Mercurial > libavcodec.hg
annotate dsputil.c @ 6051:1e3b5597505a libavcodec
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
array read.
author | michael |
---|---|
date | Fri, 21 Dec 2007 10:01:04 +0000 |
parents | f4859c13426b |
children | c90798ac28ee |
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 | |
4176 | 47 uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP] = {0, }; |
4179 | 48 uint32_t ff_squareTbl[512] = {0, }; |
0 | 49 |
1064 | 50 const uint8_t ff_zigzag_direct[64] = { |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
51 0, 1, 8, 16, 9, 2, 3, 10, |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
52 17, 24, 32, 25, 18, 11, 4, 5, |
34 | 53 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
|
54 27, 20, 13, 6, 7, 14, 21, 28, |
34 | 55 35, 42, 49, 56, 57, 50, 43, 36, |
56 29, 22, 15, 23, 30, 37, 44, 51, | |
57 58, 59, 52, 45, 38, 31, 39, 46, | |
58 53, 60, 61, 54, 47, 55, 62, 63 | |
59 }; | |
60 | |
1567 | 61 /* Specific zigzag scan for 248 idct. NOTE that unlike the |
62 specification, we interleave the fields */ | |
63 const uint8_t ff_zigzag248_direct[64] = { | |
64 0, 8, 1, 9, 16, 24, 2, 10, | |
65 17, 25, 32, 40, 48, 56, 33, 41, | |
66 18, 26, 3, 11, 4, 12, 19, 27, | |
67 34, 42, 49, 57, 50, 58, 35, 43, | |
68 20, 28, 5, 13, 6, 14, 21, 29, | |
69 36, 44, 51, 59, 52, 60, 37, 45, | |
70 22, 30, 7, 15, 23, 31, 38, 46, | |
71 53, 61, 54, 62, 39, 47, 55, 63, | |
72 }; | |
73 | |
220 | 74 /* not permutated inverse zigzag_direct + 1 for MMX quantizer */ |
4197 | 75 DECLARE_ALIGNED_8(uint16_t, inv_zigzag_direct16[64]) = {0, }; |
220 | 76 |
1064 | 77 const uint8_t ff_alternate_horizontal_scan[64] = { |
2967 | 78 0, 1, 2, 3, 8, 9, 16, 17, |
34 | 79 10, 11, 4, 5, 6, 7, 15, 14, |
2967 | 80 13, 12, 19, 18, 24, 25, 32, 33, |
34 | 81 26, 27, 20, 21, 22, 23, 28, 29, |
2967 | 82 30, 31, 34, 35, 40, 41, 48, 49, |
34 | 83 42, 43, 36, 37, 38, 39, 44, 45, |
2967 | 84 46, 47, 50, 51, 56, 57, 58, 59, |
34 | 85 52, 53, 54, 55, 60, 61, 62, 63, |
86 }; | |
87 | |
1064 | 88 const uint8_t ff_alternate_vertical_scan[64] = { |
2967 | 89 0, 8, 16, 24, 1, 9, 2, 10, |
34 | 90 17, 25, 32, 40, 48, 56, 57, 49, |
2967 | 91 41, 33, 26, 18, 3, 11, 4, 12, |
34 | 92 19, 27, 34, 42, 50, 58, 35, 43, |
2967 | 93 51, 59, 20, 28, 5, 13, 6, 14, |
34 | 94 21, 29, 36, 44, 52, 60, 37, 45, |
2967 | 95 53, 61, 22, 30, 7, 15, 23, 31, |
34 | 96 38, 46, 54, 62, 39, 47, 55, 63, |
97 }; | |
98 | |
220 | 99 /* a*inverse[b]>>32 == a/b for all 0<=a<=65536 && 2<=b<=255 */ |
4174 | 100 const uint32_t ff_inverse[256]={ |
2967 | 101 0, 4294967295U,2147483648U,1431655766, 1073741824, 858993460, 715827883, 613566757, |
102 536870912, 477218589, 429496730, 390451573, 357913942, 330382100, 306783379, 286331154, | |
103 268435456, 252645136, 238609295, 226050911, 214748365, 204522253, 195225787, 186737709, | |
104 178956971, 171798692, 165191050, 159072863, 153391690, 148102321, 143165577, 138547333, | |
105 134217728, 130150525, 126322568, 122713352, 119304648, 116080198, 113025456, 110127367, | |
106 107374183, 104755300, 102261127, 99882961, 97612894, 95443718, 93368855, 91382283, | |
107 89478486, 87652394, 85899346, 84215046, 82595525, 81037119, 79536432, 78090315, | |
108 76695845, 75350304, 74051161, 72796056, 71582789, 70409300, 69273667, 68174085, | |
109 67108864, 66076420, 65075263, 64103990, 63161284, 62245903, 61356676, 60492498, | |
110 59652324, 58835169, 58040099, 57266231, 56512728, 55778797, 55063684, 54366675, | |
111 53687092, 53024288, 52377650, 51746594, 51130564, 50529028, 49941481, 49367441, | |
112 48806447, 48258060, 47721859, 47197443, 46684428, 46182445, 45691142, 45210183, | |
113 44739243, 44278014, 43826197, 43383509, 42949673, 42524429, 42107523, 41698712, | |
114 41297763, 40904451, 40518560, 40139882, 39768216, 39403370, 39045158, 38693400, | |
115 38347923, 38008561, 37675152, 37347542, 37025581, 36709123, 36398028, 36092163, | |
116 35791395, 35495598, 35204650, 34918434, 34636834, 34359739, 34087043, 33818641, | |
117 33554432, 33294321, 33038210, 32786010, 32537632, 32292988, 32051995, 31814573, | |
118 31580642, 31350127, 31122952, 30899046, 30678338, 30460761, 30246249, 30034737, | |
119 29826162, 29620465, 29417585, 29217465, 29020050, 28825284, 28633116, 28443493, | |
120 28256364, 28071682, 27889399, 27709467, 27531842, 27356480, 27183338, 27012373, | |
121 26843546, 26676816, 26512144, 26349493, 26188825, 26030105, 25873297, 25718368, | |
122 25565282, 25414008, 25264514, 25116768, 24970741, 24826401, 24683721, 24542671, | |
123 24403224, 24265352, 24129030, 23994231, 23860930, 23729102, 23598722, 23469767, | |
124 23342214, 23216040, 23091223, 22967740, 22845571, 22724695, 22605092, 22486740, | |
125 22369622, 22253717, 22139007, 22025474, 21913099, 21801865, 21691755, 21582751, | |
126 21474837, 21367997, 21262215, 21157475, 21053762, 20951060, 20849356, 20748635, | |
127 20648882, 20550083, 20452226, 20355296, 20259280, 20164166, 20069941, 19976593, | |
128 19884108, 19792477, 19701685, 19611723, 19522579, 19434242, 19346700, 19259944, | |
129 19173962, 19088744, 19004281, 18920561, 18837576, 18755316, 18673771, 18592933, | |
130 18512791, 18433337, 18354562, 18276457, 18199014, 18122225, 18046082, 17970575, | |
131 17895698, 17821442, 17747799, 17674763, 17602325, 17530479, 17459217, 17388532, | |
220 | 132 17318417, 17248865, 17179870, 17111424, 17043522, 16976156, 16909321, 16843010, |
133 }; | |
134 | |
1092 | 135 /* Input permutation for the simple_idct_mmx */ |
136 static const uint8_t simple_mmx_permutation[64]={ | |
2979 | 137 0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, |
138 0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, | |
139 0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, | |
140 0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, | |
141 0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, | |
142 0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, | |
143 0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, | |
144 0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F, | |
1092 | 145 }; |
146 | |
1064 | 147 static int pix_sum_c(uint8_t * pix, int line_size) |
612 | 148 { |
149 int s, i, j; | |
150 | |
151 s = 0; | |
152 for (i = 0; i < 16; i++) { | |
2979 | 153 for (j = 0; j < 16; j += 8) { |
154 s += pix[0]; | |
155 s += pix[1]; | |
156 s += pix[2]; | |
157 s += pix[3]; | |
158 s += pix[4]; | |
159 s += pix[5]; | |
160 s += pix[6]; | |
161 s += pix[7]; | |
162 pix += 8; | |
163 } | |
164 pix += line_size - 16; | |
612 | 165 } |
166 return s; | |
167 } | |
168 | |
1064 | 169 static int pix_norm1_c(uint8_t * pix, int line_size) |
612 | 170 { |
171 int s, i, j; | |
4179 | 172 uint32_t *sq = ff_squareTbl + 256; |
612 | 173 |
174 s = 0; | |
175 for (i = 0; i < 16; i++) { | |
2979 | 176 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
|
177 #if 0 |
2979 | 178 s += sq[pix[0]]; |
179 s += sq[pix[1]]; | |
180 s += sq[pix[2]]; | |
181 s += sq[pix[3]]; | |
182 s += sq[pix[4]]; | |
183 s += sq[pix[5]]; | |
184 s += sq[pix[6]]; | |
185 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
|
186 #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
|
187 #if LONG_MAX > 2147483647 |
2979 | 188 register uint64_t x=*(uint64_t*)pix; |
189 s += sq[x&0xff]; | |
190 s += sq[(x>>8)&0xff]; | |
191 s += sq[(x>>16)&0xff]; | |
192 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
|
193 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
|
194 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
|
195 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
|
196 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
|
197 #else |
2979 | 198 register uint32_t x=*(uint32_t*)pix; |
199 s += sq[x&0xff]; | |
200 s += sq[(x>>8)&0xff]; | |
201 s += sq[(x>>16)&0xff]; | |
202 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
|
203 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
|
204 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
|
205 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
|
206 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
|
207 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
|
208 #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
|
209 #endif |
2979 | 210 pix += 8; |
211 } | |
212 pix += line_size - 16; | |
612 | 213 } |
214 return s; | |
215 } | |
216 | |
1273 | 217 static void bswap_buf(uint32_t *dst, uint32_t *src, int w){ |
218 int i; | |
2967 | 219 |
1273 | 220 for(i=0; i+8<=w; i+=8){ |
221 dst[i+0]= bswap_32(src[i+0]); | |
222 dst[i+1]= bswap_32(src[i+1]); | |
223 dst[i+2]= bswap_32(src[i+2]); | |
224 dst[i+3]= bswap_32(src[i+3]); | |
225 dst[i+4]= bswap_32(src[i+4]); | |
226 dst[i+5]= bswap_32(src[i+5]); | |
227 dst[i+6]= bswap_32(src[i+6]); | |
228 dst[i+7]= bswap_32(src[i+7]); | |
229 } | |
230 for(;i<w; i++){ | |
231 dst[i+0]= bswap_32(src[i+0]); | |
232 } | |
233 } | |
612 | 234 |
2184 | 235 static int sse4_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
236 { | |
237 int s, i; | |
4179 | 238 uint32_t *sq = ff_squareTbl + 256; |
2184 | 239 |
240 s = 0; | |
241 for (i = 0; i < h; i++) { | |
242 s += sq[pix1[0] - pix2[0]]; | |
243 s += sq[pix1[1] - pix2[1]]; | |
244 s += sq[pix1[2] - pix2[2]]; | |
245 s += sq[pix1[3] - pix2[3]]; | |
246 pix1 += line_size; | |
247 pix2 += line_size; | |
248 } | |
249 return s; | |
250 } | |
251 | |
1708 | 252 static int sse8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
936 | 253 { |
254 int s, i; | |
4179 | 255 uint32_t *sq = ff_squareTbl + 256; |
936 | 256 |
257 s = 0; | |
1708 | 258 for (i = 0; i < h; i++) { |
936 | 259 s += sq[pix1[0] - pix2[0]]; |
260 s += sq[pix1[1] - pix2[1]]; | |
261 s += sq[pix1[2] - pix2[2]]; | |
262 s += sq[pix1[3] - pix2[3]]; | |
263 s += sq[pix1[4] - pix2[4]]; | |
264 s += sq[pix1[5] - pix2[5]]; | |
265 s += sq[pix1[6] - pix2[6]]; | |
266 s += sq[pix1[7] - pix2[7]]; | |
267 pix1 += line_size; | |
268 pix2 += line_size; | |
269 } | |
270 return s; | |
271 } | |
272 | |
1708 | 273 static int sse16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
884 | 274 { |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
275 int s, i; |
4179 | 276 uint32_t *sq = ff_squareTbl + 256; |
884 | 277 |
278 s = 0; | |
1708 | 279 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
|
280 s += sq[pix1[ 0] - pix2[ 0]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
281 s += sq[pix1[ 1] - pix2[ 1]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
282 s += sq[pix1[ 2] - pix2[ 2]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
283 s += sq[pix1[ 3] - pix2[ 3]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
284 s += sq[pix1[ 4] - pix2[ 4]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
285 s += sq[pix1[ 5] - pix2[ 5]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
286 s += sq[pix1[ 6] - pix2[ 6]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
287 s += sq[pix1[ 7] - pix2[ 7]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
288 s += sq[pix1[ 8] - pix2[ 8]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
289 s += sq[pix1[ 9] - pix2[ 9]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
290 s += sq[pix1[10] - pix2[10]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
291 s += sq[pix1[11] - pix2[11]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
292 s += sq[pix1[12] - pix2[12]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
293 s += sq[pix1[13] - pix2[13]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
294 s += sq[pix1[14] - pix2[14]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
295 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
|
296 |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
297 pix1 += line_size; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
298 pix2 += line_size; |
884 | 299 } |
300 return s; | |
301 } | |
302 | |
2184 | 303 |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
304 #ifdef CONFIG_SNOW_ENCODER //dwt is in snow.c |
2184 | 305 static inline int w_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int w, int h, int type){ |
306 int s, i, j; | |
307 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
|
308 int tmp[32*32]; |
2184 | 309 int level, ori; |
2967 | 310 static const int scale[2][2][4][4]={ |
2184 | 311 { |
312 { | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
313 // 9/7 8x8 dec=3 |
2184 | 314 {268, 239, 239, 213}, |
315 { 0, 224, 224, 152}, | |
316 { 0, 135, 135, 110}, | |
317 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
318 // 9/7 16x16 or 32x32 dec=4 |
2184 | 319 {344, 310, 310, 280}, |
320 { 0, 320, 320, 228}, | |
321 { 0, 175, 175, 136}, | |
322 { 0, 129, 129, 102}, | |
323 } | |
324 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
325 { |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
326 // 5/3 8x8 dec=3 |
2184 | 327 {275, 245, 245, 218}, |
328 { 0, 230, 230, 156}, | |
329 { 0, 138, 138, 113}, | |
330 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
331 // 5/3 16x16 or 32x32 dec=4 |
2184 | 332 {352, 317, 317, 286}, |
333 { 0, 328, 328, 233}, | |
334 { 0, 180, 180, 140}, | |
335 { 0, 132, 132, 105}, | |
336 } | |
337 } | |
338 }; | |
339 | |
340 for (i = 0; i < h; i++) { | |
341 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
|
342 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
|
343 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
|
344 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
|
345 tmp[32*i+j+3] = (pix1[j+3] - pix2[j+3])<<4; |
2184 | 346 } |
347 pix1 += line_size; | |
348 pix2 += line_size; | |
349 } | |
2639 | 350 |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
351 ff_spatial_dwt(tmp, w, h, 32, type, dec_count); |
2184 | 352 |
353 s=0; | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
354 assert(w==h); |
2184 | 355 for(level=0; level<dec_count; level++){ |
356 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
|
357 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
|
358 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
|
359 int stride= 32<<(dec_count-level); |
2184 | 360 int sy= (ori&2) ? stride>>1 : 0; |
2967 | 361 |
2184 | 362 for(i=0; i<size; i++){ |
363 for(j=0; j<size; j++){ | |
364 int v= tmp[sx + sy + i*stride + j] * scale[type][dec_count-3][level][ori]; | |
4001 | 365 s += FFABS(v); |
2184 | 366 } |
367 } | |
368 } | |
369 } | |
2967 | 370 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
|
371 return s>>9; |
2184 | 372 } |
373 | |
374 static int w53_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
375 return w_c(v, pix1, pix2, line_size, 8, h, 1); | |
376 } | |
377 | |
378 static int w97_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
379 return w_c(v, pix1, pix2, line_size, 8, h, 0); | |
380 } | |
381 | |
382 static int w53_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
383 return w_c(v, pix1, pix2, line_size, 16, h, 1); | |
384 } | |
385 | |
386 static int w97_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
387 return w_c(v, pix1, pix2, line_size, 16, h, 0); | |
388 } | |
389 | |
4197 | 390 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
|
391 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
|
392 } |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
393 |
4197 | 394 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
|
395 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
|
396 } |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
397 #endif |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
398 |
1064 | 399 static void get_pixels_c(DCTELEM *restrict block, const uint8_t *pixels, int line_size) |
0 | 400 { |
401 int i; | |
402 | |
403 /* read the pixels */ | |
404 for(i=0;i<8;i++) { | |
516 | 405 block[0] = pixels[0]; |
406 block[1] = pixels[1]; | |
407 block[2] = pixels[2]; | |
408 block[3] = pixels[3]; | |
409 block[4] = pixels[4]; | |
410 block[5] = pixels[5]; | |
411 block[6] = pixels[6]; | |
412 block[7] = pixels[7]; | |
413 pixels += line_size; | |
414 block += 8; | |
0 | 415 } |
416 } | |
417 | |
1064 | 418 static void diff_pixels_c(DCTELEM *restrict block, const uint8_t *s1, |
2979 | 419 const uint8_t *s2, int stride){ |
324 | 420 int i; |
421 | |
422 /* read the pixels */ | |
423 for(i=0;i<8;i++) { | |
516 | 424 block[0] = s1[0] - s2[0]; |
425 block[1] = s1[1] - s2[1]; | |
426 block[2] = s1[2] - s2[2]; | |
427 block[3] = s1[3] - s2[3]; | |
428 block[4] = s1[4] - s2[4]; | |
429 block[5] = s1[5] - s2[5]; | |
430 block[6] = s1[6] - s2[6]; | |
431 block[7] = s1[7] - s2[7]; | |
324 | 432 s1 += stride; |
433 s2 += stride; | |
516 | 434 block += 8; |
324 | 435 } |
436 } | |
437 | |
438 | |
1064 | 439 static void put_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 440 int line_size) |
0 | 441 { |
442 int i; | |
4176 | 443 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 444 |
0 | 445 /* read the pixels */ |
446 for(i=0;i<8;i++) { | |
516 | 447 pixels[0] = cm[block[0]]; |
448 pixels[1] = cm[block[1]]; | |
449 pixels[2] = cm[block[2]]; | |
450 pixels[3] = cm[block[3]]; | |
451 pixels[4] = cm[block[4]]; | |
452 pixels[5] = cm[block[5]]; | |
453 pixels[6] = cm[block[6]]; | |
454 pixels[7] = cm[block[7]]; | |
455 | |
456 pixels += line_size; | |
457 block += 8; | |
0 | 458 } |
459 } | |
460 | |
2256 | 461 static void put_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 462 int line_size) |
2256 | 463 { |
464 int i; | |
4176 | 465 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 466 |
2256 | 467 /* read the pixels */ |
