Mercurial > libavcodec.hg
annotate dsputil.c @ 6384:0a403ade8c81 libavcodec
simd and unroll png_filter_row
cycles per 1000 pixels on core2:
left: 9211->5170
top: 9283->2138
avg: 12215->7611
paeth: 64024->17360
overall rgb png decoding speed: +45%
overall greyscale png decoding speed: +6%
author | lorenm |
---|---|
date | Thu, 21 Feb 2008 07:10:46 +0000 |
parents | 2799f65a24de |
children | 40fbc878ce3f |
rev | line source |
---|---|
0 | 1 /* |
2 * DSP utils | |
429 | 3 * Copyright (c) 2000, 2001 Fabrice Bellard. |
1739
07a484280a82
copyright year update of the files i touched and remembered, things look annoyingly unmaintained otherwise
michael
parents:
1729
diff
changeset
|
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> |
0 | 5 * |
5214 | 6 * gmc & q-pel & 32/64 bit based MC by Michael Niedermayer <michaelni@gmx.at> |
7 * | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
8 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
9 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
10 * FFmpeg is free software; you can redistribute it and/or |
429 | 11 * modify it under the terms of the GNU Lesser General Public |
12 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
13 * version 2.1 of the License, or (at your option) any later version. |
0 | 14 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
15 * FFmpeg is distributed in the hope that it will be useful, |
0 | 16 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
429 | 17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
18 * Lesser General Public License for more details. | |
0 | 19 * |
429 | 20 * You should have received a copy of the GNU Lesser General Public |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
21 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3029
diff
changeset
|
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
0 | 23 */ |
2967 | 24 |
1106 | 25 /** |
26 * @file dsputil.c | |
27 * DSP utils | |
28 */ | |
2967 | 29 |
0 | 30 #include "avcodec.h" |
31 #include "dsputil.h" | |
936 | 32 #include "mpegvideo.h" |
1092 | 33 #include "simple_idct.h" |
1557 | 34 #include "faandct.h" |
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 |
1064 | 53 const uint8_t ff_zigzag_direct[64] = { |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
54 0, 1, 8, 16, 9, 2, 3, 10, |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
55 17, 24, 32, 25, 18, 11, 4, 5, |
34 | 56 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
|
57 27, 20, 13, 6, 7, 14, 21, 28, |
34 | 58 35, 42, 49, 56, 57, 50, 43, 36, |
59 29, 22, 15, 23, 30, 37, 44, 51, | |
60 58, 59, 52, 45, 38, 31, 39, 46, | |
61 53, 60, 61, 54, 47, 55, 62, 63 | |
62 }; | |
63 | |
1567 | 64 /* Specific zigzag scan for 248 idct. NOTE that unlike the |
65 specification, we interleave the fields */ | |
66 const uint8_t ff_zigzag248_direct[64] = { | |
67 0, 8, 1, 9, 16, 24, 2, 10, | |
68 17, 25, 32, 40, 48, 56, 33, 41, | |
69 18, 26, 3, 11, 4, 12, 19, 27, | |
70 34, 42, 49, 57, 50, 58, 35, 43, | |
71 20, 28, 5, 13, 6, 14, 21, 29, | |
72 36, 44, 51, 59, 52, 60, 37, 45, | |
73 22, 30, 7, 15, 23, 31, 38, 46, | |
74 53, 61, 54, 62, 39, 47, 55, 63, | |
75 }; | |
76 | |
220 | 77 /* not permutated inverse zigzag_direct + 1 for MMX quantizer */ |
4197 | 78 DECLARE_ALIGNED_8(uint16_t, inv_zigzag_direct16[64]) = {0, }; |
220 | 79 |
1064 | 80 const uint8_t ff_alternate_horizontal_scan[64] = { |
2967 | 81 0, 1, 2, 3, 8, 9, 16, 17, |
34 | 82 10, 11, 4, 5, 6, 7, 15, 14, |
2967 | 83 13, 12, 19, 18, 24, 25, 32, 33, |
34 | 84 26, 27, 20, 21, 22, 23, 28, 29, |
2967 | 85 30, 31, 34, 35, 40, 41, 48, 49, |
34 | 86 42, 43, 36, 37, 38, 39, 44, 45, |
2967 | 87 46, 47, 50, 51, 56, 57, 58, 59, |
34 | 88 52, 53, 54, 55, 60, 61, 62, 63, |
89 }; | |
90 | |
1064 | 91 const uint8_t ff_alternate_vertical_scan[64] = { |
2967 | 92 0, 8, 16, 24, 1, 9, 2, 10, |
34 | 93 17, 25, 32, 40, 48, 56, 57, 49, |
2967 | 94 41, 33, 26, 18, 3, 11, 4, 12, |
34 | 95 19, 27, 34, 42, 50, 58, 35, 43, |
2967 | 96 51, 59, 20, 28, 5, 13, 6, 14, |
34 | 97 21, 29, 36, 44, 52, 60, 37, 45, |
2967 | 98 53, 61, 22, 30, 7, 15, 23, 31, |
34 | 99 38, 46, 54, 62, 39, 47, 55, 63, |
100 }; | |
101 | |
220 | 102 /* a*inverse[b]>>32 == a/b for all 0<=a<=65536 && 2<=b<=255 */ |
4174 | 103 const uint32_t ff_inverse[256]={ |
2967 | 104 0, 4294967295U,2147483648U,1431655766, 1073741824, 858993460, 715827883, 613566757, |
105 536870912, 477218589, 429496730, 390451573, 357913942, 330382100, 306783379, 286331154, | |
106 268435456, 252645136, 238609295, 226050911, 214748365, 204522253, 195225787, 186737709, | |
107 178956971, 171798692, 165191050, 159072863, 153391690, 148102321, 143165577, 138547333, | |
108 134217728, 130150525, 126322568, 122713352, 119304648, 116080198, 113025456, 110127367, | |
109 107374183, 104755300, 102261127, 99882961, 97612894, 95443718, 93368855, 91382283, | |
110 89478486, 87652394, 85899346, 84215046, 82595525, 81037119, 79536432, 78090315, | |
111 76695845, 75350304, 74051161, 72796056, 71582789, 70409300, 69273667, 68174085, | |
112 67108864, 66076420, 65075263, 64103990, 63161284, 62245903, 61356676, 60492498, | |
113 59652324, 58835169, 58040099, 57266231, 56512728, 55778797, 55063684, 54366675, | |
114 53687092, 53024288, 52377650, 51746594, 51130564, 50529028, 49941481, 49367441, | |
115 48806447, 48258060, 47721859, 47197443, 46684428, 46182445, 45691142, 45210183, | |
116 44739243, 44278014, 43826197, 43383509, 42949673, 42524429, 42107523, 41698712, | |
117 41297763, 40904451, 40518560, 40139882, 39768216, 39403370, 39045158, 38693400, | |
118 38347923, 38008561, 37675152, 37347542, 37025581, 36709123, 36398028, 36092163, | |
119 35791395, 35495598, 35204650, 34918434, 34636834, 34359739, 34087043, 33818641, | |
120 33554432, 33294321, 33038210, 32786010, 32537632, 32292988, 32051995, 31814573, | |
121 31580642, 31350127, 31122952, 30899046, 30678338, 30460761, 30246249, 30034737, | |
122 29826162, 29620465, 29417585, 29217465, 29020050, 28825284, 28633116, 28443493, | |
123 28256364, 28071682, 27889399, 27709467, 27531842, 27356480, 27183338, 27012373, | |
124 26843546, 26676816, 26512144, 26349493, 26188825, 26030105, 25873297, 25718368, | |
125 25565282, 25414008, 25264514, 25116768, 24970741, 24826401, 24683721, 24542671, | |
126 24403224, 24265352, 24129030, 23994231, 23860930, 23729102, 23598722, 23469767, | |
127 23342214, 23216040, 23091223, 22967740, 22845571, 22724695, 22605092, 22486740, | |
128 22369622, 22253717, 22139007, 22025474, 21913099, 21801865, 21691755, 21582751, | |
129 21474837, 21367997, 21262215, 21157475, 21053762, 20951060, 20849356, 20748635, | |
130 20648882, 20550083, 20452226, 20355296, 20259280, 20164166, 20069941, 19976593, | |
131 19884108, 19792477, 19701685, 19611723, 19522579, 19434242, 19346700, 19259944, | |
132 19173962, 19088744, 19004281, 18920561, 18837576, 18755316, 18673771, 18592933, | |
133 18512791, 18433337, 18354562, 18276457, 18199014, 18122225, 18046082, 17970575, | |
134 17895698, 17821442, 17747799, 17674763, 17602325, 17530479, 17459217, 17388532, | |
220 | 135 17318417, 17248865, 17179870, 17111424, 17043522, 16976156, 16909321, 16843010, |
136 }; | |
137 | |
1092 | 138 /* Input permutation for the simple_idct_mmx */ |
139 static const uint8_t simple_mmx_permutation[64]={ | |
2979 | 140 0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, |
141 0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, | |
142 0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, | |
143 0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, | |
144 0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, | |
145 0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, | |
146 0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, | |
147 0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F, | |
1092 | 148 }; |
149 | |
1064 | 150 static int pix_sum_c(uint8_t * pix, int line_size) |
612 | 151 { |
152 int s, i, j; | |
153 | |
154 s = 0; | |
155 for (i = 0; i < 16; i++) { | |
2979 | 156 for (j = 0; j < 16; j += 8) { |
157 s += pix[0]; | |
158 s += pix[1]; | |
159 s += pix[2]; | |
160 s += pix[3]; | |
161 s += pix[4]; | |
162 s += pix[5]; | |
163 s += pix[6]; | |
164 s += pix[7]; | |
165 pix += 8; | |
166 } | |
167 pix += line_size - 16; | |
612 | 168 } |
169 return s; | |
170 } | |
171 | |
1064 | 172 static int pix_norm1_c(uint8_t * pix, int line_size) |
612 | 173 { |
174 int s, i, j; | |
4179 | 175 uint32_t *sq = ff_squareTbl + 256; |
612 | 176 |
177 s = 0; | |
178 for (i = 0; i < 16; i++) { | |
2979 | 179 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
|
180 #if 0 |
2979 | 181 s += sq[pix[0]]; |
182 s += sq[pix[1]]; | |
183 s += sq[pix[2]]; | |
184 s += sq[pix[3]]; | |
185 s += sq[pix[4]]; | |
186 s += sq[pix[5]]; | |
187 s += sq[pix[6]]; | |
188 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
|
189 #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
|
190 #if LONG_MAX > 2147483647 |
2979 | 191 register uint64_t x=*(uint64_t*)pix; |
192 s += sq[x&0xff]; | |
193 s += sq[(x>>8)&0xff]; | |
194 s += sq[(x>>16)&0xff]; | |
195 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
|
196 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
|
197 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
|
198 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
|
199 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
|
200 #else |
2979 | 201 register uint32_t x=*(uint32_t*)pix; |
202 s += sq[x&0xff]; | |
203 s += sq[(x>>8)&0xff]; | |
204 s += sq[(x>>16)&0xff]; | |
205 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
|
206 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
|
207 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
|
208 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
|
209 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
|
210 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
|
211 #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
|
212 #endif |
2979 | 213 pix += 8; |
214 } | |
215 pix += line_size - 16; | |
612 | 216 } |
217 return s; | |
218 } | |
219 | |
6241 | 220 static void bswap_buf(uint32_t *dst, const uint32_t *src, int w){ |
1273 | 221 int i; |
2967 | 222 |
1273 | 223 for(i=0; i+8<=w; i+=8){ |
224 dst[i+0]= bswap_32(src[i+0]); | |
225 dst[i+1]= bswap_32(src[i+1]); | |
226 dst[i+2]= bswap_32(src[i+2]); | |
227 dst[i+3]= bswap_32(src[i+3]); | |
228 dst[i+4]= bswap_32(src[i+4]); | |
229 dst[i+5]= bswap_32(src[i+5]); | |
230 dst[i+6]= bswap_32(src[i+6]); | |
231 dst[i+7]= bswap_32(src[i+7]); | |
232 } | |
233 for(;i<w; i++){ | |
234 dst[i+0]= bswap_32(src[i+0]); | |
235 } | |
236 } | |
612 | 237 |
2184 | 238 static int sse4_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
239 { | |
240 int s, i; | |
4179 | 241 uint32_t *sq = ff_squareTbl + 256; |
2184 | 242 |
243 s = 0; | |
244 for (i = 0; i < h; i++) { | |
245 s += sq[pix1[0] - pix2[0]]; | |
246 s += sq[pix1[1] - pix2[1]]; | |
247 s += sq[pix1[2] - pix2[2]]; | |
248 s += sq[pix1[3] - pix2[3]]; | |
249 pix1 += line_size; | |
250 pix2 += line_size; | |
251 } | |
252 return s; | |
253 } | |
254 | |
1708 | 255 static int sse8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
936 | 256 { |
257 int s, i; | |
4179 | 258 uint32_t *sq = ff_squareTbl + 256; |
936 | 259 |
260 s = 0; | |
1708 | 261 for (i = 0; i < h; i++) { |
936 | 262 s += sq[pix1[0] - pix2[0]]; |
263 s += sq[pix1[1] - pix2[1]]; | |
264 s += sq[pix1[2] - pix2[2]]; | |
265 s += sq[pix1[3] - pix2[3]]; | |
266 s += sq[pix1[4] - pix2[4]]; | |
267 s += sq[pix1[5] - pix2[5]]; | |
268 s += sq[pix1[6] - pix2[6]]; | |
269 s += sq[pix1[7] - pix2[7]]; | |
270 pix1 += line_size; | |
271 pix2 += line_size; | |
272 } | |
273 return s; | |
274 } | |
275 | |
1708 | 276 static int sse16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
884 | 277 { |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
278 int s, i; |
4179 | 279 uint32_t *sq = ff_squareTbl + 256; |
884 | 280 |
281 s = 0; | |
1708 | 282 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
|
283 s += sq[pix1[ 0] - pix2[ 0]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
284 s += sq[pix1[ 1] - pix2[ 1]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
285 s += sq[pix1[ 2] - pix2[ 2]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
286 s += sq[pix1[ 3] - pix2[ 3]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
287 s += sq[pix1[ 4] - pix2[ 4]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
288 s += sq[pix1[ 5] - pix2[ 5]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
289 s += sq[pix1[ 6] - pix2[ 6]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
290 s += sq[pix1[ 7] - pix2[ 7]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
291 s += sq[pix1[ 8] - pix2[ 8]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
292 s += sq[pix1[ 9] - pix2[ 9]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
293 s += sq[pix1[10] - pix2[10]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
294 s += sq[pix1[11] - pix2[11]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
295 s += sq[pix1[12] - pix2[12]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
296 s += sq[pix1[13] - pix2[13]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
297 s += sq[pix1[14] - pix2[14]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
298 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
|
299 |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
300 pix1 += line_size; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
301 pix2 += line_size; |
884 | 302 } |
303 return s; | |
304 } | |
305 | |
2184 | 306 |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
307 #ifdef CONFIG_SNOW_ENCODER //dwt is in snow.c |
2184 | 308 static inline int w_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int w, int h, int type){ |
309 int s, i, j; | |
310 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
|
311 int tmp[32*32]; |
2184 | 312 int level, ori; |
2967 | 313 static const int scale[2][2][4][4]={ |
2184 | 314 { |
315 { | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
316 // 9/7 8x8 dec=3 |
2184 | 317 {268, 239, 239, 213}, |
318 { 0, 224, 224, 152}, | |
319 { 0, 135, 135, 110}, | |
320 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
321 // 9/7 16x16 or 32x32 dec=4 |
2184 | 322 {344, 310, 310, 280}, |
323 { 0, 320, 320, 228}, | |
324 { 0, 175, 175, 136}, | |
325 { 0, 129, 129, 102}, | |
326 } | |
327 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
328 { |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
329 // 5/3 8x8 dec=3 |
2184 | 330 {275, 245, 245, 218}, |
331 { 0, 230, 230, 156}, | |
332 { 0, 138, 138, 113}, | |
333 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
334 // 5/3 16x16 or 32x32 dec=4 |
2184 | 335 {352, 317, 317, 286}, |
336 { 0, 328, 328, 233}, | |
337 { 0, 180, 180, 140}, | |
338 { 0, 132, 132, 105}, | |
339 } | |
340 } | |
341 }; | |
342 | |
343 for (i = 0; i < h; i++) { | |
344 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
|
345 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
|
346 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
|
347 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
|
348 tmp[32*i+j+3] = (pix1[j+3] - pix2[j+3])<<4; |
2184 | 349 } |
350 pix1 += line_size; | |
351 pix2 += line_size; | |
352 } | |
2639 | 353 |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
354 ff_spatial_dwt(tmp, w, h, 32, type, dec_count); |
2184 | 355 |
356 s=0; | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
357 assert(w==h); |
2184 | 358 for(level=0; level<dec_count; level++){ |
359 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
|
360 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
|
361 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
|
362 int stride= 32<<(dec_count-level); |
2184 | 363 int sy= (ori&2) ? stride>>1 : 0; |
2967 | 364 |
2184 | 365 for(i=0; i<size; i++){ |
366 for(j=0; j<size; j++){ | |
367 int v= tmp[sx + sy + i*stride + j] * scale[type][dec_count-3][level][ori]; | |
4001 | 368 s += FFABS(v); |
2184 | 369 } |
370 } | |
371 } | |
372 } | |
2967 | 373 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
|
374 return s>>9; |
2184 | 375 } |
376 | |
377 static int w53_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
378 return w_c(v, pix1, pix2, line_size, 8, h, 1); | |
379 } | |
380 | |
381 static int w97_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
382 return w_c(v, pix1, pix2, line_size, 8, h, 0); | |
383 } | |
384 | |
385 static int w53_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
386 return w_c(v, pix1, pix2, line_size, 16, h, 1); | |
387 } | |
388 | |
389 static int w97_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
390 return w_c(v, pix1, pix2, line_size, 16, h, 0); | |
391 } | |
392 | |
4197 | 393 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
|
394 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
|
395 } |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
396 |
4197 | 397 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
|
398 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
|
399 } |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
400 #endif |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
401 |
1064 | 402 static void get_pixels_c(DCTELEM *restrict block, const uint8_t *pixels, int line_size) |
0 | 403 { |
404 int i; | |
405 | |
406 /* read the pixels */ | |
407 for(i=0;i<8;i++) { | |
516 | 408 block[0] = pixels[0]; |
409 block[1] = pixels[1]; | |
410 block[2] = pixels[2]; | |
411 block[3] = pixels[3]; | |
412 block[4] = pixels[4]; | |
413 block[5] = pixels[5]; | |
414 block[6] = pixels[6]; | |
415 block[7] = pixels[7]; | |
416 pixels += line_size; | |
417 block += 8; | |
0 | 418 } |
419 } | |
420 | |
1064 | 421 static void diff_pixels_c(DCTELEM *restrict block, const uint8_t *s1, |
2979 | 422 const uint8_t *s2, int stride){ |
324 | 423 int i; |
424 | |
425 /* read the pixels */ | |
426 for(i=0;i<8;i++) { | |
516 | 427 block[0] = s1[0] - s2[0]; |
428 block[1] = s1[1] - s2[1]; | |
429 block[2] = s1[2] - s2[2]; | |
430 block[3] = s1[3] - s2[3]; | |
431 block[4] = s1[4] - s2[4]; | |
432 block[5] = s1[5] - s2[5]; | |
433 block[6] = s1[6] - s2[6]; | |
434 block[7] = s1[7] - s2[7]; | |
324 | 435 s1 += stride; |
436 s2 += stride; | |
516 | 437 block += 8; |
324 | 438 } |
439 } | |
440 | |
441 | |
1064 | 442 static void put_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 443 int line_size) |
0 | 444 { |
445 int i; | |
4176 | 446 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 447 |
0 | 448 /* read the pixels */ |
449 for(i=0;i<8;i++) { | |
516 | 450 pixels[0] = cm[block[0]]; |
451 pixels[1] = cm[block[1]]; | |
452 pixels[2] = cm[block[2]]; | |
453 pixels[3] = cm[block[3]]; | |
454 pixels[4] = cm[block[4]]; | |
455 pixels[5] = cm[block[5]]; | |
456 pixels[6] = cm[block[6]]; | |
457 pixels[7] = cm[block[7]]; | |
458 | |
459 pixels += line_size; | |
460 block += 8; | |
0 | 461 } |
462 } | |
463 | |
2256 | 464 static void put_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 465 int line_size) |
2256 | 466 { |
467 int i; | |
4176 | 468 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 469 |
2256 | 470 /* read the pixels */ |
471 for(i=0;i<4;i++) { | |
472 pixels[0] = cm[block[0]]; | |
473 pixels[1] = cm[block[1]]; | |
474 pixels[2] = cm[block[2]]; | |
475 pixels[3] = cm[block[3]]; | |
476 | |
477 pixels += line_size; | |
478 block += 8; | |
479 } | |
480 } | |
481 | |
2257 | 482 static void put_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 483 int line_size) |
2257 | 484 { |
485 int i; | |
4176 | 486 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 487 |
2257 | 488 /* read the pixels */ |
489 for(i=0;i<2;i++) { | |
490 pixels[0] = cm[block[0]]; | |
491 pixels[1] = cm[block[1]]; | |
492 | |
493 pixels += line_size; | |
494 block += 8; | |
495 } | |
496 } | |
497 | |
2967 | 498 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
|
499 uint8_t *restrict pixels, |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
500 int line_size) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
501 { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
502 int i, j; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
503 |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
504 for (i = 0; i < 8; i++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
505 for (j = 0; j < 8; j++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
506 if (*block < -128) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
507 *pixels = 0; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
508 else if (*block > 127) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
509 *pixels = 255; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
510 else |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
511 *pixels = (uint8_t)(*block + 128); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
512 block++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
513 pixels++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
514 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
515 pixels += (line_size - 8); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
516 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
517 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
518 |
1064 | 519 static void add_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
516 | 520 int line_size) |
0 | 521 { |
522 int i; | |
4176 | 523 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 524 |
0 | 525 /* read the pixels */ |
526 for(i=0;i<8;i++) { | |
516 | 527 pixels[0] = cm[pixels[0] + block[0]]; |
528 pixels[1] = cm[pixels[1] + block[1]]; | |
529 pixels[2] = cm[pixels[2] + block[2]]; | |
530 pixels[3] = cm[pixels[3] + block[3]]; | |
531 pixels[4] = cm[pixels[4] + block[4]]; | |
532 pixels[5] = cm[pixels[5] + block[5]]; | |
533 pixels[6] = cm[pixels[6] + block[6]]; | |
534 pixels[7] = cm[pixels[7] + block[7]]; | |
535 pixels += line_size; | |
536 block += 8; | |
0 | 537 } |
538 } | |
2256 | 539 |
540 static void add_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, | |
541 int line_size) | |
542 { | |
543 int i; | |
4176 | 544 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 545 |
2256 | 546 /* read the pixels */ |
547 for(i=0;i<4;i++) { | |
548 pixels[0] = cm[pixels[0] + block[0]]; | |
549 pixels[1] = cm[pixels[1] + block[1]]; | |
550 pixels[2] = cm[pixels[2] + block[2]]; | |
551 pixels[3] = cm[pixels[3] + block[3]]; | |
552 pixels += line_size; | |
553 block += 8; | |
554 } | |
555 } | |
2257 | 556 |
557 static void add_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, | |
558 int line_size) | |
559 { | |
560 int i; | |
4176 | 561 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 562 |
2257 | 563 /* read the pixels */ |
564 for(i=0;i<2;i++) { | |
565 pixels[0] = cm[pixels[0] + block[0]]; | |
566 pixels[1] = cm[pixels[1] + block[1]]; | |
567 pixels += line_size; | |
568 block += 8; | |
569 } | |
570 } | |
2763 | 571 |
572 static void add_pixels8_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
573 { | |
574 int i; | |
575 for(i=0;i<8;i++) { | |
576 pixels[0] += block[0]; | |
577 pixels[1] += block[1]; | |
578 pixels[2] += block[2]; | |
579 pixels[3] += block[3]; | |
580 pixels[4] += block[4]; | |
581 pixels[5] += block[5]; | |
582 pixels[6] += block[6]; | |
583 pixels[7] += block[7]; | |
584 pixels += line_size; | |
585 block += 8; | |
586 } | |
587 } | |
588 | |
589 static void add_pixels4_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
590 { | |
591 int i; | |
592 for(i=0;i<4;i++) { | |
593 pixels[0] += block[0]; | |
594 pixels[1] += block[1]; | |
595 pixels[2] += block[2]; | |
596 pixels[3] += block[3]; | |
597 pixels += line_size; | |
598 block += 4; | |
599 } | |
600 } | |
601 | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
602 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
|
603 { |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
