Mercurial > libavcodec.hg
annotate dsputil.c @ 6550:3de38485c56d libavcodec
help gcc-2.95
author | michael |
---|---|
date | Mon, 31 Mar 2008 16:43:27 +0000 |
parents | c32be43b52b2 |
children | c3213c91124c |
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" | |
1092 | 32 #include "simple_idct.h" |
1557 | 33 #include "faandct.h" |
6407 | 34 #include "faanidct.h" |
5277
7b3fcb7c61ce
Avoid linking with h263.c functions when the relevant codecs
aurel
parents:
5256
diff
changeset
|
35 #include "h263.h" |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
36 #include "snow.h" |
676 | 37 |
2522
e25782262d7d
kill warnings patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2448
diff
changeset
|
38 /* snow.c */ |
e25782262d7d
kill warnings patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2448
diff
changeset
|
39 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
|
40 |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
41 /* vorbis.c */ |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
42 void vorbis_inverse_coupling(float *mag, float *ang, int blocksize); |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
43 |
5737 | 44 /* flacenc.c */ |
45 void ff_flac_compute_autocorr(const int32_t *data, int len, int lag, double *autoc); | |
46 | |
6384 | 47 /* pngdec.c */ |
48 void ff_add_png_paeth_prediction(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp); | |
49 | |
4176 | 50 uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP] = {0, }; |
4179 | 51 uint32_t ff_squareTbl[512] = {0, }; |
0 | 52 |
6387 | 53 // 0x7f7f7f7f or 0x7f7f7f7f7f7f7f7f or whatever, depending on the cpu's native arithmetic size |
54 #define pb_7f (~0UL/255 * 0x7f) | |
55 #define pb_80 (~0UL/255 * 0x80) | |
6385 | 56 |
1064 | 57 const uint8_t ff_zigzag_direct[64] = { |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
58 0, 1, 8, 16, 9, 2, 3, 10, |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
59 17, 24, 32, 25, 18, 11, 4, 5, |
34 | 60 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
|
61 27, 20, 13, 6, 7, 14, 21, 28, |
34 | 62 35, 42, 49, 56, 57, 50, 43, 36, |
63 29, 22, 15, 23, 30, 37, 44, 51, | |
64 58, 59, 52, 45, 38, 31, 39, 46, | |
65 53, 60, 61, 54, 47, 55, 62, 63 | |
66 }; | |
67 | |
1567 | 68 /* Specific zigzag scan for 248 idct. NOTE that unlike the |
69 specification, we interleave the fields */ | |
70 const uint8_t ff_zigzag248_direct[64] = { | |
71 0, 8, 1, 9, 16, 24, 2, 10, | |
72 17, 25, 32, 40, 48, 56, 33, 41, | |
73 18, 26, 3, 11, 4, 12, 19, 27, | |
74 34, 42, 49, 57, 50, 58, 35, 43, | |
75 20, 28, 5, 13, 6, 14, 21, 29, | |
76 36, 44, 51, 59, 52, 60, 37, 45, | |
77 22, 30, 7, 15, 23, 31, 38, 46, | |
78 53, 61, 54, 62, 39, 47, 55, 63, | |
79 }; | |
80 | |
220 | 81 /* not permutated inverse zigzag_direct + 1 for MMX quantizer */ |
4197 | 82 DECLARE_ALIGNED_8(uint16_t, inv_zigzag_direct16[64]) = {0, }; |
220 | 83 |
1064 | 84 const uint8_t ff_alternate_horizontal_scan[64] = { |
2967 | 85 0, 1, 2, 3, 8, 9, 16, 17, |
34 | 86 10, 11, 4, 5, 6, 7, 15, 14, |
2967 | 87 13, 12, 19, 18, 24, 25, 32, 33, |
34 | 88 26, 27, 20, 21, 22, 23, 28, 29, |
2967 | 89 30, 31, 34, 35, 40, 41, 48, 49, |
34 | 90 42, 43, 36, 37, 38, 39, 44, 45, |
2967 | 91 46, 47, 50, 51, 56, 57, 58, 59, |
34 | 92 52, 53, 54, 55, 60, 61, 62, 63, |
93 }; | |
94 | |
1064 | 95 const uint8_t ff_alternate_vertical_scan[64] = { |
2967 | 96 0, 8, 16, 24, 1, 9, 2, 10, |
34 | 97 17, 25, 32, 40, 48, 56, 57, 49, |
2967 | 98 41, 33, 26, 18, 3, 11, 4, 12, |
34 | 99 19, 27, 34, 42, 50, 58, 35, 43, |
2967 | 100 51, 59, 20, 28, 5, 13, 6, 14, |
34 | 101 21, 29, 36, 44, 52, 60, 37, 45, |
2967 | 102 53, 61, 22, 30, 7, 15, 23, 31, |
34 | 103 38, 46, 54, 62, 39, 47, 55, 63, |
104 }; | |
105 | |
220 | 106 /* a*inverse[b]>>32 == a/b for all 0<=a<=65536 && 2<=b<=255 */ |
4174 | 107 const uint32_t ff_inverse[256]={ |
2967 | 108 0, 4294967295U,2147483648U,1431655766, 1073741824, 858993460, 715827883, 613566757, |
109 536870912, 477218589, 429496730, 390451573, 357913942, 330382100, 306783379, 286331154, | |
110 268435456, 252645136, 238609295, 226050911, 214748365, 204522253, 195225787, 186737709, | |
111 178956971, 171798692, 165191050, 159072863, 153391690, 148102321, 143165577, 138547333, | |
112 134217728, 130150525, 126322568, 122713352, 119304648, 116080198, 113025456, 110127367, | |
113 107374183, 104755300, 102261127, 99882961, 97612894, 95443718, 93368855, 91382283, | |
114 89478486, 87652394, 85899346, 84215046, 82595525, 81037119, 79536432, 78090315, | |
115 76695845, 75350304, 74051161, 72796056, 71582789, 70409300, 69273667, 68174085, | |
116 67108864, 66076420, 65075263, 64103990, 63161284, 62245903, 61356676, 60492498, | |
117 59652324, 58835169, 58040099, 57266231, 56512728, 55778797, 55063684, 54366675, | |
118 53687092, 53024288, 52377650, 51746594, 51130564, 50529028, 49941481, 49367441, | |
119 48806447, 48258060, 47721859, 47197443, 46684428, 46182445, 45691142, 45210183, | |
120 44739243, 44278014, 43826197, 43383509, 42949673, 42524429, 42107523, 41698712, | |
121 41297763, 40904451, 40518560, 40139882, 39768216, 39403370, 39045158, 38693400, | |
122 38347923, 38008561, 37675152, 37347542, 37025581, 36709123, 36398028, 36092163, | |
123 35791395, 35495598, 35204650, 34918434, 34636834, 34359739, 34087043, 33818641, | |
124 33554432, 33294321, 33038210, 32786010, 32537632, 32292988, 32051995, 31814573, | |
125 31580642, 31350127, 31122952, 30899046, 30678338, 30460761, 30246249, 30034737, | |
126 29826162, 29620465, 29417585, 29217465, 29020050, 28825284, 28633116, 28443493, | |
127 28256364, 28071682, 27889399, 27709467, 27531842, 27356480, 27183338, 27012373, | |
128 26843546, 26676816, 26512144, 26349493, 26188825, 26030105, 25873297, 25718368, | |
129 25565282, 25414008, 25264514, 25116768, 24970741, 24826401, 24683721, 24542671, | |
130 24403224, 24265352, 24129030, 23994231, 23860930, 23729102, 23598722, 23469767, | |
131 23342214, 23216040, 23091223, 22967740, 22845571, 22724695, 22605092, 22486740, | |
132 22369622, 22253717, 22139007, 22025474, 21913099, 21801865, 21691755, 21582751, | |
133 21474837, 21367997, 21262215, 21157475, 21053762, 20951060, 20849356, 20748635, | |
134 20648882, 20550083, 20452226, 20355296, 20259280, 20164166, 20069941, 19976593, | |
135 19884108, 19792477, 19701685, 19611723, 19522579, 19434242, 19346700, 19259944, | |
136 19173962, 19088744, 19004281, 18920561, 18837576, 18755316, 18673771, 18592933, | |
137 18512791, 18433337, 18354562, 18276457, 18199014, 18122225, 18046082, 17970575, | |
138 17895698, 17821442, 17747799, 17674763, 17602325, 17530479, 17459217, 17388532, | |
220 | 139 17318417, 17248865, 17179870, 17111424, 17043522, 16976156, 16909321, 16843010, |
140 }; | |
141 | |
1092 | 142 /* Input permutation for the simple_idct_mmx */ |
143 static const uint8_t simple_mmx_permutation[64]={ | |
2979 | 144 0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, |
145 0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, | |
146 0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, | |
147 0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, | |
148 0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, | |
149 0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, | |
150 0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, | |
151 0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F, | |
1092 | 152 }; |
153 | |
6438 | 154 void ff_init_scantable(uint8_t *permutation, ScanTable *st, const uint8_t *src_scantable){ |
155 int i; | |
156 int end; | |
157 | |
158 st->scantable= src_scantable; | |
159 | |
160 for(i=0; i<64; i++){ | |
161 int j; | |
162 j = src_scantable[i]; | |
163 st->permutated[i] = permutation[j]; | |
164 #ifdef ARCH_POWERPC | |
165 st->inverse[j] = i; | |
166 #endif | |
167 } | |
168 | |
169 end=-1; | |
170 for(i=0; i<64; i++){ | |
171 int j; | |
172 j = st->permutated[i]; | |
173 if(j>end) end=j; | |
174 st->raster_end[i]= end; | |
175 } | |
176 } | |
177 | |
1064 | 178 static int pix_sum_c(uint8_t * pix, int line_size) |
612 | 179 { |
180 int s, i, j; | |
181 | |
182 s = 0; | |
183 for (i = 0; i < 16; i++) { | |
2979 | 184 for (j = 0; j < 16; j += 8) { |
185 s += pix[0]; | |
186 s += pix[1]; | |
187 s += pix[2]; | |
188 s += pix[3]; | |
189 s += pix[4]; | |
190 s += pix[5]; | |
191 s += pix[6]; | |
192 s += pix[7]; | |
193 pix += 8; | |
194 } | |
195 pix += line_size - 16; | |
612 | 196 } |
197 return s; | |
198 } | |
199 | |
1064 | 200 static int pix_norm1_c(uint8_t * pix, int line_size) |
612 | 201 { |
202 int s, i, j; | |
4179 | 203 uint32_t *sq = ff_squareTbl + 256; |
612 | 204 |
205 s = 0; | |
206 for (i = 0; i < 16; i++) { | |
2979 | 207 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
|
208 #if 0 |
2979 | 209 s += sq[pix[0]]; |
210 s += sq[pix[1]]; | |
211 s += sq[pix[2]]; | |
212 s += sq[pix[3]]; | |
213 s += sq[pix[4]]; | |
214 s += sq[pix[5]]; | |
215 s += sq[pix[6]]; | |
216 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
|
217 #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
|
218 #if LONG_MAX > 2147483647 |
2979 | 219 register uint64_t x=*(uint64_t*)pix; |
220 s += sq[x&0xff]; | |
221 s += sq[(x>>8)&0xff]; | |
222 s += sq[(x>>16)&0xff]; | |
223 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
|
224 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
|
225 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
|
226 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
|
227 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
|
228 #else |
2979 | 229 register uint32_t x=*(uint32_t*)pix; |
230 s += sq[x&0xff]; | |
231 s += sq[(x>>8)&0xff]; | |
232 s += sq[(x>>16)&0xff]; | |
233 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
|
234 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
|
235 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
|
236 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
|
237 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
|
238 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
|
239 #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
|
240 #endif |
2979 | 241 pix += 8; |
242 } | |
243 pix += line_size - 16; | |
612 | 244 } |
245 return s; | |
246 } | |
247 | |
6241 | 248 static void bswap_buf(uint32_t *dst, const uint32_t *src, int w){ |
1273 | 249 int i; |
2967 | 250 |
1273 | 251 for(i=0; i+8<=w; i+=8){ |
252 dst[i+0]= bswap_32(src[i+0]); | |
253 dst[i+1]= bswap_32(src[i+1]); | |
254 dst[i+2]= bswap_32(src[i+2]); | |
255 dst[i+3]= bswap_32(src[i+3]); | |
256 dst[i+4]= bswap_32(src[i+4]); | |
257 dst[i+5]= bswap_32(src[i+5]); | |
258 dst[i+6]= bswap_32(src[i+6]); | |
259 dst[i+7]= bswap_32(src[i+7]); | |
260 } | |
261 for(;i<w; i++){ | |
262 dst[i+0]= bswap_32(src[i+0]); | |
263 } | |
264 } | |
612 | 265 |
2184 | 266 static int sse4_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
267 { | |
268 int s, i; | |
4179 | 269 uint32_t *sq = ff_squareTbl + 256; |
2184 | 270 |
271 s = 0; | |
272 for (i = 0; i < h; i++) { | |
273 s += sq[pix1[0] - pix2[0]]; | |
274 s += sq[pix1[1] - pix2[1]]; | |
275 s += sq[pix1[2] - pix2[2]]; | |
276 s += sq[pix1[3] - pix2[3]]; | |
277 pix1 += line_size; | |
278 pix2 += line_size; | |
279 } | |
280 return s; | |
281 } | |
282 | |
1708 | 283 static int sse8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
936 | 284 { |
285 int s, i; | |
4179 | 286 uint32_t *sq = ff_squareTbl + 256; |
936 | 287 |
288 s = 0; | |
1708 | 289 for (i = 0; i < h; i++) { |
936 | 290 s += sq[pix1[0] - pix2[0]]; |
291 s += sq[pix1[1] - pix2[1]]; | |
292 s += sq[pix1[2] - pix2[2]]; | |
293 s += sq[pix1[3] - pix2[3]]; | |
294 s += sq[pix1[4] - pix2[4]]; | |
295 s += sq[pix1[5] - pix2[5]]; | |
296 s += sq[pix1[6] - pix2[6]]; | |
297 s += sq[pix1[7] - pix2[7]]; | |
298 pix1 += line_size; | |
299 pix2 += line_size; | |
300 } | |
301 return s; | |
302 } | |
303 | |
1708 | 304 static int sse16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
884 | 305 { |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
306 int s, i; |
4179 | 307 uint32_t *sq = ff_squareTbl + 256; |
884 | 308 |
309 s = 0; | |
1708 | 310 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
|
311 s += sq[pix1[ 0] - pix2[ 0]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
312 s += sq[pix1[ 1] - pix2[ 1]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
313 s += sq[pix1[ 2] - pix2[ 2]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
314 s += sq[pix1[ 3] - pix2[ 3]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
315 s += sq[pix1[ 4] - pix2[ 4]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
316 s += sq[pix1[ 5] - pix2[ 5]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
317 s += sq[pix1[ 6] - pix2[ 6]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
318 s += sq[pix1[ 7] - pix2[ 7]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
319 s += sq[pix1[ 8] - pix2[ 8]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
320 s += sq[pix1[ 9] - pix2[ 9]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
321 s += sq[pix1[10] - pix2[10]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
322 s += sq[pix1[11] - pix2[11]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
323 s += sq[pix1[12] - pix2[12]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
324 s += sq[pix1[13] - pix2[13]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
325 s += sq[pix1[14] - pix2[14]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
326 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
|
327 |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
328 pix1 += line_size; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
329 pix2 += line_size; |
884 | 330 } |
331 return s; | |
332 } | |
333 | |
2184 | 334 |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
335 #ifdef CONFIG_SNOW_ENCODER //dwt is in snow.c |
2184 | 336 static inline int w_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int w, int h, int type){ |
337 int s, i, j; | |
338 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
|
339 int tmp[32*32]; |
2184 | 340 int level, ori; |
2967 | 341 static const int scale[2][2][4][4]={ |
2184 | 342 { |
343 { | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
344 // 9/7 8x8 dec=3 |
2184 | 345 {268, 239, 239, 213}, |
346 { 0, 224, 224, 152}, | |
347 { 0, 135, 135, 110}, | |
348 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
349 // 9/7 16x16 or 32x32 dec=4 |
2184 | 350 {344, 310, 310, 280}, |
351 { 0, 320, 320, 228}, | |
352 { 0, 175, 175, 136}, | |
353 { 0, 129, 129, 102}, | |
354 } | |
355 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
356 { |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
357 // 5/3 8x8 dec=3 |
2184 | 358 {275, 245, 245, 218}, |
359 { 0, 230, 230, 156}, | |
360 { 0, 138, 138, 113}, | |
361 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
362 // 5/3 16x16 or 32x32 dec=4 |
2184 | 363 {352, 317, 317, 286}, |
364 { 0, 328, 328, 233}, | |
365 { 0, 180, 180, 140}, | |
366 { 0, 132, 132, 105}, | |
367 } | |
368 } | |
369 }; | |
370 | |
371 for (i = 0; i < h; i++) { | |
372 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
|
373 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
|
374 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
|
375 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
|
376 tmp[32*i+j+3] = (pix1[j+3] - pix2[j+3])<<4; |
2184 | 377 } |
378 pix1 += line_size; | |
379 pix2 += line_size; | |
380 } | |
2639 | 381 |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
382 ff_spatial_dwt(tmp, w, h, 32, type, dec_count); |
2184 | 383 |
384 s=0; | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
385 assert(w==h); |
2184 | 386 for(level=0; level<dec_count; level++){ |
387 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
|
388 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
|
389 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
|
390 int stride= 32<<(dec_count-level); |
2184 | 391 int sy= (ori&2) ? stride>>1 : 0; |
2967 | 392 |
2184 | 393 for(i=0; i<size; i++){ |
394 for(j=0; j<size; j++){ | |
395 int v= tmp[sx + sy + i*stride + j] * scale[type][dec_count-3][level][ori]; | |
4001 | 396 s += FFABS(v); |
2184 | 397 } |
398 } | |
399 } | |
400 } | |
2967 | 401 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
|
402 return s>>9; |
2184 | 403 } |
404 | |
405 static int w53_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
406 return w_c(v, pix1, pix2, line_size, 8, h, 1); | |
407 } | |
408 | |
409 static int w97_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
410 return w_c(v, pix1, pix2, line_size, 8, h, 0); | |
411 } | |
412 | |
413 static int w53_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
414 return w_c(v, pix1, pix2, line_size, 16, h, 1); | |
415 } | |
416 | |
417 static int w97_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
418 return w_c(v, pix1, pix2, line_size, 16, h, 0); | |
419 } | |
420 | |
4197 | 421 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
|
422 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
|
423 } |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
424 |
4197 | 425 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
|
426 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
|
427 } |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
428 #endif |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
429 |
6437 | 430 /* draw the edges of width 'w' of an image of size width, height */ |
431 //FIXME check that this is ok for mpeg4 interlaced | |
432 static void draw_edges_c(uint8_t *buf, int wrap, int width, int height, int w) | |
433 { | |
434 uint8_t *ptr, *last_line; | |
435 int i; | |
436 | |
437 last_line = buf + (height - 1) * wrap; | |
438 for(i=0;i<w;i++) { | |
439 /* top and bottom */ | |
440 memcpy(buf - (i + 1) * wrap, buf, width); | |
441 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
442 } | |
443 /* left and right */ | |
444 ptr = buf; | |
445 for(i=0;i<height;i++) { | |
446 memset(ptr - w, ptr[0], w); | |
447 memset(ptr + width, ptr[width-1], w); | |
448 ptr += wrap; | |
449 } | |
450 /* corners */ | |
451 for(i=0;i<w;i++) { | |
452 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
453 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
454 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
455 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
456 } | |
457 } | |
458 | |
6445 | 459 /** |
460 * Copies a rectangular area of samples to a temporary buffer and replicates the boarder samples. | |
461 * @param buf destination buffer | |
462 * @param src source buffer | |
463 * @param linesize number of bytes between 2 vertically adjacent samples in both the source and destination buffers | |
464 * @param block_w width of block | |
465 * @param block_h height of block | |
466 * @param src_x x coordinate of the top left sample of the block in the source buffer | |
467 * @param src_y y coordinate of the top left sample of the block in the source buffer | |
468 * @param w width of the source buffer | |
469 * @param h height of the source buffer | |
470 */ | |
471 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize, int block_w, int block_h, | |
472 int src_x, int src_y, int w, int h){ | |
473 int x, y; | |
474 int start_y, start_x, end_y, end_x; | |
475 | |
476 if(src_y>= h){ | |
477 src+= (h-1-src_y)*linesize; | |
478 src_y=h-1; | |
479 }else if(src_y<=-block_h){ | |
480 src+= (1-block_h-src_y)*linesize; | |
481 src_y=1-block_h; | |
482 } | |
483 if(src_x>= w){ | |
484 src+= (w-1-src_x); | |
485 src_x=w-1; | |
486 }else if(src_x<=-block_w){ | |
487 src+= (1-block_w-src_x); | |
488 src_x=1-block_w; | |
489 } | |
490 | |
491 start_y= FFMAX(0, -src_y); | |
492 start_x= FFMAX(0, -src_x); | |
493 end_y= FFMIN(block_h, h-src_y); | |
494 end_x= FFMIN(block_w, w-src_x); | |
495 | |
496 // copy existing part | |
497 for(y=start_y; y<end_y; y++){ | |
498 for(x=start_x; x<end_x; x++){ | |
499 buf[x + y*linesize]= src[x + y*linesize]; | |
500 } | |
501 } | |
502 | |
503 //top | |
504 for(y=0; y<start_y; y++){ | |
505 for(x=start_x; x<end_x; x++){ | |
506 buf[x + y*linesize]= buf[x + start_y*linesize]; | |
507 } | |
508 } | |
509 | |
510 //bottom | |
511 for(y=end_y; y<block_h; y++){ | |
512 for(x=start_x; x<end_x; x++){ | |
513 buf[x + y*linesize]= buf[x + (end_y-1)*linesize]; | |
514 } | |
515 } | |
516 | |
517 for(y=0; y<block_h; y++){ | |
518 //left | |
519 for(x=0; x<start_x; x++){ | |
520 buf[x + y*linesize]= buf[start_x + y*linesize]; | |
521 } | |
522 | |
523 //right | |
524 for(x=end_x; x<block_w; x++){ | |
525 buf[x + y*linesize]= buf[end_x - 1 + y*linesize]; | |
526 } | |
527 } | |
528 } | |
529 | |
1064 | 530 static void get_pixels_c(DCTELEM *restrict block, const uint8_t *pixels, int line_size) |
0 | 531 { |
532 int i; | |
533 | |
534 /* read the pixels */ | |
535 for(i=0;i<8;i++) { | |
516 | 536 block[0] = pixels[0]; |
537 block[1] = pixels[1]; | |
538 block[2] = pixels[2]; | |
539 block[3] = pixels[3]; | |
540 block[4] = pixels[4]; | |
541 block[5] = pixels[5]; | |
542 block[6] = pixels[6]; | |
543 block[7] = pixels[7]; | |
544 pixels += line_size; | |
545 block += 8; | |
0 | 546 } |
547 } | |
548 | |
1064 | 549 static void diff_pixels_c(DCTELEM *restrict block, const uint8_t *s1, |
2979 | 550 const uint8_t *s2, int stride){ |
324 | 551 int i; |
552 | |
553 /* read the pixels */ | |
554 for(i=0;i<8;i++) { | |
516 | 555 block[0] = s1[0] - s2[0]; |
556 block[1] = s1[1] - s2[1]; | |
557 block[2] = s1[2] - s2[2]; | |
558 block[3] = s1[3] - s2[3]; | |
559 block[4] = s1[4] - s2[4]; | |
560 block[5] = s1[5] - s2[5]; | |
561 block[6] = s1[6] - s2[6]; | |
562 block[7] = s1[7] - s2[7]; | |
324 | 563 s1 += stride; |
564 s2 += stride; | |
516 | 565 block += 8; |
324 | 566 } |
567 } | |
568 | |
569 | |
1064 | 570 static void put_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 571 int line_size) |
0 | 572 { |
573 int i; | |
4176 | 574 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 575 |
0 | 576 /* read the pixels */ |
577 for(i=0;i<8;i++) { | |
516 | 578 pixels[0] = cm[block[0]]; |
579 pixels[1] = cm[block[1]]; | |
580 pixels[2] = cm[block[2]]; | |
581 pixels[3] = cm[block[3]]; | |
582 pixels[4] = cm[block[4]]; | |
583 pixels[5] = cm[block[5]]; | |
584 pixels[6] = cm[block[6]]; | |
585 pixels[7] = cm[block[7]]; | |
586 | |
587 pixels += line_size; | |
588 block += 8; | |
0 | 589 } |
590 } | |
591 | |
2256 | 592 static void put_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 593 int line_size) |
2256 | 594 { |
595 int i; | |
4176 | 596 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 597 |
2256 | 598 /* read the pixels */ |
599 for(i=0;i<4;i++) { | |
600 pixels[0] = cm[block[0]]; | |
601 pixels[1] = cm[block[1]]; | |
602 pixels[2] = cm[block[2]]; | |
603 pixels[3] = cm[block[3]]; | |
604 | |
605 pixels += line_size; | |
606 block += 8; | |
607 } | |
608 } | |
609 | |
2257 | 610 static void put_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 611 int line_size) |
2257 | 612 { |
613 int i; | |
4176 | 614 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 615 |
2257 | 616 /* read the pixels */ |
617 for(i=0;i<2;i++) { | |
618 pixels[0] = cm[block[0]]; | |
619 pixels[1] = cm[block[1]]; | |
620 | |
621 pixels += line_size; | |
622 block += 8; | |
623 } | |
624 } | |
625 | |
2967 | 626 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
|
627 uint8_t *restrict pixels, |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
628 int line_size) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
629 { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
630 int i, j; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
631 |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
632 for (i = 0; i < 8; i++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
633 for (j = 0; j < 8; j++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
634 if (*block < -128) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
635 *pixels = 0; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
636 else if (*block > 127) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
637 *pixels = 255; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
638 else |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
639 *pixels = (uint8_t)(*block + 128); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
640 block++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