468 for(i=0;i<4;i++) { | |
469 pixels[0] = cm[block[0]]; | |
470 pixels[1] = cm[block[1]]; | |
471 pixels[2] = cm[block[2]]; | |
472 pixels[3] = cm[block[3]]; | |
473 | |
474 pixels += line_size; | |
475 block += 8; | |
476 } | |
477 } | |
478 | |
2257 | 479 static void put_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 480 int line_size) |
2257 | 481 { |
482 int i; | |
4176 | 483 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 484 |
2257 | 485 /* read the pixels */ |
486 for(i=0;i<2;i++) { | |
487 pixels[0] = cm[block[0]]; | |
488 pixels[1] = cm[block[1]]; | |
489 | |
490 pixels += line_size; | |
491 block += 8; | |
492 } | |
493 } | |
494 | |
2967 | 495 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
|
496 uint8_t *restrict pixels, |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
497 int line_size) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
498 { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
499 int i, j; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
500 |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
501 for (i = 0; i < 8; i++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
502 for (j = 0; j < 8; j++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
503 if (*block < -128) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
504 *pixels = 0; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
505 else if (*block > 127) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
506 *pixels = 255; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
507 else |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
508 *pixels = (uint8_t)(*block + 128); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
509 block++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
510 pixels++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
511 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
512 pixels += (line_size - 8); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
513 } |
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 |
1064 | 516 static void add_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
516 | 517 int line_size) |
0 | 518 { |
519 int i; | |
4176 | 520 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 521 |
0 | 522 /* read the pixels */ |
523 for(i=0;i<8;i++) { | |
516 | 524 pixels[0] = cm[pixels[0] + block[0]]; |
525 pixels[1] = cm[pixels[1] + block[1]]; | |
526 pixels[2] = cm[pixels[2] + block[2]]; | |
527 pixels[3] = cm[pixels[3] + block[3]]; | |
528 pixels[4] = cm[pixels[4] + block[4]]; | |
529 pixels[5] = cm[pixels[5] + block[5]]; | |
530 pixels[6] = cm[pixels[6] + block[6]]; | |
531 pixels[7] = cm[pixels[7] + block[7]]; | |
532 pixels += line_size; | |
533 block += 8; | |
0 | 534 } |
535 } | |
2256 | 536 |
537 static void add_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, | |
538 int line_size) | |
539 { | |
540 int i; | |
4176 | 541 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 542 |
2256 | 543 /* read the pixels */ |
544 for(i=0;i<4;i++) { | |
545 pixels[0] = cm[pixels[0] + block[0]]; | |
546 pixels[1] = cm[pixels[1] + block[1]]; | |
547 pixels[2] = cm[pixels[2] + block[2]]; | |
548 pixels[3] = cm[pixels[3] + block[3]]; | |
549 pixels += line_size; | |
550 block += 8; | |
551 } | |
552 } | |
2257 | 553 |
554 static void add_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, | |
555 int line_size) | |
556 { | |
557 int i; | |
4176 | 558 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 559 |
2257 | 560 /* read the pixels */ |
561 for(i=0;i<2;i++) { | |
562 pixels[0] = cm[pixels[0] + block[0]]; | |
563 pixels[1] = cm[pixels[1] + block[1]]; | |
564 pixels += line_size; | |
565 block += 8; | |
566 } | |
567 } | |
2763 | 568 |
569 static void add_pixels8_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
570 { | |
571 int i; | |
572 for(i=0;i<8;i++) { | |
573 pixels[0] += block[0]; | |
574 pixels[1] += block[1]; | |
575 pixels[2] += block[2]; | |
576 pixels[3] += block[3]; | |
577 pixels[4] += block[4]; | |
578 pixels[5] += block[5]; | |
579 pixels[6] += block[6]; | |
580 pixels[7] += block[7]; | |
581 pixels += line_size; | |
582 block += 8; | |
583 } | |
584 } | |
585 | |
586 static void add_pixels4_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
587 { | |
588 int i; | |
589 for(i=0;i<4;i++) { | |
590 pixels[0] += block[0]; | |
591 pixels[1] += block[1]; | |
592 pixels[2] += block[2]; | |
593 pixels[3] += block[3]; | |
594 pixels += line_size; | |
595 block += 4; | |
596 } | |
597 } | |
598 | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
599 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
|
600 { |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
601 int sum=0, i; |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
602 for(i=0; i<64; i++) |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
603 sum+= FFABS(block[i]); |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
604 return sum; |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
605 } |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
606 |
385 | 607 #if 0 |
608 | |
609 #define PIXOP2(OPNAME, OP) \ | |
651 | 610 static void OPNAME ## _pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 611 {\ |
612 int i;\ | |
613 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
|
614 OP(*((uint64_t*)block), AV_RN64(pixels));\ |
385 | 615 pixels+=line_size;\ |
616 block +=line_size;\ | |
617 }\ | |
618 }\ | |
619 \ | |
859 | 620 static void OPNAME ## _no_rnd_pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 621 {\ |
622 int i;\ | |
623 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
|
624 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
|
625 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 626 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
627 pixels+=line_size;\ | |
628 block +=line_size;\ | |
629 }\ | |
630 }\ | |
631 \ | |
859 | 632 static void OPNAME ## _pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 633 {\ |
634 int i;\ | |
635 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
|
636 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
|
637 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 638 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
639 pixels+=line_size;\ | |
640 block +=line_size;\ | |
641 }\ | |
642 }\ | |
643 \ | |
859 | 644 static void OPNAME ## _no_rnd_pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 645 {\ |
646 int i;\ | |
647 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
|
648 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
|
649 const uint64_t b= AV_RN64(pixels+line_size);\ |
385 | 650 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
651 pixels+=line_size;\ | |
652 block +=line_size;\ | |
653 }\ | |
654 }\ | |
655 \ | |
859 | 656 static void OPNAME ## _pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 657 {\ |
658 int i;\ | |
659 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
|
660 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
|
661 const uint64_t b= AV_RN64(pixels+line_size);\ |
385 | 662 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
663 pixels+=line_size;\ | |
664 block +=line_size;\ | |
665 }\ | |
666 }\ | |
667 \ | |
859 | 668 static void OPNAME ## _pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 669 {\ |
670 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
671 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
|
672 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 673 uint64_t l0= (a&0x0303030303030303ULL)\ |
674 + (b&0x0303030303030303ULL)\ | |
675 + 0x0202020202020202ULL;\ | |
676 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
677 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
678 uint64_t l1,h1;\ | |
679 \ | |
680 pixels+=line_size;\ | |
681 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
|
682 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
|
683 uint64_t b= AV_RN64(pixels+1);\ |
385 | 684 l1= (a&0x0303030303030303ULL)\ |
685 + (b&0x0303030303030303ULL);\ | |
686 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
687 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
688 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
689 pixels+=line_size;\ | |
690 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
691 a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
692 b= AV_RN64(pixels+1);\ |
385 | 693 l0= (a&0x0303030303030303ULL)\ |
694 + (b&0x0303030303030303ULL)\ | |
695 + 0x0202020202020202ULL;\ | |
696 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
697 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
698 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
699 pixels+=line_size;\ | |
700 block +=line_size;\ | |
701 }\ | |
702 }\ | |
703 \ | |
859 | 704 static void OPNAME ## _no_rnd_pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 705 {\ |
706 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
707 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
|
708 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 709 uint64_t l0= (a&0x0303030303030303ULL)\ |
710 + (b&0x0303030303030303ULL)\ | |
711 + 0x0101010101010101ULL;\ | |
712 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
713 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
714 uint64_t l1,h1;\ | |
715 \ | |
716 pixels+=line_size;\ | |
717 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
|
718 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
|
719 uint64_t b= AV_RN64(pixels+1);\ |
385 | 720 l1= (a&0x0303030303030303ULL)\ |
721 + (b&0x0303030303030303ULL);\ | |
722 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
723 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
724 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
725 pixels+=line_size;\ | |
726 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
727 a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
728 b= AV_RN64(pixels+1);\ |
385 | 729 l0= (a&0x0303030303030303ULL)\ |
730 + (b&0x0303030303030303ULL)\ | |
731 + 0x0101010101010101ULL;\ | |
732 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
733 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
734 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
735 pixels+=line_size;\ | |
736 block +=line_size;\ | |
737 }\ | |
738 }\ | |
739 \ | |
859 | 740 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels_c , 8)\ |
741 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels_x2_c , 8)\ | |
742 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels_y2_c , 8)\ | |
743 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels_xy2_c, 8)\ | |
744 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels_x2_c , 8)\ | |
745 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels_y2_c , 8)\ | |
746 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels_xy2_c, 8) | |
385 | 747 |
748 #define op_avg(a, b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEFEFEFEFEULL)>>1) ) | |
749 #else // 64 bit variant | |
750 | |
751 #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
|
752 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
|
753 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
|
754 for(i=0; i<h; i++){\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
755 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
|
756 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
|
757 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
|
758 }\ |
85b71f9f7450
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 }\ |
1168 | 760 static void OPNAME ## _pixels4_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
761 int i;\ | |
762 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
|
763 OP(*((uint32_t*)(block )), AV_RN32(pixels ));\ |
1168 | 764 pixels+=line_size;\ |
765 block +=line_size;\ | |
766 }\ | |
767 }\ | |
859 | 768 static void OPNAME ## _pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
385 | 769 int i;\ |
770 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
|
771 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
|
772 OP(*((uint32_t*)(block+4)), AV_RN32(pixels+4));\ |
385 | 773 pixels+=line_size;\ |
774 block +=line_size;\ | |
775 }\ | |
776 }\ | |
859 | 777 static inline void OPNAME ## _no_rnd_pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
778 OPNAME ## _pixels8_c(block, pixels, line_size, h);\ | |
651 | 779 }\ |
385 | 780 \ |
651 | 781 static inline void OPNAME ## _no_rnd_pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
782 int src_stride1, int src_stride2, int h){\ | |
385 | 783 int i;\ |
784 for(i=0; i<h; i++){\ | |
651 | 785 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
786 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
|
787 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 788 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
|
789 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
|
790 b= AV_RN32(&src2[i*src_stride2+4]);\ |
1264 | 791 OP(*((uint32_t*)&dst[i*dst_stride+4]), no_rnd_avg32(a, b));\ |
385 | 792 }\ |
793 }\ | |
794 \ | |
651 | 795 static inline void OPNAME ## _pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
796 int src_stride1, int src_stride2, int h){\ | |
385 | 797 int i;\ |
798 for(i=0; i<h; i++){\ | |
651 | 799 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
800 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
|
801 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 802 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
|
803 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
|
804 b= AV_RN32(&src2[i*src_stride2+4]);\ |
1264 | 805 OP(*((uint32_t*)&dst[i*dst_stride+4]), rnd_avg32(a, b));\ |
385 | 806 }\ |
807 }\ | |
808 \ | |
1168 | 809 static inline void OPNAME ## _pixels4_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
810 int src_stride1, int src_stride2, int h){\ | |
811 int i;\ | |
812 for(i=0; i<h; i++){\ | |
813 uint32_t a,b;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
814 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
|
815 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 816 OP(*((uint32_t*)&dst[i*dst_stride ]), rnd_avg32(a, b));\ |
1168 | 817 }\ |
818 }\ | |
819 \ | |
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
|
820 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
|
821 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
|
822 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
|
823 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
|
824 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
825 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
|
826 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
|
827 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
|
828 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
829 }\ |
85b71f9f7450
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 \ |
651 | 831 static inline void OPNAME ## _pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
832 int src_stride1, int src_stride2, int h){\ | |
833 OPNAME ## _pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
834 OPNAME ## _pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
835 }\ | |
836 \ | |
837 static inline void OPNAME ## _no_rnd_pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ | |
838 int src_stride1, int src_stride2, int h){\ | |
839 OPNAME ## _no_rnd_pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
840 OPNAME ## _no_rnd_pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
841 }\ | |
842 \ | |
859 | 843 static inline void OPNAME ## _no_rnd_pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 844 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
845 }\ | |
846 \ | |
859 | 847 static inline void OPNAME ## _pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 848 OPNAME ## _pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
849 }\ | |
850 \ | |
859 | 851 static inline void OPNAME ## _no_rnd_pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 852 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
853 }\ | |
854 \ | |
859 | 855 static inline void OPNAME ## _pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 856 OPNAME ## _pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
385 | 857 }\ |
858 \ | |
651 | 859 static inline void OPNAME ## _pixels8_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
860 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
861 int i;\ | |
862 for(i=0; i<h; i++){\ | |
863 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
|
864 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
|
865 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
|
866 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
|
867 d= AV_RN32(&src4[i*src_stride4]);\ |
651 | 868 l0= (a&0x03030303UL)\ |
869 + (b&0x03030303UL)\ | |
870 + 0x02020202UL;\ | |
871 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
872 + ((b&0xFCFCFCFCUL)>>2);\ | |
873 l1= (c&0x03030303UL)\ | |
874 + (d&0x03030303UL);\ | |
875 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
876 + ((d&0xFCFCFCFCUL)>>2);\ | |
877 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
|
878 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
|
879 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
|
880 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
|
881 d= AV_RN32(&src4[i*src_stride4+4]);\ |
651 | 882 l0= (a&0x03030303UL)\ |
883 + (b&0x03030303UL)\ | |
884 + 0x02020202UL;\ | |
885 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
886 + ((b&0xFCFCFCFCUL)>>2);\ | |