604 int sum=0, i; |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
605 for(i=0; i<64; i++) |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
606 sum+= FFABS(block[i]); |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
607 return sum; |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
608 } |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
609 |
385 | 610 #if 0 |
611 | |
612 #define PIXOP2(OPNAME, OP) \ | |
651 | 613 static void OPNAME ## _pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 614 {\ |
615 int i;\ | |
616 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
|
617 OP(*((uint64_t*)block), AV_RN64(pixels));\ |
385 | 618 pixels+=line_size;\ |
619 block +=line_size;\ | |
620 }\ | |
621 }\ | |
622 \ | |
859 | 623 static void OPNAME ## _no_rnd_pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 624 {\ |
625 int i;\ | |
626 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
|
627 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
|
628 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 629 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
630 pixels+=line_size;\ | |
631 block +=line_size;\ | |
632 }\ | |
633 }\ | |
634 \ | |
859 | 635 static void OPNAME ## _pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 636 {\ |
637 int i;\ | |
638 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
|
639 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
|
640 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 641 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
642 pixels+=line_size;\ | |
643 block +=line_size;\ | |
644 }\ | |
645 }\ | |
646 \ | |
859 | 647 static void OPNAME ## _no_rnd_pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 648 {\ |
649 int i;\ | |
650 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
|
651 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
|
652 const uint64_t b= AV_RN64(pixels+line_size);\ |
385 | 653 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
654 pixels+=line_size;\ | |
655 block +=line_size;\ | |
656 }\ | |
657 }\ | |
658 \ | |
859 | 659 static void OPNAME ## _pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 660 {\ |
661 int i;\ | |
662 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
|
663 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
|
664 const uint64_t b= AV_RN64(pixels+line_size);\ |
385 | 665 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
666 pixels+=line_size;\ | |
667 block +=line_size;\ | |
668 }\ | |
669 }\ | |
670 \ | |
859 | 671 static void OPNAME ## _pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 672 {\ |
673 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
674 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
|
675 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 676 uint64_t l0= (a&0x0303030303030303ULL)\ |
677 + (b&0x0303030303030303ULL)\ | |
678 + 0x0202020202020202ULL;\ | |
679 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
680 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
681 uint64_t l1,h1;\ | |
682 \ | |
683 pixels+=line_size;\ | |
684 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
|
685 uint64_t a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
686 uint64_t b= AV_RN64(pixels+1);\ |
385 | 687 l1= (a&0x0303030303030303ULL)\ |
688 + (b&0x0303030303030303ULL);\ | |
689 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
690 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
691 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
692 pixels+=line_size;\ | |
693 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
694 a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
695 b= AV_RN64(pixels+1);\ |
385 | 696 l0= (a&0x0303030303030303ULL)\ |
697 + (b&0x0303030303030303ULL)\ | |
698 + 0x0202020202020202ULL;\ | |
699 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
700 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
701 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
702 pixels+=line_size;\ | |
703 block +=line_size;\ | |
704 }\ | |
705 }\ | |
706 \ | |
859 | 707 static void OPNAME ## _no_rnd_pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 708 {\ |
709 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
710 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
|
711 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 712 uint64_t l0= (a&0x0303030303030303ULL)\ |
713 + (b&0x0303030303030303ULL)\ | |
714 + 0x0101010101010101ULL;\ | |
715 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
716 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
717 uint64_t l1,h1;\ | |
718 \ | |
719 pixels+=line_size;\ | |
720 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
|
721 uint64_t a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
722 uint64_t b= AV_RN64(pixels+1);\ |
385 | 723 l1= (a&0x0303030303030303ULL)\ |
724 + (b&0x0303030303030303ULL);\ | |
725 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
726 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
727 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
728 pixels+=line_size;\ | |
729 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
730 a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
731 b= AV_RN64(pixels+1);\ |
385 | 732 l0= (a&0x0303030303030303ULL)\ |
733 + (b&0x0303030303030303ULL)\ | |
734 + 0x0101010101010101ULL;\ | |
735 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
736 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
737 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
738 pixels+=line_size;\ | |
739 block +=line_size;\ | |
740 }\ | |
741 }\ | |
742 \ | |
859 | 743 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels_c , 8)\ |
744 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels_x2_c , 8)\ | |
745 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels_y2_c , 8)\ | |
746 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels_xy2_c, 8)\ | |
747 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels_x2_c , 8)\ | |
748 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels_y2_c , 8)\ | |
749 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels_xy2_c, 8) | |
385 | 750 |
751 #define op_avg(a, b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEFEFEFEFEULL)>>1) ) | |
752 #else // 64 bit variant | |
753 | |
754 #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
|
755 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
|
756 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
|
757 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
|
758 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
|
759 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
|
760 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
|
761 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
762 }\ |
1168 | 763 static void OPNAME ## _pixels4_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
764 int i;\ | |
765 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
|
766 OP(*((uint32_t*)(block )), AV_RN32(pixels ));\ |
1168 | 767 pixels+=line_size;\ |
768 block +=line_size;\ | |
769 }\ | |
770 }\ | |
859 | 771 static void OPNAME ## _pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
385 | 772 int i;\ |
773 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
|
774 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
|
775 OP(*((uint32_t*)(block+4)), AV_RN32(pixels+4));\ |
385 | 776 pixels+=line_size;\ |
777 block +=line_size;\ | |
778 }\ | |
779 }\ | |
859 | 780 static inline void OPNAME ## _no_rnd_pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
781 OPNAME ## _pixels8_c(block, pixels, line_size, h);\ | |
651 | 782 }\ |
385 | 783 \ |
651 | 784 static inline void OPNAME ## _no_rnd_pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
785 int src_stride1, int src_stride2, int h){\ | |
385 | 786 int i;\ |
787 for(i=0; i<h; i++){\ | |
651 | 788 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
789 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
|
790 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 791 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
|
792 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
|
793 b= AV_RN32(&src2[i*src_stride2+4]);\ |
1264 | 794 OP(*((uint32_t*)&dst[i*dst_stride+4]), no_rnd_avg32(a, b));\ |
385 | 795 }\ |
796 }\ | |
797 \ | |
651 | 798 static inline void OPNAME ## _pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
799 int src_stride1, int src_stride2, int h){\ | |
385 | 800 int i;\ |
801 for(i=0; i<h; i++){\ | |
651 | 802 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
803 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
|
804 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 805 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
|
806 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
|
807 b= AV_RN32(&src2[i*src_stride2+4]);\ |
1264 | 808 OP(*((uint32_t*)&dst[i*dst_stride+4]), rnd_avg32(a, b));\ |
385 | 809 }\ |
810 }\ | |
811 \ | |
1168 | 812 static inline void OPNAME ## _pixels4_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
813 int src_stride1, int src_stride2, int h){\ | |
814 int i;\ | |
815 for(i=0; i<h; i++){\ | |
816 uint32_t a,b;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
817 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
|
818 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 819 OP(*((uint32_t*)&dst[i*dst_stride ]), rnd_avg32(a, b));\ |
1168 | 820 }\ |
821 }\ | |
822 \ | |
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
|
823 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
|
824 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
|
825 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
|
826 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
|
827 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
828 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
|
829 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
|
830 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
|
831 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
832 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
833 \ |
651 | 834 static inline void OPNAME ## _pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
835 int src_stride1, int src_stride2, int h){\ | |
836 OPNAME ## _pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
837 OPNAME ## _pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
838 }\ | |
839 \ | |
840 static inline void OPNAME ## _no_rnd_pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ | |
841 int src_stride1, int src_stride2, int h){\ | |
842 OPNAME ## _no_rnd_pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
843 OPNAME ## _no_rnd_pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
844 }\ | |
845 \ | |
859 | 846 static inline void OPNAME ## _no_rnd_pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 847 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
848 }\ | |
849 \ | |
859 | 850 static inline void OPNAME ## _pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 851 OPNAME ## _pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
852 }\ | |
853 \ | |
859 | 854 static inline void OPNAME ## _no_rnd_pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 855 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
856 }\ | |
857 \ | |
859 | 858 static inline void OPNAME ## _pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 859 OPNAME ## _pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
385 | 860 }\ |
861 \ | |
651 | 862 static inline void OPNAME ## _pixels8_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
863 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
864 int i;\ | |
865 for(i=0; i<h; i++){\ | |
866 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
|
867 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
|
868 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
|
869 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
|
870 d= AV_RN32(&src4[i*src_stride4]);\ |
651 | 871 l0= (a&0x03030303UL)\ |
872 + (b&0x03030303UL)\ | |
873 + 0x02020202UL;\ | |
874 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
875 + ((b&0xFCFCFCFCUL)>>2);\ | |
876 l1= (c&0x03030303UL)\ | |
877 + (d&0x03030303UL);\ | |
878 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
879 + ((d&0xFCFCFCFCUL)>>2);\ | |
880 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
|
881 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
|
882 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
|
883 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
|
884 d= AV_RN32(&src4[i*src_stride4+4]);\ |
651 | 885 l0= (a&0x03030303UL)\ |
886 + (b&0x03030303UL)\ | |
887 + 0x02020202UL;\ | |
888 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
889 + ((b&0xFCFCFCFCUL)>>2);\ | |
890 l1= (c&0x03030303UL)\ | |
891 + (d&0x03030303UL);\ | |
892 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
893 + ((d&0xFCFCFCFCUL)>>2);\ | |
894 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
895 }\ | |
896 }\ | |
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
|
897 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
898 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
|
899 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
|
900 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
901 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
902 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
|
903 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
|
904 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
905 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
906 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
|
907 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
|
908 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
909 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
910 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
|
911 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
|
912 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
913 \ |
651 | 914 static inline void OPNAME ## _no_rnd_pixels8_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
915 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
385 | 916 int i;\ |
917 for(i=0; i<h; i++){\ | |
651 | 918 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
|
919 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
|
920 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
|
921 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
|
922 d= AV_RN32(&src4[i*src_stride4]);\ |
651 | 923 l0= (a&0x03030303UL)\ |
924 + (b&0x03030303UL)\ | |
925 + 0x01010101UL;\ | |
926 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
927 + ((b&0xFCFCFCFCUL)>>2);\ | |
928 l1= (c&0x03030303UL)\ | |
929 + (d&0x03030303UL);\ | |
930 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
931 + ((d&0xFCFCFCFCUL)>>2);\ | |
932 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
|
933 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
|
934 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
|
935 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
|
936 d= AV_RN32(&src4[i*src_stride4+4]);\ |
651 | 937 l0= (a&0x03030303UL)\ |
938 + (b&0x03030303UL)\ | |
939 + 0x01010101UL;\ | |
940 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
941 + ((b&0xFCFCFCFCUL)>>2);\ | |
942 l1= (c&0x03030303UL)\ | |
943 + (d&0x03030303UL);\ | |
944 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
945 + ((d&0xFCFCFCFCUL)>>2);\ | |
946 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
385 | 947 }\ |
948 }\ | |
651 | 949 static inline void OPNAME ## _pixels16_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
950 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
951 OPNAME ## _pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
952 OPNAME ## _pixels8_l4(dst+8, src1+8, src2+8, src3+8, src4+8, dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
953 }\ | |
954 static inline void OPNAME ## _no_rnd_pixels16_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ | |
955 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
956 OPNAME ## _no_rnd_pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
957 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);\ | |
958 }\ | |
385 | 959 \ |
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
|
960 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
|
961 {\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
962 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
|
963 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
|
964 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
|
965 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
|
966 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
|
967 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
968 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
|
969 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
|
970 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
|
971 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
|
972 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
|
973 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
|
974 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
975 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
|
976 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
|
977 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
978 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
|
979 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
|
980 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
981 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
|
982 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
|
983 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
|
984 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
|
985 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
986 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
|
987 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
|
988 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
|
989 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
|
990 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
991 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
992 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
993 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
|
994 {\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
995 int i;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
996 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
|
997 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
|
998 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
|
999 + (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
|
1000 + 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
|
1001 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
|
1002 + ((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
|
1003 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
|
1004 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1005 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
|
1006 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
|
1007 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
|
1008 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
|
1009 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
|
1010 + (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
|
1011 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
|
1012 + ((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
|
1013 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
|
1014 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
|
1015 block +=line_size;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1016 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1017 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
|
1018 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
|
1019 + (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
|
1020 + 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
|
1021 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
|
1022 + ((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
|
1023 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
|
1024 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
|
1025 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
|
1026 }\ |
85b71f9f7450
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 }\ |
85b71f9f7450
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 \ |
859 | 1029 static inline void OPNAME ## _pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1030 {\ |
1031 int j;\ | |
1032 for(j=0; j<2; j++){\ | |
1033 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1034 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
|
1035 const uint32_t b= AV_RN32(pixels+1);\ |
385 | 1036 uint32_t l0= (a&0x03030303UL)\ |
1037 + (b&0x03030303UL)\ | |
1038 + 0x02020202UL;\ | |
1039 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1040 + ((b&0xFCFCFCFCUL)>>2);\ | |
1041 uint32_t l1,h1;\ | |
1042 \ | |
1043 pixels+=line_size;\ | |
1044 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
|
1045 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
|
1046 uint32_t b= AV_RN32(pixels+1);\ |
385 | 1047 l1= (a&0x03030303UL)\ |
1048 + (b&0x03030303UL);\ | |
1049 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1050 + ((b&0xFCFCFCFCUL)>>2);\ | |
1051 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1052 pixels+=line_size;\ | |
1053 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1054 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1055 b= AV_RN32(pixels+1);\ |
385 | 1056 l0= (a&0x03030303UL)\ |
1057 + (b&0x03030303UL)\ | |
1058 + 0x02020202UL;\ | |
1059 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1060 + ((b&0xFCFCFCFCUL)>>2);\ | |
1061 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1062 pixels+=line_size;\ | |
1063 block +=line_size;\ | |
1064 }\ | |
1065 pixels+=4-line_size*(h+1);\ | |
1066 block +=4-line_size*h;\ | |
1067 }\ | |
1068 }\ | |
1069 \ | |
859 | 1070 static inline void OPNAME ## _no_rnd_pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1071 {\ |
1072 int j;\ | |
1073 for(j=0; j<2; j++){\ | |
1074 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1075 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
|
1076 const uint32_t b= AV_RN32(pixels+1);\ |
385 | 1077 uint32_t l0= (a&0x03030303UL)\ |
1078 + (b&0x03030303UL)\ | |
1079 + 0x01010101UL;\ | |
1080 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1081 + ((b&0xFCFCFCFCUL)>>2);\ | |
1082 uint32_t l1,h1;\ | |
1083 \ | |
1084 pixels+=line_size;\ | |
1085 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
|
1086 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
|
1087 uint32_t b= AV_RN32(pixels+1);\ |
385 | 1088 l1= (a&0x03030303UL)\ |
1089 + (b&0x03030303UL);\ | |
1090 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1091 + ((b&0xFCFCFCFCUL)>>2);\ | |
1092 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1093 pixels+=line_size;\ | |
1094 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1095 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1096 b= AV_RN32(pixels+1);\ |
385 | 1097 l0= (a&0x03030303UL)\ |
1098 + (b&0x03030303UL)\ | |
1099 + 0x01010101UL;\ | |
1100 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1101 + ((b&0xFCFCFCFCUL)>>2);\ | |
1102 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1103 pixels+=line_size;\ | |
1104 block +=line_size;\ | |
1105 }\ | |
1106 pixels+=4-line_size*(h+1);\ | |
1107 block +=4-line_size*h;\ | |
1108 }\ | |
1109 }\ | |
1110 \ | |
859 | 1111 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels8_c , 8)\ |
1112 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels8_x2_c , 8)\ | |
1113 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels8_y2_c , 8)\ | |
1114 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels8_xy2_c, 8)\ | |
1115 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_c , OPNAME ## _pixels8_c , 8)\ | |