641 pixels++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
642 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
643 pixels += (line_size - 8); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
644 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
645 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
646 |
1064 | 647 static void add_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
516 | 648 int line_size) |
0 | 649 { |
650 int i; | |
4176 | 651 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 652 |
0 | 653 /* read the pixels */ |
654 for(i=0;i<8;i++) { | |
516 | 655 pixels[0] = cm[pixels[0] + block[0]]; |
656 pixels[1] = cm[pixels[1] + block[1]]; | |
657 pixels[2] = cm[pixels[2] + block[2]]; | |
658 pixels[3] = cm[pixels[3] + block[3]]; | |
659 pixels[4] = cm[pixels[4] + block[4]]; | |
660 pixels[5] = cm[pixels[5] + block[5]]; | |
661 pixels[6] = cm[pixels[6] + block[6]]; | |
662 pixels[7] = cm[pixels[7] + block[7]]; | |
663 pixels += line_size; | |
664 block += 8; | |
0 | 665 } |
666 } | |
2256 | 667 |
668 static void add_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, | |
669 int line_size) | |
670 { | |
671 int i; | |
4176 | 672 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 673 |
2256 | 674 /* read the pixels */ |
675 for(i=0;i<4;i++) { | |
676 pixels[0] = cm[pixels[0] + block[0]]; | |
677 pixels[1] = cm[pixels[1] + block[1]]; | |
678 pixels[2] = cm[pixels[2] + block[2]]; | |
679 pixels[3] = cm[pixels[3] + block[3]]; | |
680 pixels += line_size; | |
681 block += 8; | |
682 } | |
683 } | |
2257 | 684 |
685 static void add_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, | |
686 int line_size) | |
687 { | |
688 int i; | |
4176 | 689 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 690 |
2257 | 691 /* read the pixels */ |
692 for(i=0;i<2;i++) { | |
693 pixels[0] = cm[pixels[0] + block[0]]; | |
694 pixels[1] = cm[pixels[1] + block[1]]; | |
695 pixels += line_size; | |
696 block += 8; | |
697 } | |
698 } | |
2763 | 699 |
700 static void add_pixels8_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
701 { | |
702 int i; | |
703 for(i=0;i<8;i++) { | |
704 pixels[0] += block[0]; | |
705 pixels[1] += block[1]; | |
706 pixels[2] += block[2]; | |
707 pixels[3] += block[3]; | |
708 pixels[4] += block[4]; | |
709 pixels[5] += block[5]; | |
710 pixels[6] += block[6]; | |
711 pixels[7] += block[7]; | |
712 pixels += line_size; | |
713 block += 8; | |
714 } | |
715 } | |
716 | |
717 static void add_pixels4_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
718 { | |
719 int i; | |
720 for(i=0;i<4;i++) { | |
721 pixels[0] += block[0]; | |
722 pixels[1] += block[1]; | |
723 pixels[2] += block[2]; | |
724 pixels[3] += block[3]; | |
725 pixels += line_size; | |
726 block += 4; | |
727 } | |
728 } | |
729 | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
730 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
|
731 { |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
732 int sum=0, i; |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
733 for(i=0; i<64; i++) |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
734 sum+= FFABS(block[i]); |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
735 return sum; |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
736 } |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
737 |
385 | 738 #if 0 |
739 | |
740 #define PIXOP2(OPNAME, OP) \ | |
651 | 741 static void OPNAME ## _pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 742 {\ |
743 int i;\ | |
744 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
|
745 OP(*((uint64_t*)block), AV_RN64(pixels));\ |
385 | 746 pixels+=line_size;\ |
747 block +=line_size;\ | |
748 }\ | |
749 }\ | |
750 \ | |
859 | 751 static void OPNAME ## _no_rnd_pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 752 {\ |
753 int i;\ | |
754 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
|
755 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
|
756 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 757 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
758 pixels+=line_size;\ | |
759 block +=line_size;\ | |
760 }\ | |
761 }\ | |
762 \ | |
859 | 763 static void OPNAME ## _pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 764 {\ |
765 int i;\ | |
766 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
|
767 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
|
768 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 769 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
770 pixels+=line_size;\ | |
771 block +=line_size;\ | |
772 }\ | |
773 }\ | |
774 \ | |
859 | 775 static void OPNAME ## _no_rnd_pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 776 {\ |
777 int i;\ | |
778 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
|
779 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
|
780 const uint64_t b= AV_RN64(pixels+line_size);\ |
385 | 781 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
782 pixels+=line_size;\ | |
783 block +=line_size;\ | |
784 }\ | |
785 }\ | |
786 \ | |
859 | 787 static void OPNAME ## _pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 788 {\ |
789 int i;\ | |
790 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
|
791 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
|
792 const uint64_t b= AV_RN64(pixels+line_size);\ |
385 | 793 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
794 pixels+=line_size;\ | |
795 block +=line_size;\ | |
796 }\ | |
797 }\ | |
798 \ | |
859 | 799 static void OPNAME ## _pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 800 {\ |
801 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
802 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
|
803 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 804 uint64_t l0= (a&0x0303030303030303ULL)\ |
805 + (b&0x0303030303030303ULL)\ | |
806 + 0x0202020202020202ULL;\ | |
807 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
808 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
809 uint64_t l1,h1;\ | |
810 \ | |
811 pixels+=line_size;\ | |
812 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
|
813 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
|
814 uint64_t b= AV_RN64(pixels+1);\ |
385 | 815 l1= (a&0x0303030303030303ULL)\ |
816 + (b&0x0303030303030303ULL);\ | |
817 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
818 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
819 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
820 pixels+=line_size;\ | |
821 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
822 a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
823 b= AV_RN64(pixels+1);\ |
385 | 824 l0= (a&0x0303030303030303ULL)\ |
825 + (b&0x0303030303030303ULL)\ | |
826 + 0x0202020202020202ULL;\ | |
827 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
828 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
829 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
830 pixels+=line_size;\ | |
831 block +=line_size;\ | |
832 }\ | |
833 }\ | |
834 \ | |
859 | 835 static void OPNAME ## _no_rnd_pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 836 {\ |
837 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
838 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
|
839 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 840 uint64_t l0= (a&0x0303030303030303ULL)\ |
841 + (b&0x0303030303030303ULL)\ | |
842 + 0x0101010101010101ULL;\ | |
843 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
844 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
845 uint64_t l1,h1;\ | |
846 \ | |
847 pixels+=line_size;\ | |
848 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
|
849 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
|
850 uint64_t b= AV_RN64(pixels+1);\ |
385 | 851 l1= (a&0x0303030303030303ULL)\ |
852 + (b&0x0303030303030303ULL);\ | |
853 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
854 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
855 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
856 pixels+=line_size;\ | |
857 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
858 a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
859 b= AV_RN64(pixels+1);\ |
385 | 860 l0= (a&0x0303030303030303ULL)\ |
861 + (b&0x0303030303030303ULL)\ | |
862 + 0x0101010101010101ULL;\ | |
863 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
864 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
865 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
866 pixels+=line_size;\ | |
867 block +=line_size;\ | |
868 }\ | |
869 }\ | |
870 \ | |
859 | 871 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels_c , 8)\ |
872 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels_x2_c , 8)\ | |
873 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels_y2_c , 8)\ | |
874 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels_xy2_c, 8)\ | |
875 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels_x2_c , 8)\ | |
876 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels_y2_c , 8)\ | |
877 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels_xy2_c, 8) | |
385 | 878 |
879 #define op_avg(a, b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEFEFEFEFEULL)>>1) ) | |
880 #else // 64 bit variant | |
881 | |
882 #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
|
883 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
|
884 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
|
885 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
|
886 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
|
887 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
|
888 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
|
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 }\ |
1168 | 891 static void OPNAME ## _pixels4_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
892 int i;\ | |
893 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
|
894 OP(*((uint32_t*)(block )), AV_RN32(pixels ));\ |
1168 | 895 pixels+=line_size;\ |
896 block +=line_size;\ | |
897 }\ | |
898 }\ | |
859 | 899 static void OPNAME ## _pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
385 | 900 int i;\ |
901 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
|
902 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
|
903 OP(*((uint32_t*)(block+4)), AV_RN32(pixels+4));\ |
385 | 904 pixels+=line_size;\ |
905 block +=line_size;\ | |
906 }\ | |
907 }\ | |
859 | 908 static inline void OPNAME ## _no_rnd_pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
909 OPNAME ## _pixels8_c(block, pixels, line_size, h);\ | |
651 | 910 }\ |
385 | 911 \ |
651 | 912 static inline void OPNAME ## _no_rnd_pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
913 int src_stride1, int src_stride2, int h){\ | |
385 | 914 int i;\ |
915 for(i=0; i<h; i++){\ | |
651 | 916 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
917 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
|
918 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 919 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
|
920 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
|
921 b= AV_RN32(&src2[i*src_stride2+4]);\ |
1264 | 922 OP(*((uint32_t*)&dst[i*dst_stride+4]), no_rnd_avg32(a, b));\ |
385 | 923 }\ |
924 }\ | |
925 \ | |
651 | 926 static inline void OPNAME ## _pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
927 int src_stride1, int src_stride2, int h){\ | |
385 | 928 int i;\ |
929 for(i=0; i<h; i++){\ | |
651 | 930 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
931 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
|
932 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 933 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
|
934 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
|
935 b= AV_RN32(&src2[i*src_stride2+4]);\ |
1264 | 936 OP(*((uint32_t*)&dst[i*dst_stride+4]), rnd_avg32(a, b));\ |
385 | 937 }\ |
938 }\ | |
939 \ | |
1168 | 940 static inline void OPNAME ## _pixels4_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
941 int src_stride1, int src_stride2, int h){\ | |
942 int i;\ | |
943 for(i=0; i<h; i++){\ | |
944 uint32_t a,b;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
945 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
|
946 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 947 OP(*((uint32_t*)&dst[i*dst_stride ]), rnd_avg32(a, b));\ |
1168 | 948 }\ |
949 }\ | |
950 \ | |
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
|
951 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
|
952 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
|
953 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
|
954 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
|
955 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
956 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
|
957 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
|
958 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
|
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 }\ |
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 \ |
651 | 962 static inline void OPNAME ## _pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
963 int src_stride1, int src_stride2, int h){\ | |
964 OPNAME ## _pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
965 OPNAME ## _pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
966 }\ | |
967 \ | |
968 static inline void OPNAME ## _no_rnd_pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ | |
969 int src_stride1, int src_stride2, int h){\ | |
970 OPNAME ## _no_rnd_pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
971 OPNAME ## _no_rnd_pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
972 }\ | |
973 \ | |
859 | 974 static inline void OPNAME ## _no_rnd_pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 975 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
976 }\ | |
977 \ | |
859 | 978 static inline void OPNAME ## _pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 979 OPNAME ## _pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
980 }\ | |
981 \ | |
859 | 982 static inline void OPNAME ## _no_rnd_pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 983 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
984 }\ | |
985 \ | |
859 | 986 static inline void OPNAME ## _pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 987 OPNAME ## _pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
385 | 988 }\ |
989 \ | |
651 | 990 static inline void OPNAME ## _pixels8_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
991 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
992 int i;\ | |
993 for(i=0; i<h; i++){\ | |
994 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
|
995 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
|
996 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
|
997 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
|
998 d= AV_RN32(&src4[i*src_stride4]);\ |
651 | 999 l0= (a&0x03030303UL)\ |
1000 + (b&0x03030303UL)\ | |
1001 + 0x02020202UL;\ | |
1002 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1003 + ((b&0xFCFCFCFCUL)>>2);\ | |
1004 l1= (c&0x03030303UL)\ | |
1005 + (d&0x03030303UL);\ | |
1006 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
1007 + ((d&0xFCFCFCFCUL)>>2);\ | |
1008 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
|
1009 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
|
1010 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
|
1011 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
|
1012 d= AV_RN32(&src4[i*src_stride4+4]);\ |
651 | 1013 l0= (a&0x03030303UL)\ |
1014 + (b&0x03030303UL)\ | |
1015 + 0x02020202UL;\ | |
1016 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1017 + ((b&0xFCFCFCFCUL)>>2);\ | |
1018 l1= (c&0x03030303UL)\ | |
1019 + (d&0x03030303UL);\ | |
1020 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
1021 + ((d&0xFCFCFCFCUL)>>2);\ | |
1022 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1023 }\ | |
1024 }\ | |
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
|
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 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
|
1027 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
|
1028 }\ |
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 \ |
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 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
|
1031 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
|
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 \ |
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 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
|
1035 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
|
1036 }\ |
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 \ |
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 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
|
1039 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
|
1040 }\ |
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 \ |
651 | 1042 static inline void OPNAME ## _no_rnd_pixels8_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
1043 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
385 | 1044 int i;\ |
1045 for(i=0; i<h; i++){\ | |
651 | 1046 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
|
1047 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
|
1048 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
|
1049 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
|
1050 d= AV_RN32(&src4[i*src_stride4]);\ |
651 | 1051 l0= (a&0x03030303UL)\ |
1052 + (b&0x03030303UL)\ | |
1053 + 0x01010101UL;\ | |
1054 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1055 + ((b&0xFCFCFCFCUL)>>2);\ | |
1056 l1= (c&0x03030303UL)\ | |
1057 + (d&0x03030303UL);\ | |
1058 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
1059 + ((d&0xFCFCFCFCUL)>>2);\ | |
1060 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
|
1061 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
|
1062 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
|
1063 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
|
1064 d= AV_RN32(&src4[i*src_stride4+4]);\ |
651 | 1065 l0= (a&0x03030303UL)\ |
1066 + (b&0x03030303UL)\ | |
1067 + 0x01010101UL;\ | |
1068 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1069 + ((b&0xFCFCFCFCUL)>>2);\ | |
1070 l1= (c&0x03030303UL)\ | |
1071 + (d&0x03030303UL);\ | |
1072 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
1073 + ((d&0xFCFCFCFCUL)>>2);\ | |
1074 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
385 | 1075 }\ |
1076 }\ | |
651 | 1077 static inline void OPNAME ## _pixels16_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
1078 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
1079 OPNAME ## _pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
1080 OPNAME ## _pixels8_l4(dst+8, src1+8, src2+8, src3+8, src4+8, dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
1081 }\ | |
1082 static inline void OPNAME ## _no_rnd_pixels16_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ | |
1083 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
1084 OPNAME ## _no_rnd_pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
1085 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);\ | |
1086 }\ | |
385 | 1087 \ |
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
|
1088 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
|
1089 {\ |
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
|
1090 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
|
1091 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
|
1092 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
|
1093 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
|
1094 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
|
1095 \ |
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
|
1096 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
|
1097 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
|
1098 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
|
1099 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
|
1100 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
|
1101 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
|
1102 \ |
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
|
1103 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
|
1104 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
|
1105 \ |
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
|
1106 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
|
1107 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
|
1108 \ |
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
|
1109 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
|
1110 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
|
1111 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
|
1112 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
|
1113 \ |
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
|
1114 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
|
1115 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
|
1116 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
|
1117 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
|
1118 }\ |
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
|
1119 }\ |
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
|
1120 \ |
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
|
1121 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
|
1122 {\ |
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
|
1123 int i;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1124 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
|
1125 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
|
1126 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
|
1127 + (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
|
1128 + 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
|
1129 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
|
1130 + ((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
|
1131 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
|
1132 \ |
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
|
1133 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
|
1134 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
|
1135 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
|
1136 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
|
1137 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
|
1138 + (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
|
1139 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
|
1140 + ((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
|
1141 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
|
1142 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
|
1143 block +=line_size;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1144 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1145 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
|
1146 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
|
1147 + (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
|
1148 + 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
|
1149 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
|
1150 + ((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
|
1151 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
|
1152 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
|
1153 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
|
1154 }\ |
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
|
1155 }\ |
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
|
1156 \ |
859 | 1157 static inline void OPNAME ## _pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1158 {\ |
1159 int j;\ | |
1160 for(j=0; j<2; j++){\ | |
1161 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1162 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
|
1163 const uint32_t b= AV_RN32(pixels+1);\ |
385 | 1164 uint32_t l0= (a&0x03030303UL)\ |
1165 + (b&0x03030303UL)\ | |
1166 + 0x02020202UL;\ | |
1167 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1168 + ((b&0xFCFCFCFCUL)>>2);\ | |
1169 uint32_t l1,h1;\ | |
1170 \ | |
1171 pixels+=line_size;\ | |
1172 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
|
1173 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
|
1174 uint32_t b= AV_RN32(pixels+1);\ |
385 | 1175 l1= (a&0x03030303UL)\ |
1176 + (b&0x03030303UL);\ | |