887 l1= (c&0x03030303UL)\ | |
888 + (d&0x03030303UL);\ | |
889 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
890 + ((d&0xFCFCFCFCUL)>>2);\ | |
891 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
892 }\ | |
893 }\ | |
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
|
894 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
895 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
|
896 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
|
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 \ |
85b71f9f7450
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 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
|
900 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
|
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 \ |
85b71f9f7450
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 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
|
904 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
|
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 \ |
85b71f9f7450
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 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
|
908 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
|
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 \ |
651 | 911 static inline void OPNAME ## _no_rnd_pixels8_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
912 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
385 | 913 int i;\ |
914 for(i=0; i<h; i++){\ | |
651 | 915 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
|
916 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
|
917 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
|
918 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
|
919 d= AV_RN32(&src4[i*src_stride4]);\ |
651 | 920 l0= (a&0x03030303UL)\ |
921 + (b&0x03030303UL)\ | |
922 + 0x01010101UL;\ | |
923 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
924 + ((b&0xFCFCFCFCUL)>>2);\ | |
925 l1= (c&0x03030303UL)\ | |
926 + (d&0x03030303UL);\ | |
927 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
928 + ((d&0xFCFCFCFCUL)>>2);\ | |
929 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
|
930 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
|
931 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
|
932 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
|
933 d= AV_RN32(&src4[i*src_stride4+4]);\ |
651 | 934 l0= (a&0x03030303UL)\ |
935 + (b&0x03030303UL)\ | |
936 + 0x01010101UL;\ | |
937 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
938 + ((b&0xFCFCFCFCUL)>>2);\ | |
939 l1= (c&0x03030303UL)\ | |
940 + (d&0x03030303UL);\ | |
941 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
942 + ((d&0xFCFCFCFCUL)>>2);\ | |
943 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
385 | 944 }\ |
945 }\ | |
651 | 946 static inline void OPNAME ## _pixels16_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
947 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
948 OPNAME ## _pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
949 OPNAME ## _pixels8_l4(dst+8, src1+8, src2+8, src3+8, src4+8, dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
950 }\ | |
951 static inline void OPNAME ## _no_rnd_pixels16_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ | |
952 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
953 OPNAME ## _no_rnd_pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
954 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);\ | |
955 }\ | |
385 | 956 \ |
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
|
957 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
|
958 {\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
959 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
|
960 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
|
961 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
|
962 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
|
963 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
|
964 \ |
85b71f9f7450
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 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
|
966 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
|
967 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
|
968 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
|
969 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
|
970 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
|
971 \ |
85b71f9f7450
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 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
|
973 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
|
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 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
|
976 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
|
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 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
|
979 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
|
980 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
|
981 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
|
982 \ |
85b71f9f7450
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 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
|
984 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
|
985 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
|
986 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
|
987 }\ |
85b71f9f7450
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 }\ |
85b71f9f7450
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 \ |
85b71f9f7450
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 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
|
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 int i;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
993 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
|
994 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
|
995 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
|
996 + (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
|
997 + 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
|
998 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
|
999 + ((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
|
1000 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
|
1001 \ |
85b71f9f7450
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 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
|
1003 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
|
1004 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
|
1005 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
|
1006 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
|
1007 + (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
|
1008 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
|
1009 + ((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
|
1010 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
|
1011 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
|
1012 block +=line_size;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1013 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1014 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
|
1015 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
|
1016 + (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
|
1017 + 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
|
1018 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
|
1019 + ((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
|
1020 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
|
1021 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
|
1022 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
|
1023 }\ |
85b71f9f7450
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 }\ |
85b71f9f7450
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 \ |
859 | 1026 static inline void OPNAME ## _pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1027 {\ |
1028 int j;\ | |
1029 for(j=0; j<2; j++){\ | |
1030 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1031 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
|
1032 const uint32_t b= AV_RN32(pixels+1);\ |
385 | 1033 uint32_t l0= (a&0x03030303UL)\ |
1034 + (b&0x03030303UL)\ | |
1035 + 0x02020202UL;\ | |
1036 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1037 + ((b&0xFCFCFCFCUL)>>2);\ | |
1038 uint32_t l1,h1;\ | |
1039 \ | |
1040 pixels+=line_size;\ | |
1041 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
|
1042 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
|
1043 uint32_t b= AV_RN32(pixels+1);\ |
385 | 1044 l1= (a&0x03030303UL)\ |
1045 + (b&0x03030303UL);\ | |
1046 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1047 + ((b&0xFCFCFCFCUL)>>2);\ | |
1048 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1049 pixels+=line_size;\ | |
1050 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1051 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1052 b= AV_RN32(pixels+1);\ |
385 | 1053 l0= (a&0x03030303UL)\ |
1054 + (b&0x03030303UL)\ | |
1055 + 0x02020202UL;\ | |
1056 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1057 + ((b&0xFCFCFCFCUL)>>2);\ | |
1058 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1059 pixels+=line_size;\ | |
1060 block +=line_size;\ | |
1061 }\ | |
1062 pixels+=4-line_size*(h+1);\ | |
1063 block +=4-line_size*h;\ | |
1064 }\ | |
1065 }\ | |
1066 \ | |
859 | 1067 static inline void OPNAME ## _no_rnd_pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1068 {\ |
1069 int j;\ | |
1070 for(j=0; j<2; j++){\ | |
1071 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1072 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
|
1073 const uint32_t b= AV_RN32(pixels+1);\ |
385 | 1074 uint32_t l0= (a&0x03030303UL)\ |
1075 + (b&0x03030303UL)\ | |
1076 + 0x01010101UL;\ | |
1077 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1078 + ((b&0xFCFCFCFCUL)>>2);\ | |
1079 uint32_t l1,h1;\ | |
1080 \ | |
1081 pixels+=line_size;\ | |
1082 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
|
1083 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
|
1084 uint32_t b= AV_RN32(pixels+1);\ |
385 | 1085 l1= (a&0x03030303UL)\ |
1086 + (b&0x03030303UL);\ | |
1087 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1088 + ((b&0xFCFCFCFCUL)>>2);\ | |
1089 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1090 pixels+=line_size;\ | |
1091 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1092 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1093 b= AV_RN32(pixels+1);\ |
385 | 1094 l0= (a&0x03030303UL)\ |
1095 + (b&0x03030303UL)\ | |
1096 + 0x01010101UL;\ | |
1097 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1098 + ((b&0xFCFCFCFCUL)>>2);\ | |
1099 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1100 pixels+=line_size;\ | |
1101 block +=line_size;\ | |
1102 }\ | |
1103 pixels+=4-line_size*(h+1);\ | |
1104 block +=4-line_size*h;\ | |
1105 }\ | |
1106 }\ | |
1107 \ | |
859 | 1108 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels8_c , 8)\ |
1109 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels8_x2_c , 8)\ | |
1110 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels8_y2_c , 8)\ | |
1111 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels8_xy2_c, 8)\ | |
1112 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_c , OPNAME ## _pixels8_c , 8)\ | |
1113 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels8_x2_c , 8)\ | |
1114 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels8_y2_c , 8)\ | |
1115 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels8_xy2_c, 8)\ | |
651 | 1116 |
1264 | 1117 #define op_avg(a, b) a = rnd_avg32(a, b) |
385 | 1118 #endif |
1119 #define op_put(a, b) a = b | |
1120 | |
1121 PIXOP2(avg, op_avg) | |
1122 PIXOP2(put, op_put) | |
1123 #undef op_avg | |
1124 #undef op_put | |
1125 | |
0 | 1126 #define avg2(a,b) ((a+b+1)>>1) |
1127 #define avg4(a,b,c,d) ((a+b+c+d+2)>>2) | |
1128 | |
1864 | 1129 static void put_no_rnd_pixels16_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ |
1130 put_no_rnd_pixels16_l2(dst, a, b, stride, stride, stride, h); | |
1131 } | |
1132 | |
1133 static void put_no_rnd_pixels8_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ | |
1134 put_no_rnd_pixels8_l2(dst, a, b, stride, stride, stride, h); | |
1135 } | |
753 | 1136 |
1064 | 1137 static void gmc1_c(uint8_t *dst, uint8_t *src, int stride, int h, int x16, int y16, int rounder) |
255 | 1138 { |
1139 const int A=(16-x16)*(16-y16); | |
1140 const int B=( x16)*(16-y16); | |
1141 const int C=(16-x16)*( y16); | |
1142 const int D=( x16)*( y16); | |
1143 int i; | |
1144 | |
1145 for(i=0; i<h; i++) | |
1146 { | |
651 | 1147 dst[0]= (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + rounder)>>8; |
1148 dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8; | |
1149 dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8; | |
1150 dst[3]= (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + rounder)>>8; | |
1151 dst[4]= (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + rounder)>>8; | |
1152 dst[5]= (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + rounder)>>8; | |
1153 dst[6]= (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + rounder)>>8; | |
1154 dst[7]= (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + rounder)>>8; | |
1155 dst+= stride; | |
1156 src+= stride; | |
255 | 1157 } |
1158 } | |
1159 | |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
1160 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, |
753 | 1161 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height) |
1162 { | |
1163 int y, vx, vy; | |
1164 const int s= 1<<shift; | |
2967 | 1165 |
753 | 1166 width--; |
1167 height--; | |
1168 | |
1169 for(y=0; y<h; y++){ | |
1170 int x; | |
1171 | |
1172 vx= ox; | |
1173 vy= oy; | |
1174 for(x=0; x<8; x++){ //XXX FIXME optimize | |
1175 int src_x, src_y, frac_x, frac_y, index; | |
1176 | |
1177 src_x= vx>>16; | |
1178 src_y= vy>>16; | |
1179 frac_x= src_x&(s-1); | |
1180 frac_y= src_y&(s-1); | |
1181 src_x>>=shift; | |
1182 src_y>>=shift; | |
2967 | 1183 |
753 | 1184 if((unsigned)src_x < width){ |
1185 if((unsigned)src_y < height){ | |
1186 index= src_x + src_y*stride; | |
1187 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) | |
1188 + src[index +1]* frac_x )*(s-frac_y) | |
1189 + ( src[index+stride ]*(s-frac_x) | |
1190 + src[index+stride+1]* frac_x )* frac_y | |
1191 + r)>>(shift*2); | |
1192 }else{ | |
4594 | 1193 index= src_x + av_clip(src_y, 0, height)*stride; |
2967 | 1194 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) |
753 | 1195 + src[index +1]* frac_x )*s |
1196 + r)>>(shift*2); | |
1197 } | |
1198 }else{ | |
1199 if((unsigned)src_y < height){ | |
4594 | 1200 index= av_clip(src_x, 0, width) + src_y*stride; |
2967 | 1201 dst[y*stride + x]= ( ( src[index ]*(s-frac_y) |
753 | 1202 + src[index+stride ]* frac_y )*s |
1203 + r)>>(shift*2); | |
1204 }else{ | |
4594 | 1205 index= av_clip(src_x, 0, width) + av_clip(src_y, 0, height)*stride; |
753 | 1206 dst[y*stride + x]= src[index ]; |
1207 } | |
1208 } | |
2967 | 1209 |
753 | 1210 vx+= dxx; |
1211 vy+= dyx; | |
1212 } | |
1213 ox += dxy; | |
1214 oy += dyy; | |
1215 } | |
1216 } | |
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
|
1217 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1218 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
|
1219 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
|
1220 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
|
1221 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
|
1222 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
|
1223 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
|
1224 } |
85b71f9f7450
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 } |
85b71f9f7450
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 |
85b71f9f7450
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 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
|
1228 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
|
1229 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
|
1230 for (j=0; j < width; j++) { |
2979 | 1231 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
|
1232 } |
85b71f9f7450
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 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
|
1234 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
|
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 } |
85b71f9f7450
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 |
85b71f9f7450
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 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
|
1239 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
|
1240 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
|
1241 for (j=0; j < width; j++) { |
2979 | 1242 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
|
1243 } |
85b71f9f7450
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 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
|
1245 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
|
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 } |
2967 | 1248 |
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
|
1249 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
|
1250 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
|
1251 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
|
1252 for (j=0; j < width; j++) { |
2979 | 1253 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
|
1254 } |
85b71f9f7450