1116 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels8_x2_c , 8)\ | |
1117 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels8_y2_c , 8)\ | |
1118 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels8_xy2_c, 8)\ | |
651 | 1119 |
1264 | 1120 #define op_avg(a, b) a = rnd_avg32(a, b) |
385 | 1121 #endif |
1122 #define op_put(a, b) a = b | |
1123 | |
1124 PIXOP2(avg, op_avg) | |
1125 PIXOP2(put, op_put) | |
1126 #undef op_avg | |
1127 #undef op_put | |
1128 | |
0 | 1129 #define avg2(a,b) ((a+b+1)>>1) |
1130 #define avg4(a,b,c,d) ((a+b+c+d+2)>>2) | |
1131 | |
1864 | 1132 static void put_no_rnd_pixels16_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ |
1133 put_no_rnd_pixels16_l2(dst, a, b, stride, stride, stride, h); | |
1134 } | |
1135 | |
1136 static void put_no_rnd_pixels8_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ | |
1137 put_no_rnd_pixels8_l2(dst, a, b, stride, stride, stride, h); | |
1138 } | |
753 | 1139 |
1064 | 1140 static void gmc1_c(uint8_t *dst, uint8_t *src, int stride, int h, int x16, int y16, int rounder) |
255 | 1141 { |
1142 const int A=(16-x16)*(16-y16); | |
1143 const int B=( x16)*(16-y16); | |
1144 const int C=(16-x16)*( y16); | |
1145 const int D=( x16)*( y16); | |
1146 int i; | |
1147 | |
1148 for(i=0; i<h; i++) | |
1149 { | |
651 | 1150 dst[0]= (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + rounder)>>8; |
1151 dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8; | |
1152 dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8; | |
1153 dst[3]= (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + rounder)>>8; | |
1154 dst[4]= (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + rounder)>>8; | |
1155 dst[5]= (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + rounder)>>8; | |
1156 dst[6]= (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + rounder)>>8; | |
1157 dst[7]= (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + rounder)>>8; | |
1158 dst+= stride; | |
1159 src+= stride; | |
255 | 1160 } |
1161 } | |
1162 | |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
1163 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, |
753 | 1164 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height) |
1165 { | |
1166 int y, vx, vy; | |
1167 const int s= 1<<shift; | |
2967 | 1168 |
753 | 1169 width--; |
1170 height--; | |
1171 | |
1172 for(y=0; y<h; y++){ | |
1173 int x; | |
1174 | |
1175 vx= ox; | |
1176 vy= oy; | |
1177 for(x=0; x<8; x++){ //XXX FIXME optimize | |
1178 int src_x, src_y, frac_x, frac_y, index; | |
1179 | |
1180 src_x= vx>>16; | |
1181 src_y= vy>>16; | |
1182 frac_x= src_x&(s-1); | |
1183 frac_y= src_y&(s-1); | |
1184 src_x>>=shift; | |
1185 src_y>>=shift; | |
2967 | 1186 |
753 | 1187 if((unsigned)src_x < width){ |
1188 if((unsigned)src_y < height){ | |
1189 index= src_x + src_y*stride; | |
1190 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) | |
1191 + src[index +1]* frac_x )*(s-frac_y) | |
1192 + ( src[index+stride ]*(s-frac_x) | |
1193 + src[index+stride+1]* frac_x )* frac_y | |
1194 + r)>>(shift*2); | |
1195 }else{ | |
4594 | 1196 index= src_x + av_clip(src_y, 0, height)*stride; |
2967 | 1197 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) |
753 | 1198 + src[index +1]* frac_x )*s |
1199 + r)>>(shift*2); | |
1200 } | |
1201 }else{ | |
1202 if((unsigned)src_y < height){ | |
4594 | 1203 index= av_clip(src_x, 0, width) + src_y*stride; |
2967 | 1204 dst[y*stride + x]= ( ( src[index ]*(s-frac_y) |
753 | 1205 + src[index+stride ]* frac_y )*s |
1206 + r)>>(shift*2); | |
1207 }else{ | |
4594 | 1208 index= av_clip(src_x, 0, width) + av_clip(src_y, 0, height)*stride; |
753 | 1209 dst[y*stride + x]= src[index ]; |
1210 } | |
1211 } | |
2967 | 1212 |
753 | 1213 vx+= dxx; |
1214 vy+= dyx; | |
1215 } | |
1216 ox += dxy; | |
1217 oy += dyy; | |
1218 } | |
1219 } | |
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
|
1220 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1221 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
|
1222 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
|
1223 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
|
1224 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
|
1225 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
|
1226 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
|
1227 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1228 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1229 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1230 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
|
1231 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
|
1232 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
|
1233 for (j=0; j < width; j++) { |
2979 | 1234 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
|
1235 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1236 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
|
1237 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
|
1238 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1239 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1240 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1241 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
|
1242 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
|
1243 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
|
1244 for (j=0; j < width; j++) { |
2979 | 1245 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
|
1246 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1247 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
|
1248 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
|
1249 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1250 } |
2967 | 1251 |
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
|
1252 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
|
1253 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
|
1254 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
|
1255 for (j=0; j < width; j++) { |
2979 | 1256 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
|
1257 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1258 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
|
1259 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
|
1260 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1261 } |
2967 | 1262 |
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
|
1263 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
|
1264 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
|
1265 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
|
1266 for (j=0; j < width; j++) { |
2979 | 1267 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
|
1268 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1269 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
|
1270 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
|
1271 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1272 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1273 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1274 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
|
1275 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
|
1276 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
|
1277 for (j=0; j < width; j++) { |
2979 | 1278 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
|
1279 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1280 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
|
1281 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
|
1282 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1283 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1284 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1285 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
|
1286 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
|
1287 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
|
1288 for (j=0; j < width; j++) { |
2979 | 1289 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
|
1290 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1291 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
|
1292 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
|
1293 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1294 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1295 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1296 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
|
1297 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
|
1298 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
|
1299 for (j=0; j < width; j++) { |
2979 | 1300 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
|
1301 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1302 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
|
1303 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
|
1304 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1305 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1306 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1307 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
|
1308 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
|
1309 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
|
1310 for (j=0; j < width; j++) { |
2979 | 1311 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
|
1312 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1313 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
|
1314 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
|
1315 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1316 } |
1319 | 1317 |
1318 static inline void avg_tpel_pixels_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1319 switch(width){ | |
1320 case 2: avg_pixels2_c (dst, src, stride, height); break; | |
1321 case 4: avg_pixels4_c (dst, src, stride, height); break; | |
1322 case 8: avg_pixels8_c (dst, src, stride, height); break; | |
1323 case 16:avg_pixels16_c(dst, src, stride, height); break; | |
1324 } | |
1325 } | |
1326 | |
1327 static inline void avg_tpel_pixels_mc10_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1328 int i,j; | |
1329 for (i=0; i < height; i++) { | |
1330 for (j=0; j < width; j++) { | |
2979 | 1331 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1332 } |
1333 src += stride; | |
1334 dst += stride; | |
1335 } | |
1336 } | |
1337 | |
1338 static inline void avg_tpel_pixels_mc20_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1339 int i,j; | |
1340 for (i=0; i < height; i++) { | |
1341 for (j=0; j < width; j++) { | |
2979 | 1342 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1343 } |
1344 src += stride; | |
1345 dst += stride; | |
1346 } | |
1347 } | |
2967 | 1348 |
1319 | 1349 static inline void avg_tpel_pixels_mc01_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1350 int i,j; | |
1351 for (i=0; i < height; i++) { | |
1352 for (j=0; j < width; j++) { | |
2979 | 1353 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1354 } |
1355 src += stride; | |
1356 dst += stride; | |
1357 } | |
1358 } | |
2967 | 1359 |
1319 | 1360 static inline void avg_tpel_pixels_mc11_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1361 int i,j; | |
1362 for (i=0; i < height; i++) { | |
1363 for (j=0; j < width; j++) { | |
2979 | 1364 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 | 1365 } |
1366 src += stride; | |
1367 dst += stride; | |
1368 } | |
1369 } | |
1370 | |
1371 static inline void avg_tpel_pixels_mc12_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1372 int i,j; | |
1373 for (i=0; i < height; i++) { | |
1374 for (j=0; j < width; j++) { | |
2979 | 1375 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 | 1376 } |
1377 src += stride; | |
1378 dst += stride; | |
1379 } | |
1380 } | |
1381 | |
1382 static inline void avg_tpel_pixels_mc02_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1383 int i,j; | |
1384 for (i=0; i < height; i++) { | |
1385 for (j=0; j < width; j++) { | |
2979 | 1386 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1387 } |
1388 src += stride; | |
1389 dst += stride; | |
1390 } | |
1391 } | |
1392 | |
1393 static inline void avg_tpel_pixels_mc21_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1394 int i,j; | |
1395 for (i=0; i < height; i++) { | |
1396 for (j=0; j < width; j++) { | |
2979 | 1397 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 | 1398 } |
1399 src += stride; | |
1400 dst += stride; | |
1401 } | |
1402 } | |
1403 | |
1404 static inline void avg_tpel_pixels_mc22_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1405 int i,j; | |
1406 for (i=0; i < height; i++) { | |
1407 for (j=0; j < width; j++) { | |
2979 | 1408 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 | 1409 } |
1410 src += stride; | |
1411 dst += stride; | |
1412 } | |
1413 } | |
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
|
1414 #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
|
1415 #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
|
1416 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
|
1417 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
|
1418 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
|
1419 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
|
1420 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
|
1421 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
|
1422 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
|
1423 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
|
1424 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
|
1425 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
|
1426 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
|
1427 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
|
1428 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
|
1429 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
|
1430 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
|
1431 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
|
1432 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
|
1433 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
|
1434 #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
|
1435 |
1168 | 1436 #define H264_CHROMA_MC(OPNAME, OP)\ |
1437 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){\ | |
1438 const int A=(8-x)*(8-y);\ | |
1439 const int B=( x)*(8-y);\ | |
1440 const int C=(8-x)*( y);\ | |
1441 const int D=( x)*( y);\ | |
1442 int i;\ | |
1443 \ | |
1444 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1445 \ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1446 if(D){\ |
6054 | 1447 for(i=0; i<h; i++){\ |
6053 | 1448 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1449 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1450 dst+= stride;\ | |
1451 src+= stride;\ | |
1452 }\ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1453 }else{\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1454 const int E= B+C;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1455 const int step= C ? stride : 1;\ |
6054 | 1456 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
|
1457 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
|
1458 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
|
1459 dst+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1460 src+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1461 }\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1462 }\ |
1168 | 1463 }\ |
1464 \ | |
1465 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){\ | |
1466 const int A=(8-x)*(8-y);\ | |
1467 const int B=( x)*(8-y);\ | |
1468 const int C=(8-x)*( y);\ | |
1469 const int D=( x)*( y);\ | |
1470 int i;\ | |
1471 \ | |
1472 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1473 \ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1474 if(D){\ |
6054 | 1475 for(i=0; i<h; i++){\ |
6053 | 1476 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1477 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1478 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1479 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1480 dst+= stride;\ | |
1481 src+= stride;\ | |
1482 }\ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1483 }else{\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1484 const int E= B+C;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1485 const int step= C ? stride : 1;\ |
6054 | 1486 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
|
1487 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
|
1488 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
|
1489 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
|
1490 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
|
1491 dst+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1492 src+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1493 }\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1494 }\ |
1168 | 1495 }\ |
1496 \ | |
1497 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){\ | |
1498 const int A=(8-x)*(8-y);\ | |
1499 const int B=( x)*(8-y);\ | |
1500 const int C=(8-x)*( y);\ | |
1501 const int D=( x)*( y);\ | |
1502 int i;\ | |
1503 \ | |
1504 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1505 \ | |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1506 if(D){\ |
6054 | 1507 for(i=0; i<h; i++){\ |
6053 | 1508 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1509 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1510 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1511 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1512 OP(dst[4], (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5]));\ | |
1513 OP(dst[5], (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6]));\ | |
1514 OP(dst[6], (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7]));\ | |
1515 OP(dst[7], (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8]));\ | |
1516 dst+= stride;\ | |
1517 src+= stride;\ | |
1518 }\ | |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1519 }else{\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1520 const int E= B+C;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1521 const int step= C ? stride : 1;\ |
6054 | 1522 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
|
1523 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
|
1524 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
|
1525 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
|
1526 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
|
1527 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
|
1528 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
|
1529 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
|
1530 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
|
1531 dst+= stride;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1532 src+= stride;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1533 }\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1534 }\ |
1168 | 1535 } |
1536 | |
1537 #define op_avg(a, b) a = (((a)+(((b) + 32)>>6)+1)>>1) | |
1538 #define op_put(a, b) a = (((b) + 32)>>6) | |
1539 | |
1540 H264_CHROMA_MC(put_ , op_put) | |
1541 H264_CHROMA_MC(avg_ , op_avg) | |
1542 #undef op_avg | |
1543 #undef op_put | |
1544 | |
3663 | 1545 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){ |
1546 const int A=(8-x)*(8-y); | |
1547 const int B=( x)*(8-y); | |
1548 const int C=(8-x)*( y); | |
1549 const int D=( x)*( y); | |
1550 int i; | |
1551 | |
1552 assert(x<8 && y<8 && x>=0 && y>=0); | |
1553 | |
1554 for(i=0; i<h; i++) | |
1555 { | |
1556 dst[0] = (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6; | |
1557 dst[1] = (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6; | |
1558 dst[2] = (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6; | |
1559 dst[3] = (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6; | |
1560 dst[4] = (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6; | |
1561 dst[5] = (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6; | |
1562 dst[6] = (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6; | |
1563 dst[7] = (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6; | |
1564 dst+= stride; | |
1565 src+= stride; | |
1566 } | |
1567 } | |
1568 | |
651 | 1569 #define QPEL_MC(r, OPNAME, RND, OP) \ |
1064 | 1570 static void OPNAME ## mpeg4_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
4176 | 1571 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1572 int i;\ |
1573 for(i=0; i<h; i++)\ | |
1574 {\ | |
1575 OP(dst[0], (src[0]+src[1])*20 - (src[0]+src[2])*6 + (src[1]+src[3])*3 - (src[2]+src[4]));\ | |
1576 OP(dst[1], (src[1]+src[2])*20 - (src[0]+src[3])*6 + (src[0]+src[4])*3 - (src[1]+src[5]));\ | |
1577 OP(dst[2], (src[2]+src[3])*20 - (src[1]+src[4])*6 + (src[0]+src[5])*3 - (src[0]+src[6]));\ | |
1578 OP(dst[3], (src[3]+src[4])*20 - (src[2]+src[5])*6 + (src[1]+src[6])*3 - (src[0]+src[7]));\ | |
1579 OP(dst[4], (src[4]+src[5])*20 - (src[3]+src[6])*6 + (src[2]+src[7])*3 - (src[1]+src[8]));\ | |
1580 OP(dst[5], (src[5]+src[6])*20 - (src[4]+src[7])*6 + (src[3]+src[8])*3 - (src[2]+src[8]));\ | |
1581 OP(dst[6], (src[6]+src[7])*20 - (src[5]+src[8])*6 + (src[4]+src[8])*3 - (src[3]+src[7]));\ | |
1582 OP(dst[7], (src[7]+src[8])*20 - (src[6]+src[8])*6 + (src[5]+src[7])*3 - (src[4]+src[6]));\ | |
1583 dst+=dstStride;\ | |
1584 src+=srcStride;\ | |
1585 }\ | |
1586 }\ | |
1587 \ | |
1064 | 1588 static void OPNAME ## mpeg4_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
984 | 1589 const int w=8;\ |
4176 | 1590 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1591 int i;\ |
1592 for(i=0; i<w; i++)\ | |
1593 {\ | |
1594 const int src0= src[0*srcStride];\ | |
1595 const int src1= src[1*srcStride];\ | |
1596 const int src2= src[2*srcStride];\ | |
1597 const int src3= src[3*srcStride];\ | |
1598 const int src4= src[4*srcStride];\ | |
1599 const int src5= src[5*srcStride];\ | |
1600 const int src6= src[6*srcStride];\ | |
1601 const int src7= src[7*srcStride];\ | |
1602 const int src8= src[8*srcStride];\ | |
1603 OP(dst[0*dstStride], (src0+src1)*20 - (src0+src2)*6 + (src1+src3)*3 - (src2+src4));\ | |
1604 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*6 + (src0+src4)*3 - (src1+src5));\ | |
1605 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*6 + (src0+src5)*3 - (src0+src6));\ | |
1606 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*6 + (src1+src6)*3 - (src0+src7));\ | |
1607 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*6 + (src2+src7)*3 - (src1+src8));\ | |
1608 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*6 + (src3+src8)*3 - (src2+src8));\ | |
1609 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*6 + (src4+src8)*3 - (src3+src7));\ | |
1610 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src8)*6 + (src5+src7)*3 - (src4+src6));\ | |
1611 dst++;\ | |
1612 src++;\ | |
1613 }\ | |
1614 }\ | |
1615 \ | |
1064 | 1616 static void OPNAME ## mpeg4_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
4176 | 1617 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1618 int i;\ |
954 | 1619 \ |
651 | 1620 for(i=0; i<h; i++)\ |
1621 {\ | |
1622 OP(dst[ 0], (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]));\ | |
1623 OP(dst[ 1], (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]));\ | |
1624 OP(dst[ 2], (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]));\ | |
1625 OP(dst[ 3], (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]));\ | |
1626 OP(dst[ 4], (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]));\ | |
1627 OP(dst[ 5], (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]));\ | |
1628 OP(dst[ 6], (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]));\ | |
1629 OP(dst[ 7], (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]));\ | |
1630 OP(dst[ 8], (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]));\ | |
1631 OP(dst[ 9], (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]));\ | |
1632 OP(dst[10], (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]));\ | |
1633 OP(dst[11], (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]));\ | |