1177 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1178 + ((b&0xFCFCFCFCUL)>>2);\ | |
1179 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1180 pixels+=line_size;\ | |
1181 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1182 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1183 b= AV_RN32(pixels+1);\ |
385 | 1184 l0= (a&0x03030303UL)\ |
1185 + (b&0x03030303UL)\ | |
1186 + 0x02020202UL;\ | |
1187 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1188 + ((b&0xFCFCFCFCUL)>>2);\ | |
1189 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1190 pixels+=line_size;\ | |
1191 block +=line_size;\ | |
1192 }\ | |
1193 pixels+=4-line_size*(h+1);\ | |
1194 block +=4-line_size*h;\ | |
1195 }\ | |
1196 }\ | |
1197 \ | |
859 | 1198 static inline void OPNAME ## _no_rnd_pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1199 {\ |
1200 int j;\ | |
1201 for(j=0; j<2; j++){\ | |
1202 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1203 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
|
1204 const uint32_t b= AV_RN32(pixels+1);\ |
385 | 1205 uint32_t l0= (a&0x03030303UL)\ |
1206 + (b&0x03030303UL)\ | |
1207 + 0x01010101UL;\ | |
1208 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1209 + ((b&0xFCFCFCFCUL)>>2);\ | |
1210 uint32_t l1,h1;\ | |
1211 \ | |
1212 pixels+=line_size;\ | |
1213 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
|
1214 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
|
1215 uint32_t b= AV_RN32(pixels+1);\ |
385 | 1216 l1= (a&0x03030303UL)\ |
1217 + (b&0x03030303UL);\ | |
1218 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1219 + ((b&0xFCFCFCFCUL)>>2);\ | |
1220 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1221 pixels+=line_size;\ | |
1222 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1223 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1224 b= AV_RN32(pixels+1);\ |
385 | 1225 l0= (a&0x03030303UL)\ |
1226 + (b&0x03030303UL)\ | |
1227 + 0x01010101UL;\ | |
1228 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1229 + ((b&0xFCFCFCFCUL)>>2);\ | |
1230 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1231 pixels+=line_size;\ | |
1232 block +=line_size;\ | |
1233 }\ | |
1234 pixels+=4-line_size*(h+1);\ | |
1235 block +=4-line_size*h;\ | |
1236 }\ | |
1237 }\ | |
1238 \ | |
859 | 1239 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels8_c , 8)\ |
1240 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels8_x2_c , 8)\ | |
1241 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels8_y2_c , 8)\ | |
1242 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels8_xy2_c, 8)\ | |
1243 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_c , OPNAME ## _pixels8_c , 8)\ | |
1244 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels8_x2_c , 8)\ | |
1245 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels8_y2_c , 8)\ | |
1246 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels8_xy2_c, 8)\ | |
651 | 1247 |
1264 | 1248 #define op_avg(a, b) a = rnd_avg32(a, b) |
385 | 1249 #endif |
1250 #define op_put(a, b) a = b | |
1251 | |
1252 PIXOP2(avg, op_avg) | |
1253 PIXOP2(put, op_put) | |
1254 #undef op_avg | |
1255 #undef op_put | |
1256 | |
0 | 1257 #define avg2(a,b) ((a+b+1)>>1) |
1258 #define avg4(a,b,c,d) ((a+b+c+d+2)>>2) | |
1259 | |
1864 | 1260 static void put_no_rnd_pixels16_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ |
1261 put_no_rnd_pixels16_l2(dst, a, b, stride, stride, stride, h); | |
1262 } | |
1263 | |
1264 static void put_no_rnd_pixels8_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ | |
1265 put_no_rnd_pixels8_l2(dst, a, b, stride, stride, stride, h); | |
1266 } | |
753 | 1267 |
1064 | 1268 static void gmc1_c(uint8_t *dst, uint8_t *src, int stride, int h, int x16, int y16, int rounder) |
255 | 1269 { |
1270 const int A=(16-x16)*(16-y16); | |
1271 const int B=( x16)*(16-y16); | |
1272 const int C=(16-x16)*( y16); | |
1273 const int D=( x16)*( y16); | |
1274 int i; | |
1275 | |
1276 for(i=0; i<h; i++) | |
1277 { | |
651 | 1278 dst[0]= (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + rounder)>>8; |
1279 dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8; | |
1280 dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8; | |
1281 dst[3]= (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + rounder)>>8; | |
1282 dst[4]= (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + rounder)>>8; | |
1283 dst[5]= (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + rounder)>>8; | |
1284 dst[6]= (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + rounder)>>8; | |
1285 dst[7]= (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + rounder)>>8; | |
1286 dst+= stride; | |
1287 src+= stride; | |
255 | 1288 } |
1289 } | |
1290 | |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
1291 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, |
753 | 1292 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height) |
1293 { | |
1294 int y, vx, vy; | |
1295 const int s= 1<<shift; | |
2967 | 1296 |
753 | 1297 width--; |
1298 height--; | |
1299 | |
1300 for(y=0; y<h; y++){ | |
1301 int x; | |
1302 | |
1303 vx= ox; | |
1304 vy= oy; | |
1305 for(x=0; x<8; x++){ //XXX FIXME optimize | |
1306 int src_x, src_y, frac_x, frac_y, index; | |
1307 | |
1308 src_x= vx>>16; | |
1309 src_y= vy>>16; | |
1310 frac_x= src_x&(s-1); | |
1311 frac_y= src_y&(s-1); | |
1312 src_x>>=shift; | |
1313 src_y>>=shift; | |
2967 | 1314 |
753 | 1315 if((unsigned)src_x < width){ |
1316 if((unsigned)src_y < height){ | |
1317 index= src_x + src_y*stride; | |
1318 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) | |
1319 + src[index +1]* frac_x )*(s-frac_y) | |
1320 + ( src[index+stride ]*(s-frac_x) | |
1321 + src[index+stride+1]* frac_x )* frac_y | |
1322 + r)>>(shift*2); | |
1323 }else{ | |
4594 | 1324 index= src_x + av_clip(src_y, 0, height)*stride; |
2967 | 1325 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) |
753 | 1326 + src[index +1]* frac_x )*s |
1327 + r)>>(shift*2); | |
1328 } | |
1329 }else{ | |
1330 if((unsigned)src_y < height){ | |
4594 | 1331 index= av_clip(src_x, 0, width) + src_y*stride; |
2967 | 1332 dst[y*stride + x]= ( ( src[index ]*(s-frac_y) |
753 | 1333 + src[index+stride ]* frac_y )*s |
1334 + r)>>(shift*2); | |
1335 }else{ | |
4594 | 1336 index= av_clip(src_x, 0, width) + av_clip(src_y, 0, height)*stride; |
753 | 1337 dst[y*stride + x]= src[index ]; |
1338 } | |
1339 } | |
2967 | 1340 |
753 | 1341 vx+= dxx; |
1342 vy+= dyx; | |
1343 } | |
1344 ox += dxy; | |
1345 oy += dyy; | |
1346 } | |
1347 } | |
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 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
|
1350 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
|
1351 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
|
1352 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
|
1353 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
|
1354 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
|
1355 } |
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 } |
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 |
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
|
1358 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
|
1359 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
|
1360 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
|
1361 for (j=0; j < width; j++) { |
2979 | 1362 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
|
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 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
|
1365 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
|
1366 } |
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 } |
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 |
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
|
1369 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
|
1370 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
|
1371 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
|
1372 for (j=0; j < width; j++) { |
2979 | 1373 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
|
1374 } |
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
|
1375 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
|
1376 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
|
1377 } |
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
|
1378 } |
2967 | 1379 |
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
|
1380 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
|
1381 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
|
1382 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
|
1383 for (j=0; j < width; j++) { |
2979 | 1384 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
|
1385 } |
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
|
1386 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
|
1387 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
|
1388 } |
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
|
1389 } |
2967 | 1390 |
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
|
1391 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
|
1392 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
|
1393 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
|
1394 for (j=0; j < width; j++) { |
2979 | 1395 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
|
1396 } |
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
|
1397 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
|
1398 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
|
1399 } |
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
|
1400 } |
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
|
1401 |
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
|
1402 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
|
1403 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
|
1404 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
|
1405 for (j=0; j < width; j++) { |
2979 | 1406 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
|
1407 } |
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
|
1408 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
|
1409 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
|
1410 } |
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
|
1411 } |
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
|
1412 |
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
|
1413 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
|
1414 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
|
1415 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
|
1416 for (j=0; j < width; j++) { |
2979 | 1417 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
|
1418 } |
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
|
1419 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
|
1420 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
|
1421 } |
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
|
1422 } |
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
|
1423 |
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
|
1424 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
|
1425 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
|
1426 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
|
1427 for (j=0; j < width; j++) { |
2979 | 1428 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
|
1429 } |
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
|
1430 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
|
1431 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
|
1432 } |
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
|
1433 } |
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
|
1434 |
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
|
1435 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
|
1436 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
|
1437 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
|
1438 for (j=0; j < width; j++) { |
2979 | 1439 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
|
1440 } |
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
|
1441 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
|
1442 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
|
1443 } |
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
|
1444 } |
1319 | 1445 |
1446 static inline void avg_tpel_pixels_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1447 switch(width){ | |
1448 case 2: avg_pixels2_c (dst, src, stride, height); break; | |
1449 case 4: avg_pixels4_c (dst, src, stride, height); break; | |
1450 case 8: avg_pixels8_c (dst, src, stride, height); break; | |
1451 case 16:avg_pixels16_c(dst, src, stride, height); break; | |
1452 } | |
1453 } | |
1454 | |
1455 static inline void avg_tpel_pixels_mc10_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1456 int i,j; | |
1457 for (i=0; i < height; i++) { | |
1458 for (j=0; j < width; j++) { | |
2979 | 1459 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1460 } |
1461 src += stride; | |
1462 dst += stride; | |
1463 } | |
1464 } | |
1465 | |
1466 static inline void avg_tpel_pixels_mc20_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1467 int i,j; | |
1468 for (i=0; i < height; i++) { | |
1469 for (j=0; j < width; j++) { | |
2979 | 1470 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1471 } |
1472 src += stride; | |
1473 dst += stride; | |
1474 } | |
1475 } | |
2967 | 1476 |
1319 | 1477 static inline void avg_tpel_pixels_mc01_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1478 int i,j; | |
1479 for (i=0; i < height; i++) { | |
1480 for (j=0; j < width; j++) { | |
2979 | 1481 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1482 } |
1483 src += stride; | |
1484 dst += stride; | |
1485 } | |
1486 } | |
2967 | 1487 |
1319 | 1488 static inline void avg_tpel_pixels_mc11_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1489 int i,j; | |
1490 for (i=0; i < height; i++) { | |
1491 for (j=0; j < width; j++) { | |
2979 | 1492 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 | 1493 } |
1494 src += stride; | |
1495 dst += stride; | |
1496 } | |
1497 } | |
1498 | |
1499 static inline void avg_tpel_pixels_mc12_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1500 int i,j; | |
1501 for (i=0; i < height; i++) { | |
1502 for (j=0; j < width; j++) { | |
2979 | 1503 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 | 1504 } |
1505 src += stride; | |
1506 dst += stride; | |
1507 } | |
1508 } | |
1509 | |
1510 static inline void avg_tpel_pixels_mc02_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1511 int i,j; | |
1512 for (i=0; i < height; i++) { | |
1513 for (j=0; j < width; j++) { | |
2979 | 1514 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1515 } |
1516 src += stride; | |
1517 dst += stride; | |
1518 } | |
1519 } | |
1520 | |
1521 static inline void avg_tpel_pixels_mc21_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1522 int i,j; | |
1523 for (i=0; i < height; i++) { | |
1524 for (j=0; j < width; j++) { | |
2979 | 1525 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 | 1526 } |
1527 src += stride; | |
1528 dst += stride; | |
1529 } | |
1530 } | |
1531 | |
1532 static inline void avg_tpel_pixels_mc22_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1533 int i,j; | |
1534 for (i=0; i < height; i++) { | |
1535 for (j=0; j < width; j++) { | |
2979 | 1536 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 | 1537 } |
1538 src += stride; | |
1539 dst += stride; | |
1540 } | |
1541 } | |
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
|
1542 #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
|
1543 #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
|
1544 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
|
1545 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
|
1546 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
|
1547 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
|
1548 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
|
1549 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
|
1550 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
|
1551 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
|
1552 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
|
1553 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
|
1554 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
|
1555 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
|
1556 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
|
1557 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
|
1558 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
|
1559 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
|
1560 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
|
1561 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
|
1562 #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
|
1563 |
1168 | 1564 #define H264_CHROMA_MC(OPNAME, OP)\ |
1565 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){\ | |
1566 const int A=(8-x)*(8-y);\ | |
1567 const int B=( x)*(8-y);\ | |
1568 const int C=(8-x)*( y);\ | |
1569 const int D=( x)*( y);\ | |
1570 int i;\ | |
1571 \ | |
1572 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1573 \ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1574 if(D){\ |
6054 | 1575 for(i=0; i<h; i++){\ |
6053 | 1576 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1577 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1578 dst+= stride;\ | |
1579 src+= stride;\ | |
1580 }\ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1581 }else{\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1582 const int E= B+C;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1583 const int step= C ? stride : 1;\ |
6054 | 1584 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
|
1585 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
|
1586 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
|
1587 dst+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1588 src+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1589 }\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1590 }\ |
1168 | 1591 }\ |
1592 \ | |
1593 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){\ | |
1594 const int A=(8-x)*(8-y);\ | |
1595 const int B=( x)*(8-y);\ | |
1596 const int C=(8-x)*( y);\ | |
1597 const int D=( x)*( y);\ | |
1598 int i;\ | |
1599 \ | |
1600 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1601 \ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1602 if(D){\ |
6054 | 1603 for(i=0; i<h; i++){\ |
6053 | 1604 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1605 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1606 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1607 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1608 dst+= stride;\ | |
1609 src+= stride;\ | |
1610 }\ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1611 }else{\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1612 const int E= B+C;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1613 const int step= C ? stride : 1;\ |
6054 | 1614 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
|
1615 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
|
1616 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
|
1617 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
|
1618 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
|
1619 dst+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1620 src+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1621 }\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1622 }\ |
1168 | 1623 }\ |
1624 \ | |
1625 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){\ | |
1626 const int A=(8-x)*(8-y);\ | |
1627 const int B=( x)*(8-y);\ | |
1628 const int C=(8-x)*( y);\ | |
1629 const int D=( x)*( y);\ | |
1630 int i;\ | |
1631 \ | |
1632 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1633 \ | |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1634 if(D){\ |
6054 | 1635 for(i=0; i<h; i++){\ |
6053 | 1636 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1637 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1638 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1639 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1640 OP(dst[4], (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5]));\ | |
1641 OP(dst[5], (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6]));\ | |
1642 OP(dst[6], (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7]));\ | |
1643 OP(dst[7], (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8]));\ | |
1644 dst+= stride;\ | |
1645 src+= stride;\ | |
1646 }\ | |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1647 }else{\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1648 const int E= B+C;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1649 const int step= C ? stride : 1;\ |
6054 | 1650 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
|
1651 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
|
1652 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
|
1653 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
|
1654 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
|
1655 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
|
1656 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
|
1657 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
|
1658 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
|
1659 dst+= stride;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1660 src+= stride;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1661 }\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1662 }\ |
1168 | 1663 } |
1664 | |
1665 #define op_avg(a, b) a = (((a)+(((b) + 32)>>6)+1)>>1) | |
1666 #define op_put(a, b) a = (((b) + 32)>>6) | |
1667 | |
1668 H264_CHROMA_MC(put_ , op_put) | |
1669 H264_CHROMA_MC(avg_ , op_avg) | |
1670 #undef op_avg | |
1671 #undef op_put | |
1672 | |
3663 | 1673 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){ |
1674 const int A=(8-x)*(8-y); | |
1675 const int B=( x)*(8-y); | |
1676 const int C=(8-x)*( y); | |
1677 const int D=( x)*( y); | |
1678 int i; | |
1679 | |
1680 assert(x<8 && y<8 && x>=0 && y>=0); | |
1681 | |
1682 for(i=0; i<h; i++) | |
1683 { | |
1684 dst[0] = (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6; | |
1685 dst[1] = (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6; | |
1686 dst[2] = (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6; | |
1687 dst[3] = (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6; | |
1688 dst[4] = (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6; | |
1689 dst[5] = (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6; | |
1690 dst[6] = (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6; | |
1691 dst[7] = (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6; | |
1692 dst+= stride; | |
1693 src+= stride; | |
1694 } | |
1695 } | |
1696 | |
651 | 1697 #define QPEL_MC(r, OPNAME, RND, OP) \ |
1064 | 1698 static void OPNAME ## mpeg4_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
4176 | 1699 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1700 int i;\ |
1701 for(i=0; i<h; i++)\ | |
1702 {\ | |
1703 OP(dst[0], (src[0]+src[1])*20 - (src[0]+src[2])*6 + (src[1]+src[3])*3 - (src[2]+src[4]));\ | |
1704 OP(dst[1], (src[1]+src[2])*20 - (src[0]+src[3])*6 + (src[0]+src[4])*3 - (src[1]+src[5]));\ | |
1705 OP(dst[2], (src[2]+src[3])*20 - (src[1]+src[4])*6 + (src[0]+src[5])*3 - (src[0]+src[6]));\ | |
1706 OP(dst[3], (src[3]+src[4])*20 - (src[2]+src[5])*6 + (src[1]+src[6])*3 - (src[0]+src[7]));\ | |
1707 OP(dst[4], (src[4]+src[5])*20 - (src[3]+src[6])*6 + (src[2]+src[7])*3 - (src[1]+src[8]));\ | |
1708 OP(dst[5], (src[5]+src[6])*20 - (src[4]+src[7])*6 + (src[3]+src[8])*3 - (src[2]+src[8]));\ | |
1709 OP(dst[6], (src[6]+src[7])*20 - (src[5]+src[8])*6 + (src[4]+src[8])*3 - (src[3]+src[7]));\ | |
1710 OP(dst[7], (src[7]+src[8])*20 - (src[6]+src[8])*6 + (src[5]+src[7])*3 - (src[4]+src[6]));\ | |
1711 dst+=dstStride;\ | |
1712 src+=srcStride;\ | |
1713 }\ | |
1714 }\ | |
1715 \ | |
1064 | 1716 static void OPNAME ## mpeg4_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
984 | 1717 const int w=8;\ |
4176 | 1718 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1719 int i;\ |
1720 for(i=0; i<w; i++)\ | |
1721 {\ | |
1722 const int src0= src[0*srcStride];\ | |
1723 const int src1= src[1*srcStride];\ | |
1724 const int src2= src[2*srcStride];\ | |
1725 const int src3= src[3*srcStride];\ | |