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 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
|
1256 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
|
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 } |
2967 | 1259 |
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
|
1260 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
|
1261 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
|
1262 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
|
1263 for (j=0; j < width; j++) { |
2979 | 1264 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
|
1265 } |
85b71f9f7450
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 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
|
1267 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
|
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 } |
85b71f9f7450
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 |
85b71f9f7450
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 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
|
1272 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
|
1273 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
|
1274 for (j=0; j < width; j++) { |
2979 | 1275 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
|
1276 } |
85b71f9f7450
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 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
|
1278 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
|
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 } |
85b71f9f7450
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 |
85b71f9f7450
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 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
|
1283 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
|
1284 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
|
1285 for (j=0; j < width; j++) { |
2979 | 1286 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
|
1287 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1288 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
|
1289 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
|
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 } |
85b71f9f7450
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 |
85b71f9f7450
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 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
|
1294 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
|
1295 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
|
1296 for (j=0; j < width; j++) { |
2979 | 1297 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
|
1298 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1299 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
|
1300 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
|
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 } |
85b71f9f7450
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 |
85b71f9f7450
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 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
|
1305 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
|
1306 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
|
1307 for (j=0; j < width; j++) { |
2979 | 1308 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
|
1309 } |
85b71f9f7450
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 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
|
1311 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
|
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 } |
1319 | 1314 |
1315 static inline void avg_tpel_pixels_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1316 switch(width){ | |
1317 case 2: avg_pixels2_c (dst, src, stride, height); break; | |
1318 case 4: avg_pixels4_c (dst, src, stride, height); break; | |
1319 case 8: avg_pixels8_c (dst, src, stride, height); break; | |
1320 case 16:avg_pixels16_c(dst, src, stride, height); break; | |
1321 } | |
1322 } | |
1323 | |
1324 static inline void avg_tpel_pixels_mc10_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1325 int i,j; | |
1326 for (i=0; i < height; i++) { | |
1327 for (j=0; j < width; j++) { | |
2979 | 1328 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1329 } |
1330 src += stride; | |
1331 dst += stride; | |
1332 } | |
1333 } | |
1334 | |
1335 static inline void avg_tpel_pixels_mc20_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1336 int i,j; | |
1337 for (i=0; i < height; i++) { | |
1338 for (j=0; j < width; j++) { | |
2979 | 1339 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1340 } |
1341 src += stride; | |
1342 dst += stride; | |
1343 } | |
1344 } | |
2967 | 1345 |
1319 | 1346 static inline void avg_tpel_pixels_mc01_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1347 int i,j; | |
1348 for (i=0; i < height; i++) { | |
1349 for (j=0; j < width; j++) { | |
2979 | 1350 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1351 } |
1352 src += stride; | |
1353 dst += stride; | |
1354 } | |
1355 } | |
2967 | 1356 |
1319 | 1357 static inline void avg_tpel_pixels_mc11_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1358 int i,j; | |
1359 for (i=0; i < height; i++) { | |
1360 for (j=0; j < width; j++) { | |
2979 | 1361 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 | 1362 } |
1363 src += stride; | |
1364 dst += stride; | |
1365 } | |
1366 } | |
1367 | |
1368 static inline void avg_tpel_pixels_mc12_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1369 int i,j; | |
1370 for (i=0; i < height; i++) { | |
1371 for (j=0; j < width; j++) { | |
2979 | 1372 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 | 1373 } |
1374 src += stride; | |
1375 dst += stride; | |
1376 } | |
1377 } | |
1378 | |
1379 static inline void avg_tpel_pixels_mc02_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1380 int i,j; | |
1381 for (i=0; i < height; i++) { | |
1382 for (j=0; j < width; j++) { | |
2979 | 1383 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1384 } |
1385 src += stride; | |
1386 dst += stride; | |
1387 } | |
1388 } | |
1389 | |
1390 static inline void avg_tpel_pixels_mc21_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1391 int i,j; | |
1392 for (i=0; i < height; i++) { | |
1393 for (j=0; j < width; j++) { | |
2979 | 1394 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 | 1395 } |
1396 src += stride; | |
1397 dst += stride; | |
1398 } | |
1399 } | |
1400 | |
1401 static inline void avg_tpel_pixels_mc22_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1402 int i,j; | |
1403 for (i=0; i < height; i++) { | |
1404 for (j=0; j < width; j++) { | |
2979 | 1405 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 | 1406 } |
1407 src += stride; | |
1408 dst += stride; | |
1409 } | |
1410 } | |
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
|
1411 #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
|
1412 #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
|
1413 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
|
1414 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
|
1415 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
|
1416 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
|
1417 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
|
1418 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
|
1419 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
|
1420 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
|
1421 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
|
1422 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
|
1423 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
|
1424 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
|
1425 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
|
1426 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
|
1427 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
|
1428 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
|
1429 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
|
1430 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
|
1431 #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
|
1432 |
1168 | 1433 #define H264_CHROMA_MC(OPNAME, OP)\ |
1434 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){\ | |
1435 const int A=(8-x)*(8-y);\ | |
1436 const int B=( x)*(8-y);\ | |
1437 const int C=(8-x)*( y);\ | |
1438 const int D=( x)*( y);\ | |
1439 int i;\ | |
1440 \ | |
1441 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1442 \ | |
1443 for(i=0; i<h; i++)\ | |
1444 {\ | |
1445 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ | |
1446 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1447 dst+= stride;\ | |
1448 src+= stride;\ | |
1449 }\ | |
1450 }\ | |
1451 \ | |
1452 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){\ | |
1453 const int A=(8-x)*(8-y);\ | |
1454 const int B=( x)*(8-y);\ | |
1455 const int C=(8-x)*( y);\ | |
1456 const int D=( x)*( y);\ | |
1457 int i;\ | |
1458 \ | |
1459 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1460 \ | |
1461 for(i=0; i<h; i++)\ | |
1462 {\ | |
1463 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ | |
1464 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1465 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1466 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1467 dst+= stride;\ | |
1468 src+= stride;\ | |
1469 }\ | |
1470 }\ | |
1471 \ | |
1472 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){\ | |
1473 const int A=(8-x)*(8-y);\ | |
1474 const int B=( x)*(8-y);\ | |
1475 const int C=(8-x)*( y);\ | |
1476 const int D=( x)*( y);\ | |
1477 int i;\ | |
1478 \ | |
1479 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1480 \ | |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1481 if(D){\ |
1168 | 1482 for(i=0; i<h; i++)\ |
1483 {\ | |
1484 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ | |
1485 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1486 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1487 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1488 OP(dst[4], (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5]));\ | |
1489 OP(dst[5], (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6]));\ | |
1490 OP(dst[6], (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7]));\ | |
1491 OP(dst[7], (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8]));\ | |
1492 dst+= stride;\ | |
1493 src+= stride;\ | |
1494 }\ | |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1495 }else{\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1496 const int E= B+C;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1497 const int step= C ? stride : 1;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1498 for(i=0; i<h; i++)\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1499 {\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1500 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
|
1501 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
|
1502 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
|
1503 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
|
1504 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
|
1505 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
|
1506 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
|
1507 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
|
1508 dst+= stride;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1509 src+= stride;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1510 }\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1511 }\ |
1168 | 1512 } |
1513 | |
1514 #define op_avg(a, b) a = (((a)+(((b) + 32)>>6)+1)>>1) | |
1515 #define op_put(a, b) a = (((b) + 32)>>6) | |
1516 | |
1517 H264_CHROMA_MC(put_ , op_put) | |
1518 H264_CHROMA_MC(avg_ , op_avg) | |
1519 #undef op_avg | |
1520 #undef op_put | |
1521 | |
3663 | 1522 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){ |
1523 const int A=(8-x)*(8-y); | |
1524 const int B=( x)*(8-y); | |
1525 const int C=(8-x)*( y); | |
1526 const int D=( x)*( y); | |
1527 int i; | |
1528 | |
1529 assert(x<8 && y<8 && x>=0 && y>=0); | |
1530 | |
1531 for(i=0; i<h; i++) | |
1532 { | |
1533 dst[0] = (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6; | |
1534 dst[1] = (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6; | |
1535 dst[2] = (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6; | |
1536 dst[3] = (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6; | |
1537 dst[4] = (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6; | |
1538 dst[5] = (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6; | |
1539 dst[6] = (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6; | |
1540 dst[7] = (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6; | |
1541 dst+= stride; | |
1542 src+= stride; | |
1543 } | |
1544 } | |
1545 | |
651 | 1546 #define QPEL_MC(r, OPNAME, RND, OP) \ |
1064 | 1547 static void OPNAME ## mpeg4_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
4176 | 1548 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1549 int i;\ |
1550 for(i=0; i<h; i++)\ | |
1551 {\ | |
1552 OP(dst[0], (src[0]+src[1])*20 - (src[0]+src[2])*6 + (src[1]+src[3])*3 - (src[2]+src[4]));\ | |
1553 OP(dst[1], (src[1]+src[2])*20 - (src[0]+src[3])*6 + (src[0]+src[4])*3 - (src[1]+src[5]));\ | |
1554 OP(dst[2], (src[2]+src[3])*20 - (src[1]+src[4])*6 + (src[0]+src[5])*3 - (src[0]+src[6]));\ | |
1555 OP(dst[3], (src[3]+src[4])*20 - (src[2]+src[5])*6 + (src[1]+src[6])*3 - (src[0]+src[7]));\ | |
1556 OP(dst[4], (src[4]+src[5])*20 - (src[3]+src[6])*6 + (src[2]+src[7])*3 - (src[1]+src[8]));\ | |
1557 OP(dst[5], (src[5]+src[6])*20 - (src[4]+src[7])*6 + (src[3]+src[8])*3 - (src[2]+src[8]));\ | |
1558 OP(dst[6], (src[6]+src[7])*20 - (src[5]+src[8])*6 + (src[4]+src[8])*3 - (src[3]+src[7]));\ | |
1559 OP(dst[7], (src[7]+src[8])*20 - (src[6]+src[8])*6 + (src[5]+src[7])*3 - (src[4]+src[6]));\ | |
1560 dst+=dstStride;\ | |
1561 src+=srcStride;\ | |
1562 }\ | |
1563 }\ | |
1564 \ | |
1064 | 1565 static void OPNAME ## mpeg4_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
984 | 1566 const int w=8;\ |
4176 | 1567 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1568 int i;\ |
1569 for(i=0; i<w; i++)\ | |
1570 {\ | |
1571 const int src0= src[0*srcStride];\ | |
1572 const int src1= src[1*srcStride];\ | |
1573 const int src2= src[2*srcStride];\ | |
1574 const int src3= src[3*srcStride];\ | |
1575 const int src4= src[4*srcStride];\ | |
1576 const int src5= src[5*srcStride];\ | |
1577 const int src6= src[6*srcStride];\ | |
1578 const int src7= src[7*srcStride];\ | |
1579 const int src8= src[8*srcStride];\ | |
1580 OP(dst[0*dstStride], (src0+src1)*20 - (src0+src2)*6 + (src1+src3)*3 - (src2+src4));\ | |
1581 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*6 + (src0+src4)*3 - (src1+src5));\ | |
1582 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*6 + (src0+src5)*3 - (src0+src6));\ | |
1583 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*6 + (src1+src6)*3 - (src0+src7));\ | |
1584 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*6 + (src2+src7)*3 - (src1+src8));\ | |
1585 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*6 + (src3+src8)*3 - (src2+src8));\ | |
1586 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*6 + (src4+src8)*3 - (src3+src7));\ | |
1587 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src8)*6 + (src5+src7)*3 - (src4+src6));\ | |
1588 dst++;\ | |
1589 src++;\ | |
1590 }\ | |
1591 }\ | |
1592 \ | |
1064 | 1593 static void OPNAME ## mpeg4_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
4176 | 1594 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1595 int i;\ |
954 | 1596 \ |
651 | 1597 for(i=0; i<h; i++)\ |
1598 {\ | |
1599 OP(dst[ 0], (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]));\ | |
1600 OP(dst[ 1], (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]));\ | |
1601 OP(dst[ 2], (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]));\ | |
1602 OP(dst[ 3], (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]));\ | |
1603 OP(dst[ 4], (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]));\ | |
1604 OP(dst[ 5], (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]));\ | |
1605 OP(dst[ 6], (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]));\ | |
1606 OP(dst[ 7], (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]));\ | |
1607 OP(dst[ 8], (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]));\ | |
1608 OP(dst[ 9], (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]));\ | |
1609 OP(dst[10], (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]));\ | |
1610 OP(dst[11], (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]));\ | |
1611 OP(dst[12], (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]));\ | |
1612 OP(dst[13], (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]));\ | |
1613 OP(dst[14], (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]));\ | |
1614 OP(dst[15], (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]));\ | |
1615 dst+=dstStride;\ | |
1616 src+=srcStride;\ | |
1617 }\ | |
255 | 1618 }\ |
1619 \ | |
1064 | 1620 static void OPNAME ## mpeg4_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
4176 | 1621 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1622 int i;\ |
954 | 1623 const int w=16;\ |
651 | 1624 for(i=0; i<w; i++)\ |
1625 {\ | |
1626 const int src0= src[0*srcStride];\ | |
1627 const int src1= src[1*srcStride];\ | |
1628 const int src2= src[2*srcStride];\ | |
1629 const int src3= src[3*srcStride];\ | |
1630 const int src4= src[4*srcStride];\ | |
1631 const int src5= src[5*srcStride];\ | |