1634 OP(dst[12], (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]));\ | |
1635 OP(dst[13], (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]));\ | |
1636 OP(dst[14], (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]));\ | |
1637 OP(dst[15], (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]));\ | |
1638 dst+=dstStride;\ | |
1639 src+=srcStride;\ | |
1640 }\ | |
255 | 1641 }\ |
1642 \ | |
1064 | 1643 static void OPNAME ## mpeg4_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
4176 | 1644 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1645 int i;\ |
954 | 1646 const int w=16;\ |
651 | 1647 for(i=0; i<w; i++)\ |
1648 {\ | |
1649 const int src0= src[0*srcStride];\ | |
1650 const int src1= src[1*srcStride];\ | |
1651 const int src2= src[2*srcStride];\ | |
1652 const int src3= src[3*srcStride];\ | |
1653 const int src4= src[4*srcStride];\ | |
1654 const int src5= src[5*srcStride];\ | |
1655 const int src6= src[6*srcStride];\ | |
1656 const int src7= src[7*srcStride];\ | |
1657 const int src8= src[8*srcStride];\ | |
1658 const int src9= src[9*srcStride];\ | |
1659 const int src10= src[10*srcStride];\ | |
1660 const int src11= src[11*srcStride];\ | |
1661 const int src12= src[12*srcStride];\ | |
1662 const int src13= src[13*srcStride];\ | |
1663 const int src14= src[14*srcStride];\ | |
1664 const int src15= src[15*srcStride];\ | |
1665 const int src16= src[16*srcStride];\ | |
1666 OP(dst[ 0*dstStride], (src0 +src1 )*20 - (src0 +src2 )*6 + (src1 +src3 )*3 - (src2 +src4 ));\ | |
1667 OP(dst[ 1*dstStride], (src1 +src2 )*20 - (src0 +src3 )*6 + (src0 +src4 )*3 - (src1 +src5 ));\ | |
1668 OP(dst[ 2*dstStride], (src2 +src3 )*20 - (src1 +src4 )*6 + (src0 +src5 )*3 - (src0 +src6 ));\ | |
1669 OP(dst[ 3*dstStride], (src3 +src4 )*20 - (src2 +src5 )*6 + (src1 +src6 )*3 - (src0 +src7 ));\ | |
1670 OP(dst[ 4*dstStride], (src4 +src5 )*20 - (src3 +src6 )*6 + (src2 +src7 )*3 - (src1 +src8 ));\ | |
1671 OP(dst[ 5*dstStride], (src5 +src6 )*20 - (src4 +src7 )*6 + (src3 +src8 )*3 - (src2 +src9 ));\ | |
1672 OP(dst[ 6*dstStride], (src6 +src7 )*20 - (src5 +src8 )*6 + (src4 +src9 )*3 - (src3 +src10));\ | |
1673 OP(dst[ 7*dstStride], (src7 +src8 )*20 - (src6 +src9 )*6 + (src5 +src10)*3 - (src4 +src11));\ | |
1674 OP(dst[ 8*dstStride], (src8 +src9 )*20 - (src7 +src10)*6 + (src6 +src11)*3 - (src5 +src12));\ | |
1675 OP(dst[ 9*dstStride], (src9 +src10)*20 - (src8 +src11)*6 + (src7 +src12)*3 - (src6 +src13));\ | |
1676 OP(dst[10*dstStride], (src10+src11)*20 - (src9 +src12)*6 + (src8 +src13)*3 - (src7 +src14));\ | |
1677 OP(dst[11*dstStride], (src11+src12)*20 - (src10+src13)*6 + (src9 +src14)*3 - (src8 +src15));\ | |
1678 OP(dst[12*dstStride], (src12+src13)*20 - (src11+src14)*6 + (src10+src15)*3 - (src9 +src16));\ | |
1679 OP(dst[13*dstStride], (src13+src14)*20 - (src12+src15)*6 + (src11+src16)*3 - (src10+src16));\ | |
1680 OP(dst[14*dstStride], (src14+src15)*20 - (src13+src16)*6 + (src12+src16)*3 - (src11+src15));\ | |
1681 OP(dst[15*dstStride], (src15+src16)*20 - (src14+src16)*6 + (src13+src15)*3 - (src12+src14));\ | |
1682 dst++;\ | |
1683 src++;\ | |
1684 }\ | |
255 | 1685 }\ |
1686 \ | |
1064 | 1687 static void OPNAME ## qpel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1688 OPNAME ## pixels8_c(dst, src, stride, 8);\ |
255 | 1689 }\ |
1690 \ | |
1064 | 1691 static void OPNAME ## qpel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
1692 uint8_t half[64];\ | |
651 | 1693 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1694 OPNAME ## pixels8_l2(dst, src, half, stride, stride, 8, 8);\ | |
1695 }\ | |
1696 \ | |
1064 | 1697 static void OPNAME ## qpel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 1698 OPNAME ## mpeg4_qpel8_h_lowpass(dst, src, stride, stride, 8);\ |
255 | 1699 }\ |
1700 \ | |
1064 | 1701 static void OPNAME ## qpel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
1702 uint8_t half[64];\ | |
651 | 1703 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1704 OPNAME ## pixels8_l2(dst, src+1, half, stride, stride, 8, 8);\ | |
1705 }\ | |
1706 \ | |
1064 | 1707 static void OPNAME ## qpel8_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
1708 uint8_t full[16*9];\ | |
1709 uint8_t half[64];\ | |
651 | 1710 copy_block9(full, src, 16, stride, 9);\ |
984 | 1711 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1712 OPNAME ## pixels8_l2(dst, full, half, stride, 16, 8, 8);\ |
1713 }\ | |
1714 \ | |
1064 | 1715 static void OPNAME ## qpel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
1716 uint8_t full[16*9];\ | |
651 | 1717 copy_block9(full, src, 16, stride, 9);\ |
984 | 1718 OPNAME ## mpeg4_qpel8_v_lowpass(dst, full, stride, 16);\ |
255 | 1719 }\ |
1720 \ | |
1064 | 1721 static void OPNAME ## qpel8_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
1722 uint8_t full[16*9];\ | |
1723 uint8_t half[64];\ | |
651 | 1724 copy_block9(full, src, 16, stride, 9);\ |
984 | 1725 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1726 OPNAME ## pixels8_l2(dst, full+16, half, stride, 16, 8, 8);\ |
1727 }\ | |
1064 | 1728 void ff_ ## OPNAME ## qpel8_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1729 uint8_t full[16*9];\ | |
1730 uint8_t halfH[72];\ | |
1731 uint8_t halfV[64];\ | |
1732 uint8_t halfHV[64];\ | |
651 | 1733 copy_block9(full, src, 16, stride, 9);\ |
1734 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1735 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1736 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1737 OPNAME ## pixels8_l4(dst, full, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1738 }\ |
1064 | 1739 static void OPNAME ## qpel8_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
1740 uint8_t full[16*9];\ | |
1741 uint8_t halfH[72];\ | |
1742 uint8_t halfHV[64];\ | |
984 | 1743 copy_block9(full, src, 16, stride, 9);\ |
1744 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1745 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1746 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1747 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1748 }\ | |
1064 | 1749 void ff_ ## OPNAME ## qpel8_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1750 uint8_t full[16*9];\ | |
1751 uint8_t halfH[72];\ | |
1752 uint8_t halfV[64];\ | |
1753 uint8_t halfHV[64];\ | |
651 | 1754 copy_block9(full, src, 16, stride, 9);\ |
1755 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1756 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1757 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1758 OPNAME ## pixels8_l4(dst, full+1, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1759 }\ |
1064 | 1760 static void OPNAME ## qpel8_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
1761 uint8_t full[16*9];\ | |
1762 uint8_t halfH[72];\ | |
1763 uint8_t halfHV[64];\ | |
984 | 1764 copy_block9(full, src, 16, stride, 9);\ |
1765 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1766 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1767 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1768 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1769 }\ | |
1064 | 1770 void ff_ ## OPNAME ## qpel8_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1771 uint8_t full[16*9];\ | |
1772 uint8_t halfH[72];\ | |
1773 uint8_t halfV[64];\ | |
1774 uint8_t halfHV[64];\ | |
651 | 1775 copy_block9(full, src, 16, stride, 9);\ |
1776 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1777 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1778 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1779 OPNAME ## pixels8_l4(dst, full+16, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
1780 }\ | |
1064 | 1781 static void OPNAME ## qpel8_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
1782 uint8_t full[16*9];\ | |
1783 uint8_t halfH[72];\ | |
1784 uint8_t halfHV[64];\ | |
984 | 1785 copy_block9(full, src, 16, stride, 9);\ |
1786 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1787 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1788 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1789 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1790 }\ | |
1064 | 1791 void ff_ ## OPNAME ## qpel8_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1792 uint8_t full[16*9];\ | |
1793 uint8_t halfH[72];\ | |
1794 uint8_t halfV[64];\ | |
1795 uint8_t halfHV[64];\ | |
651 | 1796 copy_block9(full, src, 16, stride, 9);\ |
1797 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full , 8, 16, 9);\ | |
984 | 1798 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1799 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1800 OPNAME ## pixels8_l4(dst, full+17, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1801 }\ |
1064 | 1802 static void OPNAME ## qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
1803 uint8_t full[16*9];\ | |
1804 uint8_t halfH[72];\ | |
1805 uint8_t halfHV[64];\ | |
984 | 1806 copy_block9(full, src, 16, stride, 9);\ |
1807 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1808 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1809 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1810 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1811 }\ | |
1064 | 1812 static void OPNAME ## qpel8_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
1813 uint8_t halfH[72];\ | |
1814 uint8_t halfHV[64];\ | |
651 | 1815 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1816 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1817 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ |
1818 }\ | |
1064 | 1819 static void OPNAME ## qpel8_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
1820 uint8_t halfH[72];\ | |
1821 uint8_t halfHV[64];\ | |
651 | 1822 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1823 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1824 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ |
1825 }\ | |
1064 | 1826 void ff_ ## OPNAME ## qpel8_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1827 uint8_t full[16*9];\ | |
1828 uint8_t halfH[72];\ | |
1829 uint8_t halfV[64];\ | |
1830 uint8_t halfHV[64];\ | |
651 | 1831 copy_block9(full, src, 16, stride, 9);\ |
1832 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1833 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1834 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1835 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
255 | 1836 }\ |
1064 | 1837 static void OPNAME ## qpel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
1838 uint8_t full[16*9];\ | |
1839 uint8_t halfH[72];\ | |
984 | 1840 copy_block9(full, src, 16, stride, 9);\ |
1841 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1842 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1843 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1844 }\ | |
1064 | 1845 void ff_ ## OPNAME ## qpel8_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1846 uint8_t full[16*9];\ | |
1847 uint8_t halfH[72];\ | |
1848 uint8_t halfV[64];\ | |
1849 uint8_t halfHV[64];\ | |
651 | 1850 copy_block9(full, src, 16, stride, 9);\ |
1851 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1852 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1853 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1854 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
1855 }\ | |
1064 | 1856 static void OPNAME ## qpel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
1857 uint8_t full[16*9];\ | |
1858 uint8_t halfH[72];\ | |
984 | 1859 copy_block9(full, src, 16, stride, 9);\ |
1860 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1861 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1862 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1863 }\ | |
1064 | 1864 static void OPNAME ## qpel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
1865 uint8_t halfH[72];\ | |
651 | 1866 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1867 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ |
651 | 1868 }\ |
1064 | 1869 static void OPNAME ## qpel16_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1870 OPNAME ## pixels16_c(dst, src, stride, 16);\ |
255 | 1871 }\ |
651 | 1872 \ |
1064 | 1873 static void OPNAME ## qpel16_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
1874 uint8_t half[256];\ | |
651 | 1875 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
1876 OPNAME ## pixels16_l2(dst, src, half, stride, stride, 16, 16);\ | |
1877 }\ | |
1878 \ | |
1064 | 1879 static void OPNAME ## qpel16_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 1880 OPNAME ## mpeg4_qpel16_h_lowpass(dst, src, stride, stride, 16);\ |
1881 }\ | |
1882 \ | |
1064 | 1883 static void OPNAME ## qpel16_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
1884 uint8_t half[256];\ | |
651 | 1885 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
1886 OPNAME ## pixels16_l2(dst, src+1, half, stride, stride, 16, 16);\ | |
1887 }\ | |
1888 \ | |
1064 | 1889 static void OPNAME ## qpel16_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
1890 uint8_t full[24*17];\ | |
1891 uint8_t half[256];\ | |
651 | 1892 copy_block17(full, src, 24, stride, 17);\ |
954 | 1893 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 1894 OPNAME ## pixels16_l2(dst, full, half, stride, 24, 16, 16);\ |
255 | 1895 }\ |
651 | 1896 \ |
1064 | 1897 static void OPNAME ## qpel16_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
1898 uint8_t full[24*17];\ | |
651 | 1899 copy_block17(full, src, 24, stride, 17);\ |
954 | 1900 OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24);\ |
651 | 1901 }\ |
1902 \ | |
1064 | 1903 static void OPNAME ## qpel16_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
1904 uint8_t full[24*17];\ | |
1905 uint8_t half[256];\ | |
651 | 1906 copy_block17(full, src, 24, stride, 17);\ |
954 | 1907 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 1908 OPNAME ## pixels16_l2(dst, full+24, half, stride, 24, 16, 16);\ |
255 | 1909 }\ |
1064 | 1910 void ff_ ## OPNAME ## qpel16_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1911 uint8_t full[24*17];\ | |
1912 uint8_t halfH[272];\ | |
1913 uint8_t halfV[256];\ | |
1914 uint8_t halfHV[256];\ | |
651 | 1915 copy_block17(full, src, 24, stride, 17);\ |
1916 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1917 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
1918 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1919 OPNAME ## pixels16_l4(dst, full, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
1920 }\ | |
1064 | 1921 static void OPNAME ## qpel16_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
1922 uint8_t full[24*17];\ | |
1923 uint8_t halfH[272];\ | |
1924 uint8_t halfHV[256];\ | |
984 | 1925 copy_block17(full, src, 24, stride, 17);\ |
1926 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1927 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
1928 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1929 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
1930 }\ | |
1064 | 1931 void ff_ ## OPNAME ## qpel16_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1932 uint8_t full[24*17];\ | |
1933 uint8_t halfH[272];\ | |
1934 uint8_t halfV[256];\ | |
1935 uint8_t halfHV[256];\ | |
651 | 1936 copy_block17(full, src, 24, stride, 17);\ |
1937 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1938 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
1939 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1940 OPNAME ## pixels16_l4(dst, full+1, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
1941 }\ | |
1064 | 1942 static void OPNAME ## qpel16_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
1943 uint8_t full[24*17];\ | |
1944 uint8_t halfH[272];\ | |
1945 uint8_t halfHV[256];\ | |
984 | 1946 copy_block17(full, src, 24, stride, 17);\ |
1947 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1948 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
1949 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1950 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
1951 }\ | |
1064 | 1952 void ff_ ## OPNAME ## qpel16_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1953 uint8_t full[24*17];\ | |
1954 uint8_t halfH[272];\ | |
1955 uint8_t halfV[256];\ | |
1956 uint8_t halfHV[256];\ | |
651 | 1957 copy_block17(full, src, 24, stride, 17);\ |
1958 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1959 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
1960 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1961 OPNAME ## pixels16_l4(dst, full+24, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
255 | 1962 }\ |
1064 | 1963 static void OPNAME ## qpel16_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
1964 uint8_t full[24*17];\ | |
1965 uint8_t halfH[272];\ | |
1966 uint8_t halfHV[256];\ | |
984 | 1967 copy_block17(full, src, 24, stride, 17);\ |
1968 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1969 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
1970 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1971 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
1972 }\ | |
1064 | 1973 void ff_ ## OPNAME ## qpel16_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1974 uint8_t full[24*17];\ | |
1975 uint8_t halfH[272];\ | |
1976 uint8_t halfV[256];\ | |
1977 uint8_t halfHV[256];\ | |
651 | 1978 copy_block17(full, src, 24, stride, 17);\ |
1979 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full , 16, 24, 17);\ | |
954 | 1980 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
1981 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1982 OPNAME ## pixels16_l4(dst, full+25, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
1983 }\ | |
1064 | 1984 static void OPNAME ## qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
1985 uint8_t full[24*17];\ | |
1986 uint8_t halfH[272];\ | |
1987 uint8_t halfHV[256];\ | |
984 | 1988 copy_block17(full, src, 24, stride, 17);\ |
1989 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1990 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
1991 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1992 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
1993 }\ | |
1064 | 1994 static void OPNAME ## qpel16_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
1995 uint8_t halfH[272];\ | |
1996 uint8_t halfHV[256];\ | |
651 | 1997 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 1998 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 1999 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ |
255 | 2000 }\ |
1064 | 2001 static void OPNAME ## qpel16_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
2002 uint8_t halfH[272];\ | |
2003 uint8_t halfHV[256];\ | |
651 | 2004 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2005 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 2006 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ |
2007 }\ | |
1064 | 2008 void ff_ ## OPNAME ## qpel16_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2009 uint8_t full[24*17];\ | |
2010 uint8_t halfH[272];\ | |
2011 uint8_t halfV[256];\ | |
2012 uint8_t halfHV[256];\ | |
651 | 2013 copy_block17(full, src, 24, stride, 17);\ |
2014 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2015 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
2016 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2017 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
255 | 2018 }\ |
1064 | 2019 static void OPNAME ## qpel16_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
2020 uint8_t full[24*17];\ | |
2021 uint8_t halfH[272];\ | |
984 | 2022 copy_block17(full, src, 24, stride, 17);\ |
2023 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2024 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
2025 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2026 }\ | |
1064 | 2027 void ff_ ## OPNAME ## qpel16_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2028 uint8_t full[24*17];\ | |
2029 uint8_t halfH[272];\ | |
2030 uint8_t halfV[256];\ | |
2031 uint8_t halfHV[256];\ | |
651 | 2032 copy_block17(full, src, 24, stride, 17);\ |
2033 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2034 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
2035 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2036 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
2037 }\ | |
1064 | 2038 static void OPNAME ## qpel16_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
2039 uint8_t full[24*17];\ | |
2040 uint8_t halfH[272];\ | |
984 | 2041 copy_block17(full, src, 24, stride, 17);\ |
2042 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2043 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2044 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2045 }\ | |
1064 | 2046 static void OPNAME ## qpel16_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
2047 uint8_t halfH[272];\ | |
651 | 2048 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2049 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ |
859 | 2050 } |
255 | 2051 |
651 | 2052 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) |
2053 #define op_avg_no_rnd(a, b) a = (((a)+cm[((b) + 15)>>5])>>1) | |
2054 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2055 #define op_put_no_rnd(a, b) a = cm[((b) + 15)>>5] | |
2056 | |
2057 QPEL_MC(0, put_ , _ , op_put) | |
2058 QPEL_MC(1, put_no_rnd_, _no_rnd_, op_put_no_rnd) | |
2059 QPEL_MC(0, avg_ , _ , op_avg) | |
2060 //QPEL_MC(1, avg_no_rnd , _ , op_avg) | |
2061 #undef op_avg | |
2062 #undef op_avg_no_rnd | |
2063 #undef op_put | |
2064 #undef op_put_no_rnd | |
255 | 2065 |
1168 | 2066 #if 1 |
2067 #define H264_LOWPASS(OPNAME, OP, OP2) \ | |
5151 | 2068 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
|
2069 const int h=2;\ |
4176 | 2070 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
|
2071 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2072 for(i=0; i<h; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2073 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2074 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
|
2075 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
|
2076 dst+=dstStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2077 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2078 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2079 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2080 \ |
5151 | 2081 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
|
2082 const int w=2;\ |
4176 | 2083 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
|
2084 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2085 for(i=0; i<w; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2086 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2087 const int srcB= src[-2*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2088 const int srcA= src[-1*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2089 const int src0= src[0 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2090 const int src1= src[1 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2091 const int src2= src[2 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2092 const int src3= src[3 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2093 const int src4= src[4 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2094 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
|
2095 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
|
2096 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2097 src++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2098 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2099 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2100 \ |
5151 | 2101 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
|
2102 const int h=2;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2103 const int w=2;\ |
4176 | 2104 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
|
2105 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2106 src -= 2*srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2107 for(i=0; i<h+5; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2108 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2109 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