1726 const int src4= src[4*srcStride];\ | |
1727 const int src5= src[5*srcStride];\ | |
1728 const int src6= src[6*srcStride];\ | |
1729 const int src7= src[7*srcStride];\ | |
1730 const int src8= src[8*srcStride];\ | |
1731 OP(dst[0*dstStride], (src0+src1)*20 - (src0+src2)*6 + (src1+src3)*3 - (src2+src4));\ | |
1732 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*6 + (src0+src4)*3 - (src1+src5));\ | |
1733 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*6 + (src0+src5)*3 - (src0+src6));\ | |
1734 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*6 + (src1+src6)*3 - (src0+src7));\ | |
1735 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*6 + (src2+src7)*3 - (src1+src8));\ | |
1736 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*6 + (src3+src8)*3 - (src2+src8));\ | |
1737 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*6 + (src4+src8)*3 - (src3+src7));\ | |
1738 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src8)*6 + (src5+src7)*3 - (src4+src6));\ | |
1739 dst++;\ | |
1740 src++;\ | |
1741 }\ | |
1742 }\ | |
1743 \ | |
1064 | 1744 static void OPNAME ## mpeg4_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
4176 | 1745 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1746 int i;\ |
954 | 1747 \ |
651 | 1748 for(i=0; i<h; i++)\ |
1749 {\ | |
1750 OP(dst[ 0], (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]));\ | |
1751 OP(dst[ 1], (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]));\ | |
1752 OP(dst[ 2], (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]));\ | |
1753 OP(dst[ 3], (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]));\ | |
1754 OP(dst[ 4], (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]));\ | |
1755 OP(dst[ 5], (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]));\ | |
1756 OP(dst[ 6], (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]));\ | |
1757 OP(dst[ 7], (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]));\ | |
1758 OP(dst[ 8], (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]));\ | |
1759 OP(dst[ 9], (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]));\ | |
1760 OP(dst[10], (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]));\ | |
1761 OP(dst[11], (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]));\ | |
1762 OP(dst[12], (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]));\ | |
1763 OP(dst[13], (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]));\ | |
1764 OP(dst[14], (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]));\ | |
1765 OP(dst[15], (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]));\ | |
1766 dst+=dstStride;\ | |
1767 src+=srcStride;\ | |
1768 }\ | |
255 | 1769 }\ |
1770 \ | |
1064 | 1771 static void OPNAME ## mpeg4_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
4176 | 1772 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1773 int i;\ |
954 | 1774 const int w=16;\ |
651 | 1775 for(i=0; i<w; i++)\ |
1776 {\ | |
1777 const int src0= src[0*srcStride];\ | |
1778 const int src1= src[1*srcStride];\ | |
1779 const int src2= src[2*srcStride];\ | |
1780 const int src3= src[3*srcStride];\ | |
1781 const int src4= src[4*srcStride];\ | |
1782 const int src5= src[5*srcStride];\ | |
1783 const int src6= src[6*srcStride];\ | |
1784 const int src7= src[7*srcStride];\ | |
1785 const int src8= src[8*srcStride];\ | |
1786 const int src9= src[9*srcStride];\ | |
1787 const int src10= src[10*srcStride];\ | |
1788 const int src11= src[11*srcStride];\ | |
1789 const int src12= src[12*srcStride];\ | |
1790 const int src13= src[13*srcStride];\ | |
1791 const int src14= src[14*srcStride];\ | |
1792 const int src15= src[15*srcStride];\ | |
1793 const int src16= src[16*srcStride];\ | |
1794 OP(dst[ 0*dstStride], (src0 +src1 )*20 - (src0 +src2 )*6 + (src1 +src3 )*3 - (src2 +src4 ));\ | |
1795 OP(dst[ 1*dstStride], (src1 +src2 )*20 - (src0 +src3 )*6 + (src0 +src4 )*3 - (src1 +src5 ));\ | |
1796 OP(dst[ 2*dstStride], (src2 +src3 )*20 - (src1 +src4 )*6 + (src0 +src5 )*3 - (src0 +src6 ));\ | |
1797 OP(dst[ 3*dstStride], (src3 +src4 )*20 - (src2 +src5 )*6 + (src1 +src6 )*3 - (src0 +src7 ));\ | |
1798 OP(dst[ 4*dstStride], (src4 +src5 )*20 - (src3 +src6 )*6 + (src2 +src7 )*3 - (src1 +src8 ));\ | |
1799 OP(dst[ 5*dstStride], (src5 +src6 )*20 - (src4 +src7 )*6 + (src3 +src8 )*3 - (src2 +src9 ));\ | |
1800 OP(dst[ 6*dstStride], (src6 +src7 )*20 - (src5 +src8 )*6 + (src4 +src9 )*3 - (src3 +src10));\ | |
1801 OP(dst[ 7*dstStride], (src7 +src8 )*20 - (src6 +src9 )*6 + (src5 +src10)*3 - (src4 +src11));\ | |
1802 OP(dst[ 8*dstStride], (src8 +src9 )*20 - (src7 +src10)*6 + (src6 +src11)*3 - (src5 +src12));\ | |
1803 OP(dst[ 9*dstStride], (src9 +src10)*20 - (src8 +src11)*6 + (src7 +src12)*3 - (src6 +src13));\ | |
1804 OP(dst[10*dstStride], (src10+src11)*20 - (src9 +src12)*6 + (src8 +src13)*3 - (src7 +src14));\ | |
1805 OP(dst[11*dstStride], (src11+src12)*20 - (src10+src13)*6 + (src9 +src14)*3 - (src8 +src15));\ | |
1806 OP(dst[12*dstStride], (src12+src13)*20 - (src11+src14)*6 + (src10+src15)*3 - (src9 +src16));\ | |
1807 OP(dst[13*dstStride], (src13+src14)*20 - (src12+src15)*6 + (src11+src16)*3 - (src10+src16));\ | |
1808 OP(dst[14*dstStride], (src14+src15)*20 - (src13+src16)*6 + (src12+src16)*3 - (src11+src15));\ | |
1809 OP(dst[15*dstStride], (src15+src16)*20 - (src14+src16)*6 + (src13+src15)*3 - (src12+src14));\ | |
1810 dst++;\ | |
1811 src++;\ | |
1812 }\ | |
255 | 1813 }\ |
1814 \ | |
1064 | 1815 static void OPNAME ## qpel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1816 OPNAME ## pixels8_c(dst, src, stride, 8);\ |
255 | 1817 }\ |
1818 \ | |
1064 | 1819 static void OPNAME ## qpel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
1820 uint8_t half[64];\ | |
651 | 1821 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1822 OPNAME ## pixels8_l2(dst, src, half, stride, stride, 8, 8);\ | |
1823 }\ | |
1824 \ | |
1064 | 1825 static void OPNAME ## qpel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 1826 OPNAME ## mpeg4_qpel8_h_lowpass(dst, src, stride, stride, 8);\ |
255 | 1827 }\ |
1828 \ | |
1064 | 1829 static void OPNAME ## qpel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
1830 uint8_t half[64];\ | |
651 | 1831 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1832 OPNAME ## pixels8_l2(dst, src+1, half, stride, stride, 8, 8);\ | |
1833 }\ | |
1834 \ | |
1064 | 1835 static void OPNAME ## qpel8_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
1836 uint8_t full[16*9];\ | |
1837 uint8_t half[64];\ | |
651 | 1838 copy_block9(full, src, 16, stride, 9);\ |
984 | 1839 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1840 OPNAME ## pixels8_l2(dst, full, half, stride, 16, 8, 8);\ |
1841 }\ | |
1842 \ | |
1064 | 1843 static void OPNAME ## qpel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
1844 uint8_t full[16*9];\ | |
651 | 1845 copy_block9(full, src, 16, stride, 9);\ |
984 | 1846 OPNAME ## mpeg4_qpel8_v_lowpass(dst, full, stride, 16);\ |
255 | 1847 }\ |
1848 \ | |
1064 | 1849 static void OPNAME ## qpel8_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
1850 uint8_t full[16*9];\ | |
1851 uint8_t half[64];\ | |
651 | 1852 copy_block9(full, src, 16, stride, 9);\ |
984 | 1853 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1854 OPNAME ## pixels8_l2(dst, full+16, half, stride, 16, 8, 8);\ |
1855 }\ | |
1064 | 1856 void ff_ ## OPNAME ## qpel8_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1857 uint8_t full[16*9];\ | |
1858 uint8_t halfH[72];\ | |
1859 uint8_t halfV[64];\ | |
1860 uint8_t halfHV[64];\ | |
651 | 1861 copy_block9(full, src, 16, stride, 9);\ |
1862 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1863 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1864 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1865 OPNAME ## pixels8_l4(dst, full, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1866 }\ |
1064 | 1867 static void OPNAME ## qpel8_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
1868 uint8_t full[16*9];\ | |
1869 uint8_t halfH[72];\ | |
1870 uint8_t halfHV[64];\ | |
984 | 1871 copy_block9(full, src, 16, stride, 9);\ |
1872 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1873 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1874 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1875 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1876 }\ | |
1064 | 1877 void ff_ ## OPNAME ## qpel8_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1878 uint8_t full[16*9];\ | |
1879 uint8_t halfH[72];\ | |
1880 uint8_t halfV[64];\ | |
1881 uint8_t halfHV[64];\ | |
651 | 1882 copy_block9(full, src, 16, stride, 9);\ |
1883 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1884 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1885 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1886 OPNAME ## pixels8_l4(dst, full+1, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1887 }\ |
1064 | 1888 static void OPNAME ## qpel8_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
1889 uint8_t full[16*9];\ | |
1890 uint8_t halfH[72];\ | |
1891 uint8_t halfHV[64];\ | |
984 | 1892 copy_block9(full, src, 16, stride, 9);\ |
1893 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1894 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1895 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1896 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1897 }\ | |
1064 | 1898 void ff_ ## OPNAME ## qpel8_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1899 uint8_t full[16*9];\ | |
1900 uint8_t halfH[72];\ | |
1901 uint8_t halfV[64];\ | |
1902 uint8_t halfHV[64];\ | |
651 | 1903 copy_block9(full, src, 16, stride, 9);\ |
1904 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1905 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1906 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1907 OPNAME ## pixels8_l4(dst, full+16, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
1908 }\ | |
1064 | 1909 static void OPNAME ## qpel8_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
1910 uint8_t full[16*9];\ | |
1911 uint8_t halfH[72];\ | |
1912 uint8_t halfHV[64];\ | |
984 | 1913 copy_block9(full, src, 16, stride, 9);\ |
1914 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1915 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1916 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1917 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1918 }\ | |
1064 | 1919 void ff_ ## OPNAME ## qpel8_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1920 uint8_t full[16*9];\ | |
1921 uint8_t halfH[72];\ | |
1922 uint8_t halfV[64];\ | |
1923 uint8_t halfHV[64];\ | |
651 | 1924 copy_block9(full, src, 16, stride, 9);\ |
1925 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full , 8, 16, 9);\ | |
984 | 1926 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1927 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1928 OPNAME ## pixels8_l4(dst, full+17, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1929 }\ |
1064 | 1930 static void OPNAME ## qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
1931 uint8_t full[16*9];\ | |
1932 uint8_t halfH[72];\ | |
1933 uint8_t halfHV[64];\ | |
984 | 1934 copy_block9(full, src, 16, stride, 9);\ |
1935 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1936 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1937 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1938 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1939 }\ | |
1064 | 1940 static void OPNAME ## qpel8_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
1941 uint8_t halfH[72];\ | |
1942 uint8_t halfHV[64];\ | |
651 | 1943 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1944 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1945 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ |
1946 }\ | |
1064 | 1947 static void OPNAME ## qpel8_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
1948 uint8_t halfH[72];\ | |
1949 uint8_t halfHV[64];\ | |
651 | 1950 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1951 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1952 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ |
1953 }\ | |
1064 | 1954 void ff_ ## OPNAME ## qpel8_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1955 uint8_t full[16*9];\ | |
1956 uint8_t halfH[72];\ | |
1957 uint8_t halfV[64];\ | |
1958 uint8_t halfHV[64];\ | |
651 | 1959 copy_block9(full, src, 16, stride, 9);\ |
1960 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1961 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1962 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1963 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
255 | 1964 }\ |
1064 | 1965 static void OPNAME ## qpel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
1966 uint8_t full[16*9];\ | |
1967 uint8_t halfH[72];\ | |
984 | 1968 copy_block9(full, src, 16, stride, 9);\ |
1969 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1970 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1971 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1972 }\ | |
1064 | 1973 void ff_ ## OPNAME ## qpel8_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1974 uint8_t full[16*9];\ | |
1975 uint8_t halfH[72];\ | |
1976 uint8_t halfV[64];\ | |
1977 uint8_t halfHV[64];\ | |
651 | 1978 copy_block9(full, src, 16, stride, 9);\ |
1979 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1980 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1981 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1982 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
1983 }\ | |
1064 | 1984 static void OPNAME ## qpel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
1985 uint8_t full[16*9];\ | |
1986 uint8_t halfH[72];\ | |
984 | 1987 copy_block9(full, src, 16, stride, 9);\ |
1988 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1989 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1990 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1991 }\ | |
1064 | 1992 static void OPNAME ## qpel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
1993 uint8_t halfH[72];\ | |
651 | 1994 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1995 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ |
651 | 1996 }\ |
1064 | 1997 static void OPNAME ## qpel16_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1998 OPNAME ## pixels16_c(dst, src, stride, 16);\ |
255 | 1999 }\ |
651 | 2000 \ |
1064 | 2001 static void OPNAME ## qpel16_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
2002 uint8_t half[256];\ | |
651 | 2003 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
2004 OPNAME ## pixels16_l2(dst, src, half, stride, stride, 16, 16);\ | |
2005 }\ | |
2006 \ | |
1064 | 2007 static void OPNAME ## qpel16_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 2008 OPNAME ## mpeg4_qpel16_h_lowpass(dst, src, stride, stride, 16);\ |
2009 }\ | |
2010 \ | |
1064 | 2011 static void OPNAME ## qpel16_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
2012 uint8_t half[256];\ | |
651 | 2013 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
2014 OPNAME ## pixels16_l2(dst, src+1, half, stride, stride, 16, 16);\ | |
2015 }\ | |
2016 \ | |
1064 | 2017 static void OPNAME ## qpel16_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
2018 uint8_t full[24*17];\ | |
2019 uint8_t half[256];\ | |
651 | 2020 copy_block17(full, src, 24, stride, 17);\ |
954 | 2021 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 2022 OPNAME ## pixels16_l2(dst, full, half, stride, 24, 16, 16);\ |
255 | 2023 }\ |
651 | 2024 \ |
1064 | 2025 static void OPNAME ## qpel16_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
2026 uint8_t full[24*17];\ | |
651 | 2027 copy_block17(full, src, 24, stride, 17);\ |
954 | 2028 OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24);\ |
651 | 2029 }\ |
2030 \ | |
1064 | 2031 static void OPNAME ## qpel16_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
2032 uint8_t full[24*17];\ | |
2033 uint8_t half[256];\ | |
651 | 2034 copy_block17(full, src, 24, stride, 17);\ |
954 | 2035 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 2036 OPNAME ## pixels16_l2(dst, full+24, half, stride, 24, 16, 16);\ |
255 | 2037 }\ |
1064 | 2038 void ff_ ## OPNAME ## qpel16_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2039 uint8_t full[24*17];\ | |
2040 uint8_t halfH[272];\ | |
2041 uint8_t halfV[256];\ | |
2042 uint8_t halfHV[256];\ | |
651 | 2043 copy_block17(full, src, 24, stride, 17);\ |
2044 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2045 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
2046 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2047 OPNAME ## pixels16_l4(dst, full, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
2048 }\ | |
1064 | 2049 static void OPNAME ## qpel16_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
2050 uint8_t full[24*17];\ | |
2051 uint8_t halfH[272];\ | |
2052 uint8_t halfHV[256];\ | |
984 | 2053 copy_block17(full, src, 24, stride, 17);\ |
2054 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2055 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
2056 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
2057 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
2058 }\ | |
1064 | 2059 void ff_ ## OPNAME ## qpel16_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2060 uint8_t full[24*17];\ | |
2061 uint8_t halfH[272];\ | |
2062 uint8_t halfV[256];\ | |
2063 uint8_t halfHV[256];\ | |
651 | 2064 copy_block17(full, src, 24, stride, 17);\ |
2065 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2066 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
2067 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2068 OPNAME ## pixels16_l4(dst, full+1, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
2069 }\ | |
1064 | 2070 static void OPNAME ## qpel16_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
2071 uint8_t full[24*17];\ | |
2072 uint8_t halfH[272];\ | |
2073 uint8_t halfHV[256];\ | |
984 | 2074 copy_block17(full, src, 24, stride, 17);\ |
2075 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2076 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2077 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
2078 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
2079 }\ | |
1064 | 2080 void ff_ ## OPNAME ## qpel16_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2081 uint8_t full[24*17];\ | |
2082 uint8_t halfH[272];\ | |
2083 uint8_t halfV[256];\ | |
2084 uint8_t halfHV[256];\ | |
651 | 2085 copy_block17(full, src, 24, stride, 17);\ |
2086 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2087 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
2088 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2089 OPNAME ## pixels16_l4(dst, full+24, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
255 | 2090 }\ |
1064 | 2091 static void OPNAME ## qpel16_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
2092 uint8_t full[24*17];\ | |
2093 uint8_t halfH[272];\ | |
2094 uint8_t halfHV[256];\ | |
984 | 2095 copy_block17(full, src, 24, stride, 17);\ |
2096 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2097 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
2098 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
2099 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
2100 }\ | |
1064 | 2101 void ff_ ## OPNAME ## qpel16_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2102 uint8_t full[24*17];\ | |
2103 uint8_t halfH[272];\ | |
2104 uint8_t halfV[256];\ | |
2105 uint8_t halfHV[256];\ | |
651 | 2106 copy_block17(full, src, 24, stride, 17);\ |
2107 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full , 16, 24, 17);\ | |
954 | 2108 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
2109 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2110 OPNAME ## pixels16_l4(dst, full+25, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
2111 }\ | |
1064 | 2112 static void OPNAME ## qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
2113 uint8_t full[24*17];\ | |
2114 uint8_t halfH[272];\ | |
2115 uint8_t halfHV[256];\ | |
984 | 2116 copy_block17(full, src, 24, stride, 17);\ |
2117 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2118 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2119 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
2120 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
2121 }\ | |
1064 | 2122 static void OPNAME ## qpel16_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
2123 uint8_t halfH[272];\ | |
2124 uint8_t halfHV[256];\ | |
651 | 2125 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2126 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 2127 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ |
255 | 2128 }\ |
1064 | 2129 static void OPNAME ## qpel16_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
2130 uint8_t halfH[272];\ | |
2131 uint8_t halfHV[256];\ | |
651 | 2132 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2133 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 2134 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ |
2135 }\ | |
1064 | 2136 void ff_ ## OPNAME ## qpel16_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2137 uint8_t full[24*17];\ | |
2138 uint8_t halfH[272];\ | |
2139 uint8_t halfV[256];\ | |
2140 uint8_t halfHV[256];\ | |
651 | 2141 copy_block17(full, src, 24, stride, 17);\ |
2142 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2143 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
2144 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2145 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
255 | 2146 }\ |
1064 | 2147 static void OPNAME ## qpel16_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
2148 uint8_t full[24*17];\ | |
2149 uint8_t halfH[272];\ | |
984 | 2150 copy_block17(full, src, 24, stride, 17);\ |
2151 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2152 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
2153 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2154 }\ | |
1064 | 2155 void ff_ ## OPNAME ## qpel16_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2156 uint8_t full[24*17];\ | |
2157 uint8_t halfH[272];\ | |
2158 uint8_t halfV[256];\ | |
2159 uint8_t halfHV[256];\ | |
651 | 2160 copy_block17(full, src, 24, stride, 17);\ |
2161 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2162 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
2163 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2164 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
2165 }\ | |
1064 | 2166 static void OPNAME ## qpel16_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
2167 uint8_t full[24*17];\ | |
2168 uint8_t halfH[272];\ | |
984 | 2169 copy_block17(full, src, 24, stride, 17);\ |
2170 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2171 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2172 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2173 }\ | |
1064 | 2174 static void OPNAME ## qpel16_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
2175 uint8_t halfH[272];\ | |
651 | 2176 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2177 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ |
859 | 2178 } |
255 | 2179 |
651 | 2180 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) |
2181 #define op_avg_no_rnd(a, b) a = (((a)+cm[((b) + 15)>>5])>>1) | |
2182 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2183 #define op_put_no_rnd(a, b) a = cm[((b) + 15)>>5] | |
2184 | |
2185 QPEL_MC(0, put_ , _ , op_put) | |
2186 QPEL_MC(1, put_no_rnd_, _no_rnd_, op_put_no_rnd) | |
2187 QPEL_MC(0, avg_ , _ , op_avg) | |
2188 //QPEL_MC(1, avg_no_rnd , _ , op_avg) | |
2189 #undef op_avg | |
2190 #undef op_avg_no_rnd | |
2191 #undef op_put | |
2192 #undef op_put_no_rnd | |
255 | 2193 |
1168 | 2194 #if 1 |
2195 #define H264_LOWPASS(OPNAME, OP, OP2) \ | |
5151 | 2196 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
|
2197 const int h=2;\ |
4176 | 2198 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
|
2199 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2200 for(i=0; i<h; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2201 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2202 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
|
2203 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
|