1632 const int src6= src[6*srcStride];\ | |
1633 const int src7= src[7*srcStride];\ | |
1634 const int src8= src[8*srcStride];\ | |
1635 const int src9= src[9*srcStride];\ | |
1636 const int src10= src[10*srcStride];\ | |
1637 const int src11= src[11*srcStride];\ | |
1638 const int src12= src[12*srcStride];\ | |
1639 const int src13= src[13*srcStride];\ | |
1640 const int src14= src[14*srcStride];\ | |
1641 const int src15= src[15*srcStride];\ | |
1642 const int src16= src[16*srcStride];\ | |
1643 OP(dst[ 0*dstStride], (src0 +src1 )*20 - (src0 +src2 )*6 + (src1 +src3 )*3 - (src2 +src4 ));\ | |
1644 OP(dst[ 1*dstStride], (src1 +src2 )*20 - (src0 +src3 )*6 + (src0 +src4 )*3 - (src1 +src5 ));\ | |
1645 OP(dst[ 2*dstStride], (src2 +src3 )*20 - (src1 +src4 )*6 + (src0 +src5 )*3 - (src0 +src6 ));\ | |
1646 OP(dst[ 3*dstStride], (src3 +src4 )*20 - (src2 +src5 )*6 + (src1 +src6 )*3 - (src0 +src7 ));\ | |
1647 OP(dst[ 4*dstStride], (src4 +src5 )*20 - (src3 +src6 )*6 + (src2 +src7 )*3 - (src1 +src8 ));\ | |
1648 OP(dst[ 5*dstStride], (src5 +src6 )*20 - (src4 +src7 )*6 + (src3 +src8 )*3 - (src2 +src9 ));\ | |
1649 OP(dst[ 6*dstStride], (src6 +src7 )*20 - (src5 +src8 )*6 + (src4 +src9 )*3 - (src3 +src10));\ | |
1650 OP(dst[ 7*dstStride], (src7 +src8 )*20 - (src6 +src9 )*6 + (src5 +src10)*3 - (src4 +src11));\ | |
1651 OP(dst[ 8*dstStride], (src8 +src9 )*20 - (src7 +src10)*6 + (src6 +src11)*3 - (src5 +src12));\ | |
1652 OP(dst[ 9*dstStride], (src9 +src10)*20 - (src8 +src11)*6 + (src7 +src12)*3 - (src6 +src13));\ | |
1653 OP(dst[10*dstStride], (src10+src11)*20 - (src9 +src12)*6 + (src8 +src13)*3 - (src7 +src14));\ | |
1654 OP(dst[11*dstStride], (src11+src12)*20 - (src10+src13)*6 + (src9 +src14)*3 - (src8 +src15));\ | |
1655 OP(dst[12*dstStride], (src12+src13)*20 - (src11+src14)*6 + (src10+src15)*3 - (src9 +src16));\ | |
1656 OP(dst[13*dstStride], (src13+src14)*20 - (src12+src15)*6 + (src11+src16)*3 - (src10+src16));\ | |
1657 OP(dst[14*dstStride], (src14+src15)*20 - (src13+src16)*6 + (src12+src16)*3 - (src11+src15));\ | |
1658 OP(dst[15*dstStride], (src15+src16)*20 - (src14+src16)*6 + (src13+src15)*3 - (src12+src14));\ | |
1659 dst++;\ | |
1660 src++;\ | |
1661 }\ | |
255 | 1662 }\ |
1663 \ | |
1064 | 1664 static void OPNAME ## qpel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1665 OPNAME ## pixels8_c(dst, src, stride, 8);\ |
255 | 1666 }\ |
1667 \ | |
1064 | 1668 static void OPNAME ## qpel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
1669 uint8_t half[64];\ | |
651 | 1670 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1671 OPNAME ## pixels8_l2(dst, src, half, stride, stride, 8, 8);\ | |
1672 }\ | |
1673 \ | |
1064 | 1674 static void OPNAME ## qpel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 1675 OPNAME ## mpeg4_qpel8_h_lowpass(dst, src, stride, stride, 8);\ |
255 | 1676 }\ |
1677 \ | |
1064 | 1678 static void OPNAME ## qpel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
1679 uint8_t half[64];\ | |
651 | 1680 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1681 OPNAME ## pixels8_l2(dst, src+1, half, stride, stride, 8, 8);\ | |
1682 }\ | |
1683 \ | |
1064 | 1684 static void OPNAME ## qpel8_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
1685 uint8_t full[16*9];\ | |
1686 uint8_t half[64];\ | |
651 | 1687 copy_block9(full, src, 16, stride, 9);\ |
984 | 1688 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1689 OPNAME ## pixels8_l2(dst, full, half, stride, 16, 8, 8);\ |
1690 }\ | |
1691 \ | |
1064 | 1692 static void OPNAME ## qpel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
1693 uint8_t full[16*9];\ | |
651 | 1694 copy_block9(full, src, 16, stride, 9);\ |
984 | 1695 OPNAME ## mpeg4_qpel8_v_lowpass(dst, full, stride, 16);\ |
255 | 1696 }\ |
1697 \ | |
1064 | 1698 static void OPNAME ## qpel8_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
1699 uint8_t full[16*9];\ | |
1700 uint8_t half[64];\ | |
651 | 1701 copy_block9(full, src, 16, stride, 9);\ |
984 | 1702 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1703 OPNAME ## pixels8_l2(dst, full+16, half, stride, 16, 8, 8);\ |
1704 }\ | |
1064 | 1705 void ff_ ## OPNAME ## qpel8_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1706 uint8_t full[16*9];\ | |
1707 uint8_t halfH[72];\ | |
1708 uint8_t halfV[64];\ | |
1709 uint8_t halfHV[64];\ | |
651 | 1710 copy_block9(full, src, 16, stride, 9);\ |
1711 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1712 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1713 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1714 OPNAME ## pixels8_l4(dst, full, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1715 }\ |
1064 | 1716 static void OPNAME ## qpel8_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
1717 uint8_t full[16*9];\ | |
1718 uint8_t halfH[72];\ | |
1719 uint8_t halfHV[64];\ | |
984 | 1720 copy_block9(full, src, 16, stride, 9);\ |
1721 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1722 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1723 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1724 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1725 }\ | |
1064 | 1726 void ff_ ## OPNAME ## qpel8_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1727 uint8_t full[16*9];\ | |
1728 uint8_t halfH[72];\ | |
1729 uint8_t halfV[64];\ | |
1730 uint8_t halfHV[64];\ | |
651 | 1731 copy_block9(full, src, 16, stride, 9);\ |
1732 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1733 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1734 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1735 OPNAME ## pixels8_l4(dst, full+1, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1736 }\ |
1064 | 1737 static void OPNAME ## qpel8_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
1738 uint8_t full[16*9];\ | |
1739 uint8_t halfH[72];\ | |
1740 uint8_t halfHV[64];\ | |
984 | 1741 copy_block9(full, src, 16, stride, 9);\ |
1742 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1743 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1744 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1745 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1746 }\ | |
1064 | 1747 void ff_ ## OPNAME ## qpel8_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1748 uint8_t full[16*9];\ | |
1749 uint8_t halfH[72];\ | |
1750 uint8_t halfV[64];\ | |
1751 uint8_t halfHV[64];\ | |
651 | 1752 copy_block9(full, src, 16, stride, 9);\ |
1753 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1754 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1755 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1756 OPNAME ## pixels8_l4(dst, full+16, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
1757 }\ | |
1064 | 1758 static void OPNAME ## qpel8_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
1759 uint8_t full[16*9];\ | |
1760 uint8_t halfH[72];\ | |
1761 uint8_t halfHV[64];\ | |
984 | 1762 copy_block9(full, src, 16, stride, 9);\ |
1763 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1764 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1765 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1766 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1767 }\ | |
1064 | 1768 void ff_ ## OPNAME ## qpel8_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1769 uint8_t full[16*9];\ | |
1770 uint8_t halfH[72];\ | |
1771 uint8_t halfV[64];\ | |
1772 uint8_t halfHV[64];\ | |
651 | 1773 copy_block9(full, src, 16, stride, 9);\ |
1774 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full , 8, 16, 9);\ | |
984 | 1775 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1776 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1777 OPNAME ## pixels8_l4(dst, full+17, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1778 }\ |
1064 | 1779 static void OPNAME ## qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
1780 uint8_t full[16*9];\ | |
1781 uint8_t halfH[72];\ | |
1782 uint8_t halfHV[64];\ | |
984 | 1783 copy_block9(full, src, 16, stride, 9);\ |
1784 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1785 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1786 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1787 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1788 }\ | |
1064 | 1789 static void OPNAME ## qpel8_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
1790 uint8_t halfH[72];\ | |
1791 uint8_t halfHV[64];\ | |
651 | 1792 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1793 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1794 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ |
1795 }\ | |
1064 | 1796 static void OPNAME ## qpel8_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
1797 uint8_t halfH[72];\ | |
1798 uint8_t halfHV[64];\ | |
651 | 1799 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1800 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1801 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ |
1802 }\ | |
1064 | 1803 void ff_ ## OPNAME ## qpel8_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1804 uint8_t full[16*9];\ | |
1805 uint8_t halfH[72];\ | |
1806 uint8_t halfV[64];\ | |
1807 uint8_t halfHV[64];\ | |
651 | 1808 copy_block9(full, src, 16, stride, 9);\ |
1809 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1810 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1811 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1812 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
255 | 1813 }\ |
1064 | 1814 static void OPNAME ## qpel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
1815 uint8_t full[16*9];\ | |
1816 uint8_t halfH[72];\ | |
984 | 1817 copy_block9(full, src, 16, stride, 9);\ |
1818 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1819 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1820 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1821 }\ | |
1064 | 1822 void ff_ ## OPNAME ## qpel8_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1823 uint8_t full[16*9];\ | |
1824 uint8_t halfH[72];\ | |
1825 uint8_t halfV[64];\ | |
1826 uint8_t halfHV[64];\ | |
651 | 1827 copy_block9(full, src, 16, stride, 9);\ |
1828 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1829 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1830 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1831 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
1832 }\ | |
1064 | 1833 static void OPNAME ## qpel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
1834 uint8_t full[16*9];\ | |
1835 uint8_t halfH[72];\ | |
984 | 1836 copy_block9(full, src, 16, stride, 9);\ |
1837 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1838 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1839 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1840 }\ | |
1064 | 1841 static void OPNAME ## qpel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
1842 uint8_t halfH[72];\ | |
651 | 1843 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1844 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ |
651 | 1845 }\ |
1064 | 1846 static void OPNAME ## qpel16_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1847 OPNAME ## pixels16_c(dst, src, stride, 16);\ |
255 | 1848 }\ |
651 | 1849 \ |
1064 | 1850 static void OPNAME ## qpel16_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
1851 uint8_t half[256];\ | |
651 | 1852 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
1853 OPNAME ## pixels16_l2(dst, src, half, stride, stride, 16, 16);\ | |
1854 }\ | |
1855 \ | |
1064 | 1856 static void OPNAME ## qpel16_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 1857 OPNAME ## mpeg4_qpel16_h_lowpass(dst, src, stride, stride, 16);\ |
1858 }\ | |
1859 \ | |
1064 | 1860 static void OPNAME ## qpel16_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
1861 uint8_t half[256];\ | |
651 | 1862 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
1863 OPNAME ## pixels16_l2(dst, src+1, half, stride, stride, 16, 16);\ | |
1864 }\ | |
1865 \ | |
1064 | 1866 static void OPNAME ## qpel16_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
1867 uint8_t full[24*17];\ | |
1868 uint8_t half[256];\ | |
651 | 1869 copy_block17(full, src, 24, stride, 17);\ |
954 | 1870 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 1871 OPNAME ## pixels16_l2(dst, full, half, stride, 24, 16, 16);\ |
255 | 1872 }\ |
651 | 1873 \ |
1064 | 1874 static void OPNAME ## qpel16_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
1875 uint8_t full[24*17];\ | |
651 | 1876 copy_block17(full, src, 24, stride, 17);\ |
954 | 1877 OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24);\ |
651 | 1878 }\ |
1879 \ | |
1064 | 1880 static void OPNAME ## qpel16_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
1881 uint8_t full[24*17];\ | |
1882 uint8_t half[256];\ | |
651 | 1883 copy_block17(full, src, 24, stride, 17);\ |
954 | 1884 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 1885 OPNAME ## pixels16_l2(dst, full+24, half, stride, 24, 16, 16);\ |
255 | 1886 }\ |
1064 | 1887 void ff_ ## OPNAME ## qpel16_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1888 uint8_t full[24*17];\ | |
1889 uint8_t halfH[272];\ | |
1890 uint8_t halfV[256];\ | |
1891 uint8_t halfHV[256];\ | |
651 | 1892 copy_block17(full, src, 24, stride, 17);\ |
1893 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1894 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
1895 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1896 OPNAME ## pixels16_l4(dst, full, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
1897 }\ | |
1064 | 1898 static void OPNAME ## qpel16_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
1899 uint8_t full[24*17];\ | |
1900 uint8_t halfH[272];\ | |
1901 uint8_t halfHV[256];\ | |
984 | 1902 copy_block17(full, src, 24, stride, 17);\ |
1903 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1904 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
1905 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1906 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
1907 }\ | |
1064 | 1908 void ff_ ## OPNAME ## qpel16_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1909 uint8_t full[24*17];\ | |
1910 uint8_t halfH[272];\ | |
1911 uint8_t halfV[256];\ | |
1912 uint8_t halfHV[256];\ | |
651 | 1913 copy_block17(full, src, 24, stride, 17);\ |
1914 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1915 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
1916 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1917 OPNAME ## pixels16_l4(dst, full+1, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
1918 }\ | |
1064 | 1919 static void OPNAME ## qpel16_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
1920 uint8_t full[24*17];\ | |
1921 uint8_t halfH[272];\ | |
1922 uint8_t halfHV[256];\ | |
984 | 1923 copy_block17(full, src, 24, stride, 17);\ |
1924 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1925 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
1926 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1927 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
1928 }\ | |
1064 | 1929 void ff_ ## OPNAME ## qpel16_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1930 uint8_t full[24*17];\ | |
1931 uint8_t halfH[272];\ | |
1932 uint8_t halfV[256];\ | |
1933 uint8_t halfHV[256];\ | |
651 | 1934 copy_block17(full, src, 24, stride, 17);\ |
1935 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1936 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
1937 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1938 OPNAME ## pixels16_l4(dst, full+24, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
255 | 1939 }\ |
1064 | 1940 static void OPNAME ## qpel16_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
1941 uint8_t full[24*17];\ | |
1942 uint8_t halfH[272];\ | |
1943 uint8_t halfHV[256];\ | |
984 | 1944 copy_block17(full, src, 24, stride, 17);\ |
1945 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1946 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
1947 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1948 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
1949 }\ | |
1064 | 1950 void ff_ ## OPNAME ## qpel16_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1951 uint8_t full[24*17];\ | |
1952 uint8_t halfH[272];\ | |
1953 uint8_t halfV[256];\ | |
1954 uint8_t halfHV[256];\ | |
651 | 1955 copy_block17(full, src, 24, stride, 17);\ |
1956 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full , 16, 24, 17);\ | |
954 | 1957 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
1958 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1959 OPNAME ## pixels16_l4(dst, full+25, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
1960 }\ | |
1064 | 1961 static void OPNAME ## qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
1962 uint8_t full[24*17];\ | |
1963 uint8_t halfH[272];\ | |
1964 uint8_t halfHV[256];\ | |
984 | 1965 copy_block17(full, src, 24, stride, 17);\ |
1966 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1967 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
1968 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1969 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
1970 }\ | |
1064 | 1971 static void OPNAME ## qpel16_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
1972 uint8_t halfH[272];\ | |
1973 uint8_t halfHV[256];\ | |
651 | 1974 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 1975 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 1976 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ |
255 | 1977 }\ |
1064 | 1978 static void OPNAME ## qpel16_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
1979 uint8_t halfH[272];\ | |
1980 uint8_t halfHV[256];\ | |
651 | 1981 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 1982 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 1983 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ |
1984 }\ | |
1064 | 1985 void ff_ ## OPNAME ## qpel16_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1986 uint8_t full[24*17];\ | |
1987 uint8_t halfH[272];\ | |
1988 uint8_t halfV[256];\ | |
1989 uint8_t halfHV[256];\ | |
651 | 1990 copy_block17(full, src, 24, stride, 17);\ |
1991 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1992 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
1993 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1994 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
255 | 1995 }\ |
1064 | 1996 static void OPNAME ## qpel16_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
1997 uint8_t full[24*17];\ | |
1998 uint8_t halfH[272];\ | |
984 | 1999 copy_block17(full, src, 24, stride, 17);\ |
2000 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2001 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
2002 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2003 }\ | |
1064 | 2004 void ff_ ## OPNAME ## qpel16_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2005 uint8_t full[24*17];\ | |
2006 uint8_t halfH[272];\ | |
2007 uint8_t halfV[256];\ | |
2008 uint8_t halfHV[256];\ | |
651 | 2009 copy_block17(full, src, 24, stride, 17);\ |
2010 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2011 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
2012 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2013 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
2014 }\ | |
1064 | 2015 static void OPNAME ## qpel16_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
2016 uint8_t full[24*17];\ | |
2017 uint8_t halfH[272];\ | |
984 | 2018 copy_block17(full, src, 24, stride, 17);\ |
2019 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2020 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2021 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2022 }\ | |
1064 | 2023 static void OPNAME ## qpel16_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
2024 uint8_t halfH[272];\ | |
651 | 2025 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2026 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ |
859 | 2027 } |
255 | 2028 |
651 | 2029 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) |
2030 #define op_avg_no_rnd(a, b) a = (((a)+cm[((b) + 15)>>5])>>1) | |
2031 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2032 #define op_put_no_rnd(a, b) a = cm[((b) + 15)>>5] | |
2033 | |
2034 QPEL_MC(0, put_ , _ , op_put) | |
2035 QPEL_MC(1, put_no_rnd_, _no_rnd_, op_put_no_rnd) | |
2036 QPEL_MC(0, avg_ , _ , op_avg) | |
2037 //QPEL_MC(1, avg_no_rnd , _ , op_avg) | |
2038 #undef op_avg | |
2039 #undef op_avg_no_rnd | |
2040 #undef op_put | |
2041 #undef op_put_no_rnd | |
255 | 2042 |
1168 | 2043 #if 1 |
2044 #define H264_LOWPASS(OPNAME, OP, OP2) \ | |
5151 | 2045 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
|
2046 const int h=2;\ |
4176 | 2047 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
|
2048 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2049 for(i=0; i<h; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2050 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2051 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
|
2052 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
|
2053 dst+=dstStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2054 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2055 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2056 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2057 \ |
5151 | 2058 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
|
2059 const int w=2;\ |
4176 | 2060 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
|
2061 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2062 for(i=0; i<w; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2063 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2064 const int srcB= src[-2*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2065 const int srcA= src[-1*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2066 const int src0= src[0 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2067 const int src1= src[1 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2068 const int src2= src[2 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2069 const int src3= src[3 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2070 const int src4= src[4 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2071 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
|
2072 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
|
2073 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2074 src++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2075 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2076 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2077 \ |
5151 | 2078 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
|
2079 const int h=2;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2080 const int w=2;\ |
4176 | 2081 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
|
2082 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2083 src -= 2*srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2084 for(i=0; i<h+5; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2085 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2086 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
|
2087 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
|
2088 tmp+=tmpStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2089 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2090 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2091 tmp -= tmpStride*(h+5-2);\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2092 for(i=0; i<w; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2093 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2094 const int tmpB= tmp[-2*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2095 const int tmpA= tmp[-1*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2096 const int tmp0= tmp[0 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2097 const int tmp1= tmp[1 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2098 const int tmp2= tmp[2 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2099 const int tmp3= tmp[3 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2100 const int tmp4= tmp[4 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2101 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
|
2102 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
|
2103 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2104 tmp++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2105 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2106 }\ |
1168 | 2107 static void OPNAME ## h264_qpel4_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
2108 const int h=4;\ | |
4176 | 2109 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2110 int i;\ |
2111 for(i=0; i<h; i++)\ | |
2112 {\ | |
2113 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]));\ | |
2114 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]));\ | |
2115 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]));\ | |
2116 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]));\ | |
2117 dst+=dstStride;\ | |
2118 src+=srcStride;\ | |
2119 }\ | |
2120 }\ | |
2121 \ | |
2122 static void OPNAME ## h264_qpel4_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2123 const int w=4;\ | |
4176 | 2124 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2125 int i;\ |
2126 for(i=0; i<w; i++)\ | |
2127 {\ | |
2128 const int srcB= src[-2*srcStride];\ | |
2129 const int srcA= src[-1*srcStride];\ | |
2130 const int src0= src[0 *srcStride];\ | |
2131 const int src1= src[1 *srcStride];\ | |
2132 const int src2= src[2 *srcStride];\ | |
2133 const int src3= src[3 *srcStride];\ | |
2134 const int src4= src[4 *srcStride];\ | |
2135 const int src5= src[5 *srcStride];\ | |
2136 const int src6= src[6 *srcStride];\ | |
2137 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2138 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2139 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2140 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2141 dst++;\ | |
2142 src++;\ | |
2143 }\ | |
2144 }\ | |
2145 \ | |
2146 static void OPNAME ## h264_qpel4_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2147 const int h=4;\ | |
2148 const int w=4;\ | |
4176 | 2149 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2150 int i;\ |
2151 src -= 2*srcStride;\ | |
2152 for(i=0; i<h+5; i++)\ | |
2153 {\ | |
2154 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]);\ | |
2155 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]);\ | |
2156 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]);\ | |
2157 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]);\ | |
2158 tmp+=tmpStride;\ | |
2159 src+=srcStride;\ | |
2160 }\ | |
2161 tmp -= tmpStride*(h+5-2);\ | |
2162 for(i=0; i<w; i++)\ | |
2163 {\ | |
2164 const int tmpB= tmp[-2*tmpStride];\ | |
2165 const int tmpA= tmp[-1*tmpStride];\ | |
2166 const int tmp0= tmp[0 *tmpStride];\ | |
2167 const int tmp1= tmp[1 *tmpStride];\ | |
2168 const int tmp2= tmp[2 *tmpStride];\ | |
2169 const int tmp3= tmp[3 *tmpStride];\ | |
2170 const int tmp4= tmp[4 *tmpStride];\ | |
2171 const int tmp5= tmp[5 *tmpStride];\ | |
2172 const int tmp6= tmp[6 *tmpStride];\ | |
2173 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2174 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2175 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2176 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2177 dst++;\ | |
2178 tmp++;\ | |
2179 }\ | |
2180 }\ | |
2181 \ | |
2182 static void OPNAME ## h264_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2183 const int h=8;\ | |
4176 | 2184 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2185 int i;\ |
2186 for(i=0; i<h; i++)\ | |
2187 {\ | |
2188 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]));\ | |
2189 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]));\ | |
2190 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]));\ | |
2191 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]));\ | |
2192 OP(dst[4], (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]));\ | |
2193 OP(dst[5], (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]));\ | |
2194 OP(dst[6], (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]));\ | |
2195 OP(dst[7], (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]));\ | |
2196 dst+=dstStride;\ | |
2197 src+=srcStride;\ | |
2198 }\ | |
2199 }\ | |
2200 \ | |
2201 static void OPNAME ## h264_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2202 const int w=8;\ | |
4176 | 2203 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2204 int i;\ |
2205 for(i=0; i<w; i++)\ | |
2206 {\ | |
2207 const int srcB= src[-2*srcStride];\ | |
2208 const int srcA= src[-1*srcStride];\ | |
2209 const int src0= src[0 *srcStride];\ | |
2210 const int src1= src[1 *srcStride];\ | |
2211 const int src2= src[2 *srcStride];\ | |
2212 const int src3= src[3 *srcStride];\ | |
2213 const int src4= src[4 *srcStride];\ | |
2214 const int src5= src[5 *srcStride];\ | |
2215 const int src6= src[6 *srcStride];\ | |
2216 const int src7= src[7 *srcStride];\ | |
2217 const int src8= src[8 *srcStride];\ | |
2218 const int src9= src[9 *srcStride];\ | |
2219 const int src10=src[10*srcStride];\ | |
2220 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2221 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2222 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2223 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2224 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*5 + (src2+src7));\ | |
2225 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*5 + (src3+src8));\ | |
2226 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*5 + (src4+src9));\ | |
2227 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src9)*5 + (src5+src10));\ | |
2228 dst++;\ | |
2229 src++;\ | |
2230 }\ | |
2231 }\ | |
2232 \ | |
2233 static void OPNAME ## h264_qpel8_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2234 const int h=8;\ | |
2235 const int w=8;\ | |
4176 | 2236 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2237 int i;\ |
2238 src -= 2*srcStride;\ | |
2239 for(i=0; i<h+5; i++)\ | |
2240 {\ | |
2241 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]);\ | |
2242 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]);\ | |
2243 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]);\ | |
2244 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]);\ | |
2245 tmp[4]= (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]);\ | |
2246 tmp[5]= (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]);\ | |
2247 tmp[6]= (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]);\ | |
2248 tmp[7]= (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]);\ | |
2249 tmp+=tmpStride;\ | |
2250 src+=srcStride;\ | |
2251 }\ | |
2252 tmp -= tmpStride*(h+5-2);\ | |
2253 for(i=0; i<w; i++)\ | |
2254 {\ | |
2255 const int tmpB= tmp[-2*tmpStride];\ | |
2256 const int tmpA= tmp[-1*tmpStride];\ | |
2257 const int tmp0= tmp[0 *tmpStride];\ | |
2258 const int tmp1= tmp[1 *tmpStride];\ | |
2259 const int tmp2= tmp[2 *tmpStride];\ | |
2260 const int tmp3= tmp[3 *tmpStride];\ | |
2261 const int tmp4= tmp[4 *tmpStride];\ | |
2262 const int tmp5= tmp[5 *tmpStride];\ | |
2263 const int tmp6= tmp[6 *tmpStride];\ | |
2264 const int tmp7= tmp[7 *tmpStride];\ | |
2265 const int tmp8= tmp[8 *tmpStride];\ | |
2266 const int tmp9= tmp[9 *tmpStride];\ | |
2267 const int tmp10=tmp[10*tmpStride];\ | |
2268 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2269 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2270 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2271 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2272 OP2(dst[4*dstStride], (tmp4+tmp5)*20 - (tmp3+tmp6)*5 + (tmp2+tmp7));\ | |
2273 OP2(dst[5*dstStride], (tmp5+tmp6)*20 - (tmp4+tmp7)*5 + (tmp3+tmp8));\ | |
2274 OP2(dst[6*dstStride], (tmp6+tmp7)*20 - (tmp5+tmp8)*5 + (tmp4+tmp9));\ | |
2275 OP2(dst[7*dstStride], (tmp7+tmp8)*20 - (tmp6+tmp9)*5 + (tmp5+tmp10));\ | |
2276 dst++;\ | |
2277 tmp++;\ | |
2278 }\ | |
2279 }\ | |
2280 \ | |
2281 static void OPNAME ## h264_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2282 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2283 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2284 src += 8*srcStride;\ | |
2285 dst += 8*dstStride;\ | |
2286 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2287 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2288 }\ | |
2289 \ | |
2290 static void OPNAME ## h264_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2291 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2292 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2293 src += 8*srcStride;\ | |
2294 dst += 8*dstStride;\ | |
2295 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2296 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2297 }\ | |
2298 \ | |
2299 static void OPNAME ## h264_qpel16_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2300 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2301 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2302 src += 8*srcStride;\ | |
2303 dst += 8*dstStride;\ | |
2304 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2305 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2306 }\ | |
2307 | |
2308 #define H264_MC(OPNAME, SIZE) \ | |
2309 static void OPNAME ## h264_qpel ## SIZE ## _mc00_c (uint8_t *dst, uint8_t *src, int stride){\ | |
2310 OPNAME ## pixels ## SIZE ## _c(dst, src, stride, SIZE);\ | |
2311 }\ | |
2312 \ | |
2313 static void OPNAME ## h264_qpel ## SIZE ## _mc10_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2314 uint8_t half[SIZE*SIZE];\ | |
2315 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2316 OPNAME ## pixels ## SIZE ## _l2(dst, src, half, stride, stride, SIZE, SIZE);\ | |
2317 }\ | |
2318 \ | |
2319 static void OPNAME ## h264_qpel ## SIZE ## _mc20_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2320 OPNAME ## h264_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride);\ | |
2321 }\ | |
2322 \ | |
2323 static void OPNAME ## h264_qpel ## SIZE ## _mc30_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2324 uint8_t half[SIZE*SIZE];\ | |
2325 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2326 OPNAME ## pixels ## SIZE ## _l2(dst, src+1, half, stride, stride, SIZE, SIZE);\ | |
2327 }\ | |
2328 \ | |
2329 static void OPNAME ## h264_qpel ## SIZE ## _mc01_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2330 uint8_t full[SIZE*(SIZE+5)];\ | |
2331 uint8_t * const full_mid= full + SIZE*2;\ | |
2332 uint8_t half[SIZE*SIZE];\ | |
2333 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2334 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2335 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid, half, stride, SIZE, SIZE, SIZE);\ | |
2336 }\ | |
2337 \ | |
2338 static void OPNAME ## h264_qpel ## SIZE ## _mc02_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2339 uint8_t full[SIZE*(SIZE+5)];\ | |