|
2110 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
|
2111 tmp+=tmpStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2112 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2113 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2114 tmp -= tmpStride*(h+5-2);\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2115 for(i=0; i<w; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2116 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2117 const int tmpB= tmp[-2*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2118 const int tmpA= tmp[-1*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2119 const int tmp0= tmp[0 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2120 const int tmp1= tmp[1 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2121 const int tmp2= tmp[2 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2122 const int tmp3= tmp[3 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2123 const int tmp4= tmp[4 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2124 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
|
2125 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
|
2126 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2127 tmp++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2128 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2129 }\ |
1168 | 2130 static void OPNAME ## h264_qpel4_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
2131 const int h=4;\ | |
4176 | 2132 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2133 int i;\ |
2134 for(i=0; i<h; i++)\ | |
2135 {\ | |
2136 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]));\ | |
2137 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]));\ | |
2138 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]));\ | |
2139 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]));\ | |
2140 dst+=dstStride;\ | |
2141 src+=srcStride;\ | |
2142 }\ | |
2143 }\ | |
2144 \ | |
2145 static void OPNAME ## h264_qpel4_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2146 const int w=4;\ | |
4176 | 2147 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2148 int i;\ |
2149 for(i=0; i<w; i++)\ | |
2150 {\ | |
2151 const int srcB= src[-2*srcStride];\ | |
2152 const int srcA= src[-1*srcStride];\ | |
2153 const int src0= src[0 *srcStride];\ | |
2154 const int src1= src[1 *srcStride];\ | |
2155 const int src2= src[2 *srcStride];\ | |
2156 const int src3= src[3 *srcStride];\ | |
2157 const int src4= src[4 *srcStride];\ | |
2158 const int src5= src[5 *srcStride];\ | |
2159 const int src6= src[6 *srcStride];\ | |
2160 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2161 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2162 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2163 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2164 dst++;\ | |
2165 src++;\ | |
2166 }\ | |
2167 }\ | |
2168 \ | |
2169 static void OPNAME ## h264_qpel4_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2170 const int h=4;\ | |
2171 const int w=4;\ | |
4176 | 2172 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2173 int i;\ |
2174 src -= 2*srcStride;\ | |
2175 for(i=0; i<h+5; i++)\ | |
2176 {\ | |
2177 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]);\ | |
2178 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]);\ | |
2179 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]);\ | |
2180 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]);\ | |
2181 tmp+=tmpStride;\ | |
2182 src+=srcStride;\ | |
2183 }\ | |
2184 tmp -= tmpStride*(h+5-2);\ | |
2185 for(i=0; i<w; i++)\ | |
2186 {\ | |
2187 const int tmpB= tmp[-2*tmpStride];\ | |
2188 const int tmpA= tmp[-1*tmpStride];\ | |
2189 const int tmp0= tmp[0 *tmpStride];\ | |
2190 const int tmp1= tmp[1 *tmpStride];\ | |
2191 const int tmp2= tmp[2 *tmpStride];\ | |
2192 const int tmp3= tmp[3 *tmpStride];\ | |
2193 const int tmp4= tmp[4 *tmpStride];\ | |
2194 const int tmp5= tmp[5 *tmpStride];\ | |
2195 const int tmp6= tmp[6 *tmpStride];\ | |
2196 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2197 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2198 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2199 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2200 dst++;\ | |
2201 tmp++;\ | |
2202 }\ | |
2203 }\ | |
2204 \ | |
2205 static void OPNAME ## h264_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2206 const int h=8;\ | |
4176 | 2207 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2208 int i;\ |
2209 for(i=0; i<h; i++)\ | |
2210 {\ | |
2211 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]));\ | |
2212 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]));\ | |
2213 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]));\ | |
2214 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]));\ | |
2215 OP(dst[4], (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]));\ | |
2216 OP(dst[5], (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]));\ | |
2217 OP(dst[6], (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]));\ | |
2218 OP(dst[7], (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]));\ | |
2219 dst+=dstStride;\ | |
2220 src+=srcStride;\ | |
2221 }\ | |
2222 }\ | |
2223 \ | |
2224 static void OPNAME ## h264_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2225 const int w=8;\ | |
4176 | 2226 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2227 int i;\ |
2228 for(i=0; i<w; i++)\ | |
2229 {\ | |
2230 const int srcB= src[-2*srcStride];\ | |
2231 const int srcA= src[-1*srcStride];\ | |
2232 const int src0= src[0 *srcStride];\ | |
2233 const int src1= src[1 *srcStride];\ | |
2234 const int src2= src[2 *srcStride];\ | |
2235 const int src3= src[3 *srcStride];\ | |
2236 const int src4= src[4 *srcStride];\ | |
2237 const int src5= src[5 *srcStride];\ | |
2238 const int src6= src[6 *srcStride];\ | |
2239 const int src7= src[7 *srcStride];\ | |
2240 const int src8= src[8 *srcStride];\ | |
2241 const int src9= src[9 *srcStride];\ | |
2242 const int src10=src[10*srcStride];\ | |
2243 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2244 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2245 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2246 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2247 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*5 + (src2+src7));\ | |
2248 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*5 + (src3+src8));\ | |
2249 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*5 + (src4+src9));\ | |
2250 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src9)*5 + (src5+src10));\ | |
2251 dst++;\ | |
2252 src++;\ | |
2253 }\ | |
2254 }\ | |
2255 \ | |
2256 static void OPNAME ## h264_qpel8_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2257 const int h=8;\ | |
2258 const int w=8;\ | |
4176 | 2259 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2260 int i;\ |
2261 src -= 2*srcStride;\ | |
2262 for(i=0; i<h+5; i++)\ | |
2263 {\ | |
2264 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]);\ | |
2265 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]);\ | |
2266 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]);\ | |
2267 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]);\ | |
2268 tmp[4]= (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]);\ | |
2269 tmp[5]= (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]);\ | |
2270 tmp[6]= (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]);\ | |
2271 tmp[7]= (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]);\ | |
2272 tmp+=tmpStride;\ | |
2273 src+=srcStride;\ | |
2274 }\ | |
2275 tmp -= tmpStride*(h+5-2);\ | |
2276 for(i=0; i<w; i++)\ | |
2277 {\ | |
2278 const int tmpB= tmp[-2*tmpStride];\ | |
2279 const int tmpA= tmp[-1*tmpStride];\ | |
2280 const int tmp0= tmp[0 *tmpStride];\ | |
2281 const int tmp1= tmp[1 *tmpStride];\ | |
2282 const int tmp2= tmp[2 *tmpStride];\ | |
2283 const int tmp3= tmp[3 *tmpStride];\ | |
2284 const int tmp4= tmp[4 *tmpStride];\ | |
2285 const int tmp5= tmp[5 *tmpStride];\ | |
2286 const int tmp6= tmp[6 *tmpStride];\ | |
2287 const int tmp7= tmp[7 *tmpStride];\ | |
2288 const int tmp8= tmp[8 *tmpStride];\ | |
2289 const int tmp9= tmp[9 *tmpStride];\ | |
2290 const int tmp10=tmp[10*tmpStride];\ | |
2291 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2292 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2293 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2294 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2295 OP2(dst[4*dstStride], (tmp4+tmp5)*20 - (tmp3+tmp6)*5 + (tmp2+tmp7));\ | |
2296 OP2(dst[5*dstStride], (tmp5+tmp6)*20 - (tmp4+tmp7)*5 + (tmp3+tmp8));\ | |
2297 OP2(dst[6*dstStride], (tmp6+tmp7)*20 - (tmp5+tmp8)*5 + (tmp4+tmp9));\ | |
2298 OP2(dst[7*dstStride], (tmp7+tmp8)*20 - (tmp6+tmp9)*5 + (tmp5+tmp10));\ | |
2299 dst++;\ | |
2300 tmp++;\ | |
2301 }\ | |
2302 }\ | |
2303 \ | |
2304 static void OPNAME ## h264_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2305 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2306 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2307 src += 8*srcStride;\ | |
2308 dst += 8*dstStride;\ | |
2309 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2310 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2311 }\ | |
2312 \ | |
2313 static void OPNAME ## h264_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2314 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2315 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2316 src += 8*srcStride;\ | |
2317 dst += 8*dstStride;\ | |
2318 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2319 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2320 }\ | |
2321 \ | |
2322 static void OPNAME ## h264_qpel16_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2323 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2324 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2325 src += 8*srcStride;\ | |
2326 dst += 8*dstStride;\ | |
2327 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2328 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2329 }\ | |
2330 | |
2331 #define H264_MC(OPNAME, SIZE) \ | |
2332 static void OPNAME ## h264_qpel ## SIZE ## _mc00_c (uint8_t *dst, uint8_t *src, int stride){\ | |
2333 OPNAME ## pixels ## SIZE ## _c(dst, src, stride, SIZE);\ | |
2334 }\ | |
2335 \ | |
2336 static void OPNAME ## h264_qpel ## SIZE ## _mc10_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2337 uint8_t half[SIZE*SIZE];\ | |
2338 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2339 OPNAME ## pixels ## SIZE ## _l2(dst, src, half, stride, stride, SIZE, SIZE);\ | |
2340 }\ | |
2341 \ | |
2342 static void OPNAME ## h264_qpel ## SIZE ## _mc20_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2343 OPNAME ## h264_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride);\ | |
2344 }\ | |
2345 \ | |
2346 static void OPNAME ## h264_qpel ## SIZE ## _mc30_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2347 uint8_t half[SIZE*SIZE];\ | |
2348 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2349 OPNAME ## pixels ## SIZE ## _l2(dst, src+1, half, stride, stride, SIZE, SIZE);\ | |
2350 }\ | |
2351 \ | |
2352 static void OPNAME ## h264_qpel ## SIZE ## _mc01_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2353 uint8_t full[SIZE*(SIZE+5)];\ | |
2354 uint8_t * const full_mid= full + SIZE*2;\ | |
2355 uint8_t half[SIZE*SIZE];\ | |
2356 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2357 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2358 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid, half, stride, SIZE, SIZE, SIZE);\ | |
2359 }\ | |
2360 \ | |
2361 static void OPNAME ## h264_qpel ## SIZE ## _mc02_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2362 uint8_t full[SIZE*(SIZE+5)];\ | |
2363 uint8_t * const full_mid= full + SIZE*2;\ | |
2364 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2365 OPNAME ## h264_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE);\ | |
2366 }\ | |
2367 \ | |
2368 static void OPNAME ## h264_qpel ## SIZE ## _mc03_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2369 uint8_t full[SIZE*(SIZE+5)];\ | |
2370 uint8_t * const full_mid= full + SIZE*2;\ | |
2371 uint8_t half[SIZE*SIZE];\ | |
2372 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2373 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2374 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid+SIZE, half, stride, SIZE, SIZE, SIZE);\ | |
2375 }\ | |
2376 \ | |
2377 static void OPNAME ## h264_qpel ## SIZE ## _mc11_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2378 uint8_t full[SIZE*(SIZE+5)];\ | |
2379 uint8_t * const full_mid= full + SIZE*2;\ | |
2380 uint8_t halfH[SIZE*SIZE];\ | |
2381 uint8_t halfV[SIZE*SIZE];\ | |
2382 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2383 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2384 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2385 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2386 }\ | |
2387 \ | |
2388 static void OPNAME ## h264_qpel ## SIZE ## _mc31_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2389 uint8_t full[SIZE*(SIZE+5)];\ | |
2390 uint8_t * const full_mid= full + SIZE*2;\ | |
2391 uint8_t halfH[SIZE*SIZE];\ | |
2392 uint8_t halfV[SIZE*SIZE];\ | |
2393 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2394 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2395 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2396 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2397 }\ | |
2398 \ | |
2399 static void OPNAME ## h264_qpel ## SIZE ## _mc13_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2400 uint8_t full[SIZE*(SIZE+5)];\ | |
2401 uint8_t * const full_mid= full + SIZE*2;\ | |
2402 uint8_t halfH[SIZE*SIZE];\ | |
2403 uint8_t halfV[SIZE*SIZE];\ | |
2404 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2405 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2406 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2407 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2408 }\ | |
2409 \ | |
2410 static void OPNAME ## h264_qpel ## SIZE ## _mc33_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2411 uint8_t full[SIZE*(SIZE+5)];\ | |
2412 uint8_t * const full_mid= full + SIZE*2;\ | |
2413 uint8_t halfH[SIZE*SIZE];\ | |
2414 uint8_t halfV[SIZE*SIZE];\ | |
2415 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2416 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2417 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2418 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2419 }\ | |
2420 \ | |
2421 static void OPNAME ## h264_qpel ## SIZE ## _mc22_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2422 int16_t tmp[SIZE*(SIZE+5)];\ | |
2423 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass(dst, tmp, src, stride, SIZE, stride);\ | |
2424 }\ | |
2425 \ | |
2426 static void OPNAME ## h264_qpel ## SIZE ## _mc21_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2427 int16_t tmp[SIZE*(SIZE+5)];\ | |
2428 uint8_t halfH[SIZE*SIZE];\ | |
2429 uint8_t halfHV[SIZE*SIZE];\ | |
2430 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2431 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2432 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2433 }\ | |
2434 \ | |
2435 static void OPNAME ## h264_qpel ## SIZE ## _mc23_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2436 int16_t tmp[SIZE*(SIZE+5)];\ | |
2437 uint8_t halfH[SIZE*SIZE];\ | |
2438 uint8_t halfHV[SIZE*SIZE];\ | |
2439 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2440 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2441 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2442 }\ | |
2443 \ | |
2444 static void OPNAME ## h264_qpel ## SIZE ## _mc12_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2445 uint8_t full[SIZE*(SIZE+5)];\ | |
2446 uint8_t * const full_mid= full + SIZE*2;\ | |
2447 int16_t tmp[SIZE*(SIZE+5)];\ | |
2448 uint8_t halfV[SIZE*SIZE];\ | |
2449 uint8_t halfHV[SIZE*SIZE];\ | |
2450 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2451 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2452 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2453 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2454 }\ | |
2455 \ | |
2456 static void OPNAME ## h264_qpel ## SIZE ## _mc32_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2457 uint8_t full[SIZE*(SIZE+5)];\ | |
2458 uint8_t * const full_mid= full + SIZE*2;\ | |
2459 int16_t tmp[SIZE*(SIZE+5)];\ | |
2460 uint8_t halfV[SIZE*SIZE];\ | |
2461 uint8_t halfHV[SIZE*SIZE];\ | |
2462 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2463 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2464 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2465 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2466 }\ | |
2467 | |
2468 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) | |
2469 //#define op_avg2(a, b) a = (((a)*w1+cm[((b) + 16)>>5]*w2 + o + 64)>>7) | |
2470 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2471 #define op2_avg(a, b) a = (((a)+cm[((b) + 512)>>10]+1)>>1) | |
2472 #define op2_put(a, b) a = cm[((b) + 512)>>10] | |
2473 | |
2474 H264_LOWPASS(put_ , op_put, op2_put) | |
2475 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
|
2476 H264_MC(put_, 2) |
1168 | 2477 H264_MC(put_, 4) |
2478 H264_MC(put_, 8) | |
2479 H264_MC(put_, 16) | |
2480 H264_MC(avg_, 4) | |
2481 H264_MC(avg_, 8) | |
2482 H264_MC(avg_, 16) | |
2483 | |
2484 #undef op_avg | |
2485 #undef op_put | |
2486 #undef op2_avg | |
2487 #undef op2_put | |
2488 #endif | |
2489 | |
4594 | 2490 #define op_scale1(x) block[x] = av_clip_uint8( (block[x]*weight + offset) >> log2_denom ) |
2491 #define op_scale2(x) dst[x] = av_clip_uint8( (src[x]*weights + dst[x]*weightd + offset) >> (log2_denom+1)) | |
2415 | 2492 #define H264_WEIGHT(W,H) \ |
2493 static void weight_h264_pixels ## W ## x ## H ## _c(uint8_t *block, int stride, int log2_denom, int weight, int offset){ \ | |
3029 | 2494 int y; \ |
2415 | 2495 offset <<= log2_denom; \ |
2496 if(log2_denom) offset += 1<<(log2_denom-1); \ | |
2497 for(y=0; y<H; y++, block += stride){ \ | |
2498 op_scale1(0); \ | |
2499 op_scale1(1); \ | |
2500 if(W==2) continue; \ | |
2501 op_scale1(2); \ | |
2502 op_scale1(3); \ | |
2503 if(W==4) continue; \ | |
2504 op_scale1(4); \ | |
2505 op_scale1(5); \ | |
2506 op_scale1(6); \ | |
2507 op_scale1(7); \ | |
2508 if(W==8) continue; \ | |
2509 op_scale1(8); \ | |
2510 op_scale1(9); \ | |
2511 op_scale1(10); \ | |
2512 op_scale1(11); \ | |
2513 op_scale1(12); \ | |
2514 op_scale1(13); \ | |
2515 op_scale1(14); \ | |
2516 op_scale1(15); \ | |
2517 } \ | |
2518 } \ | |
3029 | 2519 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){ \ |
2520 int y; \ | |
2521 offset = ((offset + 1) | 1) << log2_denom; \ | |
2415 | 2522 for(y=0; y<H; y++, dst += stride, src += stride){ \ |
2523 op_scale2(0); \ | |
2524 op_scale2(1); \ | |
2525 if(W==2) continue; \ | |
2526 op_scale2(2); \ | |
2527 op_scale2(3); \ | |
2528 if(W==4) continue; \ | |
2529 op_scale2(4); \ | |
2530 op_scale2(5); \ | |
2531 op_scale2(6); \ | |
2532 op_scale2(7); \ | |
2533 if(W==8) continue; \ | |
2534 op_scale2(8); \ | |
2535 op_scale2(9); \ | |
2536 op_scale2(10); \ | |
2537 op_scale2(11); \ | |
2538 op_scale2(12); \ | |
2539 op_scale2(13); \ | |
2540 op_scale2(14); \ | |
2541 op_scale2(15); \ | |
2542 } \ | |
2543 } | |
2544 | |
2545 H264_WEIGHT(16,16) | |
2546 H264_WEIGHT(16,8) | |
2547 H264_WEIGHT(8,16) | |
2548 H264_WEIGHT(8,8) | |
2549 H264_WEIGHT(8,4) | |
2550 H264_WEIGHT(4,8) | |
2551 H264_WEIGHT(4,4) | |
2552 H264_WEIGHT(4,2) | |
2553 H264_WEIGHT(2,4) | |
2554 H264_WEIGHT(2,2) | |
2555 | |
2556 #undef op_scale1 | |
2557 #undef op_scale2 | |
2558 #undef H264_WEIGHT | |
2559 | |
936 | 2560 static void wmv2_mspel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){ |
4176 | 2561 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
936 | 2562 int i; |
2563 | |
2564 for(i=0; i<h; i++){ | |
2565 dst[0]= cm[(9*(src[0] + src[1]) - (src[-1] + src[2]) + 8)>>4]; | |
2566 dst[1]= cm[(9*(src[1] + src[2]) - (src[ 0] + src[3]) + 8)>>4]; | |
2567 dst[2]= cm[(9*(src[2] + src[3]) - (src[ 1] + src[4]) + 8)>>4]; | |
2568 dst[3]= cm[(9*(src[3] + src[4]) - (src[ 2] + src[5]) + 8)>>4]; | |
2569 dst[4]= cm[(9*(src[4] + src[5]) - (src[ 3] + src[6]) + 8)>>4]; | |
2570 dst[5]= cm[(9*(src[5] + src[6]) - (src[ 4] + src[7]) + 8)>>4]; | |
2571 dst[6]= cm[(9*(src[6] + src[7]) - (src[ 5] + src[8]) + 8)>>4]; | |
2572 dst[7]= cm[(9*(src[7] + src[8]) - (src[ 6] + src[9]) + 8)>>4]; | |
2573 dst+=dstStride; | |
2967 | 2574 src+=srcStride; |
936 | 2575 } |
2576 } | |
2577 | |
3432 | 2578 #ifdef CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2579 /* AVS specific */ |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2580 void ff_cavsdsp_init(DSPContext* c, AVCodecContext *avctx); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2581 |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2582 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
|
2583 put_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2584 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2585 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
|
2586 avg_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2587 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2588 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
|
2589 put_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2590 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2591 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
|
2592 avg_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2593 } |
3432 | 2594 #endif /* CONFIG_CAVS_DECODER */ |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2595 |
3526 | 2596 #if defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
2597 /* VC-1 specific */ | |
2598 void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx); | |
2599 | |
2600 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, uint8_t *src, int stride, int rnd) { | |
2601 put_pixels8_c(dst, src, stride, 8); | |
2602 } | |
2603 #endif /* CONFIG_VC1_DECODER||CONFIG_WMV3_DECODER */ | |
2604 | |
5887 | 2605 void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx); |
5899 | 2606 |
4296 | 2607 /* H264 specific */ |
5411
362aec4ef932
Take care of some renames (Doxygen and function name) after the previous pure rename patch.
takis
parents:
5394
diff
changeset
|
2608 void ff_h264dspenc_init(DSPContext* c, AVCodecContext *avctx); |
4296 | 2609 |
936 | 2610 static void wmv2_mspel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int w){ |
4176 | 2611 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
936 | 2612 int i; |
2613 | |
2614 for(i=0; i<w; i++){ | |
2615 const int src_1= src[ -srcStride]; | |
2616 const int src0 = src[0 ]; | |
2617 const int src1 = src[ srcStride]; | |
2618 const int src2 = src[2*srcStride]; | |
2619 const int src3 = src[3*srcStride]; | |
2620 const int src4 = src[4*srcStride]; | |
2621 const int src5 = src[5*srcStride]; | |
2622 const int src6 = src[6*srcStride]; | |
2623 const int src7 = src[7*srcStride]; | |
2624 const int src8 = src[8*srcStride]; | |
2625 const int src9 = src[9*srcStride]; | |
2626 dst[0*dstStride]= cm[(9*(src0 + src1) - (src_1 + src2) + 8)>>4]; | |
2627 dst[1*dstStride]= cm[(9*(src1 + src2) - (src0 + src3) + 8)>>4]; | |
2628 dst[2*dstStride]= cm[(9*(src2 + src3) - (src1 + src4) + 8)>>4]; | |
2629 dst[3*dstStride]= cm[(9*(src3 + src4) - (src2 + src5) + 8)>>4]; | |
2630 dst[4*dstStride]= cm[(9*(src4 + src5) - (src3 + src6) + 8)>>4]; | |
2631 dst[5*dstStride]= cm[(9*(src5 + src6) - (src4 + src7) + 8)>>4]; | |
2632 dst[6*dstStride]= cm[(9*(src6 + src7) - (src5 + src8) + 8)>>4]; | |
2633 dst[7*dstStride]= cm[(9*(src7 + src8) - (src6 + src9) + 8)>>4]; | |
2634 src++; | |
2635 dst++; | |