2204 dst+=dstStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2205 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2206 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2207 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2208 \ |
5151 | 2209 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
|
2210 const int w=2;\ |
4176 | 2211 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
|
2212 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2213 for(i=0; i<w; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2214 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2215 const int srcB= src[-2*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2216 const int srcA= src[-1*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2217 const int src0= src[0 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2218 const int src1= src[1 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2219 const int src2= src[2 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2220 const int src3= src[3 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2221 const int src4= src[4 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2222 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
|
2223 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
|
2224 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2225 src++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2226 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2227 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2228 \ |
5151 | 2229 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
|
2230 const int h=2;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2231 const int w=2;\ |
4176 | 2232 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
|
2233 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2234 src -= 2*srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2235 for(i=0; i<h+5; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2236 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2237 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
|
2238 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
|
2239 tmp+=tmpStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2240 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2241 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2242 tmp -= tmpStride*(h+5-2);\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2243 for(i=0; i<w; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2244 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2245 const int tmpB= tmp[-2*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2246 const int tmpA= tmp[-1*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2247 const int tmp0= tmp[0 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2248 const int tmp1= tmp[1 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2249 const int tmp2= tmp[2 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2250 const int tmp3= tmp[3 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2251 const int tmp4= tmp[4 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2252 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
|
2253 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
|
2254 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2255 tmp++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2256 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2257 }\ |
1168 | 2258 static void OPNAME ## h264_qpel4_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
2259 const int h=4;\ | |
4176 | 2260 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2261 int i;\ |
2262 for(i=0; i<h; i++)\ | |
2263 {\ | |
2264 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]));\ | |
2265 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]));\ | |
2266 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]));\ | |
2267 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]));\ | |
2268 dst+=dstStride;\ | |
2269 src+=srcStride;\ | |
2270 }\ | |
2271 }\ | |
2272 \ | |
2273 static void OPNAME ## h264_qpel4_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2274 const int w=4;\ | |
4176 | 2275 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2276 int i;\ |
2277 for(i=0; i<w; i++)\ | |
2278 {\ | |
2279 const int srcB= src[-2*srcStride];\ | |
2280 const int srcA= src[-1*srcStride];\ | |
2281 const int src0= src[0 *srcStride];\ | |
2282 const int src1= src[1 *srcStride];\ | |
2283 const int src2= src[2 *srcStride];\ | |
2284 const int src3= src[3 *srcStride];\ | |
2285 const int src4= src[4 *srcStride];\ | |
2286 const int src5= src[5 *srcStride];\ | |
2287 const int src6= src[6 *srcStride];\ | |
2288 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2289 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2290 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2291 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2292 dst++;\ | |
2293 src++;\ | |
2294 }\ | |
2295 }\ | |
2296 \ | |
2297 static void OPNAME ## h264_qpel4_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2298 const int h=4;\ | |
2299 const int w=4;\ | |
4176 | 2300 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2301 int i;\ |
2302 src -= 2*srcStride;\ | |
2303 for(i=0; i<h+5; i++)\ | |
2304 {\ | |
2305 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]);\ | |
2306 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]);\ | |
2307 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]);\ | |
2308 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]);\ | |
2309 tmp+=tmpStride;\ | |
2310 src+=srcStride;\ | |
2311 }\ | |
2312 tmp -= tmpStride*(h+5-2);\ | |
2313 for(i=0; i<w; i++)\ | |
2314 {\ | |
2315 const int tmpB= tmp[-2*tmpStride];\ | |
2316 const int tmpA= tmp[-1*tmpStride];\ | |
2317 const int tmp0= tmp[0 *tmpStride];\ | |
2318 const int tmp1= tmp[1 *tmpStride];\ | |
2319 const int tmp2= tmp[2 *tmpStride];\ | |
2320 const int tmp3= tmp[3 *tmpStride];\ | |
2321 const int tmp4= tmp[4 *tmpStride];\ | |
2322 const int tmp5= tmp[5 *tmpStride];\ | |
2323 const int tmp6= tmp[6 *tmpStride];\ | |
2324 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2325 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2326 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2327 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2328 dst++;\ | |
2329 tmp++;\ | |
2330 }\ | |
2331 }\ | |
2332 \ | |
2333 static void OPNAME ## h264_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2334 const int h=8;\ | |
4176 | 2335 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2336 int i;\ |
2337 for(i=0; i<h; i++)\ | |
2338 {\ | |
2339 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]));\ | |
2340 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]));\ | |
2341 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]));\ | |
2342 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]));\ | |
2343 OP(dst[4], (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]));\ | |
2344 OP(dst[5], (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]));\ | |
2345 OP(dst[6], (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]));\ | |
2346 OP(dst[7], (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]));\ | |
2347 dst+=dstStride;\ | |
2348 src+=srcStride;\ | |
2349 }\ | |
2350 }\ | |
2351 \ | |
2352 static void OPNAME ## h264_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2353 const int w=8;\ | |
4176 | 2354 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2355 int i;\ |
2356 for(i=0; i<w; i++)\ | |
2357 {\ | |
2358 const int srcB= src[-2*srcStride];\ | |
2359 const int srcA= src[-1*srcStride];\ | |
2360 const int src0= src[0 *srcStride];\ | |
2361 const int src1= src[1 *srcStride];\ | |
2362 const int src2= src[2 *srcStride];\ | |
2363 const int src3= src[3 *srcStride];\ | |
2364 const int src4= src[4 *srcStride];\ | |
2365 const int src5= src[5 *srcStride];\ | |
2366 const int src6= src[6 *srcStride];\ | |
2367 const int src7= src[7 *srcStride];\ | |
2368 const int src8= src[8 *srcStride];\ | |
2369 const int src9= src[9 *srcStride];\ | |
2370 const int src10=src[10*srcStride];\ | |
2371 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2372 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2373 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2374 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2375 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*5 + (src2+src7));\ | |
2376 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*5 + (src3+src8));\ | |
2377 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*5 + (src4+src9));\ | |
2378 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src9)*5 + (src5+src10));\ | |
2379 dst++;\ | |
2380 src++;\ | |
2381 }\ | |
2382 }\ | |
2383 \ | |
2384 static void OPNAME ## h264_qpel8_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2385 const int h=8;\ | |
2386 const int w=8;\ | |
4176 | 2387 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2388 int i;\ |
2389 src -= 2*srcStride;\ | |
2390 for(i=0; i<h+5; i++)\ | |
2391 {\ | |
2392 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]);\ | |
2393 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]);\ | |
2394 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]);\ | |
2395 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]);\ | |
2396 tmp[4]= (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]);\ | |
2397 tmp[5]= (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]);\ | |
2398 tmp[6]= (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]);\ | |
2399 tmp[7]= (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]);\ | |
2400 tmp+=tmpStride;\ | |
2401 src+=srcStride;\ | |
2402 }\ | |
2403 tmp -= tmpStride*(h+5-2);\ | |
2404 for(i=0; i<w; i++)\ | |
2405 {\ | |
2406 const int tmpB= tmp[-2*tmpStride];\ | |
2407 const int tmpA= tmp[-1*tmpStride];\ | |
2408 const int tmp0= tmp[0 *tmpStride];\ | |
2409 const int tmp1= tmp[1 *tmpStride];\ | |
2410 const int tmp2= tmp[2 *tmpStride];\ | |
2411 const int tmp3= tmp[3 *tmpStride];\ | |
2412 const int tmp4= tmp[4 *tmpStride];\ | |
2413 const int tmp5= tmp[5 *tmpStride];\ | |
2414 const int tmp6= tmp[6 *tmpStride];\ | |
2415 const int tmp7= tmp[7 *tmpStride];\ | |
2416 const int tmp8= tmp[8 *tmpStride];\ | |
2417 const int tmp9= tmp[9 *tmpStride];\ | |
2418 const int tmp10=tmp[10*tmpStride];\ | |
2419 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2420 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2421 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2422 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2423 OP2(dst[4*dstStride], (tmp4+tmp5)*20 - (tmp3+tmp6)*5 + (tmp2+tmp7));\ | |
2424 OP2(dst[5*dstStride], (tmp5+tmp6)*20 - (tmp4+tmp7)*5 + (tmp3+tmp8));\ | |
2425 OP2(dst[6*dstStride], (tmp6+tmp7)*20 - (tmp5+tmp8)*5 + (tmp4+tmp9));\ | |
2426 OP2(dst[7*dstStride], (tmp7+tmp8)*20 - (tmp6+tmp9)*5 + (tmp5+tmp10));\ | |
2427 dst++;\ | |
2428 tmp++;\ | |
2429 }\ | |
2430 }\ | |
2431 \ | |
2432 static void OPNAME ## h264_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2433 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2434 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2435 src += 8*srcStride;\ | |
2436 dst += 8*dstStride;\ | |
2437 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2438 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2439 }\ | |
2440 \ | |
2441 static void OPNAME ## h264_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2442 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2443 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2444 src += 8*srcStride;\ | |
2445 dst += 8*dstStride;\ | |
2446 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2447 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2448 }\ | |
2449 \ | |
2450 static void OPNAME ## h264_qpel16_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2451 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2452 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2453 src += 8*srcStride;\ | |
2454 dst += 8*dstStride;\ | |
2455 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2456 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2457 }\ | |
2458 | |
2459 #define H264_MC(OPNAME, SIZE) \ | |
2460 static void OPNAME ## h264_qpel ## SIZE ## _mc00_c (uint8_t *dst, uint8_t *src, int stride){\ | |
2461 OPNAME ## pixels ## SIZE ## _c(dst, src, stride, SIZE);\ | |
2462 }\ | |
2463 \ | |
2464 static void OPNAME ## h264_qpel ## SIZE ## _mc10_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2465 uint8_t half[SIZE*SIZE];\ | |
2466 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2467 OPNAME ## pixels ## SIZE ## _l2(dst, src, half, stride, stride, SIZE, SIZE);\ | |
2468 }\ | |
2469 \ | |
2470 static void OPNAME ## h264_qpel ## SIZE ## _mc20_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2471 OPNAME ## h264_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride);\ | |
2472 }\ | |
2473 \ | |
2474 static void OPNAME ## h264_qpel ## SIZE ## _mc30_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2475 uint8_t half[SIZE*SIZE];\ | |
2476 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2477 OPNAME ## pixels ## SIZE ## _l2(dst, src+1, half, stride, stride, SIZE, SIZE);\ | |
2478 }\ | |
2479 \ | |
2480 static void OPNAME ## h264_qpel ## SIZE ## _mc01_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2481 uint8_t full[SIZE*(SIZE+5)];\ | |
2482 uint8_t * const full_mid= full + SIZE*2;\ | |
2483 uint8_t half[SIZE*SIZE];\ | |
2484 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2485 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2486 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid, half, stride, SIZE, SIZE, SIZE);\ | |
2487 }\ | |
2488 \ | |
2489 static void OPNAME ## h264_qpel ## SIZE ## _mc02_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2490 uint8_t full[SIZE*(SIZE+5)];\ | |
2491 uint8_t * const full_mid= full + SIZE*2;\ | |
2492 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2493 OPNAME ## h264_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE);\ | |
2494 }\ | |
2495 \ | |
2496 static void OPNAME ## h264_qpel ## SIZE ## _mc03_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2497 uint8_t full[SIZE*(SIZE+5)];\ | |
2498 uint8_t * const full_mid= full + SIZE*2;\ | |
2499 uint8_t half[SIZE*SIZE];\ | |
2500 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2501 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2502 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid+SIZE, half, stride, SIZE, SIZE, SIZE);\ | |
2503 }\ | |
2504 \ | |
2505 static void OPNAME ## h264_qpel ## SIZE ## _mc11_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2506 uint8_t full[SIZE*(SIZE+5)];\ | |
2507 uint8_t * const full_mid= full + SIZE*2;\ | |
2508 uint8_t halfH[SIZE*SIZE];\ | |
2509 uint8_t halfV[SIZE*SIZE];\ | |
2510 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2511 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2512 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2513 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2514 }\ | |
2515 \ | |
2516 static void OPNAME ## h264_qpel ## SIZE ## _mc31_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2517 uint8_t full[SIZE*(SIZE+5)];\ | |
2518 uint8_t * const full_mid= full + SIZE*2;\ | |
2519 uint8_t halfH[SIZE*SIZE];\ | |
2520 uint8_t halfV[SIZE*SIZE];\ | |
2521 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2522 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2523 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2524 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2525 }\ | |
2526 \ | |
2527 static void OPNAME ## h264_qpel ## SIZE ## _mc13_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2528 uint8_t full[SIZE*(SIZE+5)];\ | |
2529 uint8_t * const full_mid= full + SIZE*2;\ | |
2530 uint8_t halfH[SIZE*SIZE];\ | |
2531 uint8_t halfV[SIZE*SIZE];\ | |
2532 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2533 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2534 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2535 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2536 }\ | |
2537 \ | |
2538 static void OPNAME ## h264_qpel ## SIZE ## _mc33_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2539 uint8_t full[SIZE*(SIZE+5)];\ | |
2540 uint8_t * const full_mid= full + SIZE*2;\ | |
2541 uint8_t halfH[SIZE*SIZE];\ | |
2542 uint8_t halfV[SIZE*SIZE];\ | |
2543 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2544 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2545 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2546 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2547 }\ | |
2548 \ | |
2549 static void OPNAME ## h264_qpel ## SIZE ## _mc22_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2550 int16_t tmp[SIZE*(SIZE+5)];\ | |
2551 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass(dst, tmp, src, stride, SIZE, stride);\ | |
2552 }\ | |
2553 \ | |
2554 static void OPNAME ## h264_qpel ## SIZE ## _mc21_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2555 int16_t tmp[SIZE*(SIZE+5)];\ | |
2556 uint8_t halfH[SIZE*SIZE];\ | |
2557 uint8_t halfHV[SIZE*SIZE];\ | |
2558 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2559 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2560 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2561 }\ | |
2562 \ | |
2563 static void OPNAME ## h264_qpel ## SIZE ## _mc23_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2564 int16_t tmp[SIZE*(SIZE+5)];\ | |
2565 uint8_t halfH[SIZE*SIZE];\ | |
2566 uint8_t halfHV[SIZE*SIZE];\ | |
2567 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2568 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2569 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2570 }\ | |
2571 \ | |
2572 static void OPNAME ## h264_qpel ## SIZE ## _mc12_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2573 uint8_t full[SIZE*(SIZE+5)];\ | |
2574 uint8_t * const full_mid= full + SIZE*2;\ | |
2575 int16_t tmp[SIZE*(SIZE+5)];\ | |
2576 uint8_t halfV[SIZE*SIZE];\ | |
2577 uint8_t halfHV[SIZE*SIZE];\ | |
2578 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2579 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2580 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2581 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2582 }\ | |
2583 \ | |
2584 static void OPNAME ## h264_qpel ## SIZE ## _mc32_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2585 uint8_t full[SIZE*(SIZE+5)];\ | |
2586 uint8_t * const full_mid= full + SIZE*2;\ | |
2587 int16_t tmp[SIZE*(SIZE+5)];\ | |
2588 uint8_t halfV[SIZE*SIZE];\ | |
2589 uint8_t halfHV[SIZE*SIZE];\ | |
2590 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2591 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2592 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2593 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2594 }\ | |
2595 | |
2596 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) | |
2597 //#define op_avg2(a, b) a = (((a)*w1+cm[((b) + 16)>>5]*w2 + o + 64)>>7) | |
2598 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2599 #define op2_avg(a, b) a = (((a)+cm[((b) + 512)>>10]+1)>>1) | |
2600 #define op2_put(a, b) a = cm[((b) + 512)>>10] | |
2601 | |
2602 H264_LOWPASS(put_ , op_put, op2_put) | |
2603 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
|
2604 H264_MC(put_, 2) |
1168 | 2605 H264_MC(put_, 4) |
2606 H264_MC(put_, 8) | |
2607 H264_MC(put_, 16) | |
2608 H264_MC(avg_, 4) | |
2609 H264_MC(avg_, 8) | |
2610 H264_MC(avg_, 16) | |
2611 | |
2612 #undef op_avg | |
2613 #undef op_put | |
2614 #undef op2_avg | |
2615 #undef op2_put | |
2616 #endif | |
2617 | |
4594 | 2618 #define op_scale1(x) block[x] = av_clip_uint8( (block[x]*weight + offset) >> log2_denom ) |
2619 #define op_scale2(x) dst[x] = av_clip_uint8( (src[x]*weights + dst[x]*weightd + offset) >> (log2_denom+1)) | |
2415 | 2620 #define H264_WEIGHT(W,H) \ |
2621 static void weight_h264_pixels ## W ## x ## H ## _c(uint8_t *block, int stride, int log2_denom, int weight, int offset){ \ | |
3029 | 2622 int y; \ |
2415 | 2623 offset <<= log2_denom; \ |
2624 if(log2_denom) offset += 1<<(log2_denom-1); \ | |
2625 for(y=0; y<H; y++, block += stride){ \ | |
2626 op_scale1(0); \ | |
2627 op_scale1(1); \ | |
2628 if(W==2) continue; \ | |
2629 op_scale1(2); \ | |
2630 op_scale1(3); \ | |
2631 if(W==4) continue; \ | |
2632 op_scale1(4); \ | |
2633 op_scale1(5); \ | |
2634 op_scale1(6); \ | |
2635 op_scale1(7); \ | |
2636 if(W==8) continue; \ | |
2637 op_scale1(8); \ | |
2638 op_scale1(9); \ | |
2639 op_scale1(10); \ | |
2640 op_scale1(11); \ | |
2641 op_scale1(12); \ | |
2642 op_scale1(13); \ | |
2643 op_scale1(14); \ | |
2644 op_scale1(15); \ | |
2645 } \ | |
2646 } \ | |
3029 | 2647 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){ \ |
2648 int y; \ | |
2649 offset = ((offset + 1) | 1) << log2_denom; \ | |
2415 | 2650 for(y=0; y<H; y++, dst += stride, src += stride){ \ |
2651 op_scale2(0); \ | |
2652 op_scale2(1); \ | |
2653 if(W==2) continue; \ | |
2654 op_scale2(2); \ | |
2655 op_scale2(3); \ | |
2656 if(W==4) continue; \ | |
2657 op_scale2(4); \ | |
2658 op_scale2(5); \ | |
2659 op_scale2(6); \ | |
2660 op_scale2(7); \ | |
2661 if(W==8) continue; \ | |
2662 op_scale2(8); \ | |
2663 op_scale2(9); \ | |
2664 op_scale2(10); \ | |
2665 op_scale2(11); \ | |
2666 op_scale2(12); \ | |
2667 op_scale2(13); \ | |
2668 op_scale2(14); \ | |
2669 op_scale2(15); \ | |
2670 } \ | |
2671 } | |
2672 | |
2673 H264_WEIGHT(16,16) | |
2674 H264_WEIGHT(16,8) | |
2675 H264_WEIGHT(8,16) | |
2676 H264_WEIGHT(8,8) | |
2677 H264_WEIGHT(8,4) | |
2678 H264_WEIGHT(4,8) | |
2679 H264_WEIGHT(4,4) | |
2680 H264_WEIGHT(4,2) | |
2681 H264_WEIGHT(2,4) | |
2682 H264_WEIGHT(2,2) | |
2683 | |
2684 #undef op_scale1 | |
2685 #undef op_scale2 | |
2686 #undef H264_WEIGHT | |
2687 | |
936 | 2688 static void wmv2_mspel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){ |
4176 | 2689 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
936 | 2690 int i; |
2691 | |
2692 for(i=0; i<h; i++){ | |
2693 dst[0]= cm[(9*(src[0] + src[1]) - (src[-1] + src[2]) + 8)>>4]; | |
2694 dst[1]= cm[(9*(src[1] + src[2]) - (src[ 0] + src[3]) + 8)>>4]; | |
2695 dst[2]= cm[(9*(src[2] + src[3]) - (src[ 1] + src[4]) + 8)>>4]; | |
2696 dst[3]= cm[(9*(src[3] + src[4]) - (src[ 2] + src[5]) + 8)>>4]; | |
2697 dst[4]= cm[(9*(src[4] + src[5]) - (src[ 3] + src[6]) + 8)>>4]; | |
2698 dst[5]= cm[(9*(src[5] + src[6]) - (src[ 4] + src[7]) + 8)>>4]; | |
2699 dst[6]= cm[(9*(src[6] + src[7]) - (src[ 5] + src[8]) + 8)>>4]; | |
2700 dst[7]= cm[(9*(src[7] + src[8]) - (src[ 6] + src[9]) + 8)>>4]; | |
2701 dst+=dstStride; | |
2967 | 2702 src+=srcStride; |
936 | 2703 } |
2704 } | |
2705 | |
3432 | 2706 #ifdef CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2707 /* AVS specific */ |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2708 void ff_cavsdsp_init(DSPContext* c, AVCodecContext *avctx); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2709 |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2710 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
|
2711 put_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2712 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2713 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
|
2714 avg_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2715 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2716 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
|
2717 put_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2718 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2719 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
|
2720 avg_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2721 } |
3432 | 2722 #endif /* CONFIG_CAVS_DECODER */ |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2723 |
3526 | 2724 #if defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
2725 /* VC-1 specific */ | |
2726 void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx); | |
2727 | |
2728 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, uint8_t *src, int stride, int rnd) { | |
2729 put_pixels8_c(dst, src, stride, 8); | |
2730 } | |
2731 #endif /* CONFIG_VC1_DECODER||CONFIG_WMV3_DECODER */ | |
2732 | |
5887 | 2733 void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx); |
5899 | 2734 |
4296 | 2735 /* H264 specific */ |
5411
362aec4ef932
Take care of some renames (Doxygen and function name) after the previous pure rename patch.
takis
parents:
5394
diff
changeset
|
2736 void ff_h264dspenc_init(DSPContext* c, AVCodecContext *avctx); |
4296 | 2737 |
936 | 2738 static void wmv2_mspel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int w){ |
4176 | 2739 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
936 | 2740 int i; |
2741 | |
2742 for(i=0; i<w; i++){ | |
2743 const int src_1= src[ -srcStride]; | |
2744 const int src0 = src[0 ]; | |
2745 const int src1 = src[ srcStride]; | |
2746 const int src2 = src[2*srcStride]; | |
2747 const int src3 = src[3*srcStride]; | |
2748 const int src4 = src[4*srcStride]; | |
2749 const int src5 = src[5*srcStride]; | |
2750 const int src6 = src[6*srcStride]; | |
2751 const int src7 = src[7*srcStride]; | |
2752 const int src8 = src[8*srcStride]; | |
2753 const int src9 = src[9*srcStride]; | |
2754 dst[0*dstStride]= cm[(9*(src0 + src1) - (src_1 + src2) + 8)>>4]; | |
2755 dst[1*dstStride]= cm[(9*(src1 + src2) - (src0 + src3) + 8)>>4]; | |
2756 dst[2*dstStride]= cm[(9*(src2 + src3) - (src1 + src4) + 8)>>4]; | |
2757 dst[3*dstStride]= cm[(9*(src3 + src4) - (src2 + src5) + 8)>>4]; | |
2758 dst[4*dstStride]= cm[(9*(src4 + src5) - (src3 + src6) + 8)>>4]; | |
2759 dst[5*dstStride]= cm[(9*(src5 + src6) - (src4 + src7) + 8)>>4]; | |
2760 dst[6*dstStride]= cm[(9*(src6 + src7) - (src5 + src8) + 8)>>4]; | |
2761 dst[7*dstStride]= cm[(9*(src7 + src8) - (src6 + src9) + 8)>>4]; | |
2762 src++; | |
2763 dst++; | |
2764 } | |
2765 } | |
2766 | |
2767 static void put_mspel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){ | |
2768 put_pixels8_c(dst, src, stride, 8); | |
2769 } | |
2770 | |
2771 static void put_mspel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){ | |
2772 uint8_t half[64]; | |
2773 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2774 put_pixels8_l2(dst, src, half, stride, stride, 8, 8); | |
2775 } | |
2776 | |
2777 static void put_mspel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){ | |
2778 wmv2_mspel8_h_lowpass(dst, src, stride, stride, 8); | |
2779 } | |
2780 | |
2781 static void put_mspel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){ | |
2782 uint8_t half[64]; | |
2783 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2784 put_pixels8_l2(dst, src+1, half, stride, stride, 8, 8); | |
2785 } | |
2786 | |
2787 static void put_mspel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){ | |
2788 wmv2_mspel8_v_lowpass(dst, src, stride, stride, 8); | |
2789 } | |
2790 | |
2791 static void put_mspel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){ | |
2792 uint8_t halfH[88]; | |
2793 uint8_t halfV[64]; | |
2794 uint8_t halfHV[64]; | |
2795 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2796 wmv2_mspel8_v_lowpass(halfV, src, 8, stride, 8); | |
2797 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2798 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2799 } | |
2800 static void put_mspel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){ | |
2801 uint8_t halfH[88]; | |
2802 uint8_t halfV[64]; | |
2803 uint8_t halfHV[64]; | |
2804 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2805 wmv2_mspel8_v_lowpass(halfV, src+1, 8, stride, 8); | |
2806 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2807 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2808 } | |
2809 static void put_mspel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){ | |
2810 uint8_t halfH[88]; | |
2811 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2812 wmv2_mspel8_v_lowpass(dst, halfH+8, stride, 8, 8); | |
2813 } | |
2814 | |
1644 | 2815 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
|
2816 if(ENABLE_ANY_H263) { |
1644 | 2817 int x; |
2818 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2819 |
1644 | 2820 for(x=0; x<8; x++){ |
2821 int d1, d2, ad1; | |
2822 int p0= src[x-2*stride]; | |
2823 int p1= src[x-1*stride]; | |
2824 int p2= src[x+0*stride]; | |
2825 int p3= src[x+1*stride]; | |
2826 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2827 | |
2828 if (d<-2*strength) d1= 0; | |
2829 else if(d<- strength) d1=-2*strength - d; | |
2830 else if(d< strength) d1= d; | |
2831 else if(d< 2*strength) d1= 2*strength - d; | |
2832 else d1= 0; | |
2967 | 2833 |
1644 | 2834 p1 += d1; |
2835 p2 -= d1; | |
2836 if(p1&256) p1= ~(p1>>31); | |
2837 if(p2&256) p2= ~(p2>>31); | |
2967 | 2838 |
1644 | 2839 src[x-1*stride] = p1; |
2840 src[x+0*stride] = p2; | |
2841 | |
4001 | 2842 ad1= FFABS(d1)>>1; |
2967 | 2843 |
4594 | 2844 d2= av_clip((p0-p3)/4, -ad1, ad1); |
2967 | 2845 |
1644 | 2846 src[x-2*stride] = p0 - d2; |
2847 src[x+ stride] = p3 + d2; | |
2848 } | |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2849 } |
1644 | 2850 } |
2851 | |
2852 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
|
2853 if(ENABLE_ANY_H263) { |
1644 | 2854 int y; |
2855 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2856 |
1644 | 2857 for(y=0; y<8; y++){ |
2858 int d1, d2, ad1; | |
2859 int p0= src[y*stride-2]; | |
2860 int p1= src[y*stride-1]; | |
2861 int p2= src[y*stride+0]; | |
2862 int p3= src[y*stride+1]; | |
2863 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2864 | |
2865 if (d<-2*strength) d1= 0; | |
2866 else if(d<- strength) d1=-2*strength - d; | |
2867 else if(d< strength) d1= d; | |
2868 else if(d< 2*strength) d1= 2*strength - d; | |
2869 else d1= 0; | |
2967 | 2870 |
1644 | 2871 p1 += d1; |
2872 p2 -= d1; | |
2873 if(p1&256) p1= ~(p1>>31); | |
2874 if(p2&256) p2= ~(p2>>31); | |
2967 | 2875 |
1644 | 2876 src[y*stride-1] = p1; |
2877 src[y*stride+0] = p2; | |
2878 | |
4001 | 2879 ad1= FFABS(d1)>>1; |
2967 | 2880 |
4594 | 2881 d2= av_clip((p0-p3)/4, -ad1, ad1); |
2967 | 2882 |
1644 | 2883 src[y*stride-2] = p0 - d2; |
2884 src[y*stride+1] = p3 + d2; | |
2885 } | |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2886 } |
1644 | 2887 } |
936 | 2888 |
2045 | 2889 static void h261_loop_filter_c(uint8_t *src, int stride){ |
2890 int x,y,xy,yz; | |
2891 int temp[64]; | |
2892 | |
2893 for(x=0; x<8; x++){ | |
2894 temp[x ] = 4*src[x ]; | |
2895 temp[x + 7*8] = 4*src[x + 7*stride]; | |
2896 } | |
2897 for(y=1; y<7; y++){ | |
2898 for(x=0; x<8; x++){ | |
2899 xy = y * stride + x; | |
2900 yz = y * 8 + x; | |
2901 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
|
2902 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2903 } |
2967 | 2904 |
2045 | 2905 for(y=0; y<8; y++){ |
2906 src[ y*stride] = (temp[ y*8] + 2)>>2; | |
2907 src[7+y*stride] = (temp[7+y*8] + 2)>>2; | |
2908 for(x=1; x<7; x++){ | |
2909 xy = y * stride + x; | |
2910 yz = y * 8 + x; | |
2911 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
|
2912 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2913 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2914 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2915 |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2916 static inline void h264_loop_filter_luma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0) |
2633 | 2917 { |
2918 int i, d; | |
2919 for( i = 0; i < 4; i++ ) { | |
2920 if( tc0[i] < 0 ) { | |
2921 pix += 4*ystride; | |
2922 continue; | |
2923 } | |
2924 for( d = 0; d < 4; d++ ) { | |
2925 const int p0 = pix[-1*xstride]; | |
2926 const int p1 = pix[-2*xstride]; | |
2927 const int p2 = pix[-3*xstride]; | |
2928 const int q0 = pix[0]; | |
2929 const int q1 = pix[1*xstride]; | |
2930 const int q2 = pix[2*xstride]; | |
2967 | 2931 |
4001 | 2932 if( FFABS( p0 - q0 ) < alpha && |
2933 FFABS( p1 - p0 ) < beta && | |
2934 FFABS( q1 - q0 ) < beta ) { | |
2967 | 2935 |
2633 | 2936 int tc = tc0[i]; |
2937 int i_delta; | |
2967 | 2938 |
4001 | 2939 if( FFABS( p2 - p0 ) < beta ) { |
4594 | 2940 pix[-2*xstride] = p1 + av_clip( (( p2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - p1, -tc0[i], tc0[i] ); |
2633 | 2941 tc++; |
2942 } | |
4001 | 2943 if( FFABS( q2 - q0 ) < beta ) { |
4594 | 2944 pix[ xstride] = q1 + av_clip( (( q2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - q1, -tc0[i], tc0[i] ); |
2633 | 2945 tc++; |
2946 } | |
2967 | 2947 |
4594 | 2948 i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
2949 pix[-xstride] = av_clip_uint8( p0 + i_delta ); /* p0' */ | |
2950 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */ | |
2633 | 2951 } |
2952 pix += ystride; | |
2953 } | |
2954 } | |
2955 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2956 static void h264_v_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2957 { |
2958 h264_loop_filter_luma_c(pix, stride, 1, alpha, beta, tc0); | |
2959 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2960 static void h264_h_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2961 { |
2962 h264_loop_filter_luma_c(pix, 1, stride, alpha, beta, tc0); | |
2963 } | |
2964 | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2965 static inline void h264_loop_filter_chroma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0) |
2633 | 2966 { |
2967 int i, d; | |
2968 for( i = 0; i < 4; i++ ) { | |
2969 const int tc = tc0[i]; | |
2970 if( tc <= 0 ) { | |
2971 pix += 2*ystride; | |
2972 continue; | |
2973 } | |
2974 for( d = 0; d < 2; d++ ) { | |
2975 const int p0 = pix[-1*xstride]; | |
2976 const int p1 = pix[-2*xstride]; | |
2977 const int q0 = pix[0]; | |
2978 const int q1 = pix[1*xstride]; | |
2979 | |
4001 | 2980 if( FFABS( p0 - q0 ) < alpha && |
2981 FFABS( p1 - p0 ) < beta && | |
2982 FFABS( q1 - q0 ) < beta ) { | |
2633 | 2983 |
4594 | 2984 int delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
2985 | |
2986 pix[-xstride] = av_clip_uint8( p0 + delta ); /* p0' */ | |
2987 pix[0] = av_clip_uint8( q0 - delta ); /* q0' */ | |
2633 | 2988 } |
2989 pix += ystride; | |
2990 } | |
2991 } | |
2992 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2993 static void h264_v_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2994 { |
2995 h264_loop_filter_chroma_c(pix, stride, 1, alpha, beta, tc0); | |
2996 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2997 static void h264_h_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2998 { |
2999 h264_loop_filter_chroma_c(pix, 1, stride, alpha, beta, tc0); | |
3000 } | |
3001 | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
3002 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
|
3003 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
3004 int d; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
3005 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
|
3006 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
|
3007 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
|
3008 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
|
3009 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
|
3010 |
4001 | 3011 if( FFABS( p0 - q0 ) < alpha && |
3012 FFABS( p1 - p0 ) < beta && | |
3013 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
|
3014 |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
3015 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
|
3016 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
|
3017 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
3018 pix += ystride; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
3019 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
3020 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
3021 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
|
3022 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
3023 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
|
3024 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
3025 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
|
3026 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
3027 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
|
3028 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
3029 |
1708 | 3030 static inline int pix_abs16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 3031 { |
3032 int s, i; | |
3033 | |
3034 s = 0; | |
1708 | 3035 for(i=0;i<h;i++) { |
0 | 3036 s += abs(pix1[0] - pix2[0]); |
3037 s += abs(pix1[1] - pix2[1]); | |
3038 s += abs(pix1[2] - pix2[2]); | |
3039 s += abs(pix1[3] - pix2[3]); | |
3040 s += abs(pix1[4] - pix2[4]); | |
3041 s += abs(pix1[5] - pix2[5]); | |
3042 s += abs(pix1[6] - pix2[6]); | |
3043 s += abs(pix1[7] - pix2[7]); | |
3044 s += abs(pix1[8] - pix2[8]); | |
3045 s += abs(pix1[9] - pix2[9]); | |
3046 s += abs(pix1[10] - pix2[10]); | |
3047 s += abs(pix1[11] - pix2[11]); | |
3048 s += abs(pix1[12] - pix2[12]); | |
3049 s += abs(pix1[13] - pix2[13]); | |
3050 s += abs(pix1[14] - pix2[14]); | |
3051 s += abs(pix1[15] - pix2[15]); | |
3052 pix1 += line_size; | |
3053 pix2 += line_size; | |
3054 } | |
3055 return s; | |
3056 } | |
3057 | |
1708 | 3058 static int pix_abs16_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 3059 { |
3060 int s, i; | |
3061 | |
3062 s = 0; | |
1708 | 3063 for(i=0;i<h;i++) { |
0 | 3064 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
3065 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
3066 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
3067 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
3068 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
3069 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
3070 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
3071 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
3072 s += abs(pix1[8] - avg2(pix2[8], pix2[9])); | |
3073 s += abs(pix1[9] - avg2(pix2[9], pix2[10])); | |
3074 s += abs(pix1[10] - avg2(pix2[10], pix2[11])); | |
3075 s += abs(pix1[11] - avg2(pix2[11], pix2[12])); | |
3076 s += abs(pix1[12] - avg2(pix2[12], pix2[13])); | |
3077 s += abs(pix1[13] - avg2(pix2[13], pix2[14])); | |
3078 s += abs(pix1[14] - avg2(pix2[14], pix2[15])); | |
3079 s += abs(pix1[15] - avg2(pix2[15], pix2[16])); | |
3080 pix1 += line_size; | |
3081 pix2 += line_size; | |
3082 } | |
3083 return s; | |
3084 } | |
3085 | |
1708 | 3086 static int pix_abs16_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 3087 { |
3088 int s, i; | |
1064 | 3089 uint8_t *pix3 = pix2 + line_size; |
0 | 3090 |
3091 s = 0; | |
1708 | 3092 for(i=0;i<h;i++) { |
0 | 3093 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
3094 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
3095 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
3096 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
3097 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
3098 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
3099 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
3100 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
3101 s += abs(pix1[8] - avg2(pix2[8], pix3[8])); | |
3102 s += abs(pix1[9] - avg2(pix2[9], pix3[9])); | |
3103 s += abs(pix1[10] - avg2(pix2[10], pix3[10])); | |
3104 s += abs(pix1[11] - avg2(pix2[11], pix3[11])); | |
3105 s += abs(pix1[12] - avg2(pix2[12], pix3[12])); | |
3106 s += abs(pix1[13] - avg2(pix2[13], pix3[13])); | |
3107 s += abs(pix1[14] - avg2(pix2[14], pix3[14])); | |
3108 s += abs(pix1[15] - avg2(pix2[15], pix3[15])); | |
3109 pix1 += line_size; | |
3110 pix2 += line_size; | |
3111 pix3 += line_size; | |
3112 } | |
3113 return s; | |
3114 } | |
3115 | |
1708 | 3116 static int pix_abs16_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 3117 { |
3118 int s, i; | |
1064 | 3119 uint8_t *pix3 = pix2 + line_size; |
0 | 3120 |
3121 s = 0; | |
1708 | 3122 for(i=0;i<h;i++) { |
0 | 3123 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
3124 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
3125 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
3126 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
3127 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
3128 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
3129 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
3130 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
3131 s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9])); | |
3132 s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10])); | |
3133 s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11])); | |
3134 s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12])); | |
3135 s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13])); | |
3136 s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14])); | |
3137 s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15])); | |
3138 s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16])); | |
3139 pix1 += line_size; | |
3140 pix2 += line_size; | |
3141 pix3 += line_size; | |
3142 } | |
3143 return s; | |
3144 } | |
3145 | |
1708 | 3146 static inline int pix_abs8_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3147 { |
3148 int s, i; | |
3149 | |
3150 s = 0; | |
1708 | 3151 for(i=0;i<h;i++) { |
294 | 3152 s += abs(pix1[0] - pix2[0]); |
3153 s += abs(pix1[1] - pix2[1]); | |
3154 s += abs(pix1[2] - pix2[2]); | |
3155 s += abs(pix1[3] - pix2[3]); | |
3156 s += abs(pix1[4] - pix2[4]); | |
3157 s += abs(pix1[5] - pix2[5]); | |
3158 s += abs(pix1[6] - pix2[6]); | |
3159 s += abs(pix1[7] - pix2[7]); | |
3160 pix1 += line_size; | |
3161 pix2 += line_size; | |
3162 } | |
3163 return s; | |
3164 } | |
3165 | |
1708 | 3166 static int pix_abs8_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3167 { |
3168 int s, i; | |
3169 | |
3170 s = 0; | |
1708 | 3171 for(i=0;i<h;i++) { |
294 | 3172 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
3173 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
3174 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
3175 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
3176 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
3177 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
3178 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
3179 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
3180 pix1 += line_size; | |
3181 pix2 += line_size; | |
3182 } | |
3183 return s; | |
3184 } | |
3185 | |
1708 | 3186 static int pix_abs8_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3187 { |
3188 int s, i; | |
1064 | 3189 uint8_t *pix3 = pix2 + line_size; |
294 | 3190 |
3191 s = 0; | |
1708 | 3192 for(i=0;i<h;i++) { |
294 | 3193 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
3194 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
3195 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
3196 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
3197 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
3198 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
3199 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
3200 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
3201 pix1 += line_size; | |
3202 pix2 += line_size; | |
3203 pix3 += line_size; | |
3204 } | |
3205 return s; | |
3206 } | |
3207 | |
1708 | 3208 static int pix_abs8_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3209 { |
3210 int s, i; | |
1064 | 3211 uint8_t *pix3 = pix2 + line_size; |
294 | 3212 |
3213 s = 0; | |
1708 | 3214 for(i=0;i<h;i++) { |
294 | 3215 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
3216 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
3217 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
3218 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
3219 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
3220 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
3221 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
3222 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
3223 pix1 += line_size; | |
3224 pix2 += line_size; | |
3225 pix3 += line_size; | |
3226 } | |
3227 return s; | |
3228 } | |
3229 | |
2834 | 3230 static int nsse16_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3231 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3232 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3233 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3234 int x,y; |
2066 | 3235 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3236 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3237 for(x=0; x<16; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3238 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3239 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3240 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3241 for(x=0; x<15; x++){ |
4001 | 3242 score2+= FFABS( s1[x ] - s1[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3243 - s1[x+1] + s1[x+1+stride]) |
4001 | 3244 -FFABS( s2[x ] - s2[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3245 - s2[x+1] + s2[x+1+stride]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3246 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3247 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3248 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3249 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3250 } |
2066 | 3251 |
4001 | 3252 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight; |
3253 else return score1 + FFABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3254 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3255 |
2834 | 3256 static int nsse8_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3257 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3258 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3259 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3260 int x,y; |
2967 | 3261 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3262 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3263 for(x=0; x<8; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3264 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3265 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3266 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3267 for(x=0; x<7; x++){ |
4001 | 3268 score2+= FFABS( s1[x ] - s1[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3269 - s1[x+1] + s1[x+1+stride]) |
4001 | 3270 -FFABS( s2[x ] - s2[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3271 - s2[x+1] + s2[x+1+stride]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3272 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3273 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3274 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3275 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3276 } |
2967 | 3277 |
4001 | 3278 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight; |
3279 else return score1 + FFABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3280 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3281 |
1784 | 3282 static int try_8x8basis_c(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale){ |
3283 int i; | |
3284 unsigned int sum=0; | |
3285 | |
3286 for(i=0; i<8*8; i++){ | |
3287 int b= rem[i] + ((basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT)); | |
3288 int w= weight[i]; | |
3289 b>>= RECON_SHIFT; | |
3290 assert(-512<b && b<512); | |
3291 | |
3292 sum += (w*b)*(w*b)>>4; | |
3293 } | |
3294 return sum>>2; | |
3295 } | |
3296 | |
3297 static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale){ | |
3298 int i; | |
3299 | |
3300 for(i=0; i<8*8; i++){ | |
3301 rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT); | |
2967 | 3302 } |
1784 | 3303 } |
3304 | |
1100 | 3305 /** |
3306 * permutes an 8x8 block. | |
1101 | 3307 * @param block the block which will be permuted according to the given permutation vector |
1100 | 3308 * @param permutation the permutation vector |
3309 * @param last the last non zero coefficient in scantable order, used to speed the permutation up | |
2967 | 3310 * @param scantable the used scantable, this is only used to speed the permutation up, the block is not |
1101 | 3311 * (inverse) permutated to scantable order! |
1100 | 3312 */ |
1064 | 3313 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
|
3314 { |
764 | 3315 int i; |
945 | 3316 DCTELEM temp[64]; |
2967 | 3317 |
764 | 3318 if(last<=0) return; |
5129 | 3319 //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
|
3320 |
764 | 3321 for(i=0; i<=last; i++){ |
3322 const int j= scantable[i]; | |
3323 temp[j]= block[j]; | |
3324 block[j]=0; | |
3325 } | |
2967 | 3326 |
764 | 3327 for(i=0; i<=last; i++){ |
3328 const int j= scantable[i]; | |
3329 const int perm_j= permutation[j]; | |
3330 block[perm_j]= temp[j]; | |
3331 } | |
174
ac5075a55488
new IDCT code by Michael Niedermayer (michaelni@gmx.at) - #define SIMPLE_IDCT to enable
arpi_esp
parents:
88
diff
changeset
|
3332 } |
34 | 3333 |
1729 | 3334 static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride, int h){ |
3335 return 0; | |
3336 } | |
3337 | |
3338 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type){ | |
3339 int i; | |
2967 | 3340 |
1729 | 3341 memset(cmp, 0, sizeof(void*)*5); |
2967 | 3342 |
1729 | 3343 for(i=0; i<5; i++){ |
3344 switch(type&0xFF){ | |
3345 case FF_CMP_SAD: | |
3346 cmp[i]= c->sad[i]; | |
3347 break; | |
3348 case FF_CMP_SATD: | |
3349 cmp[i]= c->hadamard8_diff[i]; | |
3350 break; | |
3351 case FF_CMP_SSE: | |
3352 cmp[i]= c->sse[i]; | |
3353 break; | |
3354 case FF_CMP_DCT: | |
3355 cmp[i]= c->dct_sad[i]; | |
3356 break; | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3357 case FF_CMP_DCT264: |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3358 cmp[i]= c->dct264_sad[i]; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3359 break; |
2382 | 3360 case FF_CMP_DCTMAX: |
3361 cmp[i]= c->dct_max[i]; | |
3362 break; | |
1729 | 3363 case FF_CMP_PSNR: |
3364 cmp[i]= c->quant_psnr[i]; | |
3365 break; | |
3366 case FF_CMP_BIT: | |
3367 cmp[i]= c->bit[i]; | |
3368 break; | |
3369 case FF_CMP_RD: | |
3370 cmp[i]= c->rd[i]; | |
3371 break; | |
3372 case FF_CMP_VSAD: | |
3373 cmp[i]= c->vsad[i]; | |
3374 break; | |
3375 case FF_CMP_VSSE: | |
3376 cmp[i]= c->vsse[i]; | |
3377 break; | |
3378 case FF_CMP_ZERO: | |
3379 cmp[i]= zero_cmp; | |
3380 break; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3381 case FF_CMP_NSSE: |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3382 cmp[i]= c->nsse[i]; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3383 break; |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
3384 #ifdef CONFIG_SNOW_ENCODER |
2184 | 3385 case FF_CMP_W53: |
3386 cmp[i]= c->w53[i]; | |
3387 break; | |
3388 case FF_CMP_W97: | |
3389 cmp[i]= c->w97[i]; | |
3390 break; | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
3391 #endif |
1729 | 3392 default: |
3393 av_log(NULL, AV_LOG_ERROR,"internal error in cmp function selection\n"); | |
3394 } | |
3395 } | |
3396 } | |
3397 | |
1101 | 3398 /** |
3399 * memset(blocks, 0, sizeof(DCTELEM)*6*64) | |
3400 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
3401 static void clear_blocks_c(DCTELEM *blocks) |
296 | 3402 { |
3403 memset(blocks, 0, sizeof(DCTELEM)*6*64); | |
3404 } | |
3405 | |
866 | 3406 static void add_bytes_c(uint8_t *dst, uint8_t *src, int w){ |
6385 | 3407 long i; |
3408 for(i=0; i<=w-sizeof(long); i+=sizeof(long)){ | |
3409 long a = *(long*)(src+i); | |
3410 long b = *(long*)(dst+i); | |
3411 *(long*)(dst+i) = ((a&pb_7f) + (b&pb_7f)) ^ ((a^b)&pb_80); | |
866 | 3412 } |
3413 for(; i<w; i++) | |
3414 dst[i+0] += src[i+0]; | |
3415 } | |
3416 | |
6384 | 3417 static void add_bytes_l2_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ |
6385 | 3418 long i; |
6384 | 3419 for(i=0; i<=w-sizeof(long); i+=sizeof(long)){ |
3420 long a = *(long*)(src1+i); | |
3421 long b = *(long*)(src2+i); | |
6385 | 3422 *(long*)(dst+i) = ((a&pb_7f) + (b&pb_7f)) ^ ((a^b)&pb_80); |
6384 | 3423 } |
3424 for(; i<w; i++) | |
3425 dst[i] = src1[i]+src2[i]; | |
3426 } | |
3427 | |
866 | 3428 static void diff_bytes_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ |
6385 | 3429 long i; |
3430 #ifndef HAVE_FAST_UNALIGNED | |
3431 if((long)src2 & (sizeof(long)-1)){ | |
6386 | 3432 for(i=0; i+7<w; i+=8){ |
3433 dst[i+0] = src1[i+0]-src2[i+0]; | |
3434 dst[i+1] = src1[i+1]-src2[i+1]; | |
3435 dst[i+2] = src1[i+2]-src2[i+2]; | |
3436 dst[i+3] = src1[i+3]-src2[i+3]; | |
3437 dst[i+4] = src1[i+4]-src2[i+4]; | |
3438 dst[i+5] = src1[i+5]-src2[i+5]; | |
3439 dst[i+6] = src1[i+6]-src2[i+6]; | |
3440 dst[i+7] = src1[i+7]-src2[i+7]; | |
3441 } | |
6385 | 3442 }else |
3443 #endif | |
3444 for(i=0; i<=w-sizeof(long); i+=sizeof(long)){ | |
3445 long a = *(long*)(src1+i); | |
3446 long b = *(long*)(src2+i); | |
3447 *(long*)(dst+i) = ((a|pb_80) - (b&pb_7f)) ^ ((a^b^pb_80)&pb_80); | |
3448 } | |
866 | 3449 for(; i<w; i++) |
3450 dst[i+0] = src1[i+0]-src2[i+0]; | |
3451 } | |
3452 | |
1527 | 3453 static void sub_hfyu_median_prediction_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){ |
3454 int i; | |
3455 uint8_t l, lt; | |
3456 | |
3457 l= *left; | |
3458 lt= *left_top; | |
3459 | |
3460 for(i=0; i<w; i++){ | |
3461 const int pred= mid_pred(l, src1[i], (l + src1[i] - lt)&0xFF); | |
3462 lt= src1[i]; | |
3463 l= src2[i]; | |
3464 dst[i]= l - pred; | |
2967 | 3465 } |
1527 | 3466 |
3467 *left= l; | |
3468 *left_top= lt; | |
3469 } | |
3470 | |
936 | 3471 #define BUTTERFLY2(o1,o2,i1,i2) \ |
3472 o1= (i1)+(i2);\ | |
3473 o2= (i1)-(i2); | |
3474 | |
3475 #define BUTTERFLY1(x,y) \ | |
3476 {\ | |
3477 int a,b;\ | |
3478 a= x;\ | |
3479 b= y;\ | |
3480 x= a+b;\ | |
3481 y= a-b;\ | |
3482 } | |
3483 | |
4001 | 3484 #define BUTTERFLYA(x,y) (FFABS((x)+(y)) + FFABS((x)-(y))) |
936 | 3485 |
1708 | 3486 static int hadamard8_diff8x8_c(/*MpegEncContext*/ void *s, uint8_t *dst, uint8_t *src, int stride, int h){ |
936 | 3487 int i; |
3488 int temp[64]; | |
3489 int sum=0; | |
2967 | 3490 |
1708 | 3491 assert(h==8); |
936 | 3492 |
3493 for(i=0; i<8; i++){ | |
3494 //FIXME try pointer walks | |
3495 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]); | |
3496 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]); | |
3497 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]); | |
3498 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 | 3499 |
936 | 3500 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3501 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3502 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3503 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3504 |
936 | 3505 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3506 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3507 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3508 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3509 } | |
3510 | |
3511 for(i=0; i<8; i++){ | |
3512 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3513 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3514 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3515 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3516 |
936 | 3517 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3518 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3519 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3520 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
3521 | |
2967 | 3522 sum += |
936 | 3523 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3524 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3525 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3526 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3527 } | |
3528 #if 0 | |
3529 static int maxi=0; | |
3530 if(sum>maxi){ | |
3531 maxi=sum; | |
3532 printf("MAX:%d\n", maxi); | |
3533 } | |
3534 #endif | |
3535 return sum; | |
3536 } | |
3537 | |
1729 | 3538 static int hadamard8_intra8x8_c(/*MpegEncContext*/ void *s, uint8_t *src, uint8_t *dummy, int stride, int h){ |
936 | 3539 int i; |
3540 int temp[64]; | |
3541 int sum=0; | |
2967 | 3542 |
1729 | 3543 assert(h==8); |
2967 | 3544 |
936 | 3545 for(i=0; i<8; i++){ |
3546 //FIXME try pointer walks | |
1729 | 3547 BUTTERFLY2(temp[8*i+0], temp[8*i+1], src[stride*i+0],src[stride*i+1]); |
3548 BUTTERFLY2(temp[8*i+2], temp[8*i+3], src[stride*i+2],src[stride*i+3]); | |
3549 BUTTERFLY2(temp[8*i+4], temp[8*i+5], src[stride*i+4],src[stride*i+5]); | |
3550 BUTTERFLY2(temp[8*i+6], temp[8*i+7], src[stride*i+6],src[stride*i+7]); | |
2967 | 3551 |
936 | 3552 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3553 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3554 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3555 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3556 |
936 | 3557 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3558 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3559 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3560 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3561 } | |
3562 | |
3563 for(i=0; i<8; i++){ | |
3564 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3565 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3566 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3567 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3568 |
936 | 3569 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3570 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3571 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3572 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
2967 | 3573 |
3574 sum += | |
936 | 3575 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3576 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3577 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3578 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3579 } | |
2967 | 3580 |
4001 | 3581 sum -= FFABS(temp[8*0] + temp[8*4]); // -mean |
2967 | 3582 |
936 | 3583 return sum; |
3584 } | |
3585 | |
1708 | 3586 static int dct_sad8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3587 MpegEncContext * const s= (MpegEncContext *)c; |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
3588 DECLARE_ALIGNED_16(uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
1016 | 3589 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
2967 | 3590 |
1708 | 3591 assert(h==8); |
936 | 3592 |
3593 s->dsp.diff_pixels(temp, src1, src2, stride); | |
1092 | 3594 s->dsp.fdct(temp); |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
3595 return s->dsp.sum_abs_dctelem(temp); |
936 | 3596 } |
3597 | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3598 #ifdef CONFIG_GPL |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3599 #define DCT8_1D {\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3600 const int s07 = SRC(0) + SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3601 const int s16 = SRC(1) + SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3602 const int s25 = SRC(2) + SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3603 const int s34 = SRC(3) + SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3604 const int a0 = s07 + s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3605 const int a1 = s16 + s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3606 const int a2 = s07 - s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3607 const int a3 = s16 - s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3608 const int d07 = SRC(0) - SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3609 const int d16 = SRC(1) - SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3610 const int d25 = SRC(2) - SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3611 const int d34 = SRC(3) - SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3612 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
|
3613 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
|
3614 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
|
3615 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
|
3616 DST(0, a0 + a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3617 DST(1, a4 + (a7>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3618 DST(2, a2 + (a3>>1)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3619 DST(3, a5 + (a6>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3620 DST(4, a0 - a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3621 DST(5, a6 - (a5>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3622 DST(6, (a2>>1) - a3 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3623 DST(7, (a4>>2) - a7 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3624 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3625 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3626 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
|
3627 MpegEncContext * const s= (MpegEncContext *)c; |
5256 | 3628 DCTELEM dct[8][8]; |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3629 int i; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3630 int sum=0; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3631 |
5256 | 3632 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
|
3633 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3634 #define SRC(x) dct[i][x] |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3635 #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
|
3636 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3637 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3638 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3639 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3640 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3641 #define SRC(x) dct[x][i] |
4001 | 3642 #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
|
3643 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3644 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3645 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3646 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3647 return sum; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3648 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3649 #endif |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3650 |
2382 | 3651 static int dct_max8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
3652 MpegEncContext * const s= (MpegEncContext *)c; | |
3089 | 3653 DECLARE_ALIGNED_8(uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
2382 | 3654 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3655 int sum=0, i; | |
2967 | 3656 |
2382 | 3657 assert(h==8); |
3658 | |
3659 s->dsp.diff_pixels(temp, src1, src2, stride); | |
3660 s->dsp.fdct(temp); | |
3661 | |
3662 for(i=0; i<64; i++) | |
4001 | 3663 sum= FFMAX(sum, FFABS(temp[i])); |
2967 | 3664 |
2382 | 3665 return sum; |
3666 } | |
3667 | |
1708 | 3668 static int quant_psnr8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3669 MpegEncContext * const s= (MpegEncContext *)c; |
3089 | 3670 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64*2/8]); |
1016 | 3671 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3672 DCTELEM * const bak = ((DCTELEM*)aligned_temp)+64; | |
936 | 3673 int sum=0, i; |
3674 | |
1708 | 3675 assert(h==8); |
936 | 3676 s->mb_intra=0; |
2967 | 3677 |
936 | 3678 s->dsp.diff_pixels(temp, src1, src2, stride); |
2967 | 3679 |
936 | 3680 memcpy(bak, temp, 64*sizeof(DCTELEM)); |
2967 | 3681 |
1013 | 3682 s->block_last_index[0/*FIXME*/]= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
1689 | 3683 s->dct_unquantize_inter(s, temp, 0, s->qscale); |
6001 | 3684 ff_simple_idct(temp); //FIXME |
2967 | 3685 |
936 | 3686 for(i=0; i<64; i++) |
3687 sum+= (temp[i]-bak[i])*(temp[i]-bak[i]); | |
2967 | 3688 |
936 | 3689 return sum; |
3690 } | |
3691 | |
1708 | 3692 static int rd8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3693 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3694 const uint8_t *scantable= s->intra_scantable.permutated; |
3089 | 3695 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
3696 DECLARE_ALIGNED_8 (uint64_t, aligned_bak[stride]); | |
1016 | 3697 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3698 uint8_t * const bak= (uint8_t*)aligned_bak; | |
1007 | 3699 int i, last, run, bits, level, distoration, start_i; |
3700 const int esc_length= s->ac_esc_length; | |
3701 uint8_t * length; | |
3702 uint8_t * last_length; | |
2967 | 3703 |
1708 | 3704 assert(h==8); |
3705 | |
1007 | 3706 for(i=0; i<8; i++){ |
3707 ((uint32_t*)(bak + i*stride))[0]= ((uint32_t*)(src2 + i*stride))[0]; | |
3708 ((uint32_t*)(bak + i*stride))[1]= ((uint32_t*)(src2 + i*stride))[1]; | |
3709 } | |
3710 | |
3711 s->dsp.diff_pixels(temp, src1, src2, stride); | |
3712 | |
1013 | 3713 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3714 | |
3715 bits=0; | |
2967 | 3716 |
1013 | 3717 if (s->mb_intra) { |
2967 | 3718 start_i = 1; |
1013 | 3719 length = s->intra_ac_vlc_length; |
3720 last_length= s->intra_ac_vlc_last_length; | |
3721 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma | |
3722 } else { | |
3723 start_i = 0; | |
3724 length = s->inter_ac_vlc_length; | |
3725 last_length= s->inter_ac_vlc_last_length; | |
3726 } | |
2967 | 3727 |
1013 | 3728 if(last>=start_i){ |
1007 | 3729 run=0; |
3730 for(i=start_i; i<last; i++){ | |
3731 int j= scantable[i]; | |
3732 level= temp[j]; | |
2967 | 3733 |
1007 | 3734 if(level){ |
3735 level+=64; | |
3736 if((level&(~127)) == 0){ | |
3737 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3738 }else | |
3739 bits+= esc_length; | |
3740 run=0; | |
3741 }else | |
3742 run++; | |
3743 } | |
3744 i= scantable[last]; | |
2967 | 3745 |
1011 | 3746 level= temp[i] + 64; |
3747 | |
3748 assert(level - 64); | |
2967 | 3749 |
1007 | 3750 if((level&(~127)) == 0){ |
3751 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3752 }else | |
3753 bits+= esc_length; | |
2967 | 3754 |
1013 | 3755 } |
3756 | |
3757 if(last>=0){ | |
1689 | 3758 if(s->mb_intra) |
3759 s->dct_unquantize_intra(s, temp, 0, s->qscale); | |
3760 else | |
3761 s->dct_unquantize_inter(s, temp, 0, s->qscale); | |
1007 | 3762 } |
2967 | 3763 |
1092 | 3764 s->dsp.idct_add(bak, stride, temp); |
2967 | 3765 |
1708 | 3766 distoration= s->dsp.sse[1](NULL, bak, src1, stride, 8); |
1007 | 3767 |
1013 | 3768 return distoration + ((bits*s->qscale*s->qscale*109 + 64)>>7); |
1007 | 3769 } |
3770 | |
1708 | 3771 static int bit8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3772 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3773 const uint8_t *scantable= s->intra_scantable.permutated; |
3089 | 3774 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
1016 | 3775 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
1007 | 3776 int i, last, run, bits, level, start_i; |
3777 const int esc_length= s->ac_esc_length; | |
3778 uint8_t * length; | |
3779 uint8_t * last_length; | |
1708 | 3780 |
3781 assert(h==8); | |
2967 | 3782 |
1013 | 3783 s->dsp.diff_pixels(temp, src1, src2, stride); |
1007 | 3784 |
1013 | 3785 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3786 | |
3787 bits=0; | |
2967 | 3788 |
1007 | 3789 if (s->mb_intra) { |
2967 | 3790 start_i = 1; |
1007 | 3791 length = s->intra_ac_vlc_length; |
3792 last_length= s->intra_ac_vlc_last_length; | |
1013 | 3793 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma |
1007 | 3794 } else { |
3795 start_i = 0; | |
3796 length = s->inter_ac_vlc_length; | |
3797 last_length= s->inter_ac_vlc_last_length; | |
3798 } | |
2967 | 3799 |
1013 | 3800 if(last>=start_i){ |
1007 | 3801 run=0; |
3802 for(i=start_i; i<last; i++){ | |
3803 int j= scantable[i]; | |
3804 level= temp[j]; | |
2967 | 3805 |
1007 | 3806 if(level){ |
3807 level+=64; | |
3808 if((level&(~127)) == 0){ | |
3809 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3810 }else | |
3811 bits+= esc_length; | |
3812 run=0; | |
3813 }else | |
3814 run++; | |
3815 } | |
3816 i= scantable[last]; | |
2967 | 3817 |
1013 | 3818 level= temp[i] + 64; |
2967 | 3819 |
1013 | 3820 assert(level - 64); |
2967 | 3821 |
1007 | 3822 if((level&(~127)) == 0){ |
3823 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3824 }else | |
3825 bits+= esc_length; | |
3826 } | |
3827 | |
3828 return bits; | |
3829 } | |
3830 | |
1729 | 3831 static int vsad_intra16_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ |
3832 int score=0; | |
3833 int x,y; | |
2967 | 3834 |
1729 | 3835 for(y=1; y<h; y++){ |
3836 for(x=0; x<16; x+=4){ | |
4001 | 3837 score+= FFABS(s[x ] - s[x +stride]) + FFABS(s[x+1] - s[x+1+stride]) |
3838 +FFABS(s[x+2] - s[x+2+stride]) + FFABS(s[x+3] - s[x+3+stride]); | |
1729 | 3839 } |
3840 s+= stride; | |
3841 } | |
2967 | 3842 |
1729 | 3843 return score; |
3844 } | |
3845 | |
3846 static int vsad16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3847 int score=0; | |
3848 int x,y; | |
2967 | 3849 |
1729 | 3850 for(y=1; y<h; y++){ |
3851 for(x=0; x<16; x++){ | |
4001 | 3852 score+= FFABS(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); |
1729 | 3853 } |
3854 s1+= stride; | |
3855 s2+= stride; | |
3856 } | |
2967 | 3857 |
1729 | 3858 return score; |
3859 } | |
3860 | |
3861 #define SQ(a) ((a)*(a)) | |
3862 static int vsse_intra16_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ | |
3863 int score=0; | |
3864 int x,y; | |
2967 | 3865 |
1729 | 3866 for(y=1; y<h; y++){ |
3867 for(x=0; x<16; x+=4){ | |
2967 | 3868 score+= SQ(s[x ] - s[x +stride]) + SQ(s[x+1] - s[x+1+stride]) |
1729 | 3869 +SQ(s[x+2] - s[x+2+stride]) + SQ(s[x+3] - s[x+3+stride]); |
3870 } | |
3871 s+= stride; | |
3872 } | |
2967 | 3873 |
1729 | 3874 return score; |
3875 } | |
3876 | |
3877 static int vsse16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3878 int score=0; | |
3879 int x,y; | |
2967 | 3880 |
1729 | 3881 for(y=1; y<h; y++){ |
3882 for(x=0; x<16; x++){ | |
3883 score+= SQ(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); | |
3884 } | |
3885 s1+= stride; | |
3886 s2+= stride; | |
3887 } | |
2967 | 3888 |
1729 | 3889 return score; |
3890 } | |
3891 | |
5255 | 3892 static int ssd_int8_vs_int16_c(const int8_t *pix1, const int16_t *pix2, |
3893 int size){ | |
4749 | 3894 int score=0; |
3895 int i; | |
3896 for(i=0; i<size; i++) | |
3897 score += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]); | |
3898 return score; | |
3899 } | |
3900 | |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3901 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
|
3902 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
|
3903 WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c) |
3013 | 3904 #ifdef CONFIG_GPL |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3905 WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c) |
3013 | 3906 #endif |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3907 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
|
3908 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
|
3909 WRAPPER8_16_SQ(rd8x8_c, rd16_c) |
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3910 WRAPPER8_16_SQ(bit8x8_c, bit16_c) |
936 | 3911 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3912 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
|
3913 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3914 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3915 dst[i] *= src[i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3916 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3917 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3918 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
|
3919 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3920 src1 += len-1; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3921 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3922 dst[i] = src0[i] * src1[-i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3923 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3924 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3925 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
|
3926 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3927 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3928 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
|
3929 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3930 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3931 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
|
3932 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3933 for(i=0; i<len; i++) { |
6218 | 3934 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
|
3935 if(tmp & 0xf0000){ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3936 tmp = (0x43c0ffff - tmp)>>31; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3937 // 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
|
3938 // if(tmp > 0x43c0ffff) tmp = 0xFFFF; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3939 // else tmp = 0; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3940 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3941 dst[i] = tmp - 0x8000; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3942 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3943 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3944 |
5887 | 3945 #define W0 2048 |
3946 #define W1 2841 /* 2048*sqrt (2)*cos (1*pi/16) */ | |
3947 #define W2 2676 /* 2048*sqrt (2)*cos (2*pi/16) */ | |
3948 #define W3 2408 /* 2048*sqrt (2)*cos (3*pi/16) */ | |
3949 #define W4 2048 /* 2048*sqrt (2)*cos (4*pi/16) */ | |
3950 #define W5 1609 /* 2048*sqrt (2)*cos (5*pi/16) */ | |
3951 #define W6 1108 /* 2048*sqrt (2)*cos (6*pi/16) */ | |
3952 #define W7 565 /* 2048*sqrt (2)*cos (7*pi/16) */ | |
3953 | |
3954 static void wmv2_idct_row(short * b) | |
3955 { | |
3956 int s1,s2; | |
3957 int a0,a1,a2,a3,a4,a5,a6,a7; | |
3958 /*step 1*/ | |