2340 uint8_t * const full_mid= full + SIZE*2;\ | |
2341 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2342 OPNAME ## h264_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE);\ | |
2343 }\ | |
2344 \ | |
2345 static void OPNAME ## h264_qpel ## SIZE ## _mc03_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2346 uint8_t full[SIZE*(SIZE+5)];\ | |
2347 uint8_t * const full_mid= full + SIZE*2;\ | |
2348 uint8_t half[SIZE*SIZE];\ | |
2349 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2350 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2351 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid+SIZE, half, stride, SIZE, SIZE, SIZE);\ | |
2352 }\ | |
2353 \ | |
2354 static void OPNAME ## h264_qpel ## SIZE ## _mc11_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2355 uint8_t full[SIZE*(SIZE+5)];\ | |
2356 uint8_t * const full_mid= full + SIZE*2;\ | |
2357 uint8_t halfH[SIZE*SIZE];\ | |
2358 uint8_t halfV[SIZE*SIZE];\ | |
2359 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2360 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2361 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2362 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2363 }\ | |
2364 \ | |
2365 static void OPNAME ## h264_qpel ## SIZE ## _mc31_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2366 uint8_t full[SIZE*(SIZE+5)];\ | |
2367 uint8_t * const full_mid= full + SIZE*2;\ | |
2368 uint8_t halfH[SIZE*SIZE];\ | |
2369 uint8_t halfV[SIZE*SIZE];\ | |
2370 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2371 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2372 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2373 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2374 }\ | |
2375 \ | |
2376 static void OPNAME ## h264_qpel ## SIZE ## _mc13_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2377 uint8_t full[SIZE*(SIZE+5)];\ | |
2378 uint8_t * const full_mid= full + SIZE*2;\ | |
2379 uint8_t halfH[SIZE*SIZE];\ | |
2380 uint8_t halfV[SIZE*SIZE];\ | |
2381 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2382 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2383 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2384 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2385 }\ | |
2386 \ | |
2387 static void OPNAME ## h264_qpel ## SIZE ## _mc33_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2388 uint8_t full[SIZE*(SIZE+5)];\ | |
2389 uint8_t * const full_mid= full + SIZE*2;\ | |
2390 uint8_t halfH[SIZE*SIZE];\ | |
2391 uint8_t halfV[SIZE*SIZE];\ | |
2392 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2393 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2394 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2395 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2396 }\ | |
2397 \ | |
2398 static void OPNAME ## h264_qpel ## SIZE ## _mc22_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2399 int16_t tmp[SIZE*(SIZE+5)];\ | |
2400 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass(dst, tmp, src, stride, SIZE, stride);\ | |
2401 }\ | |
2402 \ | |
2403 static void OPNAME ## h264_qpel ## SIZE ## _mc21_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2404 int16_t tmp[SIZE*(SIZE+5)];\ | |
2405 uint8_t halfH[SIZE*SIZE];\ | |
2406 uint8_t halfHV[SIZE*SIZE];\ | |
2407 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2408 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2409 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2410 }\ | |
2411 \ | |
2412 static void OPNAME ## h264_qpel ## SIZE ## _mc23_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2413 int16_t tmp[SIZE*(SIZE+5)];\ | |
2414 uint8_t halfH[SIZE*SIZE];\ | |
2415 uint8_t halfHV[SIZE*SIZE];\ | |
2416 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2417 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2418 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2419 }\ | |
2420 \ | |
2421 static void OPNAME ## h264_qpel ## SIZE ## _mc12_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2422 uint8_t full[SIZE*(SIZE+5)];\ | |
2423 uint8_t * const full_mid= full + SIZE*2;\ | |
2424 int16_t tmp[SIZE*(SIZE+5)];\ | |
2425 uint8_t halfV[SIZE*SIZE];\ | |
2426 uint8_t halfHV[SIZE*SIZE];\ | |
2427 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2428 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2429 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2430 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2431 }\ | |
2432 \ | |
2433 static void OPNAME ## h264_qpel ## SIZE ## _mc32_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2434 uint8_t full[SIZE*(SIZE+5)];\ | |
2435 uint8_t * const full_mid= full + SIZE*2;\ | |
2436 int16_t tmp[SIZE*(SIZE+5)];\ | |
2437 uint8_t halfV[SIZE*SIZE];\ | |
2438 uint8_t halfHV[SIZE*SIZE];\ | |
2439 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2440 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2441 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2442 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2443 }\ | |
2444 | |
2445 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) | |
2446 //#define op_avg2(a, b) a = (((a)*w1+cm[((b) + 16)>>5]*w2 + o + 64)>>7) | |
2447 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2448 #define op2_avg(a, b) a = (((a)+cm[((b) + 512)>>10]+1)>>1) | |
2449 #define op2_put(a, b) a = cm[((b) + 512)>>10] | |
2450 | |
2451 H264_LOWPASS(put_ , op_put, op2_put) | |
2452 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
|
2453 H264_MC(put_, 2) |
1168 | 2454 H264_MC(put_, 4) |
2455 H264_MC(put_, 8) | |
2456 H264_MC(put_, 16) | |
2457 H264_MC(avg_, 4) | |
2458 H264_MC(avg_, 8) | |
2459 H264_MC(avg_, 16) | |
2460 | |
2461 #undef op_avg | |
2462 #undef op_put | |
2463 #undef op2_avg | |
2464 #undef op2_put | |
2465 #endif | |
2466 | |
4594 | 2467 #define op_scale1(x) block[x] = av_clip_uint8( (block[x]*weight + offset) >> log2_denom ) |
2468 #define op_scale2(x) dst[x] = av_clip_uint8( (src[x]*weights + dst[x]*weightd + offset) >> (log2_denom+1)) | |
2415 | 2469 #define H264_WEIGHT(W,H) \ |
2470 static void weight_h264_pixels ## W ## x ## H ## _c(uint8_t *block, int stride, int log2_denom, int weight, int offset){ \ | |
3029 | 2471 int y; \ |
2415 | 2472 offset <<= log2_denom; \ |
2473 if(log2_denom) offset += 1<<(log2_denom-1); \ | |
2474 for(y=0; y<H; y++, block += stride){ \ | |
2475 op_scale1(0); \ | |
2476 op_scale1(1); \ | |
2477 if(W==2) continue; \ | |
2478 op_scale1(2); \ | |
2479 op_scale1(3); \ | |
2480 if(W==4) continue; \ | |
2481 op_scale1(4); \ | |
2482 op_scale1(5); \ | |
2483 op_scale1(6); \ | |
2484 op_scale1(7); \ | |
2485 if(W==8) continue; \ | |
2486 op_scale1(8); \ | |
2487 op_scale1(9); \ | |
2488 op_scale1(10); \ | |
2489 op_scale1(11); \ | |
2490 op_scale1(12); \ | |
2491 op_scale1(13); \ | |
2492 op_scale1(14); \ | |
2493 op_scale1(15); \ | |
2494 } \ | |
2495 } \ | |
3029 | 2496 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){ \ |
2497 int y; \ | |
2498 offset = ((offset + 1) | 1) << log2_denom; \ | |
2415 | 2499 for(y=0; y<H; y++, dst += stride, src += stride){ \ |
2500 op_scale2(0); \ | |
2501 op_scale2(1); \ | |
2502 if(W==2) continue; \ | |
2503 op_scale2(2); \ | |
2504 op_scale2(3); \ | |
2505 if(W==4) continue; \ | |
2506 op_scale2(4); \ | |
2507 op_scale2(5); \ | |
2508 op_scale2(6); \ | |
2509 op_scale2(7); \ | |
2510 if(W==8) continue; \ | |
2511 op_scale2(8); \ | |
2512 op_scale2(9); \ | |
2513 op_scale2(10); \ | |
2514 op_scale2(11); \ | |
2515 op_scale2(12); \ | |
2516 op_scale2(13); \ | |
2517 op_scale2(14); \ | |
2518 op_scale2(15); \ | |
2519 } \ | |
2520 } | |
2521 | |
2522 H264_WEIGHT(16,16) | |
2523 H264_WEIGHT(16,8) | |
2524 H264_WEIGHT(8,16) | |
2525 H264_WEIGHT(8,8) | |
2526 H264_WEIGHT(8,4) | |
2527 H264_WEIGHT(4,8) | |
2528 H264_WEIGHT(4,4) | |
2529 H264_WEIGHT(4,2) | |
2530 H264_WEIGHT(2,4) | |
2531 H264_WEIGHT(2,2) | |
2532 | |
2533 #undef op_scale1 | |
2534 #undef op_scale2 | |
2535 #undef H264_WEIGHT | |
2536 | |
936 | 2537 static void wmv2_mspel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){ |
4176 | 2538 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
936 | 2539 int i; |
2540 | |
2541 for(i=0; i<h; i++){ | |
2542 dst[0]= cm[(9*(src[0] + src[1]) - (src[-1] + src[2]) + 8)>>4]; | |
2543 dst[1]= cm[(9*(src[1] + src[2]) - (src[ 0] + src[3]) + 8)>>4]; | |
2544 dst[2]= cm[(9*(src[2] + src[3]) - (src[ 1] + src[4]) + 8)>>4]; | |
2545 dst[3]= cm[(9*(src[3] + src[4]) - (src[ 2] + src[5]) + 8)>>4]; | |
2546 dst[4]= cm[(9*(src[4] + src[5]) - (src[ 3] + src[6]) + 8)>>4]; | |
2547 dst[5]= cm[(9*(src[5] + src[6]) - (src[ 4] + src[7]) + 8)>>4]; | |
2548 dst[6]= cm[(9*(src[6] + src[7]) - (src[ 5] + src[8]) + 8)>>4]; | |
2549 dst[7]= cm[(9*(src[7] + src[8]) - (src[ 6] + src[9]) + 8)>>4]; | |
2550 dst+=dstStride; | |
2967 | 2551 src+=srcStride; |
936 | 2552 } |
2553 } | |
2554 | |
3432 | 2555 #ifdef CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2556 /* AVS specific */ |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2557 void ff_cavsdsp_init(DSPContext* c, AVCodecContext *avctx); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2558 |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2559 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
|
2560 put_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2561 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2562 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
|
2563 avg_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2564 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2565 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
|
2566 put_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2567 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2568 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
|
2569 avg_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2570 } |
3432 | 2571 #endif /* CONFIG_CAVS_DECODER */ |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2572 |
3526 | 2573 #if defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
2574 /* VC-1 specific */ | |
2575 void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx); | |
2576 | |
2577 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, uint8_t *src, int stride, int rnd) { | |
2578 put_pixels8_c(dst, src, stride, 8); | |
2579 } | |
2580 #endif /* CONFIG_VC1_DECODER||CONFIG_WMV3_DECODER */ | |
2581 | |
5887 | 2582 void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx); |
5899 | 2583 |
4296 | 2584 /* H264 specific */ |
5411
362aec4ef932
Take care of some renames (Doxygen and function name) after the previous pure rename patch.
takis
parents:
5394
diff
changeset
|
2585 void ff_h264dspenc_init(DSPContext* c, AVCodecContext *avctx); |
4296 | 2586 |
936 | 2587 static void wmv2_mspel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int w){ |
4176 | 2588 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
936 | 2589 int i; |
2590 | |
2591 for(i=0; i<w; i++){ | |
2592 const int src_1= src[ -srcStride]; | |
2593 const int src0 = src[0 ]; | |
2594 const int src1 = src[ srcStride]; | |
2595 const int src2 = src[2*srcStride]; | |
2596 const int src3 = src[3*srcStride]; | |
2597 const int src4 = src[4*srcStride]; | |
2598 const int src5 = src[5*srcStride]; | |
2599 const int src6 = src[6*srcStride]; | |
2600 const int src7 = src[7*srcStride]; | |
2601 const int src8 = src[8*srcStride]; | |
2602 const int src9 = src[9*srcStride]; | |
2603 dst[0*dstStride]= cm[(9*(src0 + src1) - (src_1 + src2) + 8)>>4]; | |
2604 dst[1*dstStride]= cm[(9*(src1 + src2) - (src0 + src3) + 8)>>4]; | |
2605 dst[2*dstStride]= cm[(9*(src2 + src3) - (src1 + src4) + 8)>>4]; | |
2606 dst[3*dstStride]= cm[(9*(src3 + src4) - (src2 + src5) + 8)>>4]; | |
2607 dst[4*dstStride]= cm[(9*(src4 + src5) - (src3 + src6) + 8)>>4]; | |
2608 dst[5*dstStride]= cm[(9*(src5 + src6) - (src4 + src7) + 8)>>4]; | |
2609 dst[6*dstStride]= cm[(9*(src6 + src7) - (src5 + src8) + 8)>>4]; | |
2610 dst[7*dstStride]= cm[(9*(src7 + src8) - (src6 + src9) + 8)>>4]; | |
2611 src++; | |
2612 dst++; | |
2613 } | |
2614 } | |
2615 | |
2616 static void put_mspel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){ | |
2617 put_pixels8_c(dst, src, stride, 8); | |
2618 } | |
2619 | |
2620 static void put_mspel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){ | |
2621 uint8_t half[64]; | |
2622 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2623 put_pixels8_l2(dst, src, half, stride, stride, 8, 8); | |
2624 } | |
2625 | |
2626 static void put_mspel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){ | |
2627 wmv2_mspel8_h_lowpass(dst, src, stride, stride, 8); | |
2628 } | |
2629 | |
2630 static void put_mspel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){ | |
2631 uint8_t half[64]; | |
2632 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2633 put_pixels8_l2(dst, src+1, half, stride, stride, 8, 8); | |
2634 } | |
2635 | |
2636 static void put_mspel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){ | |
2637 wmv2_mspel8_v_lowpass(dst, src, stride, stride, 8); | |
2638 } | |
2639 | |
2640 static void put_mspel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){ | |
2641 uint8_t halfH[88]; | |
2642 uint8_t halfV[64]; | |
2643 uint8_t halfHV[64]; | |
2644 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2645 wmv2_mspel8_v_lowpass(halfV, src, 8, stride, 8); | |
2646 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2647 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2648 } | |
2649 static void put_mspel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){ | |
2650 uint8_t halfH[88]; | |
2651 uint8_t halfV[64]; | |
2652 uint8_t halfHV[64]; | |
2653 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2654 wmv2_mspel8_v_lowpass(halfV, src+1, 8, stride, 8); | |
2655 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2656 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2657 } | |
2658 static void put_mspel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){ | |
2659 uint8_t halfH[88]; | |
2660 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2661 wmv2_mspel8_v_lowpass(dst, halfH+8, stride, 8, 8); | |
2662 } | |
2663 | |
1644 | 2664 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
|
2665 if(ENABLE_ANY_H263) { |
1644 | 2666 int x; |
2667 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2668 |
1644 | 2669 for(x=0; x<8; x++){ |
2670 int d1, d2, ad1; | |
2671 int p0= src[x-2*stride]; | |
2672 int p1= src[x-1*stride]; | |
2673 int p2= src[x+0*stride]; | |
2674 int p3= src[x+1*stride]; | |
2675 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2676 | |
2677 if (d<-2*strength) d1= 0; | |
2678 else if(d<- strength) d1=-2*strength - d; | |
2679 else if(d< strength) d1= d; | |
2680 else if(d< 2*strength) d1= 2*strength - d; | |
2681 else d1= 0; | |
2967 | 2682 |
1644 | 2683 p1 += d1; |
2684 p2 -= d1; | |
2685 if(p1&256) p1= ~(p1>>31); | |
2686 if(p2&256) p2= ~(p2>>31); | |
2967 | 2687 |
1644 | 2688 src[x-1*stride] = p1; |
2689 src[x+0*stride] = p2; | |
2690 | |
4001 | 2691 ad1= FFABS(d1)>>1; |
2967 | 2692 |
4594 | 2693 d2= av_clip((p0-p3)/4, -ad1, ad1); |
2967 | 2694 |
1644 | 2695 src[x-2*stride] = p0 - d2; |
2696 src[x+ stride] = p3 + d2; | |
2697 } | |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2698 } |
1644 | 2699 } |
2700 | |
2701 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
|
2702 if(ENABLE_ANY_H263) { |
1644 | 2703 int y; |
2704 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2705 |
1644 | 2706 for(y=0; y<8; y++){ |
2707 int d1, d2, ad1; | |
2708 int p0= src[y*stride-2]; | |
2709 int p1= src[y*stride-1]; | |
2710 int p2= src[y*stride+0]; | |
2711 int p3= src[y*stride+1]; | |
2712 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2713 | |
2714 if (d<-2*strength) d1= 0; | |
2715 else if(d<- strength) d1=-2*strength - d; | |
2716 else if(d< strength) d1= d; | |
2717 else if(d< 2*strength) d1= 2*strength - d; | |
2718 else d1= 0; | |
2967 | 2719 |
1644 | 2720 p1 += d1; |
2721 p2 -= d1; | |
2722 if(p1&256) p1= ~(p1>>31); | |
2723 if(p2&256) p2= ~(p2>>31); | |
2967 | 2724 |
1644 | 2725 src[y*stride-1] = p1; |
2726 src[y*stride+0] = p2; | |
2727 | |
4001 | 2728 ad1= FFABS(d1)>>1; |
2967 | 2729 |
4594 | 2730 d2= av_clip((p0-p3)/4, -ad1, ad1); |
2967 | 2731 |
1644 | 2732 src[y*stride-2] = p0 - d2; |
2733 src[y*stride+1] = p3 + d2; | |
2734 } | |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2735 } |
1644 | 2736 } |
936 | 2737 |
2045 | 2738 static void h261_loop_filter_c(uint8_t *src, int stride){ |
2739 int x,y,xy,yz; | |
2740 int temp[64]; | |
2741 | |
2742 for(x=0; x<8; x++){ | |
2743 temp[x ] = 4*src[x ]; | |
2744 temp[x + 7*8] = 4*src[x + 7*stride]; | |
2745 } | |
2746 for(y=1; y<7; y++){ | |
2747 for(x=0; x<8; x++){ | |
2748 xy = y * stride + x; | |
2749 yz = y * 8 + x; | |
2750 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
|
2751 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2752 } |
2967 | 2753 |
2045 | 2754 for(y=0; y<8; y++){ |
2755 src[ y*stride] = (temp[ y*8] + 2)>>2; | |
2756 src[7+y*stride] = (temp[7+y*8] + 2)>>2; | |
2757 for(x=1; x<7; x++){ | |
2758 xy = y * stride + x; | |
2759 yz = y * 8 + x; | |
2760 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
|
2761 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2762 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2763 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2764 |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2765 static inline void h264_loop_filter_luma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0) |
2633 | 2766 { |
2767 int i, d; | |
2768 for( i = 0; i < 4; i++ ) { | |
2769 if( tc0[i] < 0 ) { | |
2770 pix += 4*ystride; | |
2771 continue; | |
2772 } | |
2773 for( d = 0; d < 4; d++ ) { | |
2774 const int p0 = pix[-1*xstride]; | |
2775 const int p1 = pix[-2*xstride]; | |
2776 const int p2 = pix[-3*xstride]; | |
2777 const int q0 = pix[0]; | |
2778 const int q1 = pix[1*xstride]; | |
2779 const int q2 = pix[2*xstride]; | |
2967 | 2780 |
4001 | 2781 if( FFABS( p0 - q0 ) < alpha && |
2782 FFABS( p1 - p0 ) < beta && | |
2783 FFABS( q1 - q0 ) < beta ) { | |
2967 | 2784 |
2633 | 2785 int tc = tc0[i]; |
2786 int i_delta; | |
2967 | 2787 |
4001 | 2788 if( FFABS( p2 - p0 ) < beta ) { |
4594 | 2789 pix[-2*xstride] = p1 + av_clip( (( p2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - p1, -tc0[i], tc0[i] ); |
2633 | 2790 tc++; |
2791 } | |
4001 | 2792 if( FFABS( q2 - q0 ) < beta ) { |
4594 | 2793 pix[ xstride] = q1 + av_clip( (( q2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - q1, -tc0[i], tc0[i] ); |
2633 | 2794 tc++; |
2795 } | |
2967 | 2796 |
4594 | 2797 i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
2798 pix[-xstride] = av_clip_uint8( p0 + i_delta ); /* p0' */ | |
2799 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */ | |
2633 | 2800 } |
2801 pix += ystride; | |
2802 } | |
2803 } | |
2804 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2805 static void h264_v_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2806 { |
2807 h264_loop_filter_luma_c(pix, stride, 1, alpha, beta, tc0); | |
2808 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2809 static void h264_h_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2810 { |
2811 h264_loop_filter_luma_c(pix, 1, stride, alpha, beta, tc0); | |
2812 } | |
2813 | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2814 static inline void h264_loop_filter_chroma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0) |
2633 | 2815 { |
2816 int i, d; | |
2817 for( i = 0; i < 4; i++ ) { | |
2818 const int tc = tc0[i]; | |
2819 if( tc <= 0 ) { | |
2820 pix += 2*ystride; | |
2821 continue; | |
2822 } | |
2823 for( d = 0; d < 2; d++ ) { | |
2824 const int p0 = pix[-1*xstride]; | |
2825 const int p1 = pix[-2*xstride]; | |
2826 const int q0 = pix[0]; | |
2827 const int q1 = pix[1*xstride]; | |
2828 | |
4001 | 2829 if( FFABS( p0 - q0 ) < alpha && |
2830 FFABS( p1 - p0 ) < beta && | |
2831 FFABS( q1 - q0 ) < beta ) { | |
2633 | 2832 |
4594 | 2833 int delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
2834 | |
2835 pix[-xstride] = av_clip_uint8( p0 + delta ); /* p0' */ | |
2836 pix[0] = av_clip_uint8( q0 - delta ); /* q0' */ | |
2633 | 2837 } |
2838 pix += ystride; | |
2839 } | |
2840 } | |
2841 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2842 static void h264_v_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2843 { |
2844 h264_loop_filter_chroma_c(pix, stride, 1, alpha, beta, tc0); | |
2845 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2846 static void h264_h_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2847 { |
2848 h264_loop_filter_chroma_c(pix, 1, stride, alpha, beta, tc0); | |
2849 } | |
2850 | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2851 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
|
2852 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2853 int d; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2854 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
|
2855 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
|
2856 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
|
2857 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
|
2858 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
|
2859 |
4001 | 2860 if( FFABS( p0 - q0 ) < alpha && |
2861 FFABS( p1 - p0 ) < beta && | |
2862 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
|
2863 |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2864 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
|
2865 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
|
2866 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2867 pix += ystride; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2868 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2869 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2870 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
|
2871 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2872 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
|
2873 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2874 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
|
2875 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2876 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
|
2877 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2878 |
1708 | 2879 static inline int pix_abs16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2880 { |
2881 int s, i; | |
2882 | |
2883 s = 0; | |
1708 | 2884 for(i=0;i<h;i++) { |
0 | 2885 s += abs(pix1[0] - pix2[0]); |
2886 s += abs(pix1[1] - pix2[1]); | |
2887 s += abs(pix1[2] - pix2[2]); | |
2888 s += abs(pix1[3] - pix2[3]); | |
2889 s += abs(pix1[4] - pix2[4]); | |
2890 s += abs(pix1[5] - pix2[5]); | |
2891 s += abs(pix1[6] - pix2[6]); | |
2892 s += abs(pix1[7] - pix2[7]); | |
2893 s += abs(pix1[8] - pix2[8]); | |
2894 s += abs(pix1[9] - pix2[9]); | |
2895 s += abs(pix1[10] - pix2[10]); | |
2896 s += abs(pix1[11] - pix2[11]); | |
2897 s += abs(pix1[12] - pix2[12]); | |
2898 s += abs(pix1[13] - pix2[13]); | |
2899 s += abs(pix1[14] - pix2[14]); | |
2900 s += abs(pix1[15] - pix2[15]); | |
2901 pix1 += line_size; | |
2902 pix2 += line_size; | |
2903 } | |
2904 return s; | |
2905 } | |
2906 | |
1708 | 2907 static int pix_abs16_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2908 { |
2909 int s, i; | |
2910 | |
2911 s = 0; | |
1708 | 2912 for(i=0;i<h;i++) { |
0 | 2913 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
2914 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
2915 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
2916 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
2917 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
2918 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
2919 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
2920 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
2921 s += abs(pix1[8] - avg2(pix2[8], pix2[9])); | |
2922 s += abs(pix1[9] - avg2(pix2[9], pix2[10])); | |
2923 s += abs(pix1[10] - avg2(pix2[10], pix2[11])); | |
2924 s += abs(pix1[11] - avg2(pix2[11], pix2[12])); | |
2925 s += abs(pix1[12] - avg2(pix2[12], pix2[13])); | |
2926 s += abs(pix1[13] - avg2(pix2[13], pix2[14])); | |
2927 s += abs(pix1[14] - avg2(pix2[14], pix2[15])); | |
2928 s += abs(pix1[15] - avg2(pix2[15], pix2[16])); | |
2929 pix1 += line_size; | |
2930 pix2 += line_size; | |
2931 } | |
2932 return s; | |
2933 } | |
2934 | |
1708 | 2935 static int pix_abs16_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2936 { |
2937 int s, i; | |
1064 | 2938 uint8_t *pix3 = pix2 + line_size; |
0 | 2939 |
2940 s = 0; | |
1708 | 2941 for(i=0;i<h;i++) { |
0 | 2942 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
2943 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
2944 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
2945 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
2946 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
2947 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
2948 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
2949 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
2950 s += abs(pix1[8] - avg2(pix2[8], pix3[8])); | |
2951 s += abs(pix1[9] - avg2(pix2[9], pix3[9])); | |
2952 s += abs(pix1[10] - avg2(pix2[10], pix3[10])); | |
2953 s += abs(pix1[11] - avg2(pix2[11], pix3[11])); | |
2954 s += abs(pix1[12] - avg2(pix2[12], pix3[12])); | |
2955 s += abs(pix1[13] - avg2(pix2[13], pix3[13])); | |
2956 s += abs(pix1[14] - avg2(pix2[14], pix3[14])); | |
2957 s += abs(pix1[15] - avg2(pix2[15], pix3[15])); | |
2958 pix1 += line_size; | |
2959 pix2 += line_size; | |
2960 pix3 += line_size; | |
2961 } | |
2962 return s; | |
2963 } | |
2964 | |
1708 | 2965 static int pix_abs16_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2966 { |
2967 int s, i; | |
1064 | 2968 uint8_t *pix3 = pix2 + line_size; |
0 | 2969 |
2970 s = 0; | |
1708 | 2971 for(i=0;i<h;i++) { |
0 | 2972 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
2973 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
2974 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
2975 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
2976 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
2977 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
2978 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
2979 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
2980 s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9])); | |
2981 s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10])); | |
2982 s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11])); | |
2983 s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12])); | |
2984 s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13])); | |
2985 s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14])); | |
2986 s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15])); | |
2987 s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16])); | |
2988 pix1 += line_size; | |
2989 pix2 += line_size; | |
2990 pix3 += line_size; | |
2991 } | |
2992 return s; | |
2993 } | |
2994 | |
1708 | 2995 static inline int pix_abs8_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 2996 { |
2997 int s, i; | |
2998 | |
2999 s = 0; | |
1708 | 3000 for(i=0;i<h;i++) { |
294 | 3001 s += abs(pix1[0] - pix2[0]); |
3002 s += abs(pix1[1] - pix2[1]); | |
3003 s += abs(pix1[2] - pix2[2]); | |
3004 s += abs(pix1[3] - pix2[3]); | |
3005 s += abs(pix1[4] - pix2[4]); | |
3006 s += abs(pix1[5] - pix2[5]); | |
3007 s += abs(pix1[6] - pix2[6]); | |
3008 s += abs(pix1[7] - pix2[7]); | |
3009 pix1 += line_size; | |
3010 pix2 += line_size; | |
3011 } | |
3012 return s; | |
3013 } | |
3014 | |
1708 | 3015 static int pix_abs8_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3016 { |
3017 int s, i; | |
3018 | |
3019 s = 0; | |
1708 | 3020 for(i=0;i<h;i++) { |
294 | 3021 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
3022 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
3023 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
3024 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
3025 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
3026 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
3027 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
3028 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
3029 pix1 += line_size; | |
3030 pix2 += line_size; | |
3031 } | |
3032 return s; | |
3033 } | |
3034 | |
1708 | 3035 static int pix_abs8_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3036 { |
3037 int s, i; | |
1064 | 3038 uint8_t *pix3 = pix2 + line_size; |
294 | 3039 |
3040 s = 0; | |
1708 | 3041 for(i=0;i<h;i++) { |
294 | 3042 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
3043 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
3044 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
3045 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
3046 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
3047 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
3048 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
3049 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
3050 pix1 += line_size; | |
3051 pix2 += line_size; | |
3052 pix3 += line_size; | |
3053 } | |
3054 return s; | |
3055 } | |
3056 | |
1708 | 3057 static int pix_abs8_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3058 { |
3059 int s, i; | |
1064 | 3060 uint8_t *pix3 = pix2 + line_size; |
294 | 3061 |
3062 s = 0; | |
1708 | 3063 for(i=0;i<h;i++) { |
294 | 3064 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
3065 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
3066 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
3067 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
3068 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
3069 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
3070 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
3071 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
3072 pix1 += line_size; | |
3073 pix2 += line_size; | |
3074 pix3 += line_size; | |
3075 } | |
3076 return s; | |
3077 } | |
3078 | |
2834 | 3079 static int nsse16_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3080 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3081 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3082 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3083 int x,y; |
2066 | 3084 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3085 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3086 for(x=0; x<16; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3087 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3088 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3089 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3090 for(x=0; x<15; x++){ |
4001 | 3091 score2+= FFABS( s1[x ] - s1[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3092 - s1[x+1] + s1[x+1+stride]) |
4001 | 3093 -FFABS( s2[x ] - s2[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3094 - s2[x+1] + s2[x+1+stride]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3095 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3096 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3097 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3098 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3099 } |
2066 | 3100 |
4001 | 3101 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight; |
3102 else return score1 + FFABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3103 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3104 |
2834 | 3105 static int nsse8_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3106 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3107 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3108 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3109 int x,y; |
2967 | 3110 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3111 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3112 for(x=0; x<8; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3113 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3114 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3115 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3116 for(x=0; x<7; x++){ |
4001 | 3117 score2+= FFABS( s1[x ] - s1[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3118 - s1[x+1] + s1[x+1+stride]) |
4001 | 3119 -FFABS( s2[x ] - s2[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3120 - s2[x+1] + s2[x+1+stride]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3121 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3122 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3123 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3124 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3125 } |
2967 | 3126 |
4001 | 3127 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight; |
3128 else return score1 + FFABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3129 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3130 |
1784 | 3131 static int try_8x8basis_c(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale){ |
3132 int i; | |
3133 unsigned int sum=0; | |
3134 | |
3135 for(i=0; i<8*8; i++){ | |
3136 int b= rem[i] + ((basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT)); | |
3137 int w= weight[i]; | |
3138 b>>= RECON_SHIFT; | |
3139 assert(-512<b && b<512); | |
3140 | |
3141 sum += (w*b)*(w*b)>>4; | |
3142 } | |
3143 return sum>>2; | |
3144 } | |
3145 | |
3146 static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale){ | |
3147 int i; | |
3148 | |
3149 for(i=0; i<8*8; i++){ | |
3150 rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT); | |
2967 | 3151 } |
1784 | 3152 } |
3153 | |
1100 | 3154 /** |
3155 * permutes an 8x8 block. | |
1101 | 3156 * @param block the block which will be permuted according to the given permutation vector |
1100 | 3157 * @param permutation the permutation vector |
3158 * @param last the last non zero coefficient in scantable order, used to speed the permutation up | |
2967 | 3159 * @param scantable the used scantable, this is only used to speed the permutation up, the block is not |
1101 | 3160 * (inverse) permutated to scantable order! |
1100 | 3161 */ |
1064 | 3162 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
|
3163 { |
764 | 3164 int i; |
945 | 3165 DCTELEM temp[64]; |
2967 | 3166 |
764 | 3167 if(last<=0) return; |
5129 | 3168 //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
|
3169 |
764 | 3170 for(i=0; i<=last; i++){ |
3171 const int j= scantable[i]; | |
3172 temp[j]= block[j]; | |
3173 block[j]=0; | |
3174 } | |
2967 | 3175 |
764 | 3176 for(i=0; i<=last; i++){ |
3177 const int j= scantable[i]; | |
3178 const int perm_j= permutation[j]; | |
3179 block[perm_j]= temp[j]; | |
3180 } | |
174
ac5075a55488
new IDCT code by Michael Niedermayer (michaelni@gmx.at) - #define SIMPLE_IDCT to enