1) * wrap; | |
439 for(i=0;i<w;i++) { | |
440 /* top and bottom */ | |
441 memcpy(buf - (i + 1) * wrap, buf, width); | |
442 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
443 } | |
444 /* left and right */ | |
445 ptr = buf; | |
446 for(i=0;i<height;i++) { | |
447 memset(ptr - w, ptr[0], w); | |
448 memset(ptr + width, ptr[width-1], w); | |
449 ptr += wrap; | |
450 } | |
451 /* corners */ | |
452 for(i=0;i<w;i++) { | |
453 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
454 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
455 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
456 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
457 } | |
458 } | |
459 | |
1064 | 460 static void get_pixels_c(DCTELEM *restrict block, const uint8_t *pixels, int line_size) |
0 | 461 { |
462 int i; | |
463 | |
464 /* read the pixels */ | |
465 for(i=0;i<8;i++) { | |
516 | 466 block[0] = pixels[0]; |
467 block[1] = pixels[1]; | |
468 block[2] = pixels[2]; | |
469 block[3] = pixels[3]; | |
470 block[4] = pixels[4]; | |
471 block[5] = pixels[5]; | |
472 block[6] = pixels[6]; | |
473 block[7] = pixels[7]; | |
474 pixels += line_size; | |
475 block += 8; | |
0 | 476 } |
477 } | |
478 | |
1064 | 479 static void diff_pixels_c(DCTELEM *restrict block, const uint8_t *s1, |
2979 | 480 const uint8_t *s2, int stride){ |
324 | 481 int i; |
482 | |
483 /* read the pixels */ | |
484 for(i=0;i<8;i++) { | |
516 | 485 block[0] = s1[0] - s2[0]; |
486 block[1] = s1[1] - s2[1]; | |
487 block[2] = s1[2] - s2[2]; | |
488 block[3] = s1[3] - s2[3]; | |
489 block[4] = s1[4] - s2[4]; | |
490 block[5] = s1[5] - s2[5]; | |
491 block[6] = s1[6] - s2[6]; | |
492 block[7] = s1[7] - s2[7]; | |
324 | 493 s1 += stride; |
494 s2 += stride; | |
516 | 495 block += 8; |
324 | 496 } |
497 } | |
498 | |
499 | |
1064 | 500 static void put_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 501 int line_size) |
0 | 502 { |
503 int i; | |
4176 | 504 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 505 |
0 | 506 /* read the pixels */ |
507 for(i=0;i<8;i++) { | |
516 | 508 pixels[0] = cm[block[0]]; |
509 pixels[1] = cm[block[1]]; | |
510 pixels[2] = cm[block[2]]; | |
511 pixels[3] = cm[block[3]]; | |
512 pixels[4] = cm[block[4]]; | |
513 pixels[5] = cm[block[5]]; | |
514 pixels[6] = cm[block[6]]; | |
515 pixels[7] = cm[block[7]]; | |
516 | |
517 pixels += line_size; | |
518 block += 8; | |
0 | 519 } |
520 } | |
521 | |
2256 | 522 static void put_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 523 int line_size) |
2256 | 524 { |
525 int i; | |
4176 | 526 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 527 |
2256 | 528 /* read the pixels */ |
529 for(i=0;i<4;i++) { | |
530 pixels[0] = cm[block[0]]; | |
531 pixels[1] = cm[block[1]]; | |
532 pixels[2] = cm[block[2]]; | |
533 pixels[3] = cm[block[3]]; | |
534 | |
535 pixels += line_size; | |
536 block += 8; | |
537 } | |
538 } | |
539 | |
2257 | 540 static void put_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 541 int line_size) |
2257 | 542 { |
543 int i; | |
4176 | 544 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 545 |
2257 | 546 /* read the pixels */ |
547 for(i=0;i<2;i++) { | |
548 pixels[0] = cm[block[0]]; | |
549 pixels[1] = cm[block[1]]; | |
550 | |
551 pixels += line_size; | |
552 block += 8; | |
553 } | |
554 } | |
555 | |
2967 | 556 static void put_signed_pixels_clamped_c(const DCTELEM *block, |
1984
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
557 uint8_t *restrict pixels, |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
558 int line_size) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
559 { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
560 int i, j; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
561 |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
562 for (i = 0; i < 8; i++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
563 for (j = 0; j < 8; j++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
564 if (*block < -128) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
565 *pixels = 0; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
566 else if (*block > 127) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
567 *pixels = 255; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
568 else |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
569 *pixels = (uint8_t)(*block + 128); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
570 block++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
571 pixels++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
572 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
573 pixels += (line_size - 8); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
574 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
575 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
576 |
1064 | 577 static void add_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
516 | 578 int line_size) |
0 | 579 { |
580 int i; | |
4176 | 581 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 582 |
0 | 583 /* read the pixels */ |
584 for(i=0;i<8;i++) { | |
516 | 585 pixels[0] = cm[pixels[0] + block[0]]; |
586 pixels[1] = cm[pixels[1] + block[1]]; | |
587 pixels[2] = cm[pixels[2] + block[2]]; | |
588 pixels[3] = cm[pixels[3] + block[3]]; | |
589 pixels[4] = cm[pixels[4] + block[4]]; | |
590 pixels[5] = cm[pixels[5] + block[5]]; | |
591 pixels[6] = cm[pixels[6] + block[6]]; | |
592 pixels[7] = cm[pixels[7] + block[7]]; | |
593 pixels += line_size; | |
594 block += 8; | |
0 | 595 } |
596 } | |
2256 | 597 |
598 static void add_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, | |
599 int line_size) | |
600 { | |
601 int i; | |
4176 | 602 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 603 |
2256 | 604 /* read the pixels */ |
605 for(i=0;i<4;i++) { | |
606 pixels[0] = cm[pixels[0] + block[0]]; | |
607 pixels[1] = cm[pixels[1] + block[1]]; | |
608 pixels[2] = cm[pixels[2] + block[2]]; | |
609 pixels[3] = cm[pixels[3] + block[3]]; | |
610 pixels += line_size; | |
611 block += 8; | |
612 } | |
613 } | |
2257 | 614 |
615 static void add_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, | |
616 int line_size) | |
617 { | |
618 int i; | |
4176 | 619 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 620 |
2257 | 621 /* read the pixels */ |
622 for(i=0;i<2;i++) { | |
623 pixels[0] = cm[pixels[0] + block[0]]; | |
624 pixels[1] = cm[pixels[1] + block[1]]; | |
625 pixels += line_size; | |
626 block += 8; | |
627 } | |
628 } | |
2763 | 629 |
630 static void add_pixels8_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
631 { | |
632 int i; | |
633 for(i=0;i<8;i++) { | |
634 pixels[0] += block[0]; | |
635 pixels[1] += block[1]; | |
636 pixels[2] += block[2]; | |
637 pixels[3] += block[3]; | |
638 pixels[4] += block[4]; | |
639 pixels[5] += block[5]; | |
640 pixels[6] += block[6]; | |
641 pixels[7] += block[7]; | |
642 pixels += line_size; | |
643 block += 8; | |
644 } | |
645 } | |
646 | |
647 static void add_pixels4_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
648 { | |
649 int i; | |
650 for(i=0;i<4;i++) { | |
651 pixels[0] += block[0]; | |
652 pixels[1] += block[1]; | |
653 pixels[2] += block[2]; | |
654 pixels[3] += block[3]; | |
655 pixels += line_size; | |
656 block += 4; | |
657 } | |
658 } | |
659 | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
660 static int sum_abs_dctelem_c(DCTELEM *block) |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
661 { |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
662 int sum=0, i; |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
663 for(i=0; i<64; i++) |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
664 sum+= FFABS(block[i]); |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
665 return sum; |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
666 } |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
667 |
385 | 668 #if 0 |
669 | |
670 #define PIXOP2(OPNAME, OP) \ | |
651 | 671 static void OPNAME ## _pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 672 {\ |
673 int i;\ | |
674 for(i=0; i<h; i++){\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
675 OP(*((uint64_t*)block), AV_RN64(pixels));\ |
385 | 676 pixels+=line_size;\ |
677 block +=line_size;\ | |
678 }\ | |
679 }\ | |
680 \ | |
859 | 681 static void OPNAME ## _no_rnd_pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 682 {\ |
683 int i;\ | |
684 for(i=0; i<h; i++){\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
685 const uint64_t a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
686 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 687 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
688 pixels+=line_size;\ | |
689 block +=line_size;\ | |
690 }\ | |
691 }\ | |
692 \ | |
859 | 693 static void OPNAME ## _pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 694 {\ |
695 int i;\ | |
696 for(i=0; i<h; i++){\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
697 const uint64_t a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
698 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 699 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
700 pixels+=line_size;\ | |
701 block +=line_size;\ | |
702 }\ | |
703 }\ | |
704 \ | |
859 | 705 static void OPNAME ## _no_rnd_pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 706 {\ |
707 int i;\ | |
708 for(i=0; i<h; i++){\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
709 const uint64_t a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
710 const uint64_t b= AV_RN64(pixels+line_size);\ |
385 | 711 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
712 pixels+=line_size;\ | |
713 block +=line_size;\ | |
714 }\ | |
715 }\ | |
716 \ | |
859 | 717 static void OPNAME ## _pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 718 {\ |
719 int i;\ | |
720 for(i=0; i<h; i++){\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
721 const uint64_t a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
722 const uint64_t b= AV_RN64(pixels+line_size);\ |
385 | 723 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
724 pixels+=line_size;\ | |
725 block +=line_size;\ | |
726 }\ | |
727 }\ | |
728 \ | |
859 | 729 static void OPNAME ## _pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 730 {\ |
731 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
732 const uint64_t a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
733 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 734 uint64_t l0= (a&0x0303030303030303ULL)\ |
735 + (b&0x0303030303030303ULL)\ | |
736 + 0x0202020202020202ULL;\ | |
737 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
738 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
739 uint64_t l1,h1;\ | |
740 \ | |
741 pixels+=line_size;\ | |
742 for(i=0; i<h; i+=2){\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
743 uint64_t a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
744 uint64_t b= AV_RN64(pixels+1);\ |
385 | 745 l1= (a&0x0303030303030303ULL)\ |
746 + (b&0x0303030303030303ULL);\ | |
747 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
748 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
749 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
750 pixels+=line_size;\ | |
751 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
752 a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
753 b= AV_RN64(pixels+1);\ |
385 | 754 l0= (a&0x0303030303030303ULL)\ |
755 + (b&0x0303030303030303ULL)\ | |
756 + 0x0202020202020202ULL;\ | |
757 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
758 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
759 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
760 pixels+=line_size;\ | |
761 block +=line_size;\ | |
762 }\ | |
763 }\ | |
764 \ | |
859 | 765 static void OPNAME ## _no_rnd_pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 766 {\ |
767 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
768 const uint64_t a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
769 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 770 uint64_t l0= (a&0x0303030303030303ULL)\ |
771 + (b&0x0303030303030303ULL)\ | |
772 + 0x0101010101010101ULL;\ | |
773 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
774 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
775 uint64_t l1,h1;\ | |
776 \ | |
777 pixels+=line_size;\ | |
778 for(i=0; i<h; i+=2){\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
779 uint64_t a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
780 uint64_t b= AV_RN64(pixels+1);\ |
385 | 781 l1= (a&0x0303030303030303ULL)\ |
782 + (b&0x0303030303030303ULL);\ | |
783 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
784 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
785 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
786 pixels+=line_size;\ | |
787 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
788 a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
789 b= AV_RN64(pixels+1);\ |
385 | 790 l0= (a&0x0303030303030303ULL)\ |
791 + (b&0x0303030303030303ULL)\ | |
792 + 0x0101010101010101ULL;\ | |
793 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
794 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
795 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
796 pixels+=line_size;\ | |
797 block +=line_size;\ | |
798 }\ | |
799 }\ | |
800 \ | |
859 | 801 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels_c , 8)\ |
802 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels_x2_c , 8)\ | |
803 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels_y2_c , 8)\ | |
804 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels_xy2_c, 8)\ | |
805 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels_x2_c , 8)\ | |
806 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels_y2_c , 8)\ | |
807 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels_xy2_c, 8) | |
385 | 808 |
809 #define op_avg(a, b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEFEFEFEFEULL)>>1) ) | |
810 #else // 64 bit variant | |
811 | |
812 #define PIXOP2(OPNAME, OP) \ | |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
813 static void OPNAME ## _pixels2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
814 int i;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
815 for(i=0; i<h; i++){\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
816 OP(*((uint16_t*)(block )), AV_RN16(pixels ));\ |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
817 pixels+=line_size;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
818 block +=line_size;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
819 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
820 }\ |
1168 | 821 static void OPNAME ## _pixels4_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
822 int i;\ | |
823 for(i=0; i<h; i++){\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
824 OP(*((uint32_t*)(block )), AV_RN32(pixels ));\ |
1168 | 825 pixels+=line_size;\ |
826 block +=line_size;\ | |
827 }\ | |
828 }\ | |
859 | 829 static void OPNAME ## _pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
385 | 830 int i;\ |
831 for(i=0; i<h; i++){\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
832 OP(*((uint32_t*)(block )), AV_RN32(pixels ));\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
833 OP(*((uint32_t*)(block+4)), AV_RN32(pixels+4));\ |
385 | 834 pixels+=line_size;\ |
835 block +=line_size;\ | |
836 }\ | |
837 }\ | |
859 | 838 static inline void OPNAME ## _no_rnd_pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
839 OPNAME ## _pixels8_c(block, pixels, line_size, h);\ | |
651 | 840 }\ |
385 | 841 \ |
651 | 842 static inline void OPNAME ## _no_rnd_pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
843 int src_stride1, int src_stride2, int h){\ | |
385 | 844 int i;\ |
845 for(i=0; i<h; i++){\ | |
651 | 846 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
847 a= AV_RN32(&src1[i*src_stride1 ]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
848 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 849 OP(*((uint32_t*)&dst[i*dst_stride ]), no_rnd_avg32(a, b));\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
850 a= AV_RN32(&src1[i*src_stride1+4]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
851 b= AV_RN32(&src2[i*src_stride2+4]);\ |
1264 | 852 OP(*((uint32_t*)&dst[i*dst_stride+4]), no_rnd_avg32(a, b));\ |
385 | 853 }\ |
854 }\ | |
855 \ | |
651 | 856 static inline void OPNAME ## _pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
857 int src_stride1, int src_stride2, int h){\ | |
385 | 858 int i;\ |
859 for(i=0; i<h; i++){\ | |
651 | 860 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
861 a= AV_RN32(&src1[i*src_stride1 ]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
862 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 863 OP(*((uint32_t*)&dst[i*dst_stride ]), rnd_avg32(a, b));\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
864 a= AV_RN32(&src1[i*src_stride1+4]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
865 b= AV_RN32(&src2[i*src_stride2+4]);\ |
1264 | 866 OP(*((uint32_t*)&dst[i*dst_stride+4]), rnd_avg32(a, b));\ |
385 | 867 }\ |
868 }\ | |
869 \ | |
1168 | 870 static inline void OPNAME ## _pixels4_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
871 int src_stride1, int src_stride2, int h){\ | |
872 int i;\ | |
873 for(i=0; i<h; i++){\ | |
874 uint32_t a,b;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
875 a= AV_RN32(&src1[i*src_stride1 ]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
876 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 877 OP(*((uint32_t*)&dst[i*dst_stride ]), rnd_avg32(a, b));\ |
1168 | 878 }\ |
879 }\ | |
880 \ | |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
881 static inline void OPNAME ## _pixels2_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
882 int src_stride1, int src_stride2, int h){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
883 int i;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
884 for(i=0; i<h; i++){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
885 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
886 a= AV_RN16(&src1[i*src_stride1 ]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
887 b= AV_RN16(&src2[i*src_stride2 ]);\ |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
888 OP(*((uint16_t*)&dst[i*dst_stride ]), rnd_avg32(a, b));\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
889 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
890 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
891 \ |
651 | 892 static inline void OPNAME ## _pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
893 int src_stride1, int src_stride2, int h){\ | |
894 OPNAME ## _pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
895 OPNAME ## _pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
896 }\ | |
897 \ | |
898 static inline void OPNAME ## _no_rnd_pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ | |
899 int src_stride1, int src_stride2, int h){\ | |
900 OPNAME ## _no_rnd_pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
901 OPNAME ## _no_rnd_pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
902 }\ | |
903 \ | |
859 | 904 static inline void OPNAME ## _no_rnd_pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 905 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
906 }\ | |
907 \ | |
859 | 908 static inline void OPNAME ## _pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 909 OPNAME ## _pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
910 }\ | |
911 \ | |
859 | 912 static inline void OPNAME ## _no_rnd_pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 913 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
914 }\ | |
915 \ | |
859 | 916 static inline void OPNAME ## _pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 917 OPNAME ## _pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
385 | 918 }\ |
919 \ | |
651 | 920 static inline void OPNAME ## _pixels8_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
921 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
922 int i;\ | |
923 for(i=0; i<h; i++){\ | |
924 uint32_t a, b, c, d, l0, l1, h0, h1;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
925 a= AV_RN32(&src1[i*src_stride1]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
926 b= AV_RN32(&src2[i*src_stride2]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
927 c= AV_RN32(&src3[i*src_stride3]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
928 d= AV_RN32(&src4[i*src_stride4]);\ |
651 | 929 l0= (a&0x03030303UL)\ |
930 + (b&0x03030303UL)\ | |
931 + 0x02020202UL;\ | |
932 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
933 + ((b&0xFCFCFCFCUL)>>2);\ | |
934 l1= (c&0x03030303UL)\ | |
935 + (d&0x03030303UL);\ | |
936 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
937 + ((d&0xFCFCFCFCUL)>>2);\ | |
938 OP(*((uint32_t*)&dst[i*dst_stride]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
939 a= AV_RN32(&src1[i*src_stride1+4]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
940 b= AV_RN32(&src2[i*src_stride2+4]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
941 c= AV_RN32(&src3[i*src_stride3+4]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
942 d= AV_RN32(&src4[i*src_stride4+4]);\ |
651 | 943 l0= (a&0x03030303UL)\ |
944 + (b&0x03030303UL)\ | |
945 + 0x02020202UL;\ | |
946 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
947 + ((b&0xFCFCFCFCUL)>>2);\ | |
948 l1= (c&0x03030303UL)\ | |
949 + (d&0x03030303UL);\ | |
950 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
951 + ((d&0xFCFCFCFCUL)>>2);\ | |
952 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
953 }\ | |
954 }\ | |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
955 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
956 static inline void OPNAME ## _pixels4_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
957 OPNAME ## _pixels4_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
958 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
959 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
960 static inline void OPNAME ## _pixels4_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
961 OPNAME ## _pixels4_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
962 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
963 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
964 static inline void OPNAME ## _pixels2_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
965 OPNAME ## _pixels2_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
966 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
967 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
968 static inline void OPNAME ## _pixels2_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
969 OPNAME ## _pixels2_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
970 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
971 \ |
651 | 972 static inline void OPNAME ## _no_rnd_pixels8_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
973 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
385 | 974 int i;\ |
975 for(i=0; i<h; i++){\ | |
651 | 976 uint32_t a, b, c, d, l0, l1, h0, h1;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
977 a= AV_RN32(&src1[i*src_stride1]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
978 b= AV_RN32(&src2[i*src_stride2]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
979 c= AV_RN32(&src3[i*src_stride3]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
980 d= AV_RN32(&src4[i*src_stride4]);\ |
651 | 981 l0= (a&0x03030303UL)\ |
982 + (b&0x03030303UL)\ | |
983 + 0x01010101UL;\ | |
984 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
985 + ((b&0xFCFCFCFCUL)>>2);\ | |
986 l1= (c&0x03030303UL)\ | |
987 + (d&0x03030303UL);\ | |
988 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
989 + ((d&0xFCFCFCFCUL)>>2);\ | |
990 OP(*((uint32_t*)&dst[i*dst_stride]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
991 a= AV_RN32(&src1[i*src_stride1+4]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
992 b= AV_RN32(&src2[i*src_stride2+4]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
993 c= AV_RN32(&src3[i*src_stride3+4]);\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
994 d= AV_RN32(&src4[i*src_stride4+4]);\ |
651 | 995 l0= (a&0x03030303UL)\ |
996 + (b&0x03030303UL)\ | |
997 + 0x01010101UL;\ | |
998 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
999 + ((b&0xFCFCFCFCUL)>>2);\ | |
1000 l1= (c&0x03030303UL)\ | |
1001 + (d&0x03030303UL);\ | |
1002 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
1003 + ((d&0xFCFCFCFCUL)>>2);\ | |
1004 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
385 | 1005 }\ |
1006 }\ | |
651 | 1007 static inline void OPNAME ## _pixels16_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
1008 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
1009 OPNAME ## _pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
1010 OPNAME ## _pixels8_l4(dst+8, src1+8, src2+8, src3+8, src4+8, dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
1011 }\ | |
1012 static inline void OPNAME ## _no_rnd_pixels16_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ | |
1013 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
1014 OPNAME ## _no_rnd_pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
1015 OPNAME ## _no_rnd_pixels8_l4(dst+8, src1+8, src2+8, src3+8, src4+8, dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
1016 }\ | |