3959 a1 = W1*b[1]+W7*b[7]; | |
3960 a7 = W7*b[1]-W1*b[7]; | |
3961 a5 = W5*b[5]+W3*b[3]; | |
3962 a3 = W3*b[5]-W5*b[3]; | |
3963 a2 = W2*b[2]+W6*b[6]; | |
3964 a6 = W6*b[2]-W2*b[6]; | |
3965 a0 = W0*b[0]+W0*b[4]; | |
3966 a4 = W0*b[0]-W0*b[4]; | |
3967 /*step 2*/ | |
3968 s1 = (181*(a1-a5+a7-a3)+128)>>8;//1,3,5,7, | |
3969 s2 = (181*(a1-a5-a7+a3)+128)>>8; | |
3970 /*step 3*/ | |
3971 b[0] = (a0+a2+a1+a5 + (1<<7))>>8; | |
3972 b[1] = (a4+a6 +s1 + (1<<7))>>8; | |
3973 b[2] = (a4-a6 +s2 + (1<<7))>>8; | |
3974 b[3] = (a0-a2+a7+a3 + (1<<7))>>8; | |
3975 b[4] = (a0-a2-a7-a3 + (1<<7))>>8; | |
3976 b[5] = (a4-a6 -s2 + (1<<7))>>8; | |
3977 b[6] = (a4+a6 -s1 + (1<<7))>>8; | |
3978 b[7] = (a0+a2-a1-a5 + (1<<7))>>8; | |
3979 } | |
3980 static void wmv2_idct_col(short * b) | |
3981 { | |
3982 int s1,s2; | |
3983 int a0,a1,a2,a3,a4,a5,a6,a7; | |
3984 /*step 1, with extended precision*/ | |
3985 a1 = (W1*b[8*1]+W7*b[8*7] + 4)>>3; | |
3986 a7 = (W7*b[8*1]-W1*b[8*7] + 4)>>3; | |
3987 a5 = (W5*b[8*5]+W3*b[8*3] + 4)>>3; | |
3988 a3 = (W3*b[8*5]-W5*b[8*3] + 4)>>3; | |
3989 a2 = (W2*b[8*2]+W6*b[8*6] + 4)>>3; | |
3990 a6 = (W6*b[8*2]-W2*b[8*6] + 4)>>3; | |
3991 a0 = (W0*b[8*0]+W0*b[8*4] )>>3; | |
3992 a4 = (W0*b[8*0]-W0*b[8*4] )>>3; | |
3993 /*step 2*/ | |
3994 s1 = (181*(a1-a5+a7-a3)+128)>>8; | |
3995 s2 = (181*(a1-a5-a7+a3)+128)>>8; | |
3996 /*step 3*/ | |
3997 b[8*0] = (a0+a2+a1+a5 + (1<<13))>>14; | |
3998 b[8*1] = (a4+a6 +s1 + (1<<13))>>14; | |
3999 b[8*2] = (a4-a6 +s2 + (1<<13))>>14; | |
4000 b[8*3] = (a0-a2+a7+a3 + (1<<13))>>14; | |
4001 | |
4002 b[8*4] = (a0-a2-a7-a3 + (1<<13))>>14; | |
4003 b[8*5] = (a4-a6 -s2 + (1<<13))>>14; | |
4004 b[8*6] = (a4+a6 -s1 + (1<<13))>>14; | |
4005 b[8*7] = (a0+a2-a1-a5 + (1<<13))>>14; | |
4006 } | |
4007 void ff_wmv2_idct_c(short * block){ | |
4008 int i; | |
4009 | |
4010 for(i=0;i<64;i+=8){ | |
4011 wmv2_idct_row(block+i); | |
4012 } | |
4013 for(i=0;i<8;i++){ | |
4014 wmv2_idct_col(block+i); | |
4015 } | |
4016 } | |
1092 | 4017 /* XXX: those functions should be suppressed ASAP when all IDCTs are |
4018 converted */ | |
5887 | 4019 static void ff_wmv2_idct_put_c(uint8_t *dest, int line_size, DCTELEM *block) |
4020 { | |
4021 ff_wmv2_idct_c(block); | |
4022 put_pixels_clamped_c(block, dest, line_size); | |
4023 } | |
4024 static void ff_wmv2_idct_add_c(uint8_t *dest, int line_size, DCTELEM *block) | |
4025 { | |
4026 ff_wmv2_idct_c(block); | |
4027 add_pixels_clamped_c(block, dest, line_size); | |
4028 } | |
1092 | 4029 static void ff_jref_idct_put(uint8_t *dest, int line_size, DCTELEM *block) |
4030 { | |
4031 j_rev_dct (block); | |
4032 put_pixels_clamped_c(block, dest, line_size); | |
4033 } | |
4034 static void ff_jref_idct_add(uint8_t *dest, int line_size, DCTELEM *block) | |
4035 { | |
4036 j_rev_dct (block); | |
4037 add_pixels_clamped_c(block, dest, line_size); | |
4038 } | |
4039 | |
2256 | 4040 static void ff_jref_idct4_put(uint8_t *dest, int line_size, DCTELEM *block) |
4041 { | |
4042 j_rev_dct4 (block); | |
4043 put_pixels_clamped4_c(block, dest, line_size); | |
4044 } | |
4045 static void ff_jref_idct4_add(uint8_t *dest, int line_size, DCTELEM *block) | |
4046 { | |
4047 j_rev_dct4 (block); | |
4048 add_pixels_clamped4_c(block, dest, line_size); | |
4049 } | |
4050 | |
2257 | 4051 static void ff_jref_idct2_put(uint8_t *dest, int line_size, DCTELEM *block) |
4052 { | |
4053 j_rev_dct2 (block); | |
4054 put_pixels_clamped2_c(block, dest, line_size); | |
4055 } | |
4056 static void ff_jref_idct2_add(uint8_t *dest, int line_size, DCTELEM *block) | |
4057 { | |
4058 j_rev_dct2 (block); | |
4059 add_pixels_clamped2_c(block, dest, line_size); | |
4060 } | |
4061 | |
2259 | 4062 static void ff_jref_idct1_put(uint8_t *dest, int line_size, DCTELEM *block) |
4063 { | |
4176 | 4064 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2259 | 4065 |
4066 dest[0] = cm[(block[0] + 4)>>3]; | |
4067 } | |
4068 static void ff_jref_idct1_add(uint8_t *dest, int line_size, DCTELEM *block) | |
4069 { | |
4176 | 4070 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2259 | 4071 |
4072 dest[0] = cm[dest[0] + ((block[0] + 4)>>3)]; | |
4073 } | |
4074 | |
5143 | 4075 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
|
4076 |
1201 | 4077 /* init static data */ |
4197 | 4078 void dsputil_static_init(void) |
0 | 4079 { |
751 | 4080 int i; |
0 | 4081 |
4176 | 4082 for(i=0;i<256;i++) ff_cropTbl[i + MAX_NEG_CROP] = i; |
1201 | 4083 for(i=0;i<MAX_NEG_CROP;i++) { |
4176 | 4084 ff_cropTbl[i] = 0; |
4085 ff_cropTbl[i + MAX_NEG_CROP + 256] = 255; | |
1201 | 4086 } |
2967 | 4087 |
1201 | 4088 for(i=0;i<512;i++) { |
4179 | 4089 ff_squareTbl[i] = (i - 256) * (i - 256); |
1201 | 4090 } |
2967 | 4091 |
4197 | 4092 for(i=0; i<64; i++) inv_zigzag_direct16[ff_zigzag_direct[i]]= i+1; |
1201 | 4093 } |
0 | 4094 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4095 int ff_check_alignment(void){ |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4096 static int did_fail=0; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4097 DECLARE_ALIGNED_16(int, aligned); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4098 |
5150 | 4099 if((long)&aligned & 15){ |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4100 if(!did_fail){ |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4101 #if defined(HAVE_MMX) || defined(HAVE_ALTIVEC) |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4102 av_log(NULL, AV_LOG_ERROR, |
4292 | 4103 "Compiler did not align stack variables. Libavcodec has been miscompiled\n" |
4104 "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
|
4105 "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
|
4106 "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
|
4107 #endif |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4108 did_fail=1; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4109 } |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4110 return -1; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4111 } |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4112 return 0; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4113 } |
861 | 4114 |
1201 | 4115 void dsputil_init(DSPContext* c, AVCodecContext *avctx) |
4116 { | |
4117 int i; | |
0 | 4118 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4119 ff_check_alignment(); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4120 |
1092 | 4121 #ifdef CONFIG_ENCODERS |
1567 | 4122 if(avctx->dct_algo==FF_DCT_FASTINT) { |
1092 | 4123 c->fdct = fdct_ifast; |
2979 | 4124 c->fdct248 = fdct_ifast248; |
2967 | 4125 } |
1567 | 4126 else if(avctx->dct_algo==FF_DCT_FAAN) { |
1557 | 4127 c->fdct = ff_faandct; |
2979 | 4128 c->fdct248 = ff_faandct248; |
2967 | 4129 } |
1567 | 4130 else { |
1092 | 4131 c->fdct = ff_jpeg_fdct_islow; //slow/accurate/default |
2979 | 4132 c->fdct248 = ff_fdct248_islow; |
1567 | 4133 } |
1092 | 4134 #endif //CONFIG_ENCODERS |
4135 | |
2256 | 4136 if(avctx->lowres==1){ |
5064 | 4137 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
|
4138 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
|
4139 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
|
4140 }else{ |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4141 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
|
4142 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
|
4143 } |
2256 | 4144 c->idct = j_rev_dct4; |
1092 | 4145 c->idct_permutation_type= FF_NO_IDCT_PERM; |
2257 | 4146 }else if(avctx->lowres==2){ |
4147 c->idct_put= ff_jref_idct2_put; | |
4148 c->idct_add= ff_jref_idct2_add; | |
4149 c->idct = j_rev_dct2; | |
4150 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2259 | 4151 }else if(avctx->lowres==3){ |
4152 c->idct_put= ff_jref_idct1_put; | |
4153 c->idct_add= ff_jref_idct1_add; | |
4154 c->idct = j_rev_dct1; | |
4155 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2256 | 4156 }else{ |
4157 if(avctx->idct_algo==FF_IDCT_INT){ | |
4158 c->idct_put= ff_jref_idct_put; | |
4159 c->idct_add= ff_jref_idct_add; | |
4160 c->idct = j_rev_dct; | |
4161 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM; | |
5007 | 4162 }else if((ENABLE_VP3_DECODER || ENABLE_VP5_DECODER || ENABLE_VP6_DECODER || ENABLE_THEORA_DECODER ) && |
4163 avctx->idct_algo==FF_IDCT_VP3){ | |
2693 | 4164 c->idct_put= ff_vp3_idct_put_c; |
4165 c->idct_add= ff_vp3_idct_add_c; | |
4166 c->idct = ff_vp3_idct_c; | |
4167 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
5887 | 4168 }else if(avctx->idct_algo==FF_IDCT_WMV2){ |
4169 c->idct_put= ff_wmv2_idct_put_c; | |
4170 c->idct_add= ff_wmv2_idct_add_c; | |
4171 c->idct = ff_wmv2_idct_c; | |
4172 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
6407 | 4173 }else if(avctx->idct_algo==FF_IDCT_FAAN){ |
4174 c->idct_put= ff_faanidct_put; | |
4175 c->idct_add= ff_faanidct_add; | |
4176 c->idct = ff_faanidct; | |
4177 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2256 | 4178 }else{ //accurate/default |
6001 | 4179 c->idct_put= ff_simple_idct_put; |
4180 c->idct_add= ff_simple_idct_add; | |
4181 c->idct = ff_simple_idct; | |
2256 | 4182 c->idct_permutation_type= FF_NO_IDCT_PERM; |
4183 } | |
1092 | 4184 } |
4185 | |
5064 | 4186 if (ENABLE_H264_DECODER) { |
5065 | 4187 c->h264_idct_add= ff_h264_idct_add_c; |
4188 c->h264_idct8_add= ff_h264_idct8_add_c; | |
4189 c->h264_idct_dc_add= ff_h264_idct_dc_add_c; | |
4190 c->h264_idct8_dc_add= ff_h264_idct8_dc_add_c; | |
5064 | 4191 } |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4192 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4193 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
|
4194 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
|
4195 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
|
4196 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
|
4197 c->add_pixels_clamped = add_pixels_clamped_c; |
2763 | 4198 c->add_pixels8 = add_pixels8_c; |
4199 c->add_pixels4 = add_pixels4_c; | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
4200 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
|
4201 c->gmc1 = gmc1_c; |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
4202 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
|
4203 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
|
4204 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
|
4205 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
|
4206 |
859 | 4207 /* TODO [0] 16 [1] 8 */ |
1708 | 4208 c->pix_abs[0][0] = pix_abs16_c; |
4209 c->pix_abs[0][1] = pix_abs16_x2_c; | |
4210 c->pix_abs[0][2] = pix_abs16_y2_c; | |
4211 c->pix_abs[0][3] = pix_abs16_xy2_c; | |
4212 c->pix_abs[1][0] = pix_abs8_c; | |
4213 c->pix_abs[1][1] = pix_abs8_x2_c; | |
4214 c->pix_abs[1][2] = pix_abs8_y2_c; | |
4215 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
|
4216 |
859 | 4217 #define dspfunc(PFX, IDX, NUM) \ |
4218 c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## NUM ## _c; \ | |
4219 c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## NUM ## _x2_c; \ | |
4220 c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## NUM ## _y2_c; \ | |
4221 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
|
4222 |
859 | 4223 dspfunc(put, 0, 16); |
4224 dspfunc(put_no_rnd, 0, 16); | |
4225 dspfunc(put, 1, 8); | |
4226 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
|
4227 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
|
4228 dspfunc(put, 3, 2); |
0 | 4229 |
859 | 4230 dspfunc(avg, 0, 16); |
4231 dspfunc(avg_no_rnd, 0, 16); | |
4232 dspfunc(avg, 1, 8); | |
4233 dspfunc(avg_no_rnd, 1, 8); | |
1319 | 4234 dspfunc(avg, 2, 4); |
4235 dspfunc(avg, 3, 2); | |
859 | 4236 #undef dspfunc |
857 | 4237 |
1864 | 4238 c->put_no_rnd_pixels_l2[0]= put_no_rnd_pixels16_l2_c; |
4239 c->put_no_rnd_pixels_l2[1]= put_no_rnd_pixels8_l2_c; | |
4240 | |
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
|
4241 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
|
4242 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
|
4243 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
|
4244 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
|
4245 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
|
4246 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
|
4247 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
|
4248 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
|
4249 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
|
4250 |
1319 | 4251 c->avg_tpel_pixels_tab[ 0] = avg_tpel_pixels_mc00_c; |
4252 c->avg_tpel_pixels_tab[ 1] = avg_tpel_pixels_mc10_c; | |
4253 c->avg_tpel_pixels_tab[ 2] = avg_tpel_pixels_mc20_c; | |
4254 c->avg_tpel_pixels_tab[ 4] = avg_tpel_pixels_mc01_c; | |
4255 c->avg_tpel_pixels_tab[ 5] = avg_tpel_pixels_mc11_c; | |
4256 c->avg_tpel_pixels_tab[ 6] = avg_tpel_pixels_mc21_c; | |
4257 c->avg_tpel_pixels_tab[ 8] = avg_tpel_pixels_mc02_c; | |
4258 c->avg_tpel_pixels_tab[ 9] = avg_tpel_pixels_mc12_c; | |
4259 c->avg_tpel_pixels_tab[10] = avg_tpel_pixels_mc22_c; | |
4260 | |
859 | 4261 #define dspfunc(PFX, IDX, NUM) \ |
4262 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_c; \ | |
4263 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_c; \ | |
4264 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_c; \ | |
4265 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_c; \ | |
4266 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_c; \ | |
4267 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_c; \ | |
4268 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_c; \ | |
4269 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_c; \ | |
4270 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_c; \ | |
4271 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_c; \ | |
4272 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_c; \ | |
4273 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_c; \ | |
4274 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_c; \ | |
4275 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_c; \ | |
4276 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_c; \ | |
4277 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_c | |
857 | 4278 |
859 | 4279 dspfunc(put_qpel, 0, 16); |
4280 dspfunc(put_no_rnd_qpel, 0, 16); | |
4281 | |
4282 dspfunc(avg_qpel, 0, 16); | |
4283 /* dspfunc(avg_no_rnd_qpel, 0, 16); */ | |
857 | 4284 |
859 | 4285 dspfunc(put_qpel, 1, 8); |
4286 dspfunc(put_no_rnd_qpel, 1, 8); | |
4287 | |
4288 dspfunc(avg_qpel, 1, 8); | |
4289 /* dspfunc(avg_no_rnd_qpel, 1, 8); */ | |
1168 | 4290 |
4291 dspfunc(put_h264_qpel, 0, 16); | |
4292 dspfunc(put_h264_qpel, 1, 8); | |
4293 dspfunc(put_h264_qpel, 2, 4); | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
4294 dspfunc(put_h264_qpel, 3, 2); |
1168 | 4295 dspfunc(avg_h264_qpel, 0, 16); |
4296 dspfunc(avg_h264_qpel, 1, 8); | |
4297 dspfunc(avg_h264_qpel, 2, 4); | |
4298 | |
859 | 4299 #undef dspfunc |
1168 | 4300 c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_c; |
4301 c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_c; | |
4302 c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_c; | |
4303 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_c; | |
4304 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_c; | |
4305 c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_c; | |
3663 | 4306 c->put_no_rnd_h264_chroma_pixels_tab[0]= put_no_rnd_h264_chroma_mc8_c; |
857 | 4307 |
2415 | 4308 c->weight_h264_pixels_tab[0]= weight_h264_pixels16x16_c; |
4309 c->weight_h264_pixels_tab[1]= weight_h264_pixels16x8_c; | |
4310 c->weight_h264_pixels_tab[2]= weight_h264_pixels8x16_c; | |
4311 c->weight_h264_pixels_tab[3]= weight_h264_pixels8x8_c; | |
4312 c->weight_h264_pixels_tab[4]= weight_h264_pixels8x4_c; | |
4313 c->weight_h264_pixels_tab[5]= weight_h264_pixels4x8_c; | |
4314 c->weight_h264_pixels_tab[6]= weight_h264_pixels4x4_c; | |
4315 c->weight_h264_pixels_tab[7]= weight_h264_pixels4x2_c; | |
4316 c->weight_h264_pixels_tab[8]= weight_h264_pixels2x4_c; | |
4317 c->weight_h264_pixels_tab[9]= weight_h264_pixels2x2_c; | |
4318 c->biweight_h264_pixels_tab[0]= biweight_h264_pixels16x16_c; | |
4319 c->biweight_h264_pixels_tab[1]= biweight_h264_pixels16x8_c; | |
4320 c->biweight_h264_pixels_tab[2]= biweight_h264_pixels8x16_c; | |
4321 c->biweight_h264_pixels_tab[3]= biweight_h264_pixels8x8_c; | |
4322 c->biweight_h264_pixels_tab[4]= biweight_h264_pixels8x4_c; | |
4323 c->biweight_h264_pixels_tab[5]= biweight_h264_pixels4x8_c; | |
4324 c->biweight_h264_pixels_tab[6]= biweight_h264_pixels4x4_c; | |
4325 c->biweight_h264_pixels_tab[7]= biweight_h264_pixels4x2_c; | |
4326 c->biweight_h264_pixels_tab[8]= biweight_h264_pixels2x4_c; | |
4327 c->biweight_h264_pixels_tab[9]= biweight_h264_pixels2x2_c; | |
4328 | |
6437 | 4329 c->draw_edges = draw_edges_c; |
4330 | |
3432 | 4331 #ifdef CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4332 ff_cavsdsp_init(c,avctx); |
3432 | 4333 #endif |
3526 | 4334 #if defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
4335 ff_vc1dsp_init(c,avctx); | |
4336 #endif | |
5887 | 4337 #if defined(CONFIG_WMV2_DECODER) || defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
4338 ff_intrax8dsp_init(c,avctx); | |
4339 #endif | |
4296 | 4340 #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
|
4341 ff_h264dspenc_init(c,avctx); |
4296 | 4342 #endif |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4343 |
936 | 4344 c->put_mspel_pixels_tab[0]= put_mspel8_mc00_c; |
4345 c->put_mspel_pixels_tab[1]= put_mspel8_mc10_c; | |
4346 c->put_mspel_pixels_tab[2]= put_mspel8_mc20_c; | |
4347 c->put_mspel_pixels_tab[3]= put_mspel8_mc30_c; | |
4348 c->put_mspel_pixels_tab[4]= put_mspel8_mc02_c; | |
4349 c->put_mspel_pixels_tab[5]= put_mspel8_mc12_c; | |
4350 c->put_mspel_pixels_tab[6]= put_mspel8_mc22_c; | |
4351 c->put_mspel_pixels_tab[7]= put_mspel8_mc32_c; | |
2967 | 4352 |
1708 | 4353 #define SET_CMP_FUNC(name) \ |
4354 c->name[0]= name ## 16_c;\ | |
4355 c->name[1]= name ## 8x8_c; | |
2967 | 4356 |
1708 | 4357 SET_CMP_FUNC(hadamard8_diff) |
1729 | 4358 c->hadamard8_diff[4]= hadamard8_intra16_c; |
1708 | 4359 SET_CMP_FUNC(dct_sad) |
2382 | 4360 SET_CMP_FUNC(dct_max) |
3013 | 4361 #ifdef CONFIG_GPL |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
4362 SET_CMP_FUNC(dct264_sad) |
3013 | 4363 #endif |
1708 | 4364 c->sad[0]= pix_abs16_c; |
4365 c->sad[1]= pix_abs8_c; | |
4366 c->sse[0]= sse16_c; | |
4367 c->sse[1]= sse8_c; | |
2184 | 4368 c->sse[2]= sse4_c; |
1708 | 4369 SET_CMP_FUNC(quant_psnr) |
4370 SET_CMP_FUNC(rd) | |
4371 SET_CMP_FUNC(bit) | |
1729 | 4372 c->vsad[0]= vsad16_c; |
4373 c->vsad[4]= vsad_intra16_c; | |
4374 c->vsse[0]= vsse16_c; | |
4375 c->vsse[4]= vsse_intra16_c; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4376 c->nsse[0]= nsse16_c; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4377 c->nsse[1]= nsse8_c; |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
4378 #ifdef CONFIG_SNOW_ENCODER |
2184 | 4379 c->w53[0]= w53_16_c; |
4380 c->w53[1]= w53_8_c; | |
4381 c->w97[0]= w97_16_c; | |
4382 c->w97[1]= w97_8_c; | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
4383 #endif |
2184 | 4384 |
4749 | 4385 c->ssd_int8_vs_int16 = ssd_int8_vs_int16_c; |
4386 | |
866 | 4387 c->add_bytes= add_bytes_c; |
6384 | 4388 c->add_bytes_l2= add_bytes_l2_c; |
866 | 4389 c->diff_bytes= diff_bytes_c; |
1527 | 4390 c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_c; |
1273 | 4391 c->bswap_buf= bswap_buf; |
6384 | 4392 #ifdef CONFIG_PNG_DECODER |
4393 c->add_png_paeth_prediction= ff_add_png_paeth_prediction; | |
4394 #endif | |
2633 | 4395 |
4396 c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_c; | |
4397 c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_c; | |
4398 c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_c; | |
4399 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
|
4400 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
|
4401 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
|
4402 c->h264_loop_filter_strength= NULL; |
2967 | 4403 |
5277
7b3fcb7c61ce
Avoid linking with h263.c functions when the relevant codecs
aurel
parents:
5256
diff
changeset
|
4404 if (ENABLE_ANY_H263) { |
5278 | 4405 c->h263_h_loop_filter= h263_h_loop_filter_c; |
4406 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
|
4407 } |
2967 | 4408 |
2045 | 4409 c->h261_loop_filter= h261_loop_filter_c; |
2967 | 4410 |
1784 | 4411 c->try_8x8basis= try_8x8basis_c; |
4412 c->add_8x8basis= add_8x8basis_c; | |
866 | 4413 |
4598
5111fceeb971
The Snow DEcoding routines should be under CONFIG_SNOW_DECODER.
diego
parents:
4594
diff
changeset
|
4414 #ifdef CONFIG_SNOW_DECODER |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4415 c->vertical_compose97i = ff_snow_vertical_compose97i; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4416 c->horizontal_compose97i = ff_snow_horizontal_compose97i; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4417 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
|
4418 #endif |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4419 |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4420 #ifdef CONFIG_VORBIS_DECODER |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4421 c->vorbis_inverse_coupling = vorbis_inverse_coupling; |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4422 #endif |
5737 | 4423 #ifdef CONFIG_FLAC_ENCODER |
4424 c->flac_compute_autocorr = ff_flac_compute_autocorr; | |
4425 #endif | |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4426 c->vector_fmul = vector_fmul_c; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4427 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
|
4428 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
|
4429 c->float_to_int16 = ff_float_to_int16_c; |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4430 |
3245 | 4431 c->shrink[0]= ff_img_copy_plane; |
4432 c->shrink[1]= ff_shrink22; | |
4433 c->shrink[2]= ff_shrink44; | |
4434 c->shrink[3]= ff_shrink88; | |
4435 | |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4436 c->prefetch= just_return; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4437 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4438 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
|
4439 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
|
4440 |
5146
5db8d038eb1f
use if() instead of #ifdef and ensure all possible optimisations are enabled
aurel
parents:
5143
diff
changeset
|
4441 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
|
4442 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
|
4443 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
|
4444 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
|
4445 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
|
4446 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
|
4447 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
|
4448 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
|
4449 if (ENABLE_BFIN) dsputil_init_bfin (c, avctx); |
1092 | 4450 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4451 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
|
4452 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
|
4453 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
|
4454 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
|
4455 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
|
4456 } |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4457 |
1092 | 4458 switch(c->idct_permutation_type){ |
4459 case FF_NO_IDCT_PERM: | |
4460 for(i=0; i<64; i++) | |
4461 c->idct_permutation[i]= i; | |
4462 break; | |
4463 case FF_LIBMPEG2_IDCT_PERM: | |
4464 for(i=0; i<64; i++) | |
4465 c->idct_permutation[i]= (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2); | |
4466 break; | |
4467 case FF_SIMPLE_IDCT_PERM: | |
4468 for(i=0; i<64; i++) | |
4469 c->idct_permutation[i]= simple_mmx_permutation[i]; | |
4470 break; | |
4471 case FF_TRANSPOSE_IDCT_PERM: | |
4472 for(i=0; i<64; i++) | |
4473 c->idct_permutation[i]= ((i&7)<<3) | (i>>3); | |
4474 break; | |
2696
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4475 case FF_PARTTRANS_IDCT_PERM: |
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4476 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
|
4477 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
|
4478 break; |
1092 | 4479 default: |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1571
diff
changeset
|
4480 av_log(avctx, AV_LOG_ERROR, "Internal error, IDCT permutation not set\n"); |
1092 | 4481 } |
0 | 4482 } |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
220
diff
changeset
|
4483 |