2636 } | |
2637 } | |
2638 | |
2639 static void put_mspel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){ | |
2640 put_pixels8_c(dst, src, stride, 8); | |
2641 } | |
2642 | |
2643 static void put_mspel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){ | |
2644 uint8_t half[64]; | |
2645 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2646 put_pixels8_l2(dst, src, half, stride, stride, 8, 8); | |
2647 } | |
2648 | |
2649 static void put_mspel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){ | |
2650 wmv2_mspel8_h_lowpass(dst, src, stride, stride, 8); | |
2651 } | |
2652 | |
2653 static void put_mspel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){ | |
2654 uint8_t half[64]; | |
2655 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2656 put_pixels8_l2(dst, src+1, half, stride, stride, 8, 8); | |
2657 } | |
2658 | |
2659 static void put_mspel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){ | |
2660 wmv2_mspel8_v_lowpass(dst, src, stride, stride, 8); | |
2661 } | |
2662 | |
2663 static void put_mspel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){ | |
2664 uint8_t halfH[88]; | |
2665 uint8_t halfV[64]; | |
2666 uint8_t halfHV[64]; | |
2667 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2668 wmv2_mspel8_v_lowpass(halfV, src, 8, stride, 8); | |
2669 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2670 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2671 } | |
2672 static void put_mspel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){ | |
2673 uint8_t halfH[88]; | |
2674 uint8_t halfV[64]; | |
2675 uint8_t halfHV[64]; | |
2676 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2677 wmv2_mspel8_v_lowpass(halfV, src+1, 8, stride, 8); | |
2678 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2679 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2680 } | |
2681 static void put_mspel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){ | |
2682 uint8_t halfH[88]; | |
2683 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2684 wmv2_mspel8_v_lowpass(dst, halfH+8, stride, 8, 8); | |
2685 } | |
2686 | |
1644 | 2687 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
|
2688 if(ENABLE_ANY_H263) { |
1644 | 2689 int x; |
2690 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2691 |
1644 | 2692 for(x=0; x<8; x++){ |
2693 int d1, d2, ad1; | |
2694 int p0= src[x-2*stride]; | |
2695 int p1= src[x-1*stride]; | |
2696 int p2= src[x+0*stride]; | |
2697 int p3= src[x+1*stride]; | |
2698 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2699 | |
2700 if (d<-2*strength) d1= 0; | |
2701 else if(d<- strength) d1=-2*strength - d; | |
2702 else if(d< strength) d1= d; | |
2703 else if(d< 2*strength) d1= 2*strength - d; | |
2704 else d1= 0; | |
2967 | 2705 |
1644 | 2706 p1 += d1; |
2707 p2 -= d1; | |
2708 if(p1&256) p1= ~(p1>>31); | |
2709 if(p2&256) p2= ~(p2>>31); | |
2967 | 2710 |
1644 | 2711 src[x-1*stride] = p1; |
2712 src[x+0*stride] = p2; | |
2713 | |
4001 | 2714 ad1= FFABS(d1)>>1; |
2967 | 2715 |
4594 | 2716 d2= av_clip((p0-p3)/4, -ad1, ad1); |
2967 | 2717 |
1644 | 2718 src[x-2*stride] = p0 - d2; |
2719 src[x+ stride] = p3 + d2; | |
2720 } | |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2721 } |
1644 | 2722 } |
2723 | |
2724 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
|
2725 if(ENABLE_ANY_H263) { |
1644 | 2726 int y; |
2727 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2728 |
1644 | 2729 for(y=0; y<8; y++){ |
2730 int d1, d2, ad1; | |
2731 int p0= src[y*stride-2]; | |
2732 int p1= src[y*stride-1]; | |
2733 int p2= src[y*stride+0]; | |
2734 int p3= src[y*stride+1]; | |
2735 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2736 | |
2737 if (d<-2*strength) d1= 0; | |
2738 else if(d<- strength) d1=-2*strength - d; | |
2739 else if(d< strength) d1= d; | |
2740 else if(d< 2*strength) d1= 2*strength - d; | |
2741 else d1= 0; | |
2967 | 2742 |
1644 | 2743 p1 += d1; |
2744 p2 -= d1; | |
2745 if(p1&256) p1= ~(p1>>31); | |
2746 if(p2&256) p2= ~(p2>>31); | |
2967 | 2747 |
1644 | 2748 src[y*stride-1] = p1; |
2749 src[y*stride+0] = p2; | |
2750 | |
4001 | 2751 ad1= FFABS(d1)>>1; |
2967 | 2752 |
4594 | 2753 d2= av_clip((p0-p3)/4, -ad1, ad1); |
2967 | 2754 |
1644 | 2755 src[y*stride-2] = p0 - d2; |
2756 src[y*stride+1] = p3 + d2; | |
2757 } | |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2758 } |
1644 | 2759 } |
936 | 2760 |
2045 | 2761 static void h261_loop_filter_c(uint8_t *src, int stride){ |
2762 int x,y,xy,yz; | |
2763 int temp[64]; | |
2764 | |
2765 for(x=0; x<8; x++){ | |
2766 temp[x ] = 4*src[x ]; | |
2767 temp[x + 7*8] = 4*src[x + 7*stride]; | |
2768 } | |
2769 for(y=1; y<7; y++){ | |
2770 for(x=0; x<8; x++){ | |
2771 xy = y * stride + x; | |
2772 yz = y * 8 + x; | |
2773 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
|
2774 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2775 } |
2967 | 2776 |
2045 | 2777 for(y=0; y<8; y++){ |
2778 src[ y*stride] = (temp[ y*8] + 2)>>2; | |
2779 src[7+y*stride] = (temp[7+y*8] + 2)>>2; | |
2780 for(x=1; x<7; x++){ | |
2781 xy = y * stride + x; | |
2782 yz = y * 8 + x; | |
2783 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
|
2784 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2785 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2786 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2787 |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2788 static inline void h264_loop_filter_luma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0) |
2633 | 2789 { |
2790 int i, d; | |
2791 for( i = 0; i < 4; i++ ) { | |
2792 if( tc0[i] < 0 ) { | |
2793 pix += 4*ystride; | |
2794 continue; | |
2795 } | |
2796 for( d = 0; d < 4; d++ ) { | |
2797 const int p0 = pix[-1*xstride]; | |
2798 const int p1 = pix[-2*xstride]; | |
2799 const int p2 = pix[-3*xstride]; | |
2800 const int q0 = pix[0]; | |
2801 const int q1 = pix[1*xstride]; | |
2802 const int q2 = pix[2*xstride]; | |
2967 | 2803 |
4001 | 2804 if( FFABS( p0 - q0 ) < alpha && |
2805 FFABS( p1 - p0 ) < beta && | |
2806 FFABS( q1 - q0 ) < beta ) { | |
2967 | 2807 |
2633 | 2808 int tc = tc0[i]; |
2809 int i_delta; | |
2967 | 2810 |
4001 | 2811 if( FFABS( p2 - p0 ) < beta ) { |
4594 | 2812 pix[-2*xstride] = p1 + av_clip( (( p2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - p1, -tc0[i], tc0[i] ); |
2633 | 2813 tc++; |
2814 } | |
4001 | 2815 if( FFABS( q2 - q0 ) < beta ) { |
4594 | 2816 pix[ xstride] = q1 + av_clip( (( q2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - q1, -tc0[i], tc0[i] ); |
2633 | 2817 tc++; |
2818 } | |
2967 | 2819 |
4594 | 2820 i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
2821 pix[-xstride] = av_clip_uint8( p0 + i_delta ); /* p0' */ | |
2822 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */ | |
2633 | 2823 } |
2824 pix += ystride; | |
2825 } | |
2826 } | |
2827 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2828 static void h264_v_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2829 { |
2830 h264_loop_filter_luma_c(pix, stride, 1, alpha, beta, tc0); | |
2831 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2832 static void h264_h_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2833 { |
2834 h264_loop_filter_luma_c(pix, 1, stride, alpha, beta, tc0); | |
2835 } | |
2836 | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2837 static inline void h264_loop_filter_chroma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0) |
2633 | 2838 { |
2839 int i, d; | |
2840 for( i = 0; i < 4; i++ ) { | |
2841 const int tc = tc0[i]; | |
2842 if( tc <= 0 ) { | |
2843 pix += 2*ystride; | |
2844 continue; | |
2845 } | |
2846 for( d = 0; d < 2; d++ ) { | |
2847 const int p0 = pix[-1*xstride]; | |
2848 const int p1 = pix[-2*xstride]; | |
2849 const int q0 = pix[0]; | |
2850 const int q1 = pix[1*xstride]; | |
2851 | |
4001 | 2852 if( FFABS( p0 - q0 ) < alpha && |
2853 FFABS( p1 - p0 ) < beta && | |
2854 FFABS( q1 - q0 ) < beta ) { | |
2633 | 2855 |
4594 | 2856 int delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
2857 | |
2858 pix[-xstride] = av_clip_uint8( p0 + delta ); /* p0' */ | |
2859 pix[0] = av_clip_uint8( q0 - delta ); /* q0' */ | |
2633 | 2860 } |
2861 pix += ystride; | |
2862 } | |
2863 } | |
2864 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2865 static void h264_v_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2866 { |
2867 h264_loop_filter_chroma_c(pix, stride, 1, alpha, beta, tc0); | |
2868 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2869 static void h264_h_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2870 { |
2871 h264_loop_filter_chroma_c(pix, 1, stride, alpha, beta, tc0); | |
2872 } | |
2873 | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2874 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
|
2875 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2876 int d; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2877 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
|
2878 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
|
2879 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
|
2880 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
|
2881 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
|
2882 |
4001 | 2883 if( FFABS( p0 - q0 ) < alpha && |
2884 FFABS( p1 - p0 ) < beta && | |
2885 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
|
2886 |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2887 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
|
2888 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
|
2889 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2890 pix += ystride; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2891 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2892 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2893 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
|
2894 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2895 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
|
2896 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2897 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
|
2898 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2899 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
|
2900 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2901 |
1708 | 2902 static inline int pix_abs16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2903 { |
2904 int s, i; | |
2905 | |
2906 s = 0; | |
1708 | 2907 for(i=0;i<h;i++) { |
0 | 2908 s += abs(pix1[0] - pix2[0]); |
2909 s += abs(pix1[1] - pix2[1]); | |
2910 s += abs(pix1[2] - pix2[2]); | |
2911 s += abs(pix1[3] - pix2[3]); | |
2912 s += abs(pix1[4] - pix2[4]); | |
2913 s += abs(pix1[5] - pix2[5]); | |
2914 s += abs(pix1[6] - pix2[6]); | |
2915 s += abs(pix1[7] - pix2[7]); | |
2916 s += abs(pix1[8] - pix2[8]); | |
2917 s += abs(pix1[9] - pix2[9]); | |
2918 s += abs(pix1[10] - pix2[10]); | |
2919 s += abs(pix1[11] - pix2[11]); | |
2920 s += abs(pix1[12] - pix2[12]); | |
2921 s += abs(pix1[13] - pix2[13]); | |
2922 s += abs(pix1[14] - pix2[14]); | |
2923 s += abs(pix1[15] - pix2[15]); | |
2924 pix1 += line_size; | |
2925 pix2 += line_size; | |
2926 } | |
2927 return s; | |
2928 } | |
2929 | |
1708 | 2930 static int pix_abs16_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2931 { |
2932 int s, i; | |
2933 | |
2934 s = 0; | |
1708 | 2935 for(i=0;i<h;i++) { |
0 | 2936 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
2937 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
2938 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
2939 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
2940 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
2941 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
2942 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
2943 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
2944 s += abs(pix1[8] - avg2(pix2[8], pix2[9])); | |
2945 s += abs(pix1[9] - avg2(pix2[9], pix2[10])); | |
2946 s += abs(pix1[10] - avg2(pix2[10], pix2[11])); | |
2947 s += abs(pix1[11] - avg2(pix2[11], pix2[12])); | |
2948 s += abs(pix1[12] - avg2(pix2[12], pix2[13])); | |
2949 s += abs(pix1[13] - avg2(pix2[13], pix2[14])); | |
2950 s += abs(pix1[14] - avg2(pix2[14], pix2[15])); | |
2951 s += abs(pix1[15] - avg2(pix2[15], pix2[16])); | |
2952 pix1 += line_size; | |
2953 pix2 += line_size; | |
2954 } | |
2955 return s; | |
2956 } | |
2957 | |
1708 | 2958 static int pix_abs16_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2959 { |
2960 int s, i; | |
1064 | 2961 uint8_t *pix3 = pix2 + line_size; |
0 | 2962 |
2963 s = 0; | |
1708 | 2964 for(i=0;i<h;i++) { |
0 | 2965 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
2966 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
2967 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
2968 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
2969 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
2970 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
2971 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
2972 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
2973 s += abs(pix1[8] - avg2(pix2[8], pix3[8])); | |
2974 s += abs(pix1[9] - avg2(pix2[9], pix3[9])); | |
2975 s += abs(pix1[10] - avg2(pix2[10], pix3[10])); | |
2976 s += abs(pix1[11] - avg2(pix2[11], pix3[11])); | |
2977 s += abs(pix1[12] - avg2(pix2[12], pix3[12])); | |
2978 s += abs(pix1[13] - avg2(pix2[13], pix3[13])); | |
2979 s += abs(pix1[14] - avg2(pix2[14], pix3[14])); | |
2980 s += abs(pix1[15] - avg2(pix2[15], pix3[15])); | |
2981 pix1 += line_size; | |
2982 pix2 += line_size; | |
2983 pix3 += line_size; | |
2984 } | |
2985 return s; | |
2986 } | |
2987 | |
1708 | 2988 static int pix_abs16_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2989 { |
2990 int s, i; | |
1064 | 2991 uint8_t *pix3 = pix2 + line_size; |
0 | 2992 |
2993 s = 0; | |
1708 | 2994 for(i=0;i<h;i++) { |
0 | 2995 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
2996 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
2997 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
2998 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
2999 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
3000 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
3001 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
3002 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
3003 s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9])); | |
3004 s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10])); | |
3005 s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11])); | |
3006 s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12])); | |
3007 s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13])); | |
3008 s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14])); | |
3009 s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15])); | |
3010 s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16])); | |
3011 pix1 += line_size; | |
3012 pix2 += line_size; | |
3013 pix3 += line_size; | |
3014 } | |
3015 return s; | |
3016 } | |
3017 | |
1708 | 3018 static inline int pix_abs8_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3019 { |
3020 int s, i; | |
3021 | |
3022 s = 0; | |
1708 | 3023 for(i=0;i<h;i++) { |
294 | 3024 s += abs(pix1[0] - pix2[0]); |
3025 s += abs(pix1[1] - pix2[1]); | |
3026 s += abs(pix1[2] - pix2[2]); | |
3027 s += abs(pix1[3] - pix2[3]); | |
3028 s += abs(pix1[4] - pix2[4]); | |
3029 s += abs(pix1[5] - pix2[5]); | |
3030 s += abs(pix1[6] - pix2[6]); | |
3031 s += abs(pix1[7] - pix2[7]); | |
3032 pix1 += line_size; | |
3033 pix2 += line_size; | |
3034 } | |
3035 return s; | |
3036 } | |
3037 | |
1708 | 3038 static int pix_abs8_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3039 { |
3040 int s, i; | |
3041 | |
3042 s = 0; | |
1708 | 3043 for(i=0;i<h;i++) { |
294 | 3044 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
3045 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
3046 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
3047 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
3048 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
3049 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
3050 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
3051 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
3052 pix1 += line_size; | |
3053 pix2 += line_size; | |
3054 } | |
3055 return s; | |
3056 } | |
3057 | |
1708 | 3058 static int pix_abs8_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3059 { |
3060 int s, i; | |
1064 | 3061 uint8_t *pix3 = pix2 + line_size; |
294 | 3062 |
3063 s = 0; | |
1708 | 3064 for(i=0;i<h;i++) { |
294 | 3065 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
3066 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
3067 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
3068 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
3069 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
3070 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
3071 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
3072 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
3073 pix1 += line_size; | |
3074 pix2 += line_size; | |
3075 pix3 += line_size; | |
3076 } | |
3077 return s; | |
3078 } | |
3079 | |
1708 | 3080 static int pix_abs8_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3081 { |
3082 int s, i; | |
1064 | 3083 uint8_t *pix3 = pix2 + line_size; |
294 | 3084 |
3085 s = 0; | |
1708 | 3086 for(i=0;i<h;i++) { |
294 | 3087 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
3088 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
3089 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
3090 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
3091 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
3092 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
3093 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
3094 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
3095 pix1 += line_size; | |
3096 pix2 += line_size; | |
3097 pix3 += line_size; | |
3098 } | |
3099 return s; | |
3100 } | |
3101 | |
2834 | 3102 static int nsse16_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3103 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3104 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3105 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3106 int x,y; |
2066 | 3107 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3108 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3109 for(x=0; x<16; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3110 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3111 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3112 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3113 for(x=0; x<15; x++){ |
4001 | 3114 score2+= FFABS( s1[x ] - s1[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3115 - s1[x+1] + s1[x+1+stride]) |
4001 | 3116 -FFABS( s2[x ] - s2[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3117 - s2[x+1] + s2[x+1+stride]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3118 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3119 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3120 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3121 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3122 } |
2066 | 3123 |
4001 | 3124 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight; |
3125 else return score1 + FFABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3126 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3127 |
2834 | 3128 static int nsse8_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3129 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3130 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3131 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3132 int x,y; |
2967 | 3133 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3134 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3135 for(x=0; x<8; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3136 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3137 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3138 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3139 for(x=0; x<7; x++){ |
4001 | 3140 score2+= FFABS( s1[x ] - s1[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3141 - s1[x+1] + s1[x+1+stride]) |
4001 | 3142 -FFABS( s2[x ] - s2[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3143 - s2[x+1] + s2[x+1+stride]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3144 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3145 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3146 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3147 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3148 } |
2967 | 3149 |
4001 | 3150 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight; |
3151 else return score1 + FFABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3152 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3153 |
1784 | 3154 static int try_8x8basis_c(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale){ |
3155 int i; | |
3156 unsigned int sum=0; | |
3157 | |
3158 for(i=0; i<8*8; i++){ | |
3159 int b= rem[i] + ((basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT)); | |
3160 int w= weight[i]; | |
3161 b>>= RECON_SHIFT; | |
3162 assert(-512<b && b<512); | |
3163 | |
3164 sum += (w*b)*(w*b)>>4; | |
3165 } | |
3166 return sum>>2; | |
3167 } | |
3168 | |
3169 static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale){ | |
3170 int i; | |
3171 | |
3172 for(i=0; i<8*8; i++){ | |
3173 rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT); | |
2967 | 3174 } |
1784 | 3175 } |
3176 | |
1100 | 3177 /** |
3178 * permutes an 8x8 block. | |
1101 | 3179 * @param block the block which will be permuted according to the given permutation vector |
1100 | 3180 * @param permutation the permutation vector |
3181 * @param last the last non zero coefficient in scantable order, used to speed the permutation up | |
2967 | 3182 * @param scantable the used scantable, this is only used to speed the permutation up, the block is not |
1101 | 3183 * (inverse) permutated to scantable order! |
1100 | 3184 */ |
1064 | 3185 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
|
3186 { |
764 | 3187 int i; |
945 | 3188 DCTELEM temp[64]; |
2967 | 3189 |
764 | 3190 if(last<=0) return; |
5129 | 3191 //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
|
3192 |
764 | 3193 for(i=0; i<=last; i++){ |
3194 const int j= scantable[i]; | |
3195 temp[j]= block[j]; | |
3196 block[j]=0; | |
3197 } | |
2967 | 3198 |
764 | 3199 for(i=0; i<=last; i++){ |
3200 const int j= scantable[i]; | |
3201 const int perm_j= permutation[j]; | |
3202 block[perm_j]= temp[j]; | |
3203 } | |
174
ac5075a55488
new IDCT code by Michael Niedermayer (michaelni@gmx.at) - #define SIMPLE_IDCT to enable
arpi_esp
parents:
88
diff
changeset
|
3204 } |
34 | 3205 |
1729 | 3206 static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride, int h){ |
3207 return 0; | |
3208 } | |
3209 | |
3210 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type){ | |
3211 int i; | |
2967 | 3212 |
1729 | 3213 memset(cmp, 0, sizeof(void*)*5); |
2967 | 3214 |
1729 | 3215 for(i=0; i<5; i++){ |
3216 switch(type&0xFF){ | |
3217 case FF_CMP_SAD: | |
3218 cmp[i]= c->sad[i]; | |
3219 break; | |
3220 case FF_CMP_SATD: | |
3221 cmp[i]= c->hadamard8_diff[i]; | |
3222 break; | |
3223 case FF_CMP_SSE: | |
3224 cmp[i]= c->sse[i]; | |
3225 break; | |
3226 case FF_CMP_DCT: | |
3227 cmp[i]= c->dct_sad[i]; | |
3228 break; | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3229 case FF_CMP_DCT264: |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3230 cmp[i]= c->dct264_sad[i]; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3231 break; |
2382 | 3232 case FF_CMP_DCTMAX: |
3233 cmp[i]= c->dct_max[i]; | |
3234 break; | |
1729 | 3235 case FF_CMP_PSNR: |
3236 cmp[i]= c->quant_psnr[i]; | |
3237 break; | |
3238 case FF_CMP_BIT: | |
3239 cmp[i]= c->bit[i]; | |
3240 break; | |
3241 case FF_CMP_RD: | |
3242 cmp[i]= c->rd[i]; | |
3243 break; | |
3244 case FF_CMP_VSAD: | |
3245 cmp[i]= c->vsad[i]; | |
3246 break; | |
3247 case FF_CMP_VSSE: | |
3248 cmp[i]= c->vsse[i]; | |
3249 break; | |
3250 case FF_CMP_ZERO: | |
3251 cmp[i]= zero_cmp; | |
3252 break; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3253 case FF_CMP_NSSE: |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3254 cmp[i]= c->nsse[i]; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3255 break; |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
3256 #ifdef CONFIG_SNOW_ENCODER |
2184 | 3257 case FF_CMP_W53: |
3258 cmp[i]= c->w53[i]; | |
3259 break; | |
3260 case FF_CMP_W97: | |
3261 cmp[i]= c->w97[i]; | |
3262 break; | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
3263 #endif |
1729 | 3264 default: |
3265 av_log(NULL, AV_LOG_ERROR,"internal error in cmp function selection\n"); | |
3266 } | |
3267 } | |
3268 } | |
3269 | |
1101 | 3270 /** |
3271 * memset(blocks, 0, sizeof(DCTELEM)*6*64) | |
3272 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
3273 static void clear_blocks_c(DCTELEM *blocks) |
296 | 3274 { |
3275 memset(blocks, 0, sizeof(DCTELEM)*6*64); | |
3276 } | |
3277 | |
866 | 3278 static void add_bytes_c(uint8_t *dst, uint8_t *src, int w){ |
3279 int i; | |
996
ad44196ea483
add/diff_bytes bugfix patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>)
michaelni
parents:
984
diff
changeset
|
3280 for(i=0; i+7<w; i+=8){ |
866 | 3281 dst[i+0] += src[i+0]; |
3282 dst[i+1] += src[i+1]; | |
3283 dst[i+2] += src[i+2]; | |
3284 dst[i+3] += src[i+3]; | |
3285 dst[i+4] += src[i+4]; | |
3286 dst[i+5] += src[i+5]; | |
3287 dst[i+6] += src[i+6]; | |
3288 dst[i+7] += src[i+7]; | |
3289 } | |
3290 for(; i<w; i++) | |
3291 dst[i+0] += src[i+0]; | |
3292 } | |
3293 | |
6384 | 3294 static void add_bytes_l2_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ |
3295 int i; | |
3296 for(i=0; i<=w-sizeof(long); i+=sizeof(long)){ | |
3297 long a = *(long*)(src1+i); | |
3298 long b = *(long*)(src2+i); | |
3299 *(long*)(dst+i) = ((a&0x7f7f7f7f7f7f7f7fL) + (b&0x7f7f7f7f7f7f7f7fL)) ^ ((a^b)&0x8080808080808080L); | |
3300 } | |
3301 for(; i<w; i++) | |
3302 dst[i] = src1[i]+src2[i]; | |
3303 } | |
3304 | |
866 | 3305 static void diff_bytes_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ |
3306 int i; | |
996
ad44196ea483
add/diff_bytes bugfix patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>)
michaelni
parents:
984
diff
changeset
|
3307 for(i=0; i+7<w; i+=8){ |
866 | 3308 dst[i+0] = src1[i+0]-src2[i+0]; |
3309 dst[i+1] = src1[i+1]-src2[i+1]; | |
3310 dst[i+2] = src1[i+2]-src2[i+2]; | |
3311 dst[i+3] = src1[i+3]-src2[i+3]; | |
3312 dst[i+4] = src1[i+4]-src2[i+4]; | |
3313 dst[i+5] = src1[i+5]-src2[i+5]; | |
3314 dst[i+6] = src1[i+6]-src2[i+6]; | |
3315 dst[i+7] = src1[i+7]-src2[i+7]; | |
3316 } | |
3317 for(; i<w; i++) | |
3318 dst[i+0] = src1[i+0]-src2[i+0]; | |
3319 } | |
3320 | |
1527 | 3321 static void sub_hfyu_median_prediction_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){ |
3322 int i; | |
3323 uint8_t l, lt; | |
3324 | |
3325 l= *left; | |
3326 lt= *left_top; | |
3327 | |
3328 for(i=0; i<w; i++){ | |
3329 const int pred= mid_pred(l, src1[i], (l + src1[i] - lt)&0xFF); | |
3330 lt= src1[i]; | |
3331 l= src2[i]; | |
3332 dst[i]= l - pred; | |
2967 | 3333 } |
1527 | 3334 |
3335 *left= l; | |
3336 *left_top= lt; | |
3337 } | |
3338 | |
936 | 3339 #define BUTTERFLY2(o1,o2,i1,i2) \ |
3340 o1= (i1)+(i2);\ | |
3341 o2= (i1)-(i2); | |
3342 | |
3343 #define BUTTERFLY1(x,y) \ | |
3344 {\ | |
3345 int a,b;\ | |
3346 a= x;\ | |
3347 b= y;\ | |
3348 x= a+b;\ | |
3349 y= a-b;\ | |
3350 } | |
3351 | |
4001 | 3352 #define BUTTERFLYA(x,y) (FFABS((x)+(y)) + FFABS((x)-(y))) |
936 | 3353 |
1708 | 3354 static int hadamard8_diff8x8_c(/*MpegEncContext*/ void *s, uint8_t *dst, uint8_t *src, int stride, int h){ |
936 | 3355 int i; |
3356 int temp[64]; | |
3357 int sum=0; | |
2967 | 3358 |
1708 | 3359 assert(h==8); |
936 | 3360 |
3361 for(i=0; i<8; i++){ | |
3362 //FIXME try pointer walks | |
3363 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]); | |
3364 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]); | |
3365 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]); | |
3366 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 | 3367 |
936 | 3368 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3369 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3370 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3371 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3372 |
936 | 3373 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3374 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3375 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3376 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3377 } | |
3378 | |
3379 for(i=0; i<8; i++){ | |
3380 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3381 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3382 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3383 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3384 |
936 | 3385 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3386 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3387 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3388 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
3389 | |
2967 | 3390 sum += |
936 | 3391 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3392 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3393 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3394 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3395 } | |
3396 #if 0 | |
3397 static int maxi=0; | |
3398 if(sum>maxi){ | |
3399 maxi=sum; | |
3400 printf("MAX:%d\n", maxi); | |
3401 } | |
3402 #endif | |
3403 return sum; | |
3404 } | |
3405 | |
1729 | 3406 static int hadamard8_intra8x8_c(/*MpegEncContext*/ void *s, uint8_t *src, uint8_t *dummy, int stride, int h){ |
936 | 3407 int i; |
3408 int temp[64]; | |
3409 int sum=0; | |
2967 | 3410 |
1729 | 3411 assert(h==8); |
2967 | 3412 |
936 | 3413 for(i=0; i<8; i++){ |
3414 //FIXME try pointer walks | |
1729 | 3415 BUTTERFLY2(temp[8*i+0], temp[8*i+1], src[stride*i+0],src[stride*i+1]); |
3416 BUTTERFLY2(temp[8*i+2], temp[8*i+3], src[stride*i+2],src[stride*i+3]); | |
3417 BUTTERFLY2(temp[8*i+4], temp[8*i+5], src[stride*i+4],src[stride*i+5]); | |
3418 BUTTERFLY2(temp[8*i+6], temp[8*i+7], src[stride*i+6],src[stride*i+7]); | |
2967 | 3419 |
936 | 3420 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3421 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3422 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3423 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3424 |
936 | 3425 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3426 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3427 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3428 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3429 } | |
3430 | |
3431 for(i=0; i<8; i++){ | |
3432 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3433 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3434 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3435 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3436 |
936 | 3437 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3438 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3439 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3440 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
2967 | 3441 |
3442 sum += | |
936 | 3443 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3444 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3445 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3446 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3447 } | |
2967 | 3448 |
4001 | 3449 sum -= FFABS(temp[8*0] + temp[8*4]); // -mean |
2967 | 3450 |
936 | 3451 return sum; |
3452 } | |
3453 | |
1708 | 3454 static int dct_sad8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3455 MpegEncContext * const s= (MpegEncContext *)c; |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
3456 DECLARE_ALIGNED_16(uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
1016 | 3457 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
2967 | 3458 |
1708 | 3459 assert(h==8); |
936 | 3460 |
3461 s->dsp.diff_pixels(temp, src1, src2, stride); | |
1092 | 3462 s->dsp.fdct(temp); |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
3463 return s->dsp.sum_abs_dctelem(temp); |
936 | 3464 } |
3465 | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3466 #ifdef CONFIG_GPL |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3467 #define DCT8_1D {\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3468 const int s07 = SRC(0) + SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3469 const int s16 = SRC(1) + SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3470 const int s25 = SRC(2) + SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3471 const int s34 = SRC(3) + SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3472 const int a0 = s07 + s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3473 const int a1 = s16 + s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3474 const int a2 = s07 - s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3475 const int a3 = s16 - s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3476 const int d07 = SRC(0) - SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3477 const int d16 = SRC(1) - SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3478 const int d25 = SRC(2) - SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3479 const int d34 = SRC(3) - SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3480 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
|
3481 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
|
3482 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
|
3483 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
|
3484 DST(0, a0 + a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3485 DST(1, a4 + (a7>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3486 DST(2, a2 + (a3>>1)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3487 DST(3, a5 + (a6>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3488 DST(4, a0 - a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3489 DST(5, a6 - (a5>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3490 DST(6, (a2>>1) - a3 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3491 DST(7, (a4>>2) - a7 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3492 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3493 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3494 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
|
3495 MpegEncContext * const s= (MpegEncContext *)c; |
5256 | 3496 DCTELEM dct[8][8]; |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3497 int i; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3498 int sum=0; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3499 |
5256 | 3500 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
|
3501 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3502 #define SRC(x) dct[i][x] |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3503 #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
|
3504 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3505 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3506 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3507 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3508 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3509 #define SRC(x) dct[x][i] |
4001 | 3510 #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
|
3511 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3512 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3513 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3514 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3515 return sum; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3516 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3517 #endif |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3518 |
2382 | 3519 static int dct_max8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
3520 MpegEncContext * const s= (MpegEncContext *)c; | |
3089 | 3521 DECLARE_ALIGNED_8(uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
2382 | 3522 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3523 int sum=0, i; | |
2967 | 3524 |
2382 | 3525 assert(h==8); |
3526 | |
3527 s->dsp.diff_pixels(temp, src1, src2, stride); | |
3528 s->dsp.fdct(temp); | |
3529 | |
3530 for(i=0; i<64; i++) | |
4001 | 3531 sum= FFMAX(sum, FFABS(temp[i])); |
2967 | 3532 |
2382 | 3533 return sum; |
3534 } | |
3535 | |
1708 | 3536 static int quant_psnr8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3537 MpegEncContext * const s= (MpegEncContext *)c; |
3089 | 3538 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64*2/8]); |
1016 | 3539 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3540 DCTELEM * const bak = ((DCTELEM*)aligned_temp)+64; | |
936 | 3541 int sum=0, i; |
3542 | |
1708 | 3543 assert(h==8); |
936 | 3544 s->mb_intra=0; |
2967 | 3545 |
936 | 3546 s->dsp.diff_pixels(temp, src1, src2, stride); |
2967 | 3547 |
936 | 3548 memcpy(bak, temp, 64*sizeof(DCTELEM)); |
2967 | 3549 |
1013 | 3550 s->block_last_index[0/*FIXME*/]= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
1689 | 3551 s->dct_unquantize_inter(s, temp, 0, s->qscale); |
6001 | 3552 ff_simple_idct(temp); //FIXME |
2967 | 3553 |
936 | 3554 for(i=0; i<64; i++) |
3555 sum+= (temp[i]-bak[i])*(temp[i]-bak[i]); | |
2967 | 3556 |
936 | 3557 return sum; |
3558 } | |
3559 | |
1708 | 3560 static int rd8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3561 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3562 const uint8_t *scantable= s->intra_scantable.permutated; |
3089 | 3563 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
3564 DECLARE_ALIGNED_8 (uint64_t, aligned_bak[stride]); | |
1016 | 3565 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3566 uint8_t * const bak= (uint8_t*)aligned_bak; | |
1007 | 3567 int i, last, run, bits, level, distoration, start_i; |
3568 const int esc_length= s->ac_esc_length; | |
3569 uint8_t * length; | |
3570 uint8_t * last_length; | |
2967 | 3571 |
1708 | 3572 assert(h==8); |
3573 | |
1007 | 3574 for(i=0; i<8; i++){ |
3575 ((uint32_t*)(bak + i*stride))[0]= ((uint32_t*)(src2 + i*stride))[0]; | |
3576 ((uint32_t*)(bak + i*stride))[1]= ((uint32_t*)(src2 + i*stride))[1]; | |
3577 } | |
3578 | |
3579 s->dsp.diff_pixels(temp, src1, src2, stride); | |
3580 | |
1013 | 3581 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3582 | |
3583 bits=0; | |
2967 | 3584 |
1013 | 3585 if (s->mb_intra) { |
2967 | 3586 start_i = 1; |
1013 | 3587 length = s->intra_ac_vlc_length; |
3588 last_length= s->intra_ac_vlc_last_length; | |
3589 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma | |
3590 } else { | |
3591 start_i = 0; | |
3592 length = s->inter_ac_vlc_length; | |
3593 last_length= s->inter_ac_vlc_last_length; | |
3594 } | |
2967 | 3595 |
1013 | 3596 if(last>=start_i){ |
1007 | 3597 run=0; |
3598 for(i=start_i; i<last; i++){ | |
3599 int j= scantable[i]; | |
3600 level= temp[j]; | |
2967 | 3601 |
1007 | 3602 if(level){ |
3603 level+=64; | |
3604 if((level&(~127)) == 0){ | |
3605 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3606 }else | |
3607 bits+= esc_length; | |
3608 run=0; | |
3609 }else | |
3610 run++; | |
3611 } | |
3612 i= scantable[last]; | |
2967 | 3613 |
1011 | 3614 level= temp[i] + 64; |
3615 | |
3616 assert(level - 64); | |
2967 | 3617 |
1007 | 3618 if((level&(~127)) == 0){ |
3619 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3620 }else | |
3621 bits+= esc_length; | |
2967 | 3622 |
1013 | 3623 } |
3624 | |
3625 if(last>=0){ | |
1689 | 3626 if(s->mb_intra) |
3627 s->dct_unquantize_intra(s, temp, 0, s->qscale); | |
3628 else | |
3629 s->dct_unquantize_inter(s, temp, 0, s->qscale); | |
1007 | 3630 } |
2967 | 3631 |
1092 | 3632 s->dsp.idct_add(bak, stride, temp); |
2967 | 3633 |
1708 | 3634 distoration= s->dsp.sse[1](NULL, bak, src1, stride, 8); |
1007 | 3635 |
1013 | 3636 return distoration + ((bits*s->qscale*s->qscale*109 + 64)>>7); |
1007 | 3637 } |
3638 | |
1708 | 3639 static int bit8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3640 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3641 const uint8_t *scantable= s->intra_scantable.permutated; |
3089 | 3642 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
1016 | 3643 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
1007 | 3644 int i, last, run, bits, level, start_i; |
3645 const int esc_length= s->ac_esc_length; | |
3646 uint8_t * length; | |
3647 uint8_t * last_length; | |
1708 | 3648 |
3649 assert(h==8); | |
2967 | 3650 |
1013 | 3651 s->dsp.diff_pixels(temp, src1, src2, stride); |
1007 | 3652 |
1013 | 3653 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3654 | |
3655 bits=0; | |
2967 | 3656 |
1007 | 3657 if (s->mb_intra) { |
2967 | 3658 start_i = 1; |
1007 | 3659 length = s->intra_ac_vlc_length; |
3660 last_length= s->intra_ac_vlc_last_length; | |
1013 | 3661 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma |
1007 | 3662 } else { |
3663 start_i = 0; | |
3664 length = s->inter_ac_vlc_length; | |
3665 last_length= s->inter_ac_vlc_last_length; | |
3666 } | |
2967 | 3667 |
1013 | 3668 if(last>=start_i){ |
1007 | 3669 run=0; |
3670 for(i=start_i; i<last; i++){ | |
3671 int j= scantable[i]; | |
3672 level= temp[j]; | |
2967 | 3673 |
1007 | 3674 if(level){ |
3675 level+=64; | |
3676 if((level&(~127)) == 0){ | |
3677 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3678 }else | |
3679 bits+= esc_length; | |
3680 run=0; | |
3681 }else | |
3682 run++; | |
3683 } | |
3684 i= scantable[last]; | |
2967 | 3685 |
1013 | 3686 level= temp[i] + 64; |
2967 | 3687 |
1013 | 3688 assert(level - 64); |
2967 | 3689 |
1007 | 3690 if((level&(~127)) == 0){ |
3691 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3692 }else | |
3693 bits+= esc_length; | |
3694 } | |
3695 | |
3696 return bits; | |
3697 } | |
3698 | |
1729 | 3699 static int vsad_intra16_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ |
3700 int score=0; | |
3701 int x,y; | |
2967 | 3702 |
1729 | 3703 for(y=1; y<h; y++){ |
3704 for(x=0; x<16; x+=4){ | |
4001 | 3705 score+= FFABS(s[x ] - s[x +stride]) + FFABS(s[x+1] - s[x+1+stride]) |
3706 +FFABS(s[x+2] - s[x+2+stride]) + FFABS(s[x+3] - s[x+3+stride]); | |
1729 | 3707 } |
3708 s+= stride; | |
3709 } | |
2967 | 3710 |
1729 | 3711 return score; |
3712 } | |
3713 | |
3714 static int vsad16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3715 int score=0; | |
3716 int x,y; | |
2967 | 3717 |
1729 | 3718 for(y=1; y<h; y++){ |
3719 for(x=0; x<16; x++){ | |
4001 | 3720 score+= FFABS(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); |
1729 | 3721 } |
3722 s1+= stride; | |
3723 s2+= stride; | |
3724 } | |
2967 | 3725 |
1729 | 3726 return score; |
3727 } | |
3728 | |
3729 #define SQ(a) ((a)*(a)) | |
3730 static int vsse_intra16_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ | |
3731 int score=0; | |
3732 int x,y; | |
2967 | 3733 |
1729 | 3734 for(y=1; y<h; y++){ |
3735 for(x=0; x<16; x+=4){ | |
2967 | 3736 score+= SQ(s[x ] - s[x +stride]) + SQ(s[x+1] - s[x+1+stride]) |
1729 | 3737 +SQ(s[x+2] - s[x+2+stride]) + SQ(s[x+3] - s[x+3+stride]); |
3738 } | |
3739 s+= stride; | |
3740 } | |
2967 | 3741 |
1729 | 3742 return score; |
3743 } | |
3744 | |
3745 static int vsse16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3746 int score=0; | |
3747 int x,y; | |
2967 | 3748 |
1729 | 3749 for(y=1; y<h; y++){ |
3750 for(x=0; x<16; x++){ | |
3751 score+= SQ(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); | |
3752 } | |
3753 s1+= stride; | |
3754 s2+= stride; | |
3755 } | |
2967 | 3756 |
1729 | 3757 return score; |
3758 } | |
3759 | |
5255 | 3760 static int ssd_int8_vs_int16_c(const int8_t *pix1, const int16_t *pix2, |
3761 int size){ | |
4749 | 3762 int score=0; |
3763 int i; | |
3764 for(i=0; i<size; i++) | |
3765 score += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]); | |
3766 return score; | |
3767 } | |
3768 | |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3769 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
|
3770 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
|
3771 WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c) |
3013 | 3772 #ifdef CONFIG_GPL |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3773 WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c) |
3013 | 3774 #endif |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3775 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
|
3776 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
|
3777 WRAPPER8_16_SQ(rd8x8_c, rd16_c) |
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3778 WRAPPER8_16_SQ(bit8x8_c, bit16_c) |
936 | 3779 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3780 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
|
3781 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3782 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3783 dst[i] *= src[i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3784 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3785 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3786 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
|
3787 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3788 src1 += len-1; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3789 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3790 dst[i] = src0[i] * src1[-i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3791 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3792 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3793 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
|
3794 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3795 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3796 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
|
3797 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3798 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3799 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
|
3800 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3801 for(i=0; i<len; i++) { |
6218 | 3802 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
|
3803 if(tmp & 0xf0000){ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3804 tmp = (0x43c0ffff - tmp)>>31; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3805 // 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
|
3806 // if(tmp > 0x43c0ffff) tmp = 0xFFFF; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3807 // else tmp = 0; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3808 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3809 dst[i] = tmp - 0x8000; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3810 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3811 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3812 |
5887 | 3813 #define W0 2048 |
3814 #define W1 2841 /* 2048*sqrt (2)*cos (1*pi/16) */ | |
3815 #define W2 2676 /* 2048*sqrt (2)*cos (2*pi/16) */ | |
3816 #define W3 2408 /* 2048*sqrt (2)*cos (3*pi/16) */ | |
3817 #define W4 2048 /* 2048*sqrt (2)*cos (4*pi/16) */ | |
3818 #define W5 1609 /* 2048*sqrt (2)*cos (5*pi/16) */ | |
3819 #define W6 1108 /* 2048*sqrt (2)*cos (6*pi/16) */ | |
3820 #define W7 565 /* 2048*sqrt (2)*cos (7*pi/16) */ | |
3821 | |
3822 static void wmv2_idct_row(short * b) | |
3823 { | |
3824 int s1,s2; | |
3825 int a0,a1,a2,a3,a4,a5,a6,a7; | |
3826 /*step 1*/ | |
3827 a1 = W1*b[1]+W7*b[7]; | |
3828 a7 = W7*b[1]-W1*b[7]; | |
3829 a5 = W5*b[5]+W3*b[3]; | |
3830 a3 = W3*b[5]-W5*b[3]; | |
3831 a2 = W2*b[2]+W6*b[6]; | |
3832 a6 = W6*b[2]-W2*b[6]; | |
3833 a0 = W0*b[0]+W0*b[4]; | |
3834 a4 = W0*b[0]-W0*b[4]; | |
3835 /*step 2*/ | |