385 | 1017 \ |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1018 static inline void OPNAME ## _pixels2_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1019 {\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1020 int i, a0, b0, a1, b1;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1021 a0= pixels[0];\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1022 b0= pixels[1] + 2;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1023 a0 += b0;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1024 b0 += pixels[2];\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1025 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1026 pixels+=line_size;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1027 for(i=0; i<h; i+=2){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1028 a1= pixels[0];\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1029 b1= pixels[1];\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1030 a1 += b1;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1031 b1 += pixels[2];\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1032 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1033 block[0]= (a1+a0)>>2; /* FIXME non put */\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1034 block[1]= (b1+b0)>>2;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1035 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1036 pixels+=line_size;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1037 block +=line_size;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1038 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1039 a0= pixels[0];\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1040 b0= pixels[1] + 2;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1041 a0 += b0;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1042 b0 += pixels[2];\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1043 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1044 block[0]= (a1+a0)>>2;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1045 block[1]= (b1+b0)>>2;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1046 pixels+=line_size;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1047 block +=line_size;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1048 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1049 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1050 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1051 static inline void OPNAME ## _pixels4_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1052 {\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1053 int i;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1054 const uint32_t a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1055 const uint32_t b= AV_RN32(pixels+1);\ |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1056 uint32_t l0= (a&0x03030303UL)\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1057 + (b&0x03030303UL)\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1058 + 0x02020202UL;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1059 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1060 + ((b&0xFCFCFCFCUL)>>2);\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1061 uint32_t l1,h1;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1062 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1063 pixels+=line_size;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1064 for(i=0; i<h; i+=2){\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1065 uint32_t a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1066 uint32_t b= AV_RN32(pixels+1);\ |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1067 l1= (a&0x03030303UL)\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1068 + (b&0x03030303UL);\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1069 h1= ((a&0xFCFCFCFCUL)>>2)\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1070 + ((b&0xFCFCFCFCUL)>>2);\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1071 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1072 pixels+=line_size;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1073 block +=line_size;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1074 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1075 b= AV_RN32(pixels+1);\ |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1076 l0= (a&0x03030303UL)\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1077 + (b&0x03030303UL)\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1078 + 0x02020202UL;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1079 h0= ((a&0xFCFCFCFCUL)>>2)\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1080 + ((b&0xFCFCFCFCUL)>>2);\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1081 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1082 pixels+=line_size;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1083 block +=line_size;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1084 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1085 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1086 \ |
859 | 1087 static inline void OPNAME ## _pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1088 {\ |
1089 int j;\ | |
1090 for(j=0; j<2; j++){\ | |
1091 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1092 const uint32_t a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1093 const uint32_t b= AV_RN32(pixels+1);\ |
385 | 1094 uint32_t l0= (a&0x03030303UL)\ |
1095 + (b&0x03030303UL)\ | |
1096 + 0x02020202UL;\ | |
1097 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1098 + ((b&0xFCFCFCFCUL)>>2);\ | |
1099 uint32_t l1,h1;\ | |
1100 \ | |
1101 pixels+=line_size;\ | |
1102 for(i=0; i<h; i+=2){\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1103 uint32_t a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1104 uint32_t b= AV_RN32(pixels+1);\ |
385 | 1105 l1= (a&0x03030303UL)\ |
1106 + (b&0x03030303UL);\ | |
1107 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1108 + ((b&0xFCFCFCFCUL)>>2);\ | |
1109 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1110 pixels+=line_size;\ | |
1111 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1112 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1113 b= AV_RN32(pixels+1);\ |
385 | 1114 l0= (a&0x03030303UL)\ |
1115 + (b&0x03030303UL)\ | |
1116 + 0x02020202UL;\ | |
1117 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1118 + ((b&0xFCFCFCFCUL)>>2);\ | |
1119 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1120 pixels+=line_size;\ | |
1121 block +=line_size;\ | |
1122 }\ | |
1123 pixels+=4-line_size*(h+1);\ | |
1124 block +=4-line_size*h;\ | |
1125 }\ | |
1126 }\ | |
1127 \ | |
859 | 1128 static inline void OPNAME ## _no_rnd_pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1129 {\ |
1130 int j;\ | |
1131 for(j=0; j<2; j++){\ | |
1132 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1133 const uint32_t a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1134 const uint32_t b= AV_RN32(pixels+1);\ |
385 | 1135 uint32_t l0= (a&0x03030303UL)\ |
1136 + (b&0x03030303UL)\ | |
1137 + 0x01010101UL;\ | |
1138 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1139 + ((b&0xFCFCFCFCUL)>>2);\ | |
1140 uint32_t l1,h1;\ | |
1141 \ | |
1142 pixels+=line_size;\ | |
1143 for(i=0; i<h; i+=2){\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1144 uint32_t a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1145 uint32_t b= AV_RN32(pixels+1);\ |
385 | 1146 l1= (a&0x03030303UL)\ |
1147 + (b&0x03030303UL);\ | |
1148 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1149 + ((b&0xFCFCFCFCUL)>>2);\ | |
1150 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1151 pixels+=line_size;\ | |
1152 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1153 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1154 b= AV_RN32(pixels+1);\ |
385 | 1155 l0= (a&0x03030303UL)\ |
1156 + (b&0x03030303UL)\ | |
1157 + 0x01010101UL;\ | |
1158 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1159 + ((b&0xFCFCFCFCUL)>>2);\ | |
1160 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1161 pixels+=line_size;\ | |
1162 block +=line_size;\ | |
1163 }\ | |
1164 pixels+=4-line_size*(h+1);\ | |
1165 block +=4-line_size*h;\ | |
1166 }\ | |
1167 }\ | |
1168 \ | |
859 | 1169 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels8_c , 8)\ |
1170 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels8_x2_c , 8)\ | |
1171 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels8_y2_c , 8)\ | |
1172 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels8_xy2_c, 8)\ | |
1173 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_c , OPNAME ## _pixels8_c , 8)\ | |
1174 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels8_x2_c , 8)\ | |
1175 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels8_y2_c , 8)\ | |
1176 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels8_xy2_c, 8)\ | |
651 | 1177 |
1264 | 1178 #define op_avg(a, b) a = rnd_avg32(a, b) |
385 | 1179 #endif |
1180 #define op_put(a, b) a = b | |
1181 | |
1182 PIXOP2(avg, op_avg) | |
1183 PIXOP2(put, op_put) | |
1184 #undef op_avg | |
1185 #undef op_put | |
1186 | |
0 | 1187 #define avg2(a,b) ((a+b+1)>>1) |
1188 #define avg4(a,b,c,d) ((a+b+c+d+2)>>2) | |
1189 | |
1864 | 1190 static void put_no_rnd_pixels16_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ |
1191 put_no_rnd_pixels16_l2(dst, a, b, stride, stride, stride, h); | |
1192 } | |
1193 | |
1194 static void put_no_rnd_pixels8_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ | |
1195 put_no_rnd_pixels8_l2(dst, a, b, stride, stride, stride, h); | |
1196 } | |
753 | 1197 |
1064 | 1198 static void gmc1_c(uint8_t *dst, uint8_t *src, int stride, int h, int x16, int y16, int rounder) |
255 | 1199 { |
1200 const int A=(16-x16)*(16-y16); | |
1201 const int B=( x16)*(16-y16); | |
1202 const int C=(16-x16)*( y16); | |
1203 const int D=( x16)*( y16); | |
1204 int i; | |
1205 | |
1206 for(i=0; i<h; i++) | |
1207 { | |
651 | 1208 dst[0]= (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + rounder)>>8; |
1209 dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8; | |
1210 dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8; | |
1211 dst[3]= (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + rounder)>>8; | |
1212 dst[4]= (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + rounder)>>8; | |
1213 dst[5]= (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + rounder)>>8; | |
1214 dst[6]= (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + rounder)>>8; | |
1215 dst[7]= (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + rounder)>>8; | |
1216 dst+= stride; | |
1217 src+= stride; | |
255 | 1218 } |
1219 } | |
1220 | |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
1221 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, |
753 | 1222 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height) |
1223 { | |
1224 int y, vx, vy; | |
1225 const int s= 1<<shift; | |
2967 | 1226 |
753 | 1227 width--; |
1228 height--; | |
1229 | |
1230 for(y=0; y<h; y++){ | |
1231 int x; | |
1232 | |
1233 vx= ox; | |
1234 vy= oy; | |
1235 for(x=0; x<8; x++){ //XXX FIXME optimize | |
1236 int src_x, src_y, frac_x, frac_y, index; | |
1237 | |
1238 src_x= vx>>16; | |
1239 src_y= vy>>16; | |
1240 frac_x= src_x&(s-1); | |
1241 frac_y= src_y&(s-1); | |
1242 src_x>>=shift; | |
1243 src_y>>=shift; | |
2967 | 1244 |
753 | 1245 if((unsigned)src_x < width){ |
1246 if((unsigned)src_y < height){ | |
1247 index= src_x + src_y*stride; | |
1248 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) | |
1249 + src[index +1]* frac_x )*(s-frac_y) | |
1250 + ( src[index+stride ]*(s-frac_x) | |
1251 + src[index+stride+1]* frac_x )* frac_y | |
1252 + r)>>(shift*2); | |
1253 }else{ | |
4594 | 1254 index= src_x + av_clip(src_y, 0, height)*stride; |
2967 | 1255 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) |
753 | 1256 + src[index +1]* frac_x )*s |
1257 + r)>>(shift*2); | |
1258 } | |
1259 }else{ | |
1260 if((unsigned)src_y < height){ | |
4594 | 1261 index= av_clip(src_x, 0, width) + src_y*stride; |
2967 | 1262 dst[y*stride + x]= ( ( src[index ]*(s-frac_y) |
753 | 1263 + src[index+stride ]* frac_y )*s |
1264 + r)>>(shift*2); | |
1265 }else{ | |
4594 | 1266 index= av_clip(src_x, 0, width) + av_clip(src_y, 0, height)*stride; |
753 | 1267 dst[y*stride + x]= src[index ]; |
1268 } | |
1269 } | |
2967 | 1270 |
753 | 1271 vx+= dxx; |
1272 vy+= dyx; | |
1273 } | |
1274 ox += dxy; | |
1275 oy += dyy; | |
1276 } | |
1277 } | |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1278 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1279 static inline void put_tpel_pixels_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1280 switch(width){ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1281 case 2: put_pixels2_c (dst, src, stride, height); break; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1282 case 4: put_pixels4_c (dst, src, stride, height); break; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1283 case 8: put_pixels8_c (dst, src, stride, height); break; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1284 case 16:put_pixels16_c(dst, src, stride, height); break; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1285 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1286 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1287 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1288 static inline void put_tpel_pixels_mc10_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1289 int i,j; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1290 for (i=0; i < height; i++) { |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1291 for (j=0; j < width; j++) { |
2979 | 1292 dst[j] = (683*(2*src[j] + src[j+1] + 1)) >> 11; |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1293 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1294 src += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1295 dst += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1296 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1297 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1298 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1299 static inline void put_tpel_pixels_mc20_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1300 int i,j; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1301 for (i=0; i < height; i++) { |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1302 for (j=0; j < width; j++) { |
2979 | 1303 dst[j] = (683*(src[j] + 2*src[j+1] + 1)) >> 11; |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1304 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1305 src += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1306 dst += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1307 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1308 } |
2967 | 1309 |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1310 static inline void put_tpel_pixels_mc01_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1311 int i,j; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1312 for (i=0; i < height; i++) { |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1313 for (j=0; j < width; j++) { |
2979 | 1314 dst[j] = (683*(2*src[j] + src[j+stride] + 1)) >> 11; |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1315 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1316 src += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1317 dst += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1318 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1319 } |
2967 | 1320 |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1321 static inline void put_tpel_pixels_mc11_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1322 int i,j; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1323 for (i=0; i < height; i++) { |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1324 for (j=0; j < width; j++) { |
2979 | 1325 dst[j] = (2731*(4*src[j] + 3*src[j+1] + 3*src[j+stride] + 2*src[j+stride+1] + 6)) >> 15; |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1326 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1327 src += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1328 dst += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1329 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1330 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1331 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1332 static inline void put_tpel_pixels_mc12_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1333 int i,j; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1334 for (i=0; i < height; i++) { |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1335 for (j=0; j < width; j++) { |
2979 | 1336 dst[j] = (2731*(3*src[j] + 2*src[j+1] + 4*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15; |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1337 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1338 src += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1339 dst += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1340 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1341 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1342 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1343 static inline void put_tpel_pixels_mc02_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1344 int i,j; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1345 for (i=0; i < height; i++) { |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1346 for (j=0; j < width; j++) { |
2979 | 1347 dst[j] = (683*(src[j] + 2*src[j+stride] + 1)) >> 11; |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1348 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1349 src += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1350 dst += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1351 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1352 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1353 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1354 static inline void put_tpel_pixels_mc21_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1355 int i,j; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1356 for (i=0; i < height; i++) { |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1357 for (j=0; j < width; j++) { |
2979 | 1358 dst[j] = (2731*(3*src[j] + 4*src[j+1] + 2*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15; |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1359 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1360 src += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1361 dst += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1362 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1363 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1364 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1365 static inline void put_tpel_pixels_mc22_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1366 int i,j; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1367 for (i=0; i < height; i++) { |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1368 for (j=0; j < width; j++) { |
2979 | 1369 dst[j] = (2731*(2*src[j] + 3*src[j+1] + 3*src[j+stride] + 4*src[j+stride+1] + 6)) >> 15; |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1370 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1371 src += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1372 dst += stride; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1373 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1374 } |
1319 | 1375 |
1376 static inline void avg_tpel_pixels_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1377 switch(width){ | |
1378 case 2: avg_pixels2_c (dst, src, stride, height); break; | |
1379 case 4: avg_pixels4_c (dst, src, stride, height); break; | |
1380 case 8: avg_pixels8_c (dst, src, stride, height); break; | |
1381 case 16:avg_pixels16_c(dst, src, stride, height); break; | |
1382 } | |
1383 } | |
1384 | |
1385 static inline void avg_tpel_pixels_mc10_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1386 int i,j; | |
1387 for (i=0; i < height; i++) { | |
1388 for (j=0; j < width; j++) { | |
2979 | 1389 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1390 } |
1391 src += stride; | |
1392 dst += stride; | |
1393 } | |
1394 } | |
1395 | |
1396 static inline void avg_tpel_pixels_mc20_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1397 int i,j; | |
1398 for (i=0; i < height; i++) { | |
1399 for (j=0; j < width; j++) { | |
2979 | 1400 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1401 } |
1402 src += stride; | |
1403 dst += stride; | |
1404 } | |
1405 } | |
2967 | 1406 |
1319 | 1407 static inline void avg_tpel_pixels_mc01_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1408 int i,j; | |
1409 for (i=0; i < height; i++) { | |
1410 for (j=0; j < width; j++) { | |
2979 | 1411 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1412 } |
1413 src += stride; | |
1414 dst += stride; | |
1415 } | |
1416 } | |
2967 | 1417 |
1319 | 1418 static inline void avg_tpel_pixels_mc11_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1419 int i,j; | |
1420 for (i=0; i < height; i++) { | |
1421 for (j=0; j < width; j++) { | |
2979 | 1422 dst[j] = (dst[j] + ((2731*(4*src[j] + 3*src[j+1] + 3*src[j+stride] + 2*src[j+stride+1] + 6)) >> 15) + 1) >> 1; |
1319 | 1423 } |
1424 src += stride; | |
1425 dst += stride; | |
1426 } | |
1427 } | |
1428 | |
1429 static inline void avg_tpel_pixels_mc12_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1430 int i,j; | |
1431 for (i=0; i < height; i++) { | |
1432 for (j=0; j < width; j++) { | |
2979 | 1433 dst[j] = (dst[j] + ((2731*(3*src[j] + 2*src[j+1] + 4*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15) + 1) >> 1; |
1319 | 1434 } |
1435 src += stride; | |
1436 dst += stride; | |
1437 } | |
1438 } | |
1439 | |
1440 static inline void avg_tpel_pixels_mc02_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1441 int i,j; | |
1442 for (i=0; i < height; i++) { | |
1443 for (j=0; j < width; j++) { | |
2979 | 1444 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1445 } |
1446 src += stride; | |
1447 dst += stride; | |
1448 } | |
1449 } | |
1450 | |
1451 static inline void avg_tpel_pixels_mc21_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1452 int i,j; | |
1453 for (i=0; i < height; i++) { | |
1454 for (j=0; j < width; j++) { | |
2979 | 1455 dst[j] = (dst[j] + ((2731*(3*src[j] + 4*src[j+1] + 2*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15) + 1) >> 1; |
1319 | 1456 } |
1457 src += stride; | |
1458 dst += stride; | |
1459 } | |
1460 } | |
1461 | |
1462 static inline void avg_tpel_pixels_mc22_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1463 int i,j; | |
1464 for (i=0; i < height; i++) { | |
1465 for (j=0; j < width; j++) { | |
2979 | 1466 dst[j] = (dst[j] + ((2731*(2*src[j] + 3*src[j+1] + 3*src[j+stride] + 4*src[j+stride+1] + 6)) >> 15) + 1) >> 1; |
1319 | 1467 } |
1468 src += stride; | |
1469 dst += stride; | |
1470 } | |
1471 } | |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1472 #if 0 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1473 #define TPEL_WIDTH(width)\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1474 static void put_tpel_pixels ## width ## _mc00_c(uint8_t *dst, const uint8_t *src, int stride, int height){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1475 void put_tpel_pixels_mc00_c(dst, src, stride, width, height);}\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1476 static void put_tpel_pixels ## width ## _mc10_c(uint8_t *dst, const uint8_t *src, int stride, int height){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1477 void put_tpel_pixels_mc10_c(dst, src, stride, width, height);}\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1478 static void put_tpel_pixels ## width ## _mc20_c(uint8_t *dst, const uint8_t *src, int stride, int height){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1479 void put_tpel_pixels_mc20_c(dst, src, stride, width, height);}\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1480 static void put_tpel_pixels ## width ## _mc01_c(uint8_t *dst, const uint8_t *src, int stride, int height){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1481 void put_tpel_pixels_mc01_c(dst, src, stride, width, height);}\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1482 static void put_tpel_pixels ## width ## _mc11_c(uint8_t *dst, const uint8_t *src, int stride, int height){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1483 void put_tpel_pixels_mc11_c(dst, src, stride, width, height);}\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1484 static void put_tpel_pixels ## width ## _mc21_c(uint8_t *dst, const uint8_t *src, int stride, int height){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1485 void put_tpel_pixels_mc21_c(dst, src, stride, width, height);}\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1486 static void put_tpel_pixels ## width ## _mc02_c(uint8_t *dst, const uint8_t *src, int stride, int height){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1487 void put_tpel_pixels_mc02_c(dst, src, stride, width, height);}\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1488 static void put_tpel_pixels ## width ## _mc12_c(uint8_t *dst, const uint8_t *src, int stride, int height){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1489 void put_tpel_pixels_mc12_c(dst, src, stride, width, height);}\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1490 static void put_tpel_pixels ## width ## _mc22_c(uint8_t *dst, const uint8_t *src, int stride, int height){\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1491 void put_tpel_pixels_mc22_c(dst, src, stride, width, height);} |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1492 #endif |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1493 |
1168 | 1494 #define H264_CHROMA_MC(OPNAME, OP)\ |
1495 static void OPNAME ## h264_chroma_mc2_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){\ | |
1496 const int A=(8-x)*(8-y);\ | |
1497 const int B=( x)*(8-y);\ | |
1498 const int C=(8-x)*( y);\ | |
1499 const int D=( x)*( y);\ | |
1500 int i;\ | |
1501 \ | |
1502 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1503 \ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1504 if(D){\ |
6054 | 1505 for(i=0; i<h; i++){\ |
6053 | 1506 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1507 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1508 dst+= stride;\ | |
1509 src+= stride;\ | |
1510 }\ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1511 }else{\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1512 const int E= B+C;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1513 const int step= C ? stride : 1;\ |
6054 | 1514 for(i=0; i<h; i++){\ |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1515 OP(dst[0], (A*src[0] + E*src[step+0]));\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1516 OP(dst[1], (A*src[1] + E*src[step+1]));\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1517 dst+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1518 src+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1519 }\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1520 }\ |
1168 | 1521 }\ |
1522 \ | |
1523 static void OPNAME ## h264_chroma_mc4_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){\ | |
1524 const int A=(8-x)*(8-y);\ | |
1525 const int B=( x)*(8-y);\ | |
1526 const int C=(8-x)*( y);\ | |
1527 const int D=( x)*( y);\ | |
1528 int i;\ | |
1529 \ | |
1530 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1531 \ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1532 if(D){\ |
6054 | 1533 for(i=0; i<h; i++){\ |
6053 | 1534 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1535 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1536 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1537 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1538 dst+= stride;\ | |
1539 src+= stride;\ | |
1540 }\ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1541 }else{\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1542 const int E= B+C;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1543 const int step= C ? stride : 1;\ |
6054 | 1544 for(i=0; i<h; i++){\ |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1545 OP(dst[0], (A*src[0] + E*src[step+0]));\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1546 OP(dst[1], (A*src[1] + E*src[step+1]));\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1547 OP(dst[2], (A*src[2] + E*src[step+2]));\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1548 OP(dst[3], (A*src[3] + E*src[step+3]));\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1549 dst+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1550 src+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1551 }\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1552 }\ |
1168 | 1553 }\ |
1554 \ | |
1555 static void OPNAME ## h264_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){\ | |
1556 const int A=(8-x)*(8-y);\ | |
1557 const int B=( x)*(8-y);\ | |