3836 s1 = (181*(a1-a5+a7-a3)+128)>>8;//1,3,5,7, | |
3837 s2 = (181*(a1-a5-a7+a3)+128)>>8; | |
3838 /*step 3*/ | |
3839 b[0] = (a0+a2+a1+a5 + (1<<7))>>8; | |
3840 b[1] = (a4+a6 +s1 + (1<<7))>>8; | |
3841 b[2] = (a4-a6 +s2 + (1<<7))>>8; | |
3842 b[3] = (a0-a2+a7+a3 + (1<<7))>>8; | |
3843 b[4] = (a0-a2-a7-a3 + (1<<7))>>8; | |
3844 b[5] = (a4-a6 -s2 + (1<<7))>>8; | |
3845 b[6] = (a4+a6 -s1 + (1<<7))>>8; | |
3846 b[7] = (a0+a2-a1-a5 + (1<<7))>>8; | |
3847 } | |
3848 static void wmv2_idct_col(short * b) | |
3849 { | |
3850 int s1,s2; | |
3851 int a0,a1,a2,a3,a4,a5,a6,a7; | |
3852 /*step 1, with extended precision*/ | |
3853 a1 = (W1*b[8*1]+W7*b[8*7] + 4)>>3; | |
3854 a7 = (W7*b[8*1]-W1*b[8*7] + 4)>>3; | |
3855 a5 = (W5*b[8*5]+W3*b[8*3] + 4)>>3; | |
3856 a3 = (W3*b[8*5]-W5*b[8*3] + 4)>>3; | |
3857 a2 = (W2*b[8*2]+W6*b[8*6] + 4)>>3; | |
3858 a6 = (W6*b[8*2]-W2*b[8*6] + 4)>>3; | |
3859 a0 = (W0*b[8*0]+W0*b[8*4] )>>3; | |
3860 a4 = (W0*b[8*0]-W0*b[8*4] )>>3; | |
3861 /*step 2*/ | |
3862 s1 = (181*(a1-a5+a7-a3)+128)>>8; | |
3863 s2 = (181*(a1-a5-a7+a3)+128)>>8; | |
3864 /*step 3*/ | |
3865 b[8*0] = (a0+a2+a1+a5 + (1<<13))>>14; | |
3866 b[8*1] = (a4+a6 +s1 + (1<<13))>>14; | |
3867 b[8*2] = (a4-a6 +s2 + (1<<13))>>14; | |
3868 b[8*3] = (a0-a2+a7+a3 + (1<<13))>>14; | |
3869 | |
3870 b[8*4] = (a0-a2-a7-a3 + (1<<13))>>14; | |
3871 b[8*5] = (a4-a6 -s2 + (1<<13))>>14; | |
3872 b[8*6] = (a4+a6 -s1 + (1<<13))>>14; | |
3873 b[8*7] = (a0+a2-a1-a5 + (1<<13))>>14; | |
3874 } | |
3875 void ff_wmv2_idct_c(short * block){ | |
3876 int i; | |
3877 | |
3878 for(i=0;i<64;i+=8){ | |
3879 wmv2_idct_row(block+i); | |
3880 } | |
3881 for(i=0;i<8;i++){ | |
3882 wmv2_idct_col(block+i); | |
3883 } | |
3884 } | |
1092 | 3885 /* XXX: those functions should be suppressed ASAP when all IDCTs are |
3886 converted */ | |
5887 | 3887 static void ff_wmv2_idct_put_c(uint8_t *dest, int line_size, DCTELEM *block) |
3888 { | |
3889 ff_wmv2_idct_c(block); | |
3890 put_pixels_clamped_c(block, dest, line_size); | |
3891 } | |
3892 static void ff_wmv2_idct_add_c(uint8_t *dest, int line_size, DCTELEM *block) | |
3893 { | |
3894 ff_wmv2_idct_c(block); | |
3895 add_pixels_clamped_c(block, dest, line_size); | |
3896 } | |
1092 | 3897 static void ff_jref_idct_put(uint8_t *dest, int line_size, DCTELEM *block) |
3898 { | |
3899 j_rev_dct (block); | |
3900 put_pixels_clamped_c(block, dest, line_size); | |
3901 } | |
3902 static void ff_jref_idct_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3903 { | |
3904 j_rev_dct (block); | |
3905 add_pixels_clamped_c(block, dest, line_size); | |
3906 } | |
3907 | |
2256 | 3908 static void ff_jref_idct4_put(uint8_t *dest, int line_size, DCTELEM *block) |
3909 { | |
3910 j_rev_dct4 (block); | |
3911 put_pixels_clamped4_c(block, dest, line_size); | |
3912 } | |
3913 static void ff_jref_idct4_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3914 { | |
3915 j_rev_dct4 (block); | |
3916 add_pixels_clamped4_c(block, dest, line_size); | |
3917 } | |
3918 | |
2257 | 3919 static void ff_jref_idct2_put(uint8_t *dest, int line_size, DCTELEM *block) |
3920 { | |
3921 j_rev_dct2 (block); | |
3922 put_pixels_clamped2_c(block, dest, line_size); | |
3923 } | |
3924 static void ff_jref_idct2_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3925 { | |
3926 j_rev_dct2 (block); | |
3927 add_pixels_clamped2_c(block, dest, line_size); | |
3928 } | |
3929 | |
2259 | 3930 static void ff_jref_idct1_put(uint8_t *dest, int line_size, DCTELEM *block) |
3931 { | |
4176 | 3932 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2259 | 3933 |
3934 dest[0] = cm[(block[0] + 4)>>3]; | |
3935 } | |
3936 static void ff_jref_idct1_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3937 { | |
4176 | 3938 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2259 | 3939 |
3940 dest[0] = cm[dest[0] + ((block[0] + 4)>>3)]; | |
3941 } | |
3942 | |
5143 | 3943 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
|
3944 |
1201 | 3945 /* init static data */ |
4197 | 3946 void dsputil_static_init(void) |
0 | 3947 { |
751 | 3948 int i; |
0 | 3949 |
4176 | 3950 for(i=0;i<256;i++) ff_cropTbl[i + MAX_NEG_CROP] = i; |
1201 | 3951 for(i=0;i<MAX_NEG_CROP;i++) { |
4176 | 3952 ff_cropTbl[i] = 0; |
3953 ff_cropTbl[i + MAX_NEG_CROP + 256] = 255; | |
1201 | 3954 } |
2967 | 3955 |
1201 | 3956 for(i=0;i<512;i++) { |
4179 | 3957 ff_squareTbl[i] = (i - 256) * (i - 256); |
1201 | 3958 } |
2967 | 3959 |
4197 | 3960 for(i=0; i<64; i++) inv_zigzag_direct16[ff_zigzag_direct[i]]= i+1; |
1201 | 3961 } |
0 | 3962 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3963 int ff_check_alignment(void){ |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3964 static int did_fail=0; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3965 DECLARE_ALIGNED_16(int, aligned); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3966 |
5150 | 3967 if((long)&aligned & 15){ |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3968 if(!did_fail){ |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3969 #if defined(HAVE_MMX) || defined(HAVE_ALTIVEC) |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3970 av_log(NULL, AV_LOG_ERROR, |
4292 | 3971 "Compiler did not align stack variables. Libavcodec has been miscompiled\n" |
3972 "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
|
3973 "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
|
3974 "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
|
3975 #endif |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3976 did_fail=1; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3977 } |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3978 return -1; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3979 } |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3980 return 0; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3981 } |
861 | 3982 |
1201 | 3983 void dsputil_init(DSPContext* c, AVCodecContext *avctx) |
3984 { | |
3985 int i; | |
0 | 3986 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3987 ff_check_alignment(); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3988 |
1092 | 3989 #ifdef CONFIG_ENCODERS |
1567 | 3990 if(avctx->dct_algo==FF_DCT_FASTINT) { |
1092 | 3991 c->fdct = fdct_ifast; |
2979 | 3992 c->fdct248 = fdct_ifast248; |
2967 | 3993 } |
1567 | 3994 else if(avctx->dct_algo==FF_DCT_FAAN) { |
1557 | 3995 c->fdct = ff_faandct; |
2979 | 3996 c->fdct248 = ff_faandct248; |
2967 | 3997 } |
1567 | 3998 else { |
1092 | 3999 c->fdct = ff_jpeg_fdct_islow; //slow/accurate/default |
2979 | 4000 c->fdct248 = ff_fdct248_islow; |
1567 | 4001 } |
1092 | 4002 #endif //CONFIG_ENCODERS |
4003 | |
2256 | 4004 if(avctx->lowres==1){ |
5064 | 4005 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
|
4006 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
|
4007 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
|
4008 }else{ |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4009 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
|
4010 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
|
4011 } |
2256 | 4012 c->idct = j_rev_dct4; |
1092 | 4013 c->idct_permutation_type= FF_NO_IDCT_PERM; |
2257 | 4014 }else if(avctx->lowres==2){ |
4015 c->idct_put= ff_jref_idct2_put; | |
4016 c->idct_add= ff_jref_idct2_add; | |
4017 c->idct = j_rev_dct2; | |
4018 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2259 | 4019 }else if(avctx->lowres==3){ |
4020 c->idct_put= ff_jref_idct1_put; | |
4021 c->idct_add= ff_jref_idct1_add; | |
4022 c->idct = j_rev_dct1; | |
4023 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2256 | 4024 }else{ |
4025 if(avctx->idct_algo==FF_IDCT_INT){ | |
4026 c->idct_put= ff_jref_idct_put; | |
4027 c->idct_add= ff_jref_idct_add; | |
4028 c->idct = j_rev_dct; | |
4029 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM; | |
5007 | 4030 }else if((ENABLE_VP3_DECODER || ENABLE_VP5_DECODER || ENABLE_VP6_DECODER || ENABLE_THEORA_DECODER ) && |
4031 avctx->idct_algo==FF_IDCT_VP3){ | |
2693 | 4032 c->idct_put= ff_vp3_idct_put_c; |
4033 c->idct_add= ff_vp3_idct_add_c; | |
4034 c->idct = ff_vp3_idct_c; | |
4035 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
5887 | 4036 }else if(avctx->idct_algo==FF_IDCT_WMV2){ |
4037 c->idct_put= ff_wmv2_idct_put_c; | |
4038 c->idct_add= ff_wmv2_idct_add_c; | |
4039 c->idct = ff_wmv2_idct_c; | |
4040 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2256 | 4041 }else{ //accurate/default |
6001 | 4042 c->idct_put= ff_simple_idct_put; |
4043 c->idct_add= ff_simple_idct_add; | |
4044 c->idct = ff_simple_idct; | |
2256 | 4045 c->idct_permutation_type= FF_NO_IDCT_PERM; |
4046 } | |
1092 | 4047 } |
4048 | |
5064 | 4049 if (ENABLE_H264_DECODER) { |
5065 | 4050 c->h264_idct_add= ff_h264_idct_add_c; |
4051 c->h264_idct8_add= ff_h264_idct8_add_c; | |
4052 c->h264_idct_dc_add= ff_h264_idct_dc_add_c; | |
4053 c->h264_idct8_dc_add= ff_h264_idct8_dc_add_c; | |
5064 | 4054 } |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4055 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4056 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
|
4057 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
|
4058 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
|
4059 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
|
4060 c->add_pixels_clamped = add_pixels_clamped_c; |
2763 | 4061 c->add_pixels8 = add_pixels8_c; |
4062 c->add_pixels4 = add_pixels4_c; | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
4063 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
|
4064 c->gmc1 = gmc1_c; |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
4065 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
|
4066 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
|
4067 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
|
4068 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
|
4069 |
859 | 4070 /* TODO [0] 16 [1] 8 */ |
1708 | 4071 c->pix_abs[0][0] = pix_abs16_c; |
4072 c->pix_abs[0][1] = pix_abs16_x2_c; | |
4073 c->pix_abs[0][2] = pix_abs16_y2_c; | |
4074 c->pix_abs[0][3] = pix_abs16_xy2_c; | |
4075 c->pix_abs[1][0] = pix_abs8_c; | |
4076 c->pix_abs[1][1] = pix_abs8_x2_c; | |
4077 c->pix_abs[1][2] = pix_abs8_y2_c; | |
4078 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
|
4079 |
859 | 4080 #define dspfunc(PFX, IDX, NUM) \ |
4081 c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## NUM ## _c; \ | |
4082 c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## NUM ## _x2_c; \ | |
4083 c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## NUM ## _y2_c; \ | |
4084 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
|
4085 |
859 | 4086 dspfunc(put, 0, 16); |
4087 dspfunc(put_no_rnd, 0, 16); | |
4088 dspfunc(put, 1, 8); | |
4089 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
|
4090 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
|
4091 dspfunc(put, 3, 2); |
0 | 4092 |
859 | 4093 dspfunc(avg, 0, 16); |
4094 dspfunc(avg_no_rnd, 0, 16); | |
4095 dspfunc(avg, 1, 8); | |
4096 dspfunc(avg_no_rnd, 1, 8); | |
1319 | 4097 dspfunc(avg, 2, 4); |
4098 dspfunc(avg, 3, 2); | |
859 | 4099 #undef dspfunc |
857 | 4100 |
1864 | 4101 c->put_no_rnd_pixels_l2[0]= put_no_rnd_pixels16_l2_c; |
4102 c->put_no_rnd_pixels_l2[1]= put_no_rnd_pixels8_l2_c; | |
4103 | |
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
|
4104 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
|
4105 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
|
4106 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
|
4107 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
|
4108 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
|
4109 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
|
4110 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
|
4111 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
|
4112 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
|
4113 |
1319 | 4114 c->avg_tpel_pixels_tab[ 0] = avg_tpel_pixels_mc00_c; |
4115 c->avg_tpel_pixels_tab[ 1] = avg_tpel_pixels_mc10_c; | |
4116 c->avg_tpel_pixels_tab[ 2] = avg_tpel_pixels_mc20_c; | |
4117 c->avg_tpel_pixels_tab[ 4] = avg_tpel_pixels_mc01_c; | |
4118 c->avg_tpel_pixels_tab[ 5] = avg_tpel_pixels_mc11_c; | |
4119 c->avg_tpel_pixels_tab[ 6] = avg_tpel_pixels_mc21_c; | |
4120 c->avg_tpel_pixels_tab[ 8] = avg_tpel_pixels_mc02_c; | |
4121 c->avg_tpel_pixels_tab[ 9] = avg_tpel_pixels_mc12_c; | |
4122 c->avg_tpel_pixels_tab[10] = avg_tpel_pixels_mc22_c; | |
4123 | |
859 | 4124 #define dspfunc(PFX, IDX, NUM) \ |
4125 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_c; \ | |
4126 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_c; \ | |
4127 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_c; \ | |
4128 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_c; \ | |
4129 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_c; \ | |
4130 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_c; \ | |
4131 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_c; \ | |
4132 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_c; \ | |
4133 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_c; \ | |
4134 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_c; \ | |
4135 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_c; \ | |
4136 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_c; \ | |
4137 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_c; \ | |
4138 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_c; \ | |
4139 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_c; \ | |
4140 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_c | |
857 | 4141 |
859 | 4142 dspfunc(put_qpel, 0, 16); |
4143 dspfunc(put_no_rnd_qpel, 0, 16); | |
4144 | |
4145 dspfunc(avg_qpel, 0, 16); | |
4146 /* dspfunc(avg_no_rnd_qpel, 0, 16); */ | |
857 | 4147 |
859 | 4148 dspfunc(put_qpel, 1, 8); |
4149 dspfunc(put_no_rnd_qpel, 1, 8); | |
4150 | |
4151 dspfunc(avg_qpel, 1, 8); | |
4152 /* dspfunc(avg_no_rnd_qpel, 1, 8); */ | |
1168 | 4153 |
4154 dspfunc(put_h264_qpel, 0, 16); | |
4155 dspfunc(put_h264_qpel, 1, 8); | |
4156 dspfunc(put_h264_qpel, 2, 4); | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
4157 dspfunc(put_h264_qpel, 3, 2); |
1168 | 4158 dspfunc(avg_h264_qpel, 0, 16); |
4159 dspfunc(avg_h264_qpel, 1, 8); | |
4160 dspfunc(avg_h264_qpel, 2, 4); | |
4161 | |
859 | 4162 #undef dspfunc |
1168 | 4163 c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_c; |
4164 c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_c; | |
4165 c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_c; | |
4166 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_c; | |
4167 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_c; | |
4168 c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_c; | |
3663 | 4169 c->put_no_rnd_h264_chroma_pixels_tab[0]= put_no_rnd_h264_chroma_mc8_c; |
857 | 4170 |
2415 | 4171 c->weight_h264_pixels_tab[0]= weight_h264_pixels16x16_c; |
4172 c->weight_h264_pixels_tab[1]= weight_h264_pixels16x8_c; | |
4173 c->weight_h264_pixels_tab[2]= weight_h264_pixels8x16_c; | |
4174 c->weight_h264_pixels_tab[3]= weight_h264_pixels8x8_c; | |
4175 c->weight_h264_pixels_tab[4]= weight_h264_pixels8x4_c; | |
4176 c->weight_h264_pixels_tab[5]= weight_h264_pixels4x8_c; | |
4177 c->weight_h264_pixels_tab[6]= weight_h264_pixels4x4_c; | |
4178 c->weight_h264_pixels_tab[7]= weight_h264_pixels4x2_c; | |
4179 c->weight_h264_pixels_tab[8]= weight_h264_pixels2x4_c; | |
4180 c->weight_h264_pixels_tab[9]= weight_h264_pixels2x2_c; | |
4181 c->biweight_h264_pixels_tab[0]= biweight_h264_pixels16x16_c; | |
4182 c->biweight_h264_pixels_tab[1]= biweight_h264_pixels16x8_c; | |
4183 c->biweight_h264_pixels_tab[2]= biweight_h264_pixels8x16_c; | |
4184 c->biweight_h264_pixels_tab[3]= biweight_h264_pixels8x8_c; | |
4185 c->biweight_h264_pixels_tab[4]= biweight_h264_pixels8x4_c; | |
4186 c->biweight_h264_pixels_tab[5]= biweight_h264_pixels4x8_c; | |
4187 c->biweight_h264_pixels_tab[6]= biweight_h264_pixels4x4_c; | |
4188 c->biweight_h264_pixels_tab[7]= biweight_h264_pixels4x2_c; | |
4189 c->biweight_h264_pixels_tab[8]= biweight_h264_pixels2x4_c; | |
4190 c->biweight_h264_pixels_tab[9]= biweight_h264_pixels2x2_c; | |
4191 | |
3432 | 4192 #ifdef CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4193 ff_cavsdsp_init(c,avctx); |
3432 | 4194 #endif |
3526 | 4195 #if defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
4196 ff_vc1dsp_init(c,avctx); | |
4197 #endif | |
5887 | 4198 #if defined(CONFIG_WMV2_DECODER) || defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
4199 ff_intrax8dsp_init(c,avctx); | |
4200 #endif | |
4296 | 4201 #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
|
4202 ff_h264dspenc_init(c,avctx); |
4296 | 4203 #endif |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4204 |
936 | 4205 c->put_mspel_pixels_tab[0]= put_mspel8_mc00_c; |
4206 c->put_mspel_pixels_tab[1]= put_mspel8_mc10_c; | |
4207 c->put_mspel_pixels_tab[2]= put_mspel8_mc20_c; | |
4208 c->put_mspel_pixels_tab[3]= put_mspel8_mc30_c; | |
4209 c->put_mspel_pixels_tab[4]= put_mspel8_mc02_c; | |
4210 c->put_mspel_pixels_tab[5]= put_mspel8_mc12_c; | |
4211 c->put_mspel_pixels_tab[6]= put_mspel8_mc22_c; | |
4212 c->put_mspel_pixels_tab[7]= put_mspel8_mc32_c; | |
2967 | 4213 |
1708 | 4214 #define SET_CMP_FUNC(name) \ |
4215 c->name[0]= name ## 16_c;\ | |
4216 c->name[1]= name ## 8x8_c; | |
2967 | 4217 |
1708 | 4218 SET_CMP_FUNC(hadamard8_diff) |
1729 | 4219 c->hadamard8_diff[4]= hadamard8_intra16_c; |
1708 | 4220 SET_CMP_FUNC(dct_sad) |
2382 | 4221 SET_CMP_FUNC(dct_max) |
3013 | 4222 #ifdef CONFIG_GPL |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
4223 SET_CMP_FUNC(dct264_sad) |
3013 | 4224 #endif |
1708 | 4225 c->sad[0]= pix_abs16_c; |
4226 c->sad[1]= pix_abs8_c; | |
4227 c->sse[0]= sse16_c; | |
4228 c->sse[1]= sse8_c; | |
2184 | 4229 c->sse[2]= sse4_c; |
1708 | 4230 SET_CMP_FUNC(quant_psnr) |
4231 SET_CMP_FUNC(rd) | |
4232 SET_CMP_FUNC(bit) | |
1729 | 4233 c->vsad[0]= vsad16_c; |
4234 c->vsad[4]= vsad_intra16_c; | |
4235 c->vsse[0]= vsse16_c; | |
4236 c->vsse[4]= vsse_intra16_c; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4237 c->nsse[0]= nsse16_c; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4238 c->nsse[1]= nsse8_c; |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
4239 #ifdef CONFIG_SNOW_ENCODER |
2184 | 4240 c->w53[0]= w53_16_c; |
4241 c->w53[1]= w53_8_c; | |
4242 c->w97[0]= w97_16_c; | |
4243 c->w97[1]= w97_8_c; | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
4244 #endif |
2184 | 4245 |
4749 | 4246 c->ssd_int8_vs_int16 = ssd_int8_vs_int16_c; |
4247 | |
866 | 4248 c->add_bytes= add_bytes_c; |
6384 | 4249 c->add_bytes_l2= add_bytes_l2_c; |
866 | 4250 c->diff_bytes= diff_bytes_c; |
1527 | 4251 c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_c; |
1273 | 4252 c->bswap_buf= bswap_buf; |
6384 | 4253 #ifdef CONFIG_PNG_DECODER |
4254 c->add_png_paeth_prediction= ff_add_png_paeth_prediction; | |
4255 #endif | |
2633 | 4256 |
4257 c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_c; | |
4258 c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_c; | |
4259 c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_c; | |
4260 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
|
4261 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
|
4262 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
|
4263 c->h264_loop_filter_strength= NULL; |
2967 | 4264 |
5277
7b3fcb7c61ce
Avoid linking with h263.c functions when the relevant codecs
aurel
parents:
5256
diff
changeset
|
4265 if (ENABLE_ANY_H263) { |
5278 | 4266 c->h263_h_loop_filter= h263_h_loop_filter_c; |
4267 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
|
4268 } |
2967 | 4269 |
2045 | 4270 c->h261_loop_filter= h261_loop_filter_c; |
2967 | 4271 |
1784 | 4272 c->try_8x8basis= try_8x8basis_c; |
4273 c->add_8x8basis= add_8x8basis_c; | |
866 | 4274 |
4598
5111fceeb971
The Snow DEcoding routines should be under CONFIG_SNOW_DECODER.
diego
parents:
4594
diff
changeset
|
4275 #ifdef CONFIG_SNOW_DECODER |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4276 c->vertical_compose97i = ff_snow_vertical_compose97i; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4277 c->horizontal_compose97i = ff_snow_horizontal_compose97i; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4278 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
|
4279 #endif |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4280 |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4281 #ifdef CONFIG_VORBIS_DECODER |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4282 c->vorbis_inverse_coupling = vorbis_inverse_coupling; |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4283 #endif |
5737 | 4284 #ifdef CONFIG_FLAC_ENCODER |
4285 c->flac_compute_autocorr = ff_flac_compute_autocorr; | |
4286 #endif | |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4287 c->vector_fmul = vector_fmul_c; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4288 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
|
4289 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
|
4290 c->float_to_int16 = ff_float_to_int16_c; |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4291 |
3245 | 4292 c->shrink[0]= ff_img_copy_plane; |
4293 c->shrink[1]= ff_shrink22; | |
4294 c->shrink[2]= ff_shrink44; | |
4295 c->shrink[3]= ff_shrink88; | |
4296 | |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4297 c->prefetch= just_return; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4298 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4299 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
|
4300 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
|
4301 |
5146
5db8d038eb1f
use if() instead of #ifdef and ensure all possible optimisations are enabled
aurel
parents:
5143
diff
changeset
|
4302 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
|
4303 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
|
4304 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
|
4305 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
|
4306 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
|
4307 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
|
4308 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
|
4309 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
|
4310 if (ENABLE_BFIN) dsputil_init_bfin (c, avctx); |
1092 | 4311 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4312 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
|
4313 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
|
4314 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
|
4315 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
|
4316 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
|
4317 } |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4318 |
1092 | 4319 switch(c->idct_permutation_type){ |
4320 case FF_NO_IDCT_PERM: | |
4321 for(i=0; i<64; i++) | |
4322 c->idct_permutation[i]= i; | |
4323 break; | |
4324 case FF_LIBMPEG2_IDCT_PERM: | |
4325 for(i=0; i<64; i++) | |
4326 c->idct_permutation[i]= (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2); | |
4327 break; | |
4328 case FF_SIMPLE_IDCT_PERM: | |
4329 for(i=0; i<64; i++) | |
4330 c->idct_permutation[i]= simple_mmx_permutation[i]; | |
4331 break; | |
4332 case FF_TRANSPOSE_IDCT_PERM: | |
4333 for(i=0; i<64; i++) | |
4334 c->idct_permutation[i]= ((i&7)<<3) | (i>>3); | |
4335 break; | |
2696
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4336 case FF_PARTTRANS_IDCT_PERM: |
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4337 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
|
4338 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
|
4339 break; |
1092 | 4340 default: |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1571
diff
changeset
|
4341 av_log(avctx, AV_LOG_ERROR, "Internal error, IDCT permutation not set\n"); |
1092 | 4342 } |
0 | 4343 } |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
220
diff
changeset
|
4344 |