1558 const int C=(8-x)*( y);\ | |
1559 const int D=( x)*( y);\ | |
1560 int i;\ | |
1561 \ | |
1562 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1563 \ | |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1564 if(D){\ |
6054 | 1565 for(i=0; i<h; i++){\ |
6053 | 1566 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1567 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1568 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1569 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1570 OP(dst[4], (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5]));\ | |
1571 OP(dst[5], (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6]));\ | |
1572 OP(dst[6], (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7]));\ | |
1573 OP(dst[7], (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8]));\ | |
1574 dst+= stride;\ | |
1575 src+= stride;\ | |
1576 }\ | |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1577 }else{\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1578 const int E= B+C;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1579 const int step= C ? stride : 1;\ |
6054 | 1580 for(i=0; i<h; i++){\ |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1581 OP(dst[0], (A*src[0] + E*src[step+0]));\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1582 OP(dst[1], (A*src[1] + E*src[step+1]));\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1583 OP(dst[2], (A*src[2] + E*src[step+2]));\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1584 OP(dst[3], (A*src[3] + E*src[step+3]));\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1585 OP(dst[4], (A*src[4] + E*src[step+4]));\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1586 OP(dst[5], (A*src[5] + E*src[step+5]));\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1587 OP(dst[6], (A*src[6] + E*src[step+6]));\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1588 OP(dst[7], (A*src[7] + E*src[step+7]));\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1589 dst+= stride;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1590 src+= stride;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1591 }\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1592 }\ |
1168 | 1593 } |
1594 | |
1595 #define op_avg(a, b) a = (((a)+(((b) + 32)>>6)+1)>>1) | |
1596 #define op_put(a, b) a = (((b) + 32)>>6) | |
1597 | |
1598 H264_CHROMA_MC(put_ , op_put) | |
1599 H264_CHROMA_MC(avg_ , op_avg) | |
1600 #undef op_avg | |
1601 #undef op_put | |
1602 | |
3663 | 1603 static void put_no_rnd_h264_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){ |
1604 const int A=(8-x)*(8-y); | |
1605 const int B=( x)*(8-y); | |
1606 const int C=(8-x)*( y); | |
1607 const int D=( x)*( y); | |
1608 int i; | |
1609 | |
1610 assert(x<8 && y<8 && x>=0 && y>=0); | |
1611 | |
1612 for(i=0; i<h; i++) | |
1613 { | |
1614 dst[0] = (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6; | |
1615 dst[1] = (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6; | |
1616 dst[2] = (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6; | |
1617 dst[3] = (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6; | |
1618 dst[4] = (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6; | |
1619 dst[5] = (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6; | |
1620 dst[6] = (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6; | |
1621 dst[7] = (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6; | |
1622 dst+= stride; | |
1623 src+= stride; | |
1624 } | |
1625 } | |
1626 | |
651 | 1627 #define QPEL_MC(r, OPNAME, RND, OP) \ |
1064 | 1628 static void OPNAME ## mpeg4_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
4176 | 1629 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1630 int i;\ |
1631 for(i=0; i<h; i++)\ | |
1632 {\ | |
1633 OP(dst[0], (src[0]+src[1])*20 - (src[0]+src[2])*6 + (src[1]+src[3])*3 - (src[2]+src[4]));\ | |
1634 OP(dst[1], (src[1]+src[2])*20 - (src[0]+src[3])*6 + (src[0]+src[4])*3 - (src[1]+src[5]));\ | |
1635 OP(dst[2], (src[2]+src[3])*20 - (src[1]+src[4])*6 + (src[0]+src[5])*3 - (src[0]+src[6]));\ | |
1636 OP(dst[3], (src[3]+src[4])*20 - (src[2]+src[5])*6 + (src[1]+src[6])*3 - (src[0]+src[7]));\ | |
1637 OP(dst[4], (src[4]+src[5])*20 - (src[3]+src[6])*6 + (src[2]+src[7])*3 - (src[1]+src[8]));\ | |
1638 OP(dst[5], (src[5]+src[6])*20 - (src[4]+src[7])*6 + (src[3]+src[8])*3 - (src[2]+src[8]));\ | |
1639 OP(dst[6], (src[6]+src[7])*20 - (src[5]+src[8])*6 + (src[4]+src[8])*3 - (src[3]+src[7]));\ | |
1640 OP(dst[7], (src[7]+src[8])*20 - (src[6]+src[8])*6 + (src[5]+src[7])*3 - (src[4]+src[6]));\ | |
1641 dst+=dstStride;\ | |
1642 src+=srcStride;\ | |
1643 }\ | |
1644 }\ | |
1645 \ | |
1064 | 1646 static void OPNAME ## mpeg4_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
984 | 1647 const int w=8;\ |
4176 | 1648 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1649 int i;\ |
1650 for(i=0; i<w; i++)\ | |
1651 {\ | |
1652 const int src0= src[0*srcStride];\ | |
1653 const int src1= src[1*srcStride];\ | |
1654 const int src2= src[2*srcStride];\ | |
1655 const int src3= src[3*srcStride];\ | |
1656 const int src4= src[4*srcStride];\ | |
1657 const int src5= src[5*srcStride];\ | |
1658 const int src6= src[6*srcStride];\ | |
1659 const int src7= src[7*srcStride];\ | |
1660 const int src8= src[8*srcStride];\ | |
1661 OP(dst[0*dstStride], (src0+src1)*20 - (src0+src2)*6 + (src1+src3)*3 - (src2+src4));\ | |
1662 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*6 + (src0+src4)*3 - (src1+src5));\ | |
1663 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*6 + (src0+src5)*3 - (src0+src6));\ | |
1664 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*6 + (src1+src6)*3 - (src0+src7));\ | |
1665 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*6 + (src2+src7)*3 - (src1+src8));\ | |
1666 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*6 + (src3+src8)*3 - (src2+src8));\ | |
1667 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*6 + (src4+src8)*3 - (src3+src7));\ | |
1668 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src8)*6 + (src5+src7)*3 - (src4+src6));\ | |
1669 dst++;\ | |
1670 src++;\ | |
1671 }\ | |
1672 }\ | |
1673 \ | |
1064 | 1674 static void OPNAME ## mpeg4_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
4176 | 1675 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1676 int i;\ |
954 | 1677 \ |
651 | 1678 for(i=0; i<h; i++)\ |
1679 {\ | |
1680 OP(dst[ 0], (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]));\ | |
1681 OP(dst[ 1], (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]));\ | |
1682 OP(dst[ 2], (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]));\ | |
1683 OP(dst[ 3], (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]));\ | |
1684 OP(dst[ 4], (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]));\ | |
1685 OP(dst[ 5], (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]));\ | |
1686 OP(dst[ 6], (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]));\ | |
1687 OP(dst[ 7], (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]));\ | |
1688 OP(dst[ 8], (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]));\ | |
1689 OP(dst[ 9], (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]));\ | |
1690 OP(dst[10], (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]));\ | |
1691 OP(dst[11], (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]));\ | |
1692 OP(dst[12], (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]));\ | |
1693 OP(dst[13], (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]));\ | |
1694 OP(dst[14], (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]));\ | |
1695 OP(dst[15], (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]));\ | |
1696 dst+=dstStride;\ | |
1697 src+=srcStride;\ | |
1698 }\ | |
255 | 1699 }\ |
1700 \ | |
1064 | 1701 static void OPNAME ## mpeg4_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
4176 | 1702 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1703 int i;\ |
954 | 1704 const int w=16;\ |
651 | 1705 for(i=0; i<w; i++)\ |
1706 {\ | |
1707 const int src0= src[0*srcStride];\ | |
1708 const int src1= src[1*srcStride];\ | |
1709 const int src2= src[2*srcStride];\ | |
1710 const int src3= src[3*srcStride];\ | |
1711 const int src4= src[4*srcStride];\ | |
1712 const int src5= src[5*srcStride];\ | |
1713 const int src6= src[6*srcStride];\ | |
1714 const int src7= src[7*srcStride];\ | |
1715 const int src8= src[8*srcStride];\ | |
1716 const int src9= src[9*srcStride];\ | |
1717 const int src10= src[10*srcStride];\ | |
1718 const int src11= src[11*srcStride];\ | |
1719 const int src12= src[12*srcStride];\ | |
1720 const int src13= src[13*srcStride];\ | |
1721 const int src14= src[14*srcStride];\ | |
1722 const int src15= src[15*srcStride];\ | |
1723 const int src16= src[16*srcStride];\ | |
1724 OP(dst[ 0*dstStride], (src0 +src1 )*20 - (src0 +src2 )*6 + (src1 +src3 )*3 - (src2 +src4 ));\ | |
1725 OP(dst[ 1*dstStride], (src1 +src2 )*20 - (src0 +src3 )*6 + (src0 +src4 )*3 - (src1 +src5 ));\ | |
1726 OP(dst[ 2*dstStride], (src2 +src3 )*20 - (src1 +src4 )*6 + (src0 +src5 )*3 - (src0 +src6 ));\ | |
1727 OP(dst[ 3*dstStride], (src3 +src4 )*20 - (src2 +src5 )*6 + (src1 +src6 )*3 - (src0 +src7 ));\ | |
1728 OP(dst[ 4*dstStride], (src4 +src5 )*20 - (src3 +src6 )*6 + (src2 +src7 )*3 - (src1 +src8 ));\ | |
1729 OP(dst[ 5*dstStride], (src5 +src6 )*20 - (src4 +src7 )*6 + (src3 +src8 )*3 - (src2 +src9 ));\ | |
1730 OP(dst[ 6*dstStride], (src6 +src7 )*20 - (src5 +src8 )*6 + (src4 +src9 )*3 - (src3 +src10));\ | |
1731 OP(dst[ 7*dstStride], (src7 +src8 )*20 - (src6 +src9 )*6 + (src5 +src10)*3 - (src4 +src11));\ | |
1732 OP(dst[ 8*dstStride], (src8 +src9 )*20 - (src7 +src10)*6 + (src6 +src11)*3 - (src5 +src12));\ | |
1733 OP(dst[ 9*dstStride], (src9 +src10)*20 - (src8 +src11)*6 + (src7 +src12)*3 - (src6 +src13));\ | |
1734 OP(dst[10*dstStride], (src10+src11)*20 - (src9 +src12)*6 + (src8 +src13)*3 - (src7 +src14));\ | |
1735 OP(dst[11*dstStride], (src11+src12)*20 - (src10+src13)*6 + (src9 +src14)*3 - (src8 +src15));\ | |
1736 OP(dst[12*dstStride], (src12+src13)*20 - (src11+src14)*6 + (src10+src15)*3 - (src9 +src16));\ | |
1737 OP(dst[13*dstStride], (src13+src14)*20 - (src12+src15)*6 + (src11+src16)*3 - (src10+src16));\ | |
1738 OP(dst[14*dstStride], (src14+src15)*20 - (src13+src16)*6 + (src12+src16)*3 - (src11+src15));\ | |
1739 OP(dst[15*dstStride], (src15+src16)*20 - (src14+src16)*6 + (src13+src15)*3 - (src12+src14));\ | |
1740 dst++;\ | |
1741 src++;\ | |
1742 }\ | |
255 | 1743 }\ |
1744 \ | |
1064 | 1745 static void OPNAME ## qpel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1746 OPNAME ## pixels8_c(dst, src, stride, 8);\ |
255 | 1747 }\ |
1748 \ | |
1064 | 1749 static void OPNAME ## qpel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
1750 uint8_t half[64];\ | |
651 | 1751 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1752 OPNAME ## pixels8_l2(dst, src, half, stride, stride, 8, 8);\ | |
1753 }\ | |
1754 \ | |
1064 | 1755 static void OPNAME ## qpel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 1756 OPNAME ## mpeg4_qpel8_h_lowpass(dst, src, stride, stride, 8);\ |
255 | 1757 }\ |
1758 \ | |
1064 | 1759 static void OPNAME ## qpel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
1760 uint8_t half[64];\ | |
651 | 1761 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1762 OPNAME ## pixels8_l2(dst, src+1, half, stride, stride, 8, 8);\ | |
1763 }\ | |
1764 \ | |
1064 | 1765 static void OPNAME ## qpel8_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
1766 uint8_t full[16*9];\ | |
1767 uint8_t half[64];\ | |
651 | 1768 copy_block9(full, src, 16, stride, 9);\ |
984 | 1769 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1770 OPNAME ## pixels8_l2(dst, full, half, stride, 16, 8, 8);\ |
1771 }\ | |
1772 \ | |
1064 | 1773 static void OPNAME ## qpel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
1774 uint8_t full[16*9];\ | |
651 | 1775 copy_block9(full, src, 16, stride, 9);\ |
984 | 1776 OPNAME ## mpeg4_qpel8_v_lowpass(dst, full, stride, 16);\ |
255 | 1777 }\ |
1778 \ | |
1064 | 1779 static void OPNAME ## qpel8_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
1780 uint8_t full[16*9];\ | |
1781 uint8_t half[64];\ | |
651 | 1782 copy_block9(full, src, 16, stride, 9);\ |
984 | 1783 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1784 OPNAME ## pixels8_l2(dst, full+16, half, stride, 16, 8, 8);\ |
1785 }\ | |
1064 | 1786 void ff_ ## OPNAME ## qpel8_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1787 uint8_t full[16*9];\ | |
1788 uint8_t halfH[72];\ | |
1789 uint8_t halfV[64];\ | |
1790 uint8_t halfHV[64];\ | |
651 | 1791 copy_block9(full, src, 16, stride, 9);\ |
1792 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1793 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1794 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1795 OPNAME ## pixels8_l4(dst, full, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1796 }\ |
1064 | 1797 static void OPNAME ## qpel8_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
1798 uint8_t full[16*9];\ | |
1799 uint8_t halfH[72];\ | |
1800 uint8_t halfHV[64];\ | |
984 | 1801 copy_block9(full, src, 16, stride, 9);\ |
1802 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1803 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1804 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1805 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1806 }\ | |
1064 | 1807 void ff_ ## OPNAME ## qpel8_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1808 uint8_t full[16*9];\ | |
1809 uint8_t halfH[72];\ | |
1810 uint8_t halfV[64];\ | |
1811 uint8_t halfHV[64];\ | |
651 | 1812 copy_block9(full, src, 16, stride, 9);\ |
1813 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1814 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1815 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1816 OPNAME ## pixels8_l4(dst, full+1, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1817 }\ |
1064 | 1818 static void OPNAME ## qpel8_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
1819 uint8_t full[16*9];\ | |
1820 uint8_t halfH[72];\ | |
1821 uint8_t halfHV[64];\ | |
984 | 1822 copy_block9(full, src, 16, stride, 9);\ |
1823 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1824 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1825 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1826 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1827 }\ | |
1064 | 1828 void ff_ ## OPNAME ## qpel8_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1829 uint8_t full[16*9];\ | |
1830 uint8_t halfH[72];\ | |
1831 uint8_t halfV[64];\ | |
1832 uint8_t halfHV[64];\ | |
651 | 1833 copy_block9(full, src, 16, stride, 9);\ |
1834 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1835 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1836 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1837 OPNAME ## pixels8_l4(dst, full+16, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
1838 }\ | |
1064 | 1839 static void OPNAME ## qpel8_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
1840 uint8_t full[16*9];\ | |
1841 uint8_t halfH[72];\ | |
1842 uint8_t halfHV[64];\ | |
984 | 1843 copy_block9(full, src, 16, stride, 9);\ |
1844 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1845 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1846 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1847 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1848 }\ | |
1064 | 1849 void ff_ ## OPNAME ## qpel8_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1850 uint8_t full[16*9];\ | |
1851 uint8_t halfH[72];\ | |
1852 uint8_t halfV[64];\ | |
1853 uint8_t halfHV[64];\ | |
651 | 1854 copy_block9(full, src, 16, stride, 9);\ |
1855 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full , 8, 16, 9);\ | |
984 | 1856 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1857 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1858 OPNAME ## pixels8_l4(dst, full+17, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1859 }\ |
1064 | 1860 static void OPNAME ## qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
1861 uint8_t full[16*9];\ | |
1862 uint8_t halfH[72];\ | |
1863 uint8_t halfHV[64];\ | |
984 | 1864 copy_block9(full, src, 16, stride, 9);\ |
1865 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1866 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1867 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1868 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1869 }\ | |
1064 | 1870 static void OPNAME ## qpel8_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
1871 uint8_t halfH[72];\ | |
1872 uint8_t halfHV[64];\ | |
651 | 1873 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1874 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1875 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ |
1876 }\ | |
1064 | 1877 static void OPNAME ## qpel8_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
1878 uint8_t halfH[72];\ | |
1879 uint8_t halfHV[64];\ | |
651 | 1880 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1881 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1882 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ |
1883 }\ | |
1064 | 1884 void ff_ ## OPNAME ## qpel8_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1885 uint8_t full[16*9];\ | |
1886 uint8_t halfH[72];\ | |
1887 uint8_t halfV[64];\ | |
1888 uint8_t halfHV[64];\ | |
651 | 1889 copy_block9(full, src, 16, stride, 9);\ |
1890 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1891 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1892 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1893 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
255 | 1894 }\ |
1064 | 1895 static void OPNAME ## qpel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
1896 uint8_t full[16*9];\ | |
1897 uint8_t halfH[72];\ | |
984 | 1898 copy_block9(full, src, 16, stride, 9);\ |
1899 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1900 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1901 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1902 }\ | |
1064 | 1903 void ff_ ## OPNAME ## qpel8_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1904 uint8_t full[16*9];\ | |
1905 uint8_t halfH[72];\ | |
1906 uint8_t halfV[64];\ | |
1907 uint8_t halfHV[64];\ | |
651 | 1908 copy_block9(full, src, 16, stride, 9);\ |
1909 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1910 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1911 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1912 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
1913 }\ | |
1064 | 1914 static void OPNAME ## qpel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
1915 uint8_t full[16*9];\ | |
1916 uint8_t halfH[72];\ | |
984 | 1917 copy_block9(full, src, 16, stride, 9);\ |
1918 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1919 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1920 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1921 }\ | |
1064 | 1922 static void OPNAME ## qpel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
1923 uint8_t halfH[72];\ | |
651 | 1924 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1925 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ |
651 | 1926 }\ |
1064 | 1927 static void OPNAME ## qpel16_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1928 OPNAME ## pixels16_c(dst, src, stride, 16);\ |
255 | 1929 }\ |
651 | 1930 \ |
1064 | 1931 static void OPNAME ## qpel16_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
1932 uint8_t half[256];\ | |
651 | 1933 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
1934 OPNAME ## pixels16_l2(dst, src, half, stride, stride, 16, 16);\ | |
1935 }\ | |
1936 \ | |
1064 | 1937 static void OPNAME ## qpel16_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 1938 OPNAME ## mpeg4_qpel16_h_lowpass(dst, src, stride, stride, 16);\ |
1939 }\ | |
1940 \ | |
1064 | 1941 static void OPNAME ## qpel16_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
1942 uint8_t half[256];\ | |
651 | 1943 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
1944 OPNAME ## pixels16_l2(dst, src+1, half, stride, stride, 16, 16);\ | |
1945 }\ | |
1946 \ | |
1064 | 1947 static void OPNAME ## qpel16_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
1948 uint8_t full[24*17];\ | |
1949 uint8_t half[256];\ | |
651 | 1950 copy_block17(full, src, 24, stride, 17);\ |
954 | 1951 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 1952 OPNAME ## pixels16_l2(dst, full, half, stride, 24, 16, 16);\ |
255 | 1953 }\ |
651 | 1954 \ |
1064 | 1955 static void OPNAME ## qpel16_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
1956 uint8_t full[24*17];\ | |
651 | 1957 copy_block17(full, src, 24, stride, 17);\ |
954 | 1958 OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24);\ |
651 | 1959 }\ |
1960 \ | |
1064 | 1961 static void OPNAME ## qpel16_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
1962 uint8_t full[24*17];\ | |
1963 uint8_t half[256];\ | |
651 | 1964 copy_block17(full, src, 24, stride, 17);\ |
954 | 1965 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 1966 OPNAME ## pixels16_l2(dst, full+24, half, stride, 24, 16, 16);\ |
255 | 1967 }\ |
1064 | 1968 void ff_ ## OPNAME ## qpel16_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1969 uint8_t full[24*17];\ | |
1970 uint8_t halfH[272];\ | |
1971 uint8_t halfV[256];\ | |
1972 uint8_t halfHV[256];\ | |
651 | 1973 copy_block17(full, src, 24, stride, 17);\ |
1974 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1975 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
1976 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1977 OPNAME ## pixels16_l4(dst, full, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
1978 }\ | |
1064 | 1979 static void OPNAME ## qpel16_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
1980 uint8_t full[24*17];\ | |
1981 uint8_t halfH[272];\ | |
1982 uint8_t halfHV[256];\ | |
984 | 1983 copy_block17(full, src, 24, stride, 17);\ |
1984 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1985 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
1986 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1987 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
1988 }\ | |
1064 | 1989 void ff_ ## OPNAME ## qpel16_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1990 uint8_t full[24*17];\ | |
1991 uint8_t halfH[272];\ | |
1992 uint8_t halfV[256];\ | |
1993 uint8_t halfHV[256];\ | |
651 | 1994 copy_block17(full, src, 24, stride, 17);\ |
1995 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1996 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
1997 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1998 OPNAME ## pixels16_l4(dst, full+1, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
1999 }\ | |
1064 | 2000 static void OPNAME ## qpel16_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
2001 uint8_t full[24*17];\ | |
2002 uint8_t halfH[272];\ | |
2003 uint8_t halfHV[256];\ | |
984 | 2004 copy_block17(full, src, 24, stride, 17);\ |
2005 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2006 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2007 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
2008 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
2009 }\ | |
1064 | 2010 void ff_ ## OPNAME ## qpel16_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2011 uint8_t full[24*17];\ | |
2012 uint8_t halfH[272];\ | |
2013 uint8_t halfV[256];\ | |
2014 uint8_t halfHV[256];\ | |
651 | 2015 copy_block17(full, src, 24, stride, 17);\ |
2016 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2017 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
2018 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2019 OPNAME ## pixels16_l4(dst, full+24, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
255 | 2020 }\ |
1064 | 2021 static void OPNAME ## qpel16_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
2022 uint8_t full[24*17];\ | |
2023 uint8_t halfH[272];\ | |
2024 uint8_t halfHV[256];\ | |
984 | 2025 copy_block17(full, src, 24, stride, 17);\ |
2026 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2027 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
2028 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
2029 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
2030 }\ | |
1064 | 2031 void ff_ ## OPNAME ## qpel16_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2032 uint8_t full[24*17];\ | |
2033 uint8_t halfH[272];\ | |
2034 uint8_t halfV[256];\ | |
2035 uint8_t halfHV[256];\ | |
651 | 2036 copy_block17(full, src, 24, stride, 17);\ |
2037 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full , 16, 24, 17);\ | |
954 | 2038 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
2039 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2040 OPNAME ## pixels16_l4(dst, full+25, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
2041 }\ | |
1064 | 2042 static void OPNAME ## qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
2043 uint8_t full[24*17];\ | |
2044 uint8_t halfH[272];\ | |
2045 uint8_t halfHV[256];\ | |
984 | 2046 copy_block17(full, src, 24, stride, 17);\ |
2047 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2048 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2049 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
2050 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
2051 }\ | |
1064 | 2052 static void OPNAME ## qpel16_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
2053 uint8_t halfH[272];\ | |
2054 uint8_t halfHV[256];\ | |
651 | 2055 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2056 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 2057 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ |
255 | 2058 }\ |
1064 | 2059 static void OPNAME ## qpel16_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
2060 uint8_t halfH[272];\ | |
2061 uint8_t halfHV[256];\ | |
651 | 2062 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2063 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 2064 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ |
2065 }\ | |
1064 | 2066 void ff_ ## OPNAME ## qpel16_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2067 uint8_t full[24*17];\ | |
2068 uint8_t halfH[272];\ | |
2069 uint8_t halfV[256];\ | |
2070 uint8_t halfHV[256];\ | |
651 | 2071 copy_block17(full, src, 24, stride, 17);\ |
2072 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2073 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
2074 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2075 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
255 | 2076 }\ |
1064 | 2077 static void OPNAME ## qpel16_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
2078 uint8_t full[24*17];\ | |
2079 uint8_t halfH[272];\ | |
984 | 2080 copy_block17(full, src, 24, stride, 17);\ |
2081 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2082 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
2083 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2084 }\ | |
1064 | 2085 void ff_ ## OPNAME ## qpel16_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2086 uint8_t full[24*17];\ | |
2087 uint8_t halfH[272];\ | |
2088 uint8_t halfV[256];\ | |
2089 uint8_t halfHV[256];\ | |
651 | 2090 copy_block17(full, src, 24, stride, 17);\ |
2091 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2092 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
2093 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2094 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
2095 }\ | |
1064 | 2096 static void OPNAME ## qpel16_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
2097 uint8_t full[24*17];\ | |
2098 uint8_t halfH[272];\ | |
984 | 2099 copy_block17(full, src, 24, stride, 17);\ |
2100 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2101 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2102 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2103 }\ | |
1064 | 2104 static void OPNAME ## qpel16_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
2105 uint8_t halfH[272];\ | |
651 | 2106 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2107 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ |
859 | 2108 } |
255 | 2109 |
651 | 2110 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) |
2111 #define op_avg_no_rnd(a, b) a = (((a)+cm[((b) + 15)>>5])>>1) | |
2112 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2113 #define op_put_no_rnd(a, b) a = cm[((b) + 15)>>5] | |
2114 | |
2115 QPEL_MC(0, put_ , _ , op_put) | |
2116 QPEL_MC(1, put_no_rnd_, _no_rnd_, op_put_no_rnd) | |
2117 QPEL_MC(0, avg_ , _ , op_avg) | |
2118 //QPEL_MC(1, avg_no_rnd , _ , op_avg) | |
2119 #undef op_avg | |
2120 #undef op_avg_no_rnd | |
2121 #undef op_put | |
2122 #undef op_put_no_rnd | |
255 | 2123 |
1168 | 2124 #if 1 |
2125 #define H264_LOWPASS(OPNAME, OP, OP2) \ | |
5151 | 2126 static av_unused void OPNAME ## h264_qpel2_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2127 const int h=2;\ |
4176 | 2128 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2129 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2130 for(i=0; i<h; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2131 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2132 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]));\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2133 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]));\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2134 dst+=dstStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2135 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2136 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2137 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2138 \ |
5151 | 2139 static av_unused void OPNAME ## h264_qpel2_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2140 const int w=2;\ |
4176 | 2141 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2142 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2143 for(i=0; i<w; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2144 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2145 const int srcB= src[-2*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2146 const int srcA= src[-1*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2147 const int src0= src[0 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2148 const int src1= src[1 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2149 const int src2= src[2 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2150 const int src3= src[3 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2151 const int src4= src[4 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2152 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2153 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2154 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2155 src++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2156 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2157 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2158 \ |
5151 | 2159 static av_unused void OPNAME ## h264_qpel2_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2160 const int h=2;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2161 const int w=2;\ |
4176 | 2162 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2163 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2164 src -= 2*srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2165 for(i=0; i<h+5; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2166 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2167 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]);\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2168 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]);\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2169 tmp+=tmpStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2170 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2171 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2172 tmp -= tmpStride*(h+5-2);\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2173 for(i=0; i<w; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2174 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2175 const int tmpB= tmp[-2*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2176 const int tmpA= tmp[-1*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2177 const int tmp0= tmp[0 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2178 const int tmp1= tmp[1 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2179 const int tmp2= tmp[2 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2180 const int tmp3= tmp[3 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2181 const int tmp4= tmp[4 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2182 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2183 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2184 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2185 tmp++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2186 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2187 }\ |
1168 | 2188 static void OPNAME ## h264_qpel4_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
2189 const int h=4;\ | |
4176 | 2190 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2191 int i;\ |
2192 for(i=0; i<h; i++)\ | |
2193 {\ | |
2194 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]));\ | |
2195 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]));\ | |
2196 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]));\ | |
2197 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]));\ | |
2198 dst+=dstStride;\ | |
2199 src+=srcStride;\ | |
2200 }\ | |
2201 }\ | |
2202 \ | |
2203 static void OPNAME ## h264_qpel4_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2204 const int w=4;\ | |
4176 | 2205 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2206 int i;\ |
2207 for(i=0; i<w; i++)\ | |
2208 {\ | |
2209 const int srcB= src[-2*srcStride];\ | |
2210 const int srcA= src[-1*srcStride];\ | |
2211 const int src0= src[0 *srcStride];\ | |
2212 const int src1= src[1 *srcStride];\ | |
2213 const int src2= src[2 *srcStride];\ | |
2214 const int src3= src[3 *srcStride];\ | |
2215 const int src4= src[4 *srcStride];\ | |
2216 const int src5= src[5 *srcStride];\ | |
2217 const int src6= src[6 *srcStride];\ | |
2218 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2219 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2220 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2221 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2222 dst++;\ | |
2223 src++;\ | |
2224 }\ | |
2225 }\ | |
2226 \ | |
2227 static void OPNAME ## h264_qpel4_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2228 const int h=4;\ | |
2229 const int w=4;\ | |
4176 | 2230 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2231 int i;\ |
2232 src -= 2*srcStride;\ | |
2233 for(i=0; i<h+5; i++)\ | |
2234 {\ | |
2235 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]);\ | |
2236 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]);\ | |
2237 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]);\ | |
2238 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]);\ | |
2239 tmp+=tmpStride;\ | |
2240 src+=srcStride;\ | |
2241 }\ | |
2242 tmp -= tmpStride*(h+5-2);\ | |
2243 for(i=0; i<w; i++)\ | |
2244 {\ | |
2245 const int tmpB= tmp[-2*tmpStride];\ | |
2246 const int tmpA= tmp[-1*tmpStride];\ | |
2247 const int tmp0= tmp[0 *tmpStride];\ | |
2248 const int tmp1= tmp[1 *tmpStride];\ | |
2249 const int tmp2= tmp[2 *tmpStride];\ | |
2250 const int tmp3= tmp[3 *tmpStride];\ | |
2251 const int tmp4= tmp[4 *tmpStride];\ | |
2252 const int tmp5= tmp[5 *tmpStride];\ | |
2253 const int tmp6= tmp[6 *tmpStride];\ | |
2254 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2255 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2256 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2257 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2258 dst++;\ | |
2259 tmp++;\ | |
2260 }\ | |
2261 }\ | |
2262 \ | |
2263 static void OPNAME ## h264_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2264 const int h=8;\ | |
4176 | 2265 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2266 int i;\ |
2267 for(i=0; i<h; i++)\ | |
2268 {\ | |
2269 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]));\ | |
2270 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]));\ | |
2271 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]));\ | |
2272 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]));\ | |
2273 OP(dst[4], (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]));\ | |
2274 OP(dst[5], (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]));\ | |
2275 OP(dst[6], (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]));\ | |
2276 OP(dst[7], (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]));\ | |
2277 dst+=dstStride;\ | |
2278 src+=srcStride;\ | |
2279 }\ | |
2280 }\ | |
2281 \ | |
2282 static void OPNAME ## h264_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2283 const int w=8;\ | |
4176 | 2284 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2285 int i;\ |
2286 for(i=0; i<w; i++)\ | |
2287 {\ | |
2288 const int srcB= src[-2*srcStride];\ | |
2289 const int srcA= src[-1*srcStride];\ | |
2290 const int src0= src[0 *srcStride];\ | |
2291 const int src1= src[1 *srcStride];\ | |
2292 const int src2= src[2 *srcStride];\ | |
2293 const int src3= src[3 *srcStride];\ | |
2294 const int src4= src[4 *srcStride];\ | |
2295 const int src5= src[5 *srcStride];\ | |
2296 const int src6= src[6 *srcStride];\ | |
2297 const int src7= src[7 *srcStride];\ | |
2298 const int src8= src[8 *srcStride];\ | |
2299 const int src9= src[9 *srcStride];\ | |
2300 const int src10=src[10*srcStride];\ | |
2301 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2302 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2303 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2304 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2305 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*5 + (src2+src7));\ | |
2306 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*5 + (src3+src8));\ | |
2307 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*5 + (src4+src9));\ | |
2308 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src9)*5 + (src5+src10));\ | |
2309 dst++;\ | |
2310 src++;\ | |
2311 }\ | |
2312 }\ | |
2313 \ | |
2314 static void OPNAME ## h264_qpel8_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2315 const int h=8;\ | |
2316 const int w=8;\ | |
4176 | 2317 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2318 int i;\ |
2319 src -= 2*srcStride;\ | |
2320 for(i=0; i<h+5; i++)\ | |
2321 {\ | |
2322 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]);\ | |
2323 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]);\ | |
2324 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]);\ | |
2325 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]);\ | |
2326 tmp[4]= (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]);\ | |
2327 tmp[5]= (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]);\ | |
2328 tmp[6]= (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]);\ | |
2329 tmp[7]= (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]);\ | |
2330 tmp+=tmpStride;\ | |
2331 src+=srcStride;\ | |
2332 }\ | |
2333 tmp -= tmpStride*(h+5-2);\ | |
2334 for(i=0; i<w; i++)\ | |
2335 {\ | |
2336 const int tmpB= tmp[-2*tmpStride];\ | |
2337 const int tmpA= tmp[-1*tmpStride];\ | |
2338 const int tmp0= tmp[0 *tmpStride];\ | |
2339 const int tmp1= tmp[1 *tmpStride];\ | |
2340 const int tmp2= tmp[2 *tmpStride];\ | |
2341 const int tmp3= tmp[3 *tmpStride];\ | |
2342 const int tmp4= tmp[4 *tmpStride];\ | |
2343 const int tmp5= tmp[5 *tmpStride];\ | |
2344 const int tmp6= tmp[6 *tmpStride];\ | |
2345 const int tmp7= tmp[7 *tmpStride];\ | |
2346 const int tmp8= tmp[8 *tmpStride];\ | |
2347 const int tmp9= tmp[9 *tmpStride];\ | |
2348 const int tmp10=tmp[10*tmpStride];\ | |
2349 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2350 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2351 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2352 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2353 OP2(dst[4*dstStride], (tmp4+tmp5)*20 - (tmp3+tmp6)*5 + (tmp2+tmp7));\ | |
2354 OP2(dst[5*dstStride], (tmp5+tmp6)*20 - (tmp4+tmp7)*5 + (tmp3+tmp8));\ | |
2355 OP2(dst[6*dstStride], (tmp6+tmp7)*20 - (tmp5+tmp8)*5 + (tmp4+tmp9));\ | |
2356 OP2(dst[7*dstStride], (tmp7+tmp8)*20 - (tmp6+tmp9)*5 + (tmp5+tmp10));\ | |
2357 dst++;\ | |
2358 tmp++;\ | |
2359 }\ | |
2360 }\ | |
2361 \ | |
2362 static void OPNAME ## h264_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2363 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2364 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2365 src += 8*srcStride;\ | |
2366 dst += 8*dstStride;\ | |
2367 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2368 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2369 }\ | |
2370 \ | |
2371 static void OPNAME ## h264_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2372 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2373 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2374 src += 8*srcStride;\ | |
2375 dst += 8*dstStride;\ | |
2376 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2377 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2378 }\ | |
2379 \ | |
2380 static void OPNAME ## h264_qpel16_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2381 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2382 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2383 src += 8*srcStride;\ | |
2384 dst += 8*dstStride;\ | |
2385 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2386 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2387 }\ | |
2388 | |
2389 #define H264_MC(OPNAME, SIZE) \ | |
2390 static void OPNAME ## h264_qpel ## SIZE ## _mc00_c (uint8_t *dst, uint8_t *src, int stride){\ | |
2391 OPNAME ## pixels ## SIZE ## _c(dst, src, stride, SIZE);\ | |
2392 }\ | |
2393 \ | |
2394 static void OPNAME ## h264_qpel ## SIZE ## _mc10_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2395 uint8_t half[SIZE*SIZE];\ | |
2396 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2397 OPNAME ## pixels ## SIZE ## _l2(dst, src, half, stride, stride, SIZE, SIZE);\ | |
2398 }\ | |
2399 \ | |
2400 static void OPNAME ## h264_qpel ## SIZE ## _mc20_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2401 OPNAME ## h264_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride);\ | |
2402 }\ | |
2403 \ | |
2404 static void OPNAME ## h264_qpel ## SIZE ## _mc30_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2405 uint8_t half[SIZE*SIZE];\ | |
2406 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2407 OPNAME ## pixels ## SIZE ## _l2(dst, src+1, half, stride, stride, SIZE, SIZE);\ | |
2408 }\ | |
2409 \ | |
2410 static void OPNAME ## h264_qpel ## SIZE ## _mc01_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2411 uint8_t full[SIZE*(SIZE+5)];\ | |
2412 uint8_t * const full_mid= full + SIZE*2;\ | |
2413 uint8_t half[SIZE*SIZE];\ | |
2414 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2415 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2416 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid, half, stride, SIZE, SIZE, SIZE);\ | |
2417 }\ | |
2418 \ | |
2419 static void OPNAME ## h264_qpel ## SIZE ## _mc02_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2420 uint8_t full[SIZE*(SIZE+5)];\ | |
2421 uint8_t * const full_mid= full + SIZE*2;\ | |
2422 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2423 OPNAME ## h264_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE);\ | |
2424 }\ | |
2425 \ | |
2426 static void OPNAME ## h264_qpel ## SIZE ## _mc03_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2427 uint8_t full[SIZE*(SIZE+5)];\ | |
2428 uint8_t * const full_mid= full + SIZE*2;\ | |
2429 uint8_t half[SIZE*SIZE];\ | |
2430 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2431 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2432 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid+SIZE, half, stride, SIZE, SIZE, SIZE);\ | |
2433 }\ | |
2434 \ | |
2435 static void OPNAME ## h264_qpel ## SIZE ## _mc11_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2436 uint8_t full[SIZE*(SIZE+5)];\ | |
2437 uint8_t * const full_mid= full + SIZE*2;\ | |
2438 uint8_t halfH[SIZE*SIZE];\ | |
2439 uint8_t halfV[SIZE*SIZE];\ | |
2440 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2441 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2442 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2443 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2444 }\ | |
2445 \ | |
2446 static void OPNAME ## h264_qpel ## SIZE ## _mc31_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2447 uint8_t full[SIZE*(SIZE+5)];\ | |
2448 uint8_t * const full_mid= full + SIZE*2;\ | |
2449 uint8_t halfH[SIZE*SIZE];\ | |
2450 uint8_t halfV[SIZE*SIZE];\ | |
2451 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2452 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2453 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2454 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2455 }\ | |
2456 \ | |
2457 static void OPNAME ## h264_qpel ## SIZE ## _mc13_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2458 uint8_t full[SIZE*(SIZE+5)];\ | |
2459 uint8_t * const full_mid= full + SIZE*2;\ | |
2460 uint8_t halfH[SIZE*SIZE];\ | |
2461 uint8_t halfV[SIZE*SIZE];\ | |
2462 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2463 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2464 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2465 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2466 }\ | |
2467 \ | |
2468 static void OPNAME ## h264_qpel ## SIZE ## _mc33_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2469 uint8_t full[SIZE*(SIZE+5)];\ | |
2470 uint8_t * const full_mid= full + SIZE*2;\ | |
2471 uint8_t halfH[SIZE*SIZE];\ | |
2472 uint8_t halfV[SIZE*SIZE];\ | |
2473 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2474 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2475 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2476 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2477 }\ | |
2478 \ | |
2479 static void OPNAME ## h264_qpel ## SIZE ## _mc22_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2480 int16_t tmp[SIZE*(SIZE+5)];\ | |
2481 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass(dst, tmp, src, stride, SIZE, stride);\ | |
2482 }\ | |
2483 \ | |
2484 static void OPNAME ## h264_qpel ## SIZE ## _mc21_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2485 int16_t tmp[SIZE*(SIZE+5)];\ | |
2486 uint8_t halfH[SIZE*SIZE];\ | |
2487 uint8_t halfHV[SIZE*SIZE];\ | |
2488 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2489 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2490 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2491 }\ | |
2492 \ | |
2493 static void OPNAME ## h264_qpel ## SIZE ## _mc23_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2494 int16_t tmp[SIZE*(SIZE+5)];\ | |
2495 uint8_t halfH[SIZE*SIZE];\ | |
2496 uint8_t halfHV[SIZE*SIZE];\ | |
2497 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2498 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2499 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2500 }\ | |
2501 \ | |
2502 static void OPNAME ## h264_qpel ## SIZE ## _mc12_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2503 uint8_t full[SIZE*(SIZE+5)];\ | |
2504 uint8_t * const full_mid= full + SIZE*2;\ | |
2505 int16_t tmp[SIZE*(SIZE+5)];\ | |
2506 uint8_t halfV[SIZE*SIZE];\ | |
2507 uint8_t halfHV[SIZE*SIZE];\ | |
2508 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2509 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2510 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2511 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2512 }\ | |
2513 \ | |
2514 static void OPNAME ## h264_qpel ## SIZE ## _mc32_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2515 uint8_t full[SIZE*(SIZE+5)];\ | |
2516 uint8_t * const full_mid= full + SIZE*2;\ | |
2517 int16_t tmp[SIZE*(SIZE+5)];\ | |
2518 uint8_t halfV[SIZE*SIZE];\ | |
2519 uint8_t halfHV[SIZE*SIZE];\ | |
2520 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2521 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2522 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2523 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2524 }\ | |
2525 | |
2526 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) | |
2527 //#define op_avg2(a, b) a = (((a)*w1+cm[((b) + 16)>>5]*w2 + o + 64)>>7) | |
2528 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2529 #define op2_avg(a, b) a = (((a)+cm[((b) + 512)>>10]+1)>>1) | |
2530 #define op2_put(a, b) a = cm[((b) + 512)>>10] | |
2531 | |
2532 H264_LOWPASS(put_ , op_put, op2_put) | |
2533 H264_LOWPASS(avg_ , op_avg, op2_avg) | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2534 H264_MC(put_, 2) |
1168 | 2535 H264_MC(put_, 4) |
2536 H264_MC(put_, 8) | |
2537 H264_MC(put_, 16) | |
2538 H264_MC(avg_, 4) | |
2539 H264_MC(avg_, 8) | |
2540 H264_MC(avg_, 16) | |
2541 | |
2542 #undef op_avg | |
2543 #undef op_put | |
2544 #undef op2_avg | |
2545 #undef op2_put | |
2546 #endif | |
2547 | |
4594 | 2548 #define op_scale1(x) block[x] = av_clip_uint8( (block[x]*weight + offset) >> log2_denom ) |
2549 #define op_scale2(x) dst[x] = av_clip_uint8( (src[x]*weights + dst[x]*weightd + offset) >> (log2_denom+1)) | |
2415 | 2550 #define H264_WEIGHT(W,H) \ |
2551 static void weight_h264_pixels ## W ## x ## H ## _c(uint8_t *block, int stride, int log2_denom, int weight, int offset){ \ | |
3029 | 2552 int y; \ |
2415 | 2553 offset <<= log2_denom; \ |
2554 if(log2_denom) offset += 1<<(log2_denom-1); \ | |
2555 for(y=0; y<H; y++, block += stride){ \ | |
2556 op_scale1(0); \ | |
2557 op_scale1(1); \ | |
2558 if(W==2) continue; \ | |
2559 op_scale1(2); \ | |
2560 op_scale1(3); \ | |
2561 if(W==4) continue; \ | |
2562 op_scale1(4); \ | |
2563 op_scale1(5); \ | |
2564 op_scale1(6); \ | |
2565 op_scale1(7); \ | |
2566 if(W==8) continue; \ | |
2567 op_scale1(8); \ | |
2568 op_scale1(9); \ | |
2569 op_scale1(10); \ | |
2570 op_scale1(11); \ | |
2571 op_scale1(12); \ | |
2572 op_scale1(13); \ | |
2573 op_scale1(14); \ | |
2574 op_scale1(15); \ | |
2575 } \ | |
2576 } \ | |
3029 | 2577 static void biweight_h264_pixels ## W ## x ## H ## _c(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset){ \ |
2578 int y; \ | |
2579 offset = ((offset + 1) | 1) << log2_denom; \ | |
2415 | 2580 for(y=0; y<H; y++, dst += stride, src += stride){ \ |
2581 op_scale2(0); \ | |
2582 op_scale2(1); \ | |
2583 if(W==2) continue; \ | |
2584 op_scale2(2); \ | |
2585 op_scale2(3); \ | |
2586 if(W==4) continue; \ | |
2587 op_scale2(4); \ | |
2588 op_scale2(5); \ | |
2589 op_scale2(6); \ | |
2590 op_scale2(7); \ | |
2591 if(W==8) continue; \ | |
2592 op_scale2(8); \ | |
2593 op_scale2(9); \ | |
2594 op_scale2(10); \ | |
2595 op_scale2(11); \ | |
2596 op_scale2(12); \ | |
2597 op_scale2(13); \ | |
2598 op_scale2(14); \ | |
2599 op_scale2(15); \ | |
2600 } \ | |
2601 } | |
2602 | |
2603 H264_WEIGHT(16,16) | |
2604 H264_WEIGHT(16,8) | |
2605 H264_WEIGHT(8,16) | |
2606 H264_WEIGHT(8,8) | |
2607 H264_WEIGHT(8,4) | |
2608 H264_WEIGHT(4,8) | |
2609 H264_WEIGHT(4,4) | |
2610 H264_WEIGHT(4,2) | |
2611 H264_WEIGHT(2,4) | |
2612 H264_WEIGHT(2,2) | |
2613 | |
2614 #undef op_scale1 | |
2615 #undef op_scale2 | |
2616 #undef H264_WEIGHT | |
2617 | |
936 | 2618 static void wmv2_mspel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){ |
4176 | 2619 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
936 | 2620 int i; |
2621 | |
2622 for(i=0; i<h; i++){ | |
2623 dst[0]= cm[(9*(src[0] + src[1]) - (src[-1] + src[2]) + 8)>>4]; | |
2624 dst[1]= cm[(9*(src[1] + src[2]) - (src[ 0] + src[3]) + 8)>>4]; | |
2625 dst[2]= cm[(9*(src[2] + src[3]) - (src[ 1] + src[4]) + 8)>>4]; | |
2626 dst[3]= cm[(9*(src[3] + src[4]) - (src[ 2] + src[5]) + 8)>>4]; | |
2627 dst[4]= cm[(9*(src[4] + src[5]) - (src[ 3] + src[6]) + 8)>>4]; | |
2628 dst[5]= cm[(9*(src[5] + src[6]) - (src[ 4] + src[7]) + 8)>>4]; | |
2629 dst[6]= cm[(9*(src[6] + src[7]) - (src[ 5] + src[8]) + 8)>>4]; | |
2630 dst[7]= cm[(9*(src[7] + src[8]) - (src[ 6] + src[9]) + 8)>>4]; | |
2631 dst+=dstStride; | |
2967 | 2632 src+=srcStride; |
936 | 2633 } |
2634 } | |
2635 | |
3432 | 2636 #ifdef CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2637 /* AVS specific */ |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2638 void ff_cavsdsp_init(DSPContext* c, AVCodecContext *avctx); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2639 |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2640 void ff_put_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride) { |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2641 put_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2642 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2643 void ff_avg_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride) { |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2644 avg_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2645 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2646 void ff_put_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride) { |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2647 put_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2648 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2649 void ff_avg_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride) { |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2650 avg_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2651 } |
3432 | 2652 #endif /* CONFIG_CAVS_DECODER */ |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2653 |
3526 | 2654 #if defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
2655 /* VC-1 specific */ | |
2656 void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx); | |
2657 | |
2658 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, uint8_t *src, int stride, int rnd) { | |
2659 put_pixels8_c(dst, src, stride, 8); | |
2660 } | |
2661 #endif /* CONFIG_VC1_DECODER||CONFIG_WMV3_DECODER */ | |
2662 | |
5887 | 2663 void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx); |
5899 | 2664 |
4296 | 2665 /* H264 specific */ |
5411
362aec4ef932
Take care of some renames (Doxygen and function name) after the previous pure rename patch.
takis
parents:
5394
diff
changeset
|
2666 void ff_h264dspenc_init(DSPContext* c, AVCodecContext *avctx); |
4296 | 2667 |
936 | 2668 static void wmv2_mspel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int w){ |
4176 | 2669 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
936 | 2670 int i; |
2671 | |
2672 for(i=0; i<w; i++){ | |
2673 const int src_1= src[ -srcStride]; | |
2674 const int src0 = src[0 ]; | |
2675 const int src1 = src[ srcStride]; | |
2676 const int src2 = src[2*srcStride]; | |
2677 const int src3 = src[3*srcStride]; | |
2678 const int src4 = src[4*srcStride]; | |
2679 const int src5 = src[5*srcStride]; | |
2680 const int src6 = src[6*srcStride]; | |
2681 const int src7 = src[7*srcStride]; | |
2682 const int src8 = src[8*srcStride]; | |
2683 const int src9 = src[9*srcStride]; | |
2684 dst[0*dstStride]= cm[(9*(src0 + src1) - (src_1 + src2) + 8)>>4]; | |
2685 dst[1*dstStride]= cm[(9*(src1 + src2) - (src0 + src3) + 8)>>4]; | |
2686 dst[2*dstStride]= cm[(9*(src2 + src3) - (src1 + src4) + 8)>>4]; | |
2687 dst[3*dstStride]= cm[(9*(src3 + src4) - (src2 + src5) + 8)>>4]; | |
2688 dst[4*dstStride]= cm[(9*(src4 + src5) - (src3 + src6) + 8)>>4]; | |
2689 dst[5*dstStride]= cm[(9*(src5 + src6) - (src4 + src7) + 8)>>4]; | |
2690 dst[6*dstStride]= cm[(9*(src6 + src7) - (src5 + src8) + 8)>>4]; | |
2691 dst[7*dstStride]= cm[(9*(src7 + src8) - (src6 + src9) + 8)>>4]; | |
2692 src++; | |
2693 dst++; | |
2694 } | |
2695 } | |
2696 | |
2697 static void put_mspel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){ | |
2698 put_pixels8_c(dst, src, stride, 8); | |
2699 } | |
2700 | |
2701 static void put_mspel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){ | |
2702 uint8_t half[64]; | |
2703 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2704 put_pixels8_l2(dst, src, half, stride, stride, 8, 8); | |
2705 } | |
2706 | |
2707 static void put_mspel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){ | |
2708 wmv2_mspel8_h_lowpass(dst, src, stride, stride, 8); | |
2709 } | |
2710 | |
2711 static void put_mspel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){ | |
2712 uint8_t half[64]; | |
2713 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2714 put_pixels8_l2(dst, src+1, half, stride, stride, 8, 8); | |
2715 } | |
2716 | |
2717 static void put_mspel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){ | |
2718 wmv2_mspel8_v_lowpass(dst, src, stride, stride, 8); | |
2719 } | |
2720 | |
2721 static void put_mspel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){ | |
2722 uint8_t halfH[88]; | |
2723 uint8_t halfV[64]; | |
2724 uint8_t halfHV[64]; | |
2725 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2726 wmv2_mspel8_v_lowpass(halfV, src, 8, stride, 8); | |
2727 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2728 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2729 } | |
2730 static void put_mspel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){ | |
2731 uint8_t halfH[88]; | |
2732 uint8_t halfV[64]; | |
2733 uint8_t halfHV[64]; | |
2734 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2735 wmv2_mspel8_v_lowpass(halfV, src+1, 8, stride, 8); | |
2736 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2737 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2738 } | |
2739 static void put_mspel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){ | |
2740 uint8_t halfH[88]; | |
2741 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2742 wmv2_mspel8_v_lowpass(dst, halfH+8, stride, 8, 8); | |
2743 } | |
2744 | |
1644 | 2745 static void h263_v_loop_filter_c(uint8_t *src, int stride, int qscale){ |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2746 if(ENABLE_ANY_H263) { |
1644 | 2747 int x; |
2748 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2749 |
1644 | 2750 for(x=0; x<8; x++){ |
2751 int d1, d2, ad1; | |
2752 int p0= src[x-2*stride]; | |
2753 int p1= src[x-1*stride]; | |
2754 int p2= src[x+0*stride]; | |
2755 int p3= src[x+1*stride]; | |
2756 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2757 | |
2758 if (d<-2*strength) d1= 0; | |
2759 else if(d<- strength) d1=-2*strength - d; | |
2760 else if(d< strength) d1= d; | |
2761 else if(d< 2*strength) d1= 2*strength - d; | |
2762 else d1= 0; | |
2967 | 2763 |
1644 | 2764 p1 += d1; |
2765 p2 -= d1; | |
2766 if(p1&256) p1= ~(p1>>31); | |
2767 if(p2&256) p2= ~(p2>>31); | |
2967 | 2768 |
1644 | 2769 src[x-1*stride] = p1; |
2770 src[x+0*stride] = p2; | |
2771 | |
4001 | 2772 ad1= FFABS(d1)>>1; |
2967 | 2773 |
4594 | 2774 d2= av_clip((p0-p3)/4, -ad1, ad1); |
2967 | 2775 |
1644 | 2776 src[x-2*stride] = p0 - d2; |
2777 src[x+ stride] = p3 + d2; | |
2778 } | |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2779 } |
1644 | 2780 } |
2781 | |
2782 static void h263_h_loop_filter_c(uint8_t *src, int stride, int qscale){ | |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2783 if(ENABLE_ANY_H263) { |
1644 | 2784 int y; |
2785 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2786 |
1644 | 2787 for(y=0; y<8; y++){ |
2788 int d1, d2, ad1; | |
2789 int p0= src[y*stride-2]; | |
2790 int p1= src[y*stride-1]; | |
2791 int p2= src[y*stride+0]; | |
2792 int p3= src[y*stride+1]; | |
2793 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2794 | |
2795 if (d<-2*strength) d1= 0; | |
2796 else if(d<- strength) d1=-2*strength - d; | |
2797 else if(d< strength) d1= d; | |
2798 else if(d< 2*strength) d1= 2*strength - d; | |
2799 else d1= 0; | |
2967 | 2800 |
1644 | 2801 p1 += d1; |
2802 p2 -= d1; | |
2803 if(p1&256) p1= ~(p1>>31); | |
2804 if(p2&256) p2= ~(p2>>31); | |
2967 | 2805 |
1644 | 2806 src[y*stride-1] = p1; |
2807 src[y*stride+0] = p2; | |
2808 | |
4001 | 2809 ad1= FFABS(d1)>>1; |
2967 | 2810 |
4594 | 2811 d2= av_clip((p0-p3)/4, -ad1, ad1); |
2967 | 2812 |
1644 | 2813 src[y*stride-2] = p0 - d2; |
2814 src[y*stride+1] = p3 + d2; | |
2815 } | |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2816 } |
1644 | 2817 } |
936 | 2818 |
2045 | 2819 static void h261_loop_filter_c(uint8_t *src, int stride){ |
2820 int x,y,xy,yz; | |
2821 int temp[64]; | |
2822 | |
2823 for(x=0; x<8; x++){ | |
2824 temp[x ] = 4*src[x ]; | |
2825 temp[x + 7*8] = 4*src[x + 7*stride]; | |
2826 } | |
2827 for(y=1; y<7; y++){ | |
2828 for(x=0; x<8; x++){ | |
2829 xy = y * stride + x; | |
2830 yz = y * 8 + x; | |
2831 temp[yz] = src[xy - stride] + 2*src[xy] + src[xy + stride]; | |
2044
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2832 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2833 } |
2967 | 2834 |
2045 | 2835 for(y=0; y<8; y++){ |
2836 src[ y*stride] = (temp[ y*8] + 2)>>2; | |
2837 src[7+y*stride] = (temp[7+y*8] + 2)>>2; | |
2838 for(x=1; x<7; x++){ | |
2839 xy = y * stride + x; | |
2840 yz = y * 8 + x; | |
2841 src[xy] = (temp[yz-1] + 2*temp[yz] + temp[yz+1] + 8)>>4; | |
2044
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2842 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2843 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2844 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2845 |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2846 static inline void h264_loop_filter_luma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0) |
2633 | 2847 { |
2848 int i, d; | |
2849 for( i = 0; i < 4; i++ ) { | |
2850 if( tc0[i] < 0 ) { | |
2851 pix += 4*ystride; | |
2852 continue; | |
2853 } | |
2854 for( d = 0; d < 4; d++ ) { | |
2855 const int p0 = pix[-1*xstride]; | |
2856 const int p1 = pix[-2*xstride]; | |
2857 const int p2 = pix[-3*xstride]; | |
2858 const int q0 = pix[0]; | |
2859 const int q1 = pix[1*xstride]; | |
2860 const int q2 = pix[2*xstride]; | |
2967 | 2861 |
4001 | 2862 if( FFABS( p0 - q0 ) < alpha && |
2863 FFABS( p1 - p0 ) < beta && | |
2864 FFABS( q1 - q0 ) < beta ) { | |
2967 | 2865 |
2633 | 2866 int tc = tc0[i]; |
2867 int i_delta; | |
2967 | 2868 |
4001 | 2869 if( FFABS( p2 - p0 ) < beta ) { |
4594 | 2870 pix[-2*xstride] = p1 + av_clip( (( p2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - p1, -tc0[i], tc0[i] ); |
2633 | 2871 tc++; |
2872 } | |
4001 | 2873 if( FFABS( q2 - q0 ) < beta ) { |
4594 | 2874 pix[ xstride] = q1 + av_clip( (( q2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - q1, -tc0[i], tc0[i] ); |
2633 | 2875 tc++; |
2876 } | |
2967 | 2877 |
4594 | 2878 i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
2879 pix[-xstride] = av_clip_uint8( p0 + i_delta ); /* p0' */ | |
2880 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */ | |
2633 | 2881 } |
2882 pix += ystride; | |
2883 } | |
2884 } | |
2885 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2886 static void h264_v_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2887 { |
2888 h264_loop_filter_luma_c(pix, stride, 1, alpha, beta, tc0); | |
2889 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2890 static void h264_h_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2891 { |
2892 h264_loop_filter_luma_c(pix, 1, stride, alpha, beta, tc0); | |
2893 } | |
2894 | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2895 static inline void h264_loop_filter_chroma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0) |
2633 | 2896 { |
2897 int i, d; | |
2898 for( i = 0; i < 4; i++ ) { | |
2899 const int tc = tc0[i]; | |
2900 if( tc <= 0 ) { | |
2901 pix += 2*ystride; | |
2902 continue; | |
2903 } | |
2904 for( d = 0; d < 2; d++ ) { | |
2905 const int p0 = pix[-1*xstride]; | |
2906 const int p1 = pix[-2*xstride]; | |
2907 const int q0 = pix[0]; | |
2908 const int q1 = pix[1*xstride]; | |
2909 | |
4001 | 2910 if( FFABS( p0 - q0 ) < alpha && |
2911 FFABS( p1 - p0 ) < beta && | |
2912 FFABS( q1 - q0 ) < beta ) { | |
2633 | 2913 |
4594 | 2914 int delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
2915 | |
2916 pix[-xstride] = av_clip_uint8( p0 + delta ); /* p0' */ | |
2917 pix[0] = av_clip_uint8( q0 - delta ); /* q0' */ | |
2633 | 2918 } |
2919 pix += ystride; | |
2920 } | |
2921 } | |
2922 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2923 static void h264_v_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2924 { |
2925 h264_loop_filter_chroma_c(pix, stride, 1, alpha, beta, tc0); | |
2926 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2927 static void h264_h_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2928 { |
2929 h264_loop_filter_chroma_c(pix, 1, stride, alpha, beta, tc0); | |
2930 } | |
2931 | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2932 static inline void h264_loop_filter_chroma_intra_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta) |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2933 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2934 int d; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2935 for( d = 0; d < 8; d++ ) { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2936 const int p0 = pix[-1*xstride]; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2937 const int p1 = pix[-2*xstride]; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2938 const int q0 = pix[0]; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2939 const int q1 = pix[1*xstride]; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2940 |
4001 | 2941 if( FFABS( p0 - q0 ) < alpha && |
2942 FFABS( p1 - p0 ) < beta && | |
2943 FFABS( q1 - q0 ) < beta ) { | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2944 |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2945 pix[-xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */ |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2946 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */ |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2947 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2948 pix += ystride; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2949 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2950 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2951 static void h264_v_loop_filter_chroma_intra_c(uint8_t *pix, int stride, int alpha, int beta) |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2952 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2953 h264_loop_filter_chroma_intra_c(pix, stride, 1, alpha, beta); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2954 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2955 static void h264_h_loop_filter_chroma_intra_c(uint8_t *pix, int stride, int alpha, int beta) |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2956 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2957 h264_loop_filter_chroma_intra_c(pix, 1, stride, alpha, beta); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2958 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2959 |
1708 | 2960 static inline int pix_abs16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2961 { |
2962 int s, i; | |
2963 | |
2964 s = 0; | |
1708 | 2965 for(i=0;i<h;i++) { |
0 | 2966 s += abs(pix1[0] - pix2[0]); |
2967 s += abs(pix1[1] - pix2[1]); | |
2968 s += abs(pix1[2] - pix2[2]); | |
2969 s += abs(pix1[3] - pix2[3]); | |
2970 s += abs(pix1[4] - pix2[4]); | |
2971 s += abs(pix1[5] - pix2[5]); | |
2972 s += abs(pix1[6] - pix2[6]); | |
2973 s += abs(pix1[7] - pix2[7]); | |
2974 s += abs(pix1[8] - pix2[8]); | |
2975 s += abs(pix1[9] - pix2[9]); | |
2976 s += abs(pix1[10] - pix2[10]); | |
2977 s += abs(pix1[11] - pix2[11]); | |
2978 s += abs(pix1[12] - pix2[12]); | |
2979 s += abs(pix1[13] - pix2[13]); | |
2980 s += abs(pix1[14] - pix2[14]); | |
2981 s += abs(pix1[15] - pix2[15]); | |
2982 pix1 += line_size; | |
2983 pix2 += line_size; | |
2984 } | |
2985 return s; | |
2986 } | |
2987 | |
1708 | 2988 static int pix_abs16_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2989 { |
2990 int s, i; | |
2991 | |
2992 s = 0; | |
1708 | 2993 for(i=0;i<h;i++) { |
0 | 2994 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
2995 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
2996 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
2997 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
2998 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
2999 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
3000 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
3001 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
3002 s += abs(pix1[8] - avg2(pix2[8], pix2[9])); | |
3003 s += abs(pix1[9] - avg2(pix2[9], pix2[10])); | |
3004 s += abs(pix1[10] - avg2(pix2[10], pix2[11])); | |
3005 s += abs(pix1[11] - avg2(pix2[11], pix2[12])); | |
3006 s += abs(pix1[12] - avg2(pix2[12], pix2[13])); | |
3007 s += abs(pix1[13] - avg2(pix2[13], pix2[14])); | |
3008 s += abs(pix1[14] - avg2(pix2[14], pix2[15])); | |
3009 s += abs(pix1[15] - avg2(pix2[15], pix2[16])); | |
3010 pix1 += line_size; | |
3011 pix2 += line_size; | |
3012 } | |
3013 return s; | |
3014 } | |
3015 | |
1708 | 3016 static int pix_abs16_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 3017 { |
3018 int s, i; | |
1064 | 3019 uint8_t *pix3 = pix2 + line_size; |
0 | 3020 |
3021 s = 0; | |
1708 | 3022 for(i=0;i<h;i++) { |
0 | 3023 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
3024 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
3025 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
3026 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
3027 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
3028 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
3029 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
3030 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
3031 s += abs(pix1[8] - avg2(pix2[8], pix3[8])); | |
3032 s += abs(pix1[9] - avg2(pix2[9], pix3[9])); | |
3033 s += abs(pix1[10] - avg2(pix2[10], pix3[10])); | |
3034 s += abs(pix1[11] - avg2(pix2[11], pix3[11])); | |
3035 s += abs(pix1[12] - avg2(pix2[12], pix3[12])); | |
3036 s += abs(pix1[13] - avg2(pix2[13], pix3[13])); | |
3037 s += abs(pix1[14] - avg2(pix2[14], pix3[14])); | |
3038 s += abs(pix1[15] - avg2(pix2[15], pix3[15])); | |
3039 pix1 += line_size; | |
3040 pix2 += line_size; | |
3041 pix3 += line_size; | |
3042 } | |
3043 return s; | |
3044 } | |
3045 | |
1708 | 3046 static int pix_abs16_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 3047 { |
3048 int s, i; | |
1064 | 3049 uint8_t *pix3 = pix2 + line_size; |
0 | 3050 |
3051 s = 0; | |
1708 | 3052 for(i=0;i<h;i++) { |
0 | 3053 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
3054 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
3055 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
3056 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
3057 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
3058 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
3059 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
3060 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
3061 s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9])); | |
3062 s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10])); | |
3063 s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11])); | |
3064 s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12])); | |
3065 s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13])); | |
3066 s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14])); | |
3067 s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15])); | |
3068 s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16])); | |
3069 pix1 += line_size; | |
3070 pix2 += line_size; | |
3071 pix3 += line_size; | |
3072 } | |
3073 return s; | |
3074 } | |
3075 | |
1708 | 3076 static inline int pix_abs8_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3077 { |
3078 int s, i; | |
3079 | |
3080 s = 0; | |
1708 | 3081 for(i=0;i<h;i++) { |
294 | 3082 s += abs(pix1[0] - pix2[0]); |
3083 s += abs(pix1[1] - pix2[1]); | |
3084 s += abs(pix1[2] - pix2[2]); | |
3085 s += abs(pix1[3] - pix2[3]); | |
3086 s += abs(pix1[4] - pix2[4]); | |
3087 s += abs(pix1[5] - pix2[5]); | |
3088 s += abs(pix1[6] - pix2[6]); | |
3089 s += abs(pix1[7] - pix2[7]); | |
3090 pix1 += line_size; | |
3091 pix2 += line_size; | |
3092 } | |
3093 return s; | |
3094 } | |
3095 | |
1708 | 3096 static int pix_abs8_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3097 { |
3098 int s, i; | |
3099 | |
3100 s = 0; | |
1708 | 3101 for(i=0;i<h;i++) { |
294 | 3102 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
3103 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
3104 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
3105 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
3106 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
3107 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
3108 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
3109 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
3110 pix1 += line_size; | |
3111 pix2 += line_size; | |
3112 } | |
3113 return s; | |
3114 } | |
3115 | |
1708 | 3116 static int pix_abs8_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3117 { |
3118 int s, i; | |
1064 | 3119 uint8_t *pix3 = pix2 + line_size; |
294 | 3120 |
3121 s = 0; | |
1708 | 3122 for(i=0;i<h;i++) { |
294 | 3123 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
3124 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
3125 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
3126 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
3127 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
3128 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
3129 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
3130 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
3131 pix1 += line_size; | |
3132 pix2 += line_size; | |
3133 pix3 += line_size; | |
3134 } | |
3135 return s; | |
3136 } | |
3137 | |
1708 | 3138 static int pix_abs8_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3139 { |
3140 int s, i; | |
1064 | 3141 uint8_t *pix3 = pix2 + line_size; |
294 | 3142 |
3143 s = 0; | |
1708 | 3144 for(i=0;i<h;i++) { |
294 | 3145 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
3146 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
3147 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
3148 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
3149 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
3150 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
3151 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
3152 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
3153 pix1 += line_size; | |
3154 pix2 += line_size; | |
3155 pix3 += line_size; | |
3156 } | |
3157 return s; | |
3158 } | |
3159 | |
2834 | 3160 static int nsse16_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3161 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3162 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3163 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3164 int x,y; |
2066 | 3165 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3166 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3167 for(x=0; x<16; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3168 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3169 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3170 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3171 for(x=0; x<15; x++){ |
4001 | 3172 score2+= FFABS( s1[x ] - s1[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3173 - s1[x+1] + s1[x+1+stride]) |
4001 | 3174 -FFABS( s2[x ] - s2[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3175 - s2[x+1] + s2[x+1+stride]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3176 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3177 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3178 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3179 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3180 } |
2066 | 3181 |
4001 | 3182 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight; |
3183 else return score1 + FFABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3184 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3185 |
2834 | 3186 static int nsse8_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3187 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3188 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3189 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3190 int x,y; |
2967 | 3191 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3192 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3193 for(x=0; x<8; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3194 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3195 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3196 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3197 for(x=0; x<7; x++){ |
4001 | 3198 score2+= FFABS( s1[x ] - s1[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3199 - s1[x+1] + s1[x+1+stride]) |
4001 | 3200 -FFABS( s2[x ] - s2[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3201 - s2[x+1] + s2[x+1+stride]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3202 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3203 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3204 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3205 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3206 } |
2967 | 3207 |
4001 | 3208 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight; |
3209 else return score1 + FFABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3210 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3211 |
1784 | 3212 static int try_8x8basis_c(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale){ |
3213 int i; | |
3214 unsigned int sum=0; | |
3215 | |
3216 for(i=0; i<8*8; i++){ | |
3217 int b= rem[i] + ((basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT)); | |
3218 int w= weight[i]; | |
3219 b>>= RECON_SHIFT; | |
3220 assert(-512<b && b<512); | |
3221 | |
3222 sum += (w*b)*(w*b)>>4; | |
3223 } | |
3224 return sum>>2; | |
3225 } | |
3226 | |
3227 static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale){ | |
3228 int i; | |
3229 | |
3230 for(i=0; i<8*8; i++){ | |
3231 rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT); | |
2967 | 3232 } |
1784 | 3233 } |
3234 | |
1100 | 3235 /** |
3236 * permutes an 8x8 block. | |
1101 | 3237 * @param block the block which will be permuted according to the given permutation vector |
1100 | 3238 * @param permutation the permutation vector |
3239 * @param last the last non zero coefficient in scantable order, used to speed the permutation up | |
2967 | 3240 * @param scantable the used scantable, this is only used to speed the permutation up, the block is not |
1101 | 3241 * (inverse) permutated to scantable order! |
1100 | 3242 */ |
1064 | 3243 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last) |
174
ac5075a55488
new IDCT code by Michael Niedermayer (michaelni@gmx.at) - #define SIMPLE_IDCT to enable
arpi_esp
parents:
88
diff
changeset
|
3244 { |
764 | 3245 int i; |
945 | 3246 DCTELEM temp[64]; |
2967 | 3247 |
764 | 3248 if(last<=0) return; |
5129 | 3249 //if(permutation[1]==1) return; //FIXME it is ok but not clean and might fail for some permutations |
174
ac5075a55488
new IDCT code by Michael Niedermayer (michaelni@gmx.at) - #define SIMPLE_IDCT to enable
arpi_esp
parents:
88
diff
changeset
|
3250 |
764 | 3251 for(i=0; i<=last; i++){ |
3252 const int j= scantable[i]; | |
3253 temp[j]= block[j]; | |
3254 block[j]=0; | |
3255 } | |
2967 | 3256 |
764 | 3257 for(i=0; i<=last; i++){ |
3258 const int j= scantable[i]; | |
3259 const int perm_j= permutation[j]; | |
3260 block[perm_j]= temp[j]; | |
3261 } | |
174
ac5075a55488
new IDCT code by Michael Niedermayer (michaelni@gmx.at) - #define SIMPLE_IDCT to enable
arpi_esp
parents:
88
diff
changeset
|
3262 } |
34 | 3263 |
1729 | 3264 static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride, int h){ |
3265 return 0; | |
3266 } | |
3267 | |
3268 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type){ | |
3269 int i; | |
2967 | 3270 |
1729 | 3271 memset(cmp, 0, sizeof(void*)*5); |
2967 | 3272 |
1729 | 3273 for(i=0; i<5; i++){ |
3274 switch(type&0xFF){ | |
3275 case FF_CMP_SAD: | |
3276 cmp[i]= c->sad[i]; | |
3277 break; | |
3278 case FF_CMP_SATD: | |
3279 cmp[i]= c->hadamard8_diff[i]; | |
3280 break; | |
3281 case FF_CMP_SSE: | |
3282 cmp[i]= c->sse[i]; | |
3283 break; | |
3284 case FF_CMP_DCT: | |
3285 cmp[i]= c->dct_sad[i]; | |
3286 break; | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3287 case FF_CMP_DCT264: |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3288 cmp[i]= c->dct264_sad[i]; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3289 break; |
2382 | 3290 case FF_CMP_DCTMAX: |
3291 cmp[i]= c->dct_max[i]; | |
3292 break; | |
1729 | 3293 case FF_CMP_PSNR: |
3294 cmp[i]= c->quant_psnr[i]; | |
3295 break; | |
3296 case FF_CMP_BIT: | |
3297 cmp[i]= c->bit[i]; | |
3298 break; | |
3299 case FF_CMP_RD: | |
3300 cmp[i]= c->rd[i]; | |
3301 break; | |
3302 case FF_CMP_VSAD: | |
3303 cmp[i]= c->vsad[i]; | |
3304 break; | |
3305 case FF_CMP_VSSE: | |
3306 cmp[i]= c->vsse[i]; | |
3307 break; | |
3308 case FF_CMP_ZERO: | |
3309 cmp[i]= zero_cmp; | |
3310 break; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3311 case FF_CMP_NSSE: |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3312 cmp[i]= c->nsse[i]; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3313 break; |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
3314 #ifdef CONFIG_SNOW_ENCODER |
2184 | 3315 case FF_CMP_W53: |
3316 cmp[i]= c->w53[i]; | |
3317 break; | |
3318 case FF_CMP_W97: | |
3319 cmp[i]= c->w97[i]; | |
3320 break; | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
3321 #endif |
1729 | 3322 default: |
3323 av_log(NULL, AV_LOG_ERROR,"internal error in cmp function selection\n"); | |
3324 } | |
3325 } | |
3326 } | |
3327 | |
1101 | 3328 /** |
3329 * memset(blocks, 0, sizeof(DCTELEM)*6*64) | |
3330 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
3331 static void clear_blocks_c(DCTELEM *blocks) |
296 | 3332 { |
3333 memset(blocks, 0, sizeof(DCTELEM)*6*64); | |
3334 } | |
3335 | |
866 | 3336 static void add_bytes_c(uint8_t *dst, uint8_t *src, int w){ |
6385 | 3337 long i; |
3338 for(i=0; i<=w-sizeof(long); i+=sizeof(long)){ | |
3339 long a = *(long*)(src+i); | |
3340 long b = *(long*)(dst+i); | |
3341 *(long*)(dst+i) = ((a&pb_7f) + (b&pb_7f)) ^ ((a^b)&pb_80); | |
866 | 3342 } |
3343 for(; i<w; i++) | |
3344 dst[i+0] += src[i+0]; | |
3345 } | |
3346 | |
6384 | 3347 static void add_bytes_l2_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ |
6385 | 3348 long i; |
6384 | 3349 for(i=0; i<=w-sizeof(long); i+=sizeof(long)){ |
3350 long a = *(long*)(src1+i); | |
3351 long b = *(long*)(src2+i); | |
6385 | 3352 *(long*)(dst+i) = ((a&pb_7f) + (b&pb_7f)) ^ ((a^b)&pb_80); |
6384 | 3353 } |
3354 for(; i<w; i++) | |
3355 dst[i] = src1[i]+src2[i]; | |
3356 } | |
3357 | |
866 | 3358 static void diff_bytes_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ |
6385 | 3359 long i; |
3360 #ifndef HAVE_FAST_UNALIGNED | |
3361 if((long)src2 & (sizeof(long)-1)){ | |
6386 | 3362 for(i=0; i+7<w; i+=8){ |
3363 dst[i+0] = src1[i+0]-src2[i+0]; | |
3364 dst[i+1] = src1[i+1]-src2[i+1]; | |
3365 dst[i+2] = src1[i+2]-src2[i+2]; | |
3366 dst[i+3] = src1[i+3]-src2[i+3]; | |
3367 dst[i+4] = src1[i+4]-src2[i+4]; | |
3368 dst[i+5] = src1[i+5]-src2[i+5]; | |
3369 dst[i+6] = src1[i+6]-src2[i+6]; | |
3370 dst[i+7] = src1[i+7]-src2[i+7]; | |
3371 } | |
6385 | 3372 }else |
3373 #endif | |
3374 for(i=0; i<=w-sizeof(long); i+=sizeof(long)){ | |
3375 long a = *(long*)(src1+i); | |
3376 long b = *(long*)(src2+i); | |
3377 *(long*)(dst+i) = ((a|pb_80) - (b&pb_7f)) ^ ((a^b^pb_80)&pb_80); | |
3378 } | |
866 | 3379 for(; i<w; i++) |
3380 dst[i+0] = src1[i+0]-src2[i+0]; | |
3381 } | |
3382 | |
1527 | 3383 static void sub_hfyu_median_prediction_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){ |
3384 int i; | |
3385 uint8_t l, lt; | |
3386 | |
3387 l= *left; | |
3388 lt= *left_top; | |
3389 | |
3390 for(i=0; i<w; i++){ | |
3391 const int pred= mid_pred(l, src1[i], (l + src1[i] - lt)&0xFF); | |
3392 lt= src1[i]; | |
3393 l= src2[i]; | |
3394 dst[i]= l - pred; | |
2967 | 3395 } |
1527 | 3396 |
3397 *left= l; | |
3398 *left_top= lt; | |
3399 } | |
3400 | |
936 | 3401 #define BUTTERFLY2(o1,o2,i1,i2) \ |
3402 o1= (i1)+(i2);\ | |
3403 o2= (i1)-(i2); | |
3404 | |
3405 #define BUTTERFLY1(x,y) \ | |
3406 {\ | |
3407 int a,b;\ | |
3408 a= x;\ | |
3409 b= y;\ | |
3410 x= a+b;\ | |
3411 y= a-b;\ | |
3412 } | |
3413 | |
4001 | 3414 #define BUTTERFLYA(x,y) (FFABS((x)+(y)) + FFABS((x)-(y))) |
936 | 3415 |
1708 | 3416 static int hadamard8_diff8x8_c(/*MpegEncContext*/ void *s, uint8_t *dst, uint8_t *src, int stride, int h){ |
936 | 3417 int i; |
3418 int temp[64]; | |
3419 int sum=0; | |
2967 | 3420 |
1708 | 3421 assert(h==8); |
936 | 3422 |
3423 for(i=0; i<8; i++){ | |
3424 //FIXME try pointer walks | |
3425 BUTTERFLY2(temp[8*i+0], temp[8*i+1], src[stride*i+0]-dst[stride*i+0],src[stride*i+1]-dst[stride*i+1]); | |
3426 BUTTERFLY2(temp[8*i+2], temp[8*i+3], src[stride*i+2]-dst[stride*i+2],src[stride*i+3]-dst[stride*i+3]); | |
3427 BUTTERFLY2(temp[8*i+4], temp[8*i+5], src[stride*i+4]-dst[stride*i+4],src[stride*i+5]-dst[stride*i+5]); | |
3428 BUTTERFLY2(temp[8*i+6], temp[8*i+7], src[stride*i+6]-dst[stride*i+6],src[stride*i+7]-dst[stride*i+7]); | |
2967 | 3429 |
936 | 3430 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3431 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3432 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3433 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3434 |
936 | 3435 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3436 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3437 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3438 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3439 } | |
3440 | |
3441 for(i=0; i<8; i++){ | |
3442 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3443 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3444 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3445 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3446 |
936 | 3447 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3448 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3449 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3450 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
3451 | |
2967 | 3452 sum += |
936 | 3453 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3454 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3455 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3456 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3457 } | |
3458 #if 0 | |
3459 static int maxi=0; | |
3460 if(sum>maxi){ | |
3461 maxi=sum; | |
3462 printf("MAX:%d\n", maxi); | |
3463 } | |
3464 #endif | |
3465 return sum; | |
3466 } | |
3467 | |
1729 | 3468 static int hadamard8_intra8x8_c(/*MpegEncContext*/ void *s, uint8_t *src, uint8_t *dummy, int stride, int h){ |
936 | 3469 int i; |
3470 int temp[64]; | |
3471 int sum=0; | |
2967 | 3472 |
1729 | 3473 assert(h==8); |
2967 | 3474 |
936 | 3475 for(i=0; i<8; i++){ |
3476 //FIXME try pointer walks | |
1729 | 3477 BUTTERFLY2(temp[8*i+0], temp[8*i+1], src[stride*i+0],src[stride*i+1]); |
3478 BUTTERFLY2(temp[8*i+2], temp[8*i+3], src[stride*i+2],src[stride*i+3]); | |
3479 BUTTERFLY2(temp[8*i+4], temp[8*i+5], src[stride*i+4],src[stride*i+5]); | |
3480 BUTTERFLY2(temp[8*i+6], temp[8*i+7], src[stride*i+6],src[stride*i+7]); | |
2967 | 3481 |
936 | 3482 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3483 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3484 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3485 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3486 |
936 | 3487 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3488 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3489 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3490 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3491 } | |
3492 | |
3493 for(i=0; i<8; i++){ | |
3494 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3495 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3496 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3497 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3498 |
936 | 3499 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3500 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3501 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3502 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
2967 | 3503 |
3504 sum += | |
936 | 3505 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3506 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3507 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3508 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3509 } | |
2967 | 3510 |
4001 | 3511 sum -= FFABS(temp[8*0] + temp[8*4]); // -mean |
2967 | 3512 |
936 | 3513 return sum; |
3514 } | |
3515 | |
1708 | 3516 static int dct_sad8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3517 MpegEncContext * const s= (MpegEncContext *)c; |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
3518 DECLARE_ALIGNED_16(uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
1016 | 3519 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
2967 | 3520 |
1708 | 3521 assert(h==8); |
936 | 3522 |
3523 s->dsp.diff_pixels(temp, src1, src2, stride); | |
1092 | 3524 s->dsp.fdct(temp); |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
3525 return s->dsp.sum_abs_dctelem(temp); |
936 | 3526 } |
3527 | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3528 #ifdef CONFIG_GPL |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3529 #define DCT8_1D {\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3530 const int s07 = SRC(0) + SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3531 const int s16 = SRC(1) + SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3532 const int s25 = SRC(2) + SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3533 const int s34 = SRC(3) + SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3534 const int a0 = s07 + s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3535 const int a1 = s16 + s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3536 const int a2 = s07 - s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3537 const int a3 = s16 - s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3538 const int d07 = SRC(0) - SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3539 const int d16 = SRC(1) - SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3540 const int d25 = SRC(2) - SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3541 const int d34 = SRC(3) - SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3542 const int a4 = d16 + d25 + (d07 + (d07>>1));\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3543 const int a5 = d07 - d34 - (d25 + (d25>>1));\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3544 const int a6 = d07 + d34 - (d16 + (d16>>1));\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3545 const int a7 = d16 - d25 + (d34 + (d34>>1));\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3546 DST(0, a0 + a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3547 DST(1, a4 + (a7>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3548 DST(2, a2 + (a3>>1)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3549 DST(3, a5 + (a6>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3550 DST(4, a0 - a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3551 DST(5, a6 - (a5>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3552 DST(6, (a2>>1) - a3 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3553 DST(7, (a4>>2) - a7 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3554 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3555 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3556 static int dct264_sad8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3557 MpegEncContext * const s= (MpegEncContext *)c; |
5256 | 3558 DCTELEM dct[8][8]; |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3559 int i; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3560 int sum=0; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3561 |
5256 | 3562 s->dsp.diff_pixels(dct[0], src1, src2, stride); |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3563 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3564 #define SRC(x) dct[i][x] |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3565 #define DST(x,v) dct[i][x]= v |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3566 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3567 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3568 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3569 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3570 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3571 #define SRC(x) dct[x][i] |
4001 | 3572 #define DST(x,v) sum += FFABS(v) |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3573 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3574 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3575 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3576 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3577 return sum; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3578 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3579 #endif |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3580 |
2382 | 3581 static int dct_max8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
3582 MpegEncContext * const s= (MpegEncContext *)c; | |
3089 | 3583 DECLARE_ALIGNED_8(uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
2382 | 3584 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3585 int sum=0, i; | |
2967 | 3586 |
2382 | 3587 assert(h==8); |
3588 | |
3589 s->dsp.diff_pixels(temp, src1, src2, stride); | |
3590 s->dsp.fdct(temp); | |
3591 | |
3592 for(i=0; i<64; i++) | |
4001 | 3593 sum= FFMAX(sum, FFABS(temp[i])); |
2967 | 3594 |
2382 | 3595 return sum; |
3596 } | |
3597 | |
1708 | 3598 static int quant_psnr8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3599 MpegEncContext * const s= (MpegEncContext *)c; |
3089 | 3600 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64*2/8]); |
1016 | 3601 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3602 DCTELEM * const bak = ((DCTELEM*)aligned_temp)+64; | |
936 | 3603 int sum=0, i; |
3604 | |
1708 | 3605 assert(h==8); |
936 | 3606 s->mb_intra=0; |
2967 | 3607 |
936 | 3608 s->dsp.diff_pixels(temp, src1, src2, stride); |
2967 | 3609 |
936 | 3610 memcpy(bak, temp, 64*sizeof(DCTELEM)); |
2967 | 3611 |
1013 | 3612 s->block_last_index[0/*FIXME*/]= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
1689 | 3613 s->dct_unquantize_inter(s, temp, 0, s->qscale); |
6001 | 3614 ff_simple_idct(temp); //FIXME |
2967 | 3615 |
936 | 3616 for(i=0; i<64; i++) |
3617 sum+= (temp[i]-bak[i])*(temp[i]-bak[i]); | |
2967 | 3618 |
936 | 3619 return sum; |
3620 } | |
3621 | |
1708 | 3622 static int rd8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3623 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3624 const uint8_t *scantable= s->intra_scantable.permutated; |
3089 | 3625 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
3626 DECLARE_ALIGNED_8 (uint64_t, aligned_bak[stride]); | |
1016 | 3627 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3628 uint8_t * const bak= (uint8_t*)aligned_bak; | |
1007 | 3629 int i, last, run, bits, level, distoration, start_i; |
3630 const int esc_length= s->ac_esc_length; | |
3631 uint8_t * length; | |
3632 uint8_t * last_length; | |
2967 | 3633 |
1708 | 3634 assert(h==8); |
3635 | |
1007 | 3636 for(i=0; i<8; i++){ |
3637 ((uint32_t*)(bak + i*stride))[0]= ((uint32_t*)(src2 + i*stride))[0]; | |
3638 ((uint32_t*)(bak + i*stride))[1]= ((uint32_t*)(src2 + i*stride))[1]; | |
3639 } | |
3640 | |
3641 s->dsp.diff_pixels(temp, src1, src2, stride); | |
3642 | |
1013 | 3643 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3644 | |
3645 bits=0; | |
2967 | 3646 |
1013 | 3647 if (s->mb_intra) { |
2967 | 3648 start_i = 1; |
1013 | 3649 length = s->intra_ac_vlc_length; |
3650 last_length= s->intra_ac_vlc_last_length; | |
3651 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma | |
3652 } else { | |
3653 start_i = 0; | |
3654 length = s->inter_ac_vlc_length; | |
3655 last_length= s->inter_ac_vlc_last_length; | |
3656 } | |
2967 | 3657 |
1013 | 3658 if(last>=start_i){ |
1007 | 3659 run=0; |
3660 for(i=start_i; i<last; i++){ | |
3661 int j= scantable[i]; | |
3662 level= temp[j]; | |
2967 | 3663 |
1007 | 3664 if(level){ |
3665 level+=64; | |
3666 if((level&(~127)) == 0){ | |
3667 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3668 }else | |
3669 bits+= esc_length; | |
3670 run=0; | |
3671 }else | |
3672 run++; | |
3673 } | |
3674 i= scantable[last]; | |
2967 | 3675 |
1011 | 3676 level= temp[i] + 64; |
3677 | |
3678 assert(level - 64); | |
2967 | 3679 |
1007 | 3680 if((level&(~127)) == 0){ |
3681 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3682 }else | |
3683 bits+= esc_length; | |
2967 | 3684 |
1013 | 3685 } |
3686 | |
3687 if(last>=0){ | |
1689 | 3688 if(s->mb_intra) |
3689 s->dct_unquantize_intra(s, temp, 0, s->qscale); | |
3690 else | |
3691 s->dct_unquantize_inter(s, temp, 0, s->qscale); | |
1007 | 3692 } |
2967 | 3693 |
1092 | 3694 s->dsp.idct_add(bak, stride, temp); |
2967 | 3695 |
1708 | 3696 distoration= s->dsp.sse[1](NULL, bak, src1, stride, 8); |
1007 | 3697 |
1013 | 3698 return distoration + ((bits*s->qscale*s->qscale*109 + 64)>>7); |
1007 | 3699 } |
3700 | |
1708 | 3701 static int bit8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3702 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3703 const uint8_t *scantable= s->intra_scantable.permutated; |
3089 | 3704 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
1016 | 3705 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
1007 | 3706 int i, last, run, bits, level, start_i; |
3707 const int esc_length= s->ac_esc_length; | |
3708 uint8_t * length; | |
3709 uint8_t * last_length; | |
1708 | 3710 |
3711 assert(h==8); | |
2967 | 3712 |
1013 | 3713 s->dsp.diff_pixels(temp, src1, src2, stride); |
1007 | 3714 |
1013 | 3715 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3716 | |
3717 bits=0; | |
2967 | 3718 |
1007 | 3719 if (s->mb_intra) { |
2967 | 3720 start_i = 1; |
1007 | 3721 length = s->intra_ac_vlc_length; |
3722 last_length= s->intra_ac_vlc_last_length; | |
1013 | 3723 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma |
1007 | 3724 } else { |
3725 start_i = 0; | |
3726 length = s->inter_ac_vlc_length; | |
3727 last_length= s->inter_ac_vlc_last_length; | |
3728 } | |
2967 | 3729 |
1013 | 3730 if(last>=start_i){ |
1007 | 3731 run=0; |
3732 for(i=start_i; i<last; i++){ | |
3733 int j= scantable[i]; | |
3734 level= temp[j]; | |
2967 | 3735 |
1007 | 3736 if(level){ |
3737 level+=64; | |
3738 if((level&(~127)) == 0){ | |
3739 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3740 }else | |
3741 bits+= esc_length; | |
3742 run=0; | |
3743 }else | |
3744 run++; | |
3745 } | |
3746 i= scantable[last]; | |
2967 | 3747 |
1013 | 3748 level= temp[i] + 64; |
2967 | 3749 |
1013 | 3750 assert(level - 64); |
2967 | 3751 |
1007 | 3752 if((level&(~127)) == 0){ |
3753 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3754 }else | |
3755 bits+= esc_length; | |
3756 } | |
3757 | |
3758 return bits; | |
3759 } | |
3760 | |
1729 | 3761 static int vsad_intra16_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ |
3762 int score=0; | |
3763 int x,y; | |
2967 | 3764 |
1729 | 3765 for(y=1; y<h; y++){ |
3766 for(x=0; x<16; x+=4){ | |
4001 | 3767 score+= FFABS(s[x ] - s[x +stride]) + FFABS(s[x+1] - s[x+1+stride]) |
3768 +FFABS(s[x+2] - s[x+2+stride]) + FFABS(s[x+3] - s[x+3+stride]); | |
1729 | 3769 } |
3770 s+= stride; | |
3771 } | |
2967 | 3772 |
1729 | 3773 return score; |
3774 } | |
3775 | |
3776 static int vsad16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3777 int score=0; | |
3778 int x,y; | |
2967 | 3779 |
1729 | 3780 for(y=1; y<h; y++){ |
3781 for(x=0; x<16; x++){ | |
4001 | 3782 score+= FFABS(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); |
1729 | 3783 } |
3784 s1+= stride; | |
3785 s2+= stride; | |
3786 } | |
2967 | 3787 |
1729 | 3788 return score; |
3789 } | |
3790 | |
3791 #define SQ(a) ((a)*(a)) | |
3792 static int vsse_intra16_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ | |
3793 int score=0; | |
3794 int x,y; | |
2967 | 3795 |
1729 | 3796 for(y=1; y<h; y++){ |
3797 for(x=0; x<16; x+=4){ | |
2967 | 3798 score+= SQ(s[x ] - s[x +stride]) + SQ(s[x+1] - s[x+1+stride]) |
1729 | 3799 +SQ(s[x+2] - s[x+2+stride]) + SQ(s[x+3] - s[x+3+stride]); |
3800 } | |
3801 s+= stride; | |
3802 } | |
2967 | 3803 |
1729 | 3804 return score; |
3805 } | |
3806 | |
3807 static int vsse16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3808 int score=0; | |
3809 int x,y; | |
2967 | 3810 |
1729 | 3811 for(y=1; y<h; y++){ |
3812 for(x=0; x<16; x++){ | |
3813 score+= SQ(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); | |
3814 } | |
3815 s1+= stride; | |
3816 s2+= stride; | |
3817 } | |
2967 | 3818 |
1729 | 3819 return score; |
3820 } | |
3821 | |
5255 | 3822 static int ssd_int8_vs_int16_c(const int8_t *pix1, const int16_t *pix2, |
3823 int size){ | |
4749 | 3824 int score=0; |
3825 int i; | |
3826 for(i=0; i<size; i++) | |
3827 score += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]); | |
3828 return score; | |
3829 } | |
3830 | |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3831 WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c) |
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3832 WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c) |
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3833 WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c) |
3013 | 3834 #ifdef CONFIG_GPL |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3835 WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c) |
3013 | 3836 #endif |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3837 WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c) |
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3838 WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c) |
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3839 WRAPPER8_16_SQ(rd8x8_c, rd16_c) |
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3840 WRAPPER8_16_SQ(bit8x8_c, bit16_c) |
936 | 3841 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3842 static void vector_fmul_c(float *dst, const float *src, int len){ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3843 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3844 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3845 dst[i] *= src[i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3846 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3847 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3848 static void vector_fmul_reverse_c(float *dst, const float *src0, const float *src1, int len){ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3849 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3850 src1 += len-1; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3851 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3852 dst[i] = src0[i] * src1[-i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3853 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3854 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3855 void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step){ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3856 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3857 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3858 dst[i*step] = src0[i] * src1[i] + src2[i] + src3; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3859 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3860 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3861 void ff_float_to_int16_c(int16_t *dst, const float *src, int len){ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3862 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3863 for(i=0; i<len; i++) { |
6218 | 3864 int_fast32_t tmp = ((const int32_t*)src)[i]; |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3865 if(tmp & 0xf0000){ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3866 tmp = (0x43c0ffff - tmp)>>31; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3867 // is this faster on some gcc/cpu combinations? |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3868 // if(tmp > 0x43c0ffff) tmp = 0xFFFF; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3869 // else tmp = 0; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3870 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3871 dst[i] = tmp - 0x8000; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3872 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3873 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3874 |
5887 | 3875 #define W0 2048 |
3876 #define W1 2841 /* 2048*sqrt (2)*cos (1*pi/16) */ | |
3877 #define W2 2676 /* 2048*sqrt (2)*cos (2*pi/16) */ | |
3878 #define W3 2408 /* 2048*sqrt (2)*cos (3*pi/16) */ | |
3879 #define W4 2048 /* 2048*sqrt (2)*cos (4*pi/16) */ | |
3880 #define W5 1609 /* 2048*sqrt (2)*cos (5*pi/16) */ | |
3881 #define W6 1108 /* 2048*sqrt (2)*cos (6*pi/16) */ | |
3882 #define W7 565 /* 2048*sqrt (2)*cos (7*pi/16) */ | |
3883 | |
3884 static void wmv2_idct_row(short * b) | |
3885 { | |
3886 int s1,s2; | |
3887 int a0,a1,a2,a3,a4,a5,a6,a7; | |
3888 /*step 1*/ | |
3889 a1 = W1*b[1]+W7*b[7]; | |
3890 a7 = W7*b[1]-W1*b[7]; | |
3891 a5 = W5*b[5]+W3*b[3]; | |
3892 a3 = W3*b[5]-W5*b[3]; | |
3893 a2 = W2*b[2]+W6*b[6]; | |
3894 a6 = W6*b[2]-W2*b[6]; | |
3895 a0 = W0*b[0]+W0*b[4]; | |
3896 a4 = W0*b[0]-W0*b[4]; | |
3897 /*step 2*/ | |
3898 s1 = (181*(a1-a5+a7-a3)+128)>>8;//1,3,5,7, | |
3899 s2 = (181*(a1-a5-a7+a3)+128)>>8; | |
3900 /*step 3*/ | |
3901 b[0] = (a0+a2+a1+a5 + (1<<7))>>8; | |
3902 b[1] = (a4+a6 +s1 + (1<<7))>>8; | |
3903 b[2] = (a4-a6 +s2 + (1<<7))>>8; | |
3904 b[3] = (a0-a2+a7+a3 + (1<<7))>>8; | |
3905 b[4] = (a0-a2-a7-a3 + (1<<7))>>8; | |
3906 b[5] = (a4-a6 -s2 + (1<<7))>>8; | |
3907 b[6] = (a4+a6 -s1 + (1<<7))>>8; | |
3908 b[7] = (a0+a2-a1-a5 + (1<<7))>>8; | |
3909 } | |
3910 static void wmv2_idct_col(short * b) | |
3911 { | |
3912 int s1,s2; | |
3913 int a0,a1,a2,a3,a4,a5,a6,a7; | |
3914 /*step 1, with extended precision*/ | |
3915 a1 = (W1*b[8*1]+W7*b[8*7] + 4)>>3; | |
3916 a7 = (W7*b[8*1]-W1*b[8*7] + 4)>>3; | |
3917 a5 = (W5*b[8*5]+W3*b[8*3] + 4)>>3; | |
3918 a3 = (W3*b[8*5]-W5*b[8*3] + 4)>>3; | |
3919 a2 = (W2*b[8*2]+W6*b[8*6] + 4)>>3; | |
3920 a6 = (W6*b[8*2]-W2*b[8*6] + 4)>>3; | |
3921 a0 = (W0*b[8*0]+W0*b[8*4] )>>3; | |
3922 a4 = (W0*b[8*0]-W0*b[8*4] )>>3; | |
3923 /*step 2*/ | |
3924 s1 = (181*(a1-a5+a7-a3)+128)>>8; | |
3925 s2 = (181*(a1-a5-a7+a3)+128)>>8; | |
3926 /*step 3*/ | |
3927 b[8*0] = (a0+a2+a1+a5 + (1<<13))>>14; | |
3928 b[8*1] = (a4+a6 +s1 + (1<<13))>>14; | |
3929 b[8*2] = (a4-a6 +s2 + (1<<13))>>14; | |
3930 b[8*3] = (a0-a2+a7+a3 + (1<<13))>>14; | |
3931 | |
3932 b[8*4] = (a0-a2-a7-a3 + (1<<13))>>14; | |
3933 b[8*5] = (a4-a6 -s2 + (1<<13))>>14; | |
3934 b[8*6] = (a4+a6 -s1 + (1<<13))>>14; | |
3935 b[8*7] = (a0+a2-a1-a5 + (1<<13))>>14; | |
3936 } | |
3937 void ff_wmv2_idct_c(short * block){ | |
3938 int i; | |
3939 | |
3940 for(i=0;i<64;i+=8){ | |
3941 wmv2_idct_row(block+i); | |
3942 } | |
3943 for(i=0;i<8;i++){ | |
3944 wmv2_idct_col(block+i); | |
3945 } | |
3946 } | |
1092 | 3947 /* XXX: those functions should be suppressed ASAP when all IDCTs are |
3948 converted */ | |
5887 | 3949 static void ff_wmv2_idct_put_c(uint8_t *dest, int line_size, DCTELEM *block) |
3950 { | |
3951 ff_wmv2_idct_c(block); | |
3952 put_pixels_clamped_c(block, dest, line_size); | |
3953 } | |
3954 static void ff_wmv2_idct_add_c(uint8_t *dest, int line_size, DCTELEM *block) | |
3955 { | |
3956 ff_wmv2_idct_c(block); | |
3957 add_pixels_clamped_c(block, dest, line_size); | |
3958 } | |
1092 | 3959 static void ff_jref_idct_put(uint8_t *dest, int line_size, DCTELEM *block) |
3960 { | |
3961 j_rev_dct (block); | |
3962 put_pixels_clamped_c(block, dest, line_size); | |
3963 } | |
3964 static void ff_jref_idct_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3965 { | |
3966 j_rev_dct (block); | |
3967 add_pixels_clamped_c(block, dest, line_size); | |
3968 } | |
3969 | |
2256 | 3970 static void ff_jref_idct4_put(uint8_t *dest, int line_size, DCTELEM *block) |
3971 { | |
3972 j_rev_dct4 (block); | |
3973 put_pixels_clamped4_c(block, dest, line_size); | |
3974 } | |
3975 static void ff_jref_idct4_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3976 { | |
3977 j_rev_dct4 (block); | |
3978 add_pixels_clamped4_c(block, dest, line_size); | |
3979 } | |
3980 | |
2257 | 3981 static void ff_jref_idct2_put(uint8_t *dest, int line_size, DCTELEM *block) |
3982 { | |
3983 j_rev_dct2 (block); | |
3984 put_pixels_clamped2_c(block, dest, line_size); | |
3985 } | |
3986 static void ff_jref_idct2_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3987 { | |
3988 j_rev_dct2 (block); | |
3989 add_pixels_clamped2_c(block, dest, line_size); | |
3990 } | |
3991 | |
2259 | 3992 static void ff_jref_idct1_put(uint8_t *dest, int line_size, DCTELEM *block) |
3993 { | |
4176 | 3994 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2259 | 3995 |
3996 dest[0] = cm[(block[0] + 4)>>3]; | |
3997 } | |
3998 static void ff_jref_idct1_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3999 { | |
4176 | 4000 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2259 | 4001 |
4002 dest[0] = cm[dest[0] + ((block[0] + 4)>>3)]; | |
4003 } | |
4004 | |
5143 | 4005 static void just_return(void *mem av_unused, int stride av_unused, int h av_unused) { return; } |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4006 |
1201 | 4007 /* init static data */ |
4197 | 4008 void dsputil_static_init(void) |
0 | 4009 { |
751 | 4010 int i; |
0 | 4011 |
4176 | 4012 for(i=0;i<256;i++) ff_cropTbl[i + MAX_NEG_CROP] = i; |
1201 | 4013 for(i=0;i<MAX_NEG_CROP;i++) { |
4176 | 4014 ff_cropTbl[i] = 0; |
4015 ff_cropTbl[i + MAX_NEG_CROP + 256] = 255; | |
1201 | 4016 } |
2967 | 4017 |
1201 | 4018 for(i=0;i<512;i++) { |
4179 | 4019 ff_squareTbl[i] = (i - 256) * (i - 256); |
1201 | 4020 } |
2967 | 4021 |
4197 | 4022 for(i=0; i<64; i++) inv_zigzag_direct16[ff_zigzag_direct[i]]= i+1; |
1201 | 4023 } |
0 | 4024 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4025 int ff_check_alignment(void){ |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4026 static int did_fail=0; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4027 DECLARE_ALIGNED_16(int, aligned); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4028 |
5150 | 4029 if((long)&aligned & 15){ |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4030 if(!did_fail){ |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4031 #if defined(HAVE_MMX) || defined(HAVE_ALTIVEC) |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4032 av_log(NULL, AV_LOG_ERROR, |
4292 | 4033 "Compiler did not align stack variables. Libavcodec has been miscompiled\n" |
4034 "and may be very slow or crash. This is not a bug in libavcodec,\n" | |
5542
b0a566346fb1
Add attribute that forces alignment of stack to functions that need it.
ramiro
parents:
5520
diff
changeset
|
4035 "but in the compiler. You may try recompiling using gcc >= 4.2.\n" |
b0a566346fb1
Add attribute that forces alignment of stack to functions that need it.
ramiro
parents:
5520
diff
changeset
|
4036 "Do not report crashes to FFmpeg developers.\n"); |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4037 #endif |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4038 did_fail=1; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4039 } |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4040 return -1; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4041 } |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4042 return 0; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4043 } |
861 | 4044 |
1201 | 4045 void dsputil_init(DSPContext* c, AVCodecContext *avctx) |
4046 { | |
4047 int i; | |
0 | 4048 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4049 ff_check_alignment(); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4050 |
1092 | 4051 #ifdef CONFIG_ENCODERS |
1567 | 4052 if(avctx->dct_algo==FF_DCT_FASTINT) { |
1092 | 4053 c->fdct = fdct_ifast; |
2979 | 4054 c->fdct248 = fdct_ifast248; |
2967 | 4055 } |
1567 | 4056 else if(avctx->dct_algo==FF_DCT_FAAN) { |
1557 | 4057 c->fdct = ff_faandct; |
2979 | 4058 c->fdct248 = ff_faandct248; |
2967 | 4059 } |
1567 | 4060 else { |
1092 | 4061 c->fdct = ff_jpeg_fdct_islow; //slow/accurate/default |
2979 | 4062 c->fdct248 = ff_fdct248_islow; |
1567 | 4063 } |
1092 | 4064 #endif //CONFIG_ENCODERS |
4065 | |
2256 | 4066 if(avctx->lowres==1){ |
5064 | 4067 if(avctx->idct_algo==FF_IDCT_INT || avctx->idct_algo==FF_IDCT_AUTO || !ENABLE_H264_DECODER){ |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4068 c->idct_put= ff_jref_idct4_put; |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4069 c->idct_add= ff_jref_idct4_add; |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4070 }else{ |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4071 c->idct_put= ff_h264_lowres_idct_put_c; |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4072 c->idct_add= ff_h264_lowres_idct_add_c; |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4073 } |
2256 | 4074 c->idct = j_rev_dct4; |
1092 | 4075 c->idct_permutation_type= FF_NO_IDCT_PERM; |
2257 | 4076 }else if(avctx->lowres==2){ |
4077 c->idct_put= ff_jref_idct2_put; | |
4078 c->idct_add= ff_jref_idct2_add; | |
4079 c->idct = j_rev_dct2; | |
4080 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2259 | 4081 }else if(avctx->lowres==3){ |
4082 c->idct_put= ff_jref_idct1_put; | |
4083 c->idct_add= ff_jref_idct1_add; | |
4084 c->idct = j_rev_dct1; | |
4085 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2256 | 4086 }else{ |
4087 if(avctx->idct_algo==FF_IDCT_INT){ | |
4088 c->idct_put= ff_jref_idct_put; | |
4089 c->idct_add= ff_jref_idct_add; | |
4090 c->idct = j_rev_dct; | |
4091 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM; | |
5007 | 4092 }else if((ENABLE_VP3_DECODER || ENABLE_VP5_DECODER || ENABLE_VP6_DECODER || ENABLE_THEORA_DECODER ) && |
4093 avctx->idct_algo==FF_IDCT_VP3){ | |
2693 | 4094 c->idct_put= ff_vp3_idct_put_c; |
4095 c->idct_add= ff_vp3_idct_add_c; | |
4096 c->idct = ff_vp3_idct_c; | |
4097 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
5887 | 4098 }else if(avctx->idct_algo==FF_IDCT_WMV2){ |
4099 c->idct_put= ff_wmv2_idct_put_c; | |
4100 c->idct_add= ff_wmv2_idct_add_c; | |
4101 c->idct = ff_wmv2_idct_c; | |
4102 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
6407 | 4103 }else if(avctx->idct_algo==FF_IDCT_FAAN){ |
4104 c->idct_put= ff_faanidct_put; | |
4105 c->idct_add= ff_faanidct_add; | |
4106 c->idct = ff_faanidct; | |
4107 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2256 | 4108 }else{ //accurate/default |
6001 | 4109 c->idct_put= ff_simple_idct_put; |
4110 c->idct_add= ff_simple_idct_add; | |
4111 c->idct = ff_simple_idct; | |
2256 | 4112 c->idct_permutation_type= FF_NO_IDCT_PERM; |
4113 } | |
1092 | 4114 } |
4115 | |
5064 | 4116 if (ENABLE_H264_DECODER) { |
5065 | 4117 c->h264_idct_add= ff_h264_idct_add_c; |
4118 c->h264_idct8_add= ff_h264_idct8_add_c; | |
4119 c->h264_idct_dc_add= ff_h264_idct_dc_add_c; | |
4120 c->h264_idct8_dc_add= ff_h264_idct8_dc_add_c; | |
5064 | 4121 } |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4122 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4123 c->get_pixels = get_pixels_c; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4124 c->diff_pixels = diff_pixels_c; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4125 c->put_pixels_clamped = put_pixels_clamped_c; |
1984
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
4126 c->put_signed_pixels_clamped = put_signed_pixels_clamped_c; |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4127 c->add_pixels_clamped = add_pixels_clamped_c; |
2763 | 4128 c->add_pixels8 = add_pixels8_c; |
4129 c->add_pixels4 = add_pixels4_c; | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
4130 c->sum_abs_dctelem = sum_abs_dctelem_c; |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4131 c->gmc1 = gmc1_c; |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
4132 c->gmc = ff_gmc_c; |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4133 c->clear_blocks = clear_blocks_c; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4134 c->pix_sum = pix_sum_c; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4135 c->pix_norm1 = pix_norm1_c; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4136 |
859 | 4137 /* TODO [0] 16 [1] 8 */ |
1708 | 4138 c->pix_abs[0][0] = pix_abs16_c; |
4139 c->pix_abs[0][1] = pix_abs16_x2_c; | |
4140 c->pix_abs[0][2] = pix_abs16_y2_c; | |
4141 c->pix_abs[0][3] = pix_abs16_xy2_c; | |
4142 c->pix_abs[1][0] = pix_abs8_c; | |
4143 c->pix_abs[1][1] = pix_abs8_x2_c; | |
4144 c->pix_abs[1][2] = pix_abs8_y2_c; | |
4145 c->pix_abs[1][3] = pix_abs8_xy2_c; | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4146 |
859 | 4147 #define dspfunc(PFX, IDX, NUM) \ |
4148 c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## NUM ## _c; \ | |
4149 c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## NUM ## _x2_c; \ | |
4150 c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## NUM ## _y2_c; \ | |
4151 c->PFX ## _pixels_tab[IDX][3] = PFX ## _pixels ## NUM ## _xy2_c | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4152 |
859 | 4153 dspfunc(put, 0, 16); |
4154 dspfunc(put_no_rnd, 0, 16); | |
4155 dspfunc(put, 1, 8); | |
4156 dspfunc(put_no_rnd, 1, 8); | |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
4157 dspfunc(put, 2, 4); |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
4158 dspfunc(put, 3, 2); |
0 | 4159 |
859 | 4160 dspfunc(avg, 0, 16); |
4161 dspfunc(avg_no_rnd, 0, 16); | |
4162 dspfunc(avg, 1, 8); | |
4163 dspfunc(avg_no_rnd, 1, 8); | |
1319 | 4164 dspfunc(avg, 2, 4); |
4165 dspfunc(avg, 3, 2); | |
859 | 4166 #undef dspfunc |
857 | 4167 |
1864 | 4168 c->put_no_rnd_pixels_l2[0]= put_no_rnd_pixels16_l2_c; |
4169 c->put_no_rnd_pixels_l2[1]= put_no_rnd_pixels8_l2_c; | |
4170 | |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
4171 c->put_tpel_pixels_tab[ 0] = put_tpel_pixels_mc00_c; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
4172 c->put_tpel_pixels_tab[ 1] = put_tpel_pixels_mc10_c; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
4173 c->put_tpel_pixels_tab[ 2] = put_tpel_pixels_mc20_c; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
4174 c->put_tpel_pixels_tab[ 4] = put_tpel_pixels_mc01_c; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
4175 c->put_tpel_pixels_tab[ 5] = put_tpel_pixels_mc11_c; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
4176 c->put_tpel_pixels_tab[ 6] = put_tpel_pixels_mc21_c; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
4177 c->put_tpel_pixels_tab[ 8] = put_tpel_pixels_mc02_c; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
4178 c->put_tpel_pixels_tab[ 9] = put_tpel_pixels_mc12_c; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
4179 c->put_tpel_pixels_tab[10] = put_tpel_pixels_mc22_c; |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
4180 |
1319 | 4181 c->avg_tpel_pixels_tab[ 0] = avg_tpel_pixels_mc00_c; |
4182 c->avg_tpel_pixels_tab[ 1] = avg_tpel_pixels_mc10_c; | |
4183 c->avg_tpel_pixels_tab[ 2] = avg_tpel_pixels_mc20_c; | |
4184 c->avg_tpel_pixels_tab[ 4] = avg_tpel_pixels_mc01_c; | |
4185 c->avg_tpel_pixels_tab[ 5] = avg_tpel_pixels_mc11_c; | |
4186 c->avg_tpel_pixels_tab[ 6] = avg_tpel_pixels_mc21_c; | |
4187 c->avg_tpel_pixels_tab[ 8] = avg_tpel_pixels_mc02_c; | |
4188 c->avg_tpel_pixels_tab[ 9] = avg_tpel_pixels_mc12_c; | |
4189 c->avg_tpel_pixels_tab[10] = avg_tpel_pixels_mc22_c; | |
4190 | |
859 | 4191 #define dspfunc(PFX, IDX, NUM) \ |
4192 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_c; \ | |
4193 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_c; \ | |
4194 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_c; \ | |
4195 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_c; \ | |
4196 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_c; \ | |
4197 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_c; \ | |
4198 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_c; \ | |
4199 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_c; \ | |
4200 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_c; \ | |
4201 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_c; \ | |
4202 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_c; \ | |
4203 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_c; \ | |
4204 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_c; \ | |
4205 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_c; \ | |
4206 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_c; \ | |
4207 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_c | |
857 | 4208 |
859 | 4209 dspfunc(put_qpel, 0, 16); |
4210 dspfunc(put_no_rnd_qpel, 0, 16); | |
4211 | |
4212 dspfunc(avg_qpel, 0, 16); | |
4213 /* dspfunc(avg_no_rnd_qpel, 0, 16); */ | |
857 | 4214 |
859 | 4215 dspfunc(put_qpel, 1, 8); |
4216 dspfunc(put_no_rnd_qpel, 1, 8); | |
4217 | |
4218 dspfunc(avg_qpel, 1, 8); | |
4219 /* dspfunc(avg_no_rnd_qpel, 1, 8); */ | |
1168 | 4220 |
4221 dspfunc(put_h264_qpel, 0, 16); | |
4222 dspfunc(put_h264_qpel, 1, 8); | |
4223 dspfunc(put_h264_qpel, 2, 4); | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
4224 dspfunc(put_h264_qpel, 3, 2); |
1168 | 4225 dspfunc(avg_h264_qpel, 0, 16); |
4226 dspfunc(avg_h264_qpel, 1, 8); | |
4227 dspfunc(avg_h264_qpel, 2, 4); | |
4228 | |
859 | 4229 #undef dspfunc |
1168 | 4230 c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_c; |
4231 c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_c; | |
4232 c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_c; | |
4233 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_c; | |
4234 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_c; | |
4235 c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_c; | |
3663 | 4236 c->put_no_rnd_h264_chroma_pixels_tab[0]= put_no_rnd_h264_chroma_mc8_c; |
857 | 4237 |
2415 | 4238 c->weight_h264_pixels_tab[0]= weight_h264_pixels16x16_c; |
4239 c->weight_h264_pixels_tab[1]= weight_h264_pixels16x8_c; | |
4240 c->weight_h264_pixels_tab[2]= weight_h264_pixels8x16_c; | |
4241 c->weight_h264_pixels_tab[3]= weight_h264_pixels8x8_c; | |
4242 c->weight_h264_pixels_tab[4]= weight_h264_pixels8x4_c; | |
4243 c->weight_h264_pixels_tab[5]= weight_h264_pixels4x8_c; | |
4244 c->weight_h264_pixels_tab[6]= weight_h264_pixels4x4_c; | |
4245 c->weight_h264_pixels_tab[7]= weight_h264_pixels4x2_c; | |
4246 c->weight_h264_pixels_tab[8]= weight_h264_pixels2x4_c; | |
4247 c->weight_h264_pixels_tab[9]= weight_h264_pixels2x2_c; | |
4248 c->biweight_h264_pixels_tab[0]= biweight_h264_pixels16x16_c; | |
4249 c->biweight_h264_pixels_tab[1]= biweight_h264_pixels16x8_c; | |
4250 c->biweight_h264_pixels_tab[2]= biweight_h264_pixels8x16_c; | |
4251 c->biweight_h264_pixels_tab[3]= biweight_h264_pixels8x8_c; | |
4252 c->biweight_h264_pixels_tab[4]= biweight_h264_pixels8x4_c; | |
4253 c->biweight_h264_pixels_tab[5]= biweight_h264_pixels4x8_c; | |
4254 c->biweight_h264_pixels_tab[6]= biweight_h264_pixels4x4_c; | |
4255 c->biweight_h264_pixels_tab[7]= biweight_h264_pixels4x2_c; | |
4256 c->biweight_h264_pixels_tab[8]= biweight_h264_pixels2x4_c; | |
4257 c->biweight_h264_pixels_tab[9]= biweight_h264_pixels2x2_c; | |
4258 | |
6437 | 4259 c->draw_edges = draw_edges_c; |
4260 | |
3432 | 4261 #ifdef CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4262 ff_cavsdsp_init(c,avctx); |
3432 | 4263 #endif |
3526 | 4264 #if defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
4265 ff_vc1dsp_init(c,avctx); | |
4266 #endif | |
5887 | 4267 #if defined(CONFIG_WMV2_DECODER) || defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
4268 ff_intrax8dsp_init(c,avctx); | |
4269 #endif | |
4296 | 4270 #if defined(CONFIG_H264_ENCODER) |
5411
362aec4ef932
Take care of some renames (Doxygen and function name) after the previous pure rename patch.
takis
parents:
5394
diff
changeset
|
4271 ff_h264dspenc_init(c,avctx); |
4296 | 4272 #endif |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4273 |
936 | 4274 c->put_mspel_pixels_tab[0]= put_mspel8_mc00_c; |
4275 c->put_mspel_pixels_tab[1]= put_mspel8_mc10_c; | |
4276 c->put_mspel_pixels_tab[2]= put_mspel8_mc20_c; | |
4277 c->put_mspel_pixels_tab[3]= put_mspel8_mc30_c; | |
4278 c->put_mspel_pixels_tab[4]= put_mspel8_mc02_c; | |
4279 c->put_mspel_pixels_tab[5]= put_mspel8_mc12_c; | |
4280 c->put_mspel_pixels_tab[6]= put_mspel8_mc22_c; | |
4281 c->put_mspel_pixels_tab[7]= put_mspel8_mc32_c; | |
2967 | 4282 |
1708 | 4283 #define SET_CMP_FUNC(name) \ |
4284 c->name[0]= name ## 16_c;\ | |
4285 c->name[1]= name ## 8x8_c; | |
2967 | 4286 |
1708 | 4287 SET_CMP_FUNC(hadamard8_diff) |
1729 | 4288 c->hadamard8_diff[4]= hadamard8_intra16_c; |
1708 | 4289 SET_CMP_FUNC(dct_sad) |
2382 | 4290 SET_CMP_FUNC(dct_max) |
3013 | 4291 #ifdef CONFIG_GPL |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
4292 SET_CMP_FUNC(dct264_sad) |
3013 | 4293 #endif |
1708 | 4294 c->sad[0]= pix_abs16_c; |
4295 c->sad[1]= pix_abs8_c; | |
4296 c->sse[0]= sse16_c; | |
4297 c->sse[1]= sse8_c; | |
2184 | 4298 c->sse[2]= sse4_c; |
1708 | 4299 SET_CMP_FUNC(quant_psnr) |
4300 SET_CMP_FUNC(rd) | |
4301 SET_CMP_FUNC(bit) | |
1729 | 4302 c->vsad[0]= vsad16_c; |
4303 c->vsad[4]= vsad_intra16_c; | |
4304 c->vsse[0]= vsse16_c; | |
4305 c->vsse[4]= vsse_intra16_c; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4306 c->nsse[0]= nsse16_c; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4307 c->nsse[1]= nsse8_c; |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
4308 #ifdef CONFIG_SNOW_ENCODER |
2184 | 4309 c->w53[0]= w53_16_c; |
4310 c->w53[1]= w53_8_c; | |
4311 c->w97[0]= w97_16_c; | |
4312 c->w97[1]= w97_8_c; | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
4313 #endif |
2184 | 4314 |
4749 | 4315 c->ssd_int8_vs_int16 = ssd_int8_vs_int16_c; |
4316 | |
866 | 4317 c->add_bytes= add_bytes_c; |
6384 | 4318 c->add_bytes_l2= add_bytes_l2_c; |
866 | 4319 c->diff_bytes= diff_bytes_c; |
1527 | 4320 c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_c; |
1273 | 4321 c->bswap_buf= bswap_buf; |
6384 | 4322 #ifdef CONFIG_PNG_DECODER |
4323 c->add_png_paeth_prediction= ff_add_png_paeth_prediction; | |
4324 #endif | |
2633 | 4325 |
4326 c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_c; | |
4327 c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_c; | |
4328 c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_c; | |
4329 c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_c; | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
4330 c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_c; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
4331 c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_c; |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3568
diff
changeset
|
4332 c->h264_loop_filter_strength= NULL; |
2967 | 4333 |
5277
7b3fcb7c61ce
Avoid linking with h263.c functions when the relevant codecs
aurel
parents:
5256
diff
changeset
|
4334 if (ENABLE_ANY_H263) { |
5278 | 4335 c->h263_h_loop_filter= h263_h_loop_filter_c; |
4336 c->h263_v_loop_filter= h263_v_loop_filter_c; | |
5277
7b3fcb7c61ce
Avoid linking with h263.c functions when the relevant codecs
aurel
parents:
5256
diff
changeset
|
4337 } |
2967 | 4338 |
2045 | 4339 c->h261_loop_filter= h261_loop_filter_c; |
2967 | 4340 |
1784 | 4341 c->try_8x8basis= try_8x8basis_c; |
4342 c->add_8x8basis= add_8x8basis_c; | |
866 | 4343 |
4598
5111fceeb971
The Snow DEcoding routines should be under CONFIG_SNOW_DECODER.
diego
parents:
4594
diff
changeset
|
4344 #ifdef CONFIG_SNOW_DECODER |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4345 c->vertical_compose97i = ff_snow_vertical_compose97i; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4346 c->horizontal_compose97i = ff_snow_horizontal_compose97i; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4347 c->inner_add_yblock = ff_snow_inner_add_yblock; |
3199
1651e69b9f7a
10l: Only set *compose97i *add_yblock to dsputils context if we are building with Snow enabled
gpoirier
parents:
3198
diff
changeset
|
4348 #endif |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4349 |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4350 #ifdef CONFIG_VORBIS_DECODER |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4351 c->vorbis_inverse_coupling = vorbis_inverse_coupling; |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4352 #endif |
5737 | 4353 #ifdef CONFIG_FLAC_ENCODER |
4354 c->flac_compute_autocorr = ff_flac_compute_autocorr; | |
4355 #endif | |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4356 c->vector_fmul = vector_fmul_c; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4357 c->vector_fmul_reverse = vector_fmul_reverse_c; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4358 c->vector_fmul_add_add = ff_vector_fmul_add_add_c; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4359 c->float_to_int16 = ff_float_to_int16_c; |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4360 |
3245 | 4361 c->shrink[0]= ff_img_copy_plane; |
4362 c->shrink[1]= ff_shrink22; | |
4363 c->shrink[2]= ff_shrink44; | |
4364 c->shrink[3]= ff_shrink88; | |
4365 | |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4366 c->prefetch= just_return; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4367 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4368 memset(c->put_2tap_qpel_pixels_tab, 0, sizeof(c->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:
3728
diff
changeset
|
4369 memset(c->avg_2tap_qpel_pixels_tab, 0, sizeof(c->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:
3728
diff
changeset
|
4370 |
5146
5db8d038eb1f
use if() instead of #ifdef and ensure all possible optimisations are enabled
aurel
parents:
5143
diff
changeset
|
4371 if (ENABLE_MMX) dsputil_init_mmx (c, avctx); |
5db8d038eb1f
use if() instead of #ifdef and ensure all possible optimisations are enabled
aurel
parents:
5143
diff
changeset
|
4372 if (ENABLE_ARMV4L) dsputil_init_armv4l(c, avctx); |
5db8d038eb1f
use if() instead of #ifdef and ensure all possible optimisations are enabled
aurel
parents:
5143
diff
changeset
|
4373 if (ENABLE_MLIB) dsputil_init_mlib (c, avctx); |
5703
ec04964a1d1a
VIS-specific code should be enabled conditional to HAVE_VIS, not ARCH_SPARC.
diego
parents:
5542
diff
changeset
|
4374 if (ENABLE_VIS) dsputil_init_vis (c, avctx); |
5146
5db8d038eb1f
use if() instead of #ifdef and ensure all possible optimisations are enabled
aurel
parents:
5143
diff
changeset
|
4375 if (ENABLE_ALPHA) dsputil_init_alpha (c, avctx); |
5db8d038eb1f
use if() instead of #ifdef and ensure all possible optimisations are enabled
aurel
parents:
5143
diff
changeset
|
4376 if (ENABLE_POWERPC) dsputil_init_ppc (c, avctx); |
5db8d038eb1f
use if() instead of #ifdef and ensure all possible optimisations are enabled
aurel
parents:
5143
diff
changeset
|
4377 if (ENABLE_MMI) dsputil_init_mmi (c, avctx); |
5db8d038eb1f
use if() instead of #ifdef and ensure all possible optimisations are enabled
aurel
parents:
5143
diff
changeset
|
4378 if (ENABLE_SH4) dsputil_init_sh4 (c, avctx); |
5db8d038eb1f
use if() instead of #ifdef and ensure all possible optimisations are enabled
aurel
parents:
5143
diff
changeset
|
4379 if (ENABLE_BFIN) dsputil_init_bfin (c, avctx); |
1092 | 4380 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4381 for(i=0; i<64; i++){ |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4382 if(!c->put_2tap_qpel_pixels_tab[0][i]) |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4383 c->put_2tap_qpel_pixels_tab[0][i]= c->put_h264_qpel_pixels_tab[0][i]; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4384 if(!c->avg_2tap_qpel_pixels_tab[0][i]) |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4385 c->avg_2tap_qpel_pixels_tab[0][i]= c->avg_h264_qpel_pixels_tab[0][i]; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4386 } |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4387 |
1092 | 4388 switch(c->idct_permutation_type){ |
4389 case FF_NO_IDCT_PERM: | |
4390 for(i=0; i<64; i++) | |
4391 c->idct_permutation[i]= i; | |
4392 break; | |
4393 case FF_LIBMPEG2_IDCT_PERM: | |
4394 for(i=0; i<64; i++) | |
4395 c->idct_permutation[i]= (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2); | |
4396 break; | |
4397 case FF_SIMPLE_IDCT_PERM: | |
4398 for(i=0; i<64; i++) | |
4399 c->idct_permutation[i]= simple_mmx_permutation[i]; | |
4400 break; | |
4401 case FF_TRANSPOSE_IDCT_PERM: | |
4402 for(i=0; i<64; i++) | |
4403 c->idct_permutation[i]= ((i&7)<<3) | (i>>3); | |
4404 break; | |
2696
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4405 case FF_PARTTRANS_IDCT_PERM: |
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4406 for(i=0; i<64; i++) |
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4407 c->idct_permutation[i]= (i&0x24) | ((i&3)<<3) | ((i>>3)&3); |
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4408 break; |
1092 | 4409 default: |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1571
diff
changeset
|
4410 av_log(avctx, AV_LOG_ERROR, "Internal error, IDCT permutation not set\n"); |
1092 | 4411 } |
0 | 4412 } |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
220
diff
changeset
|
4413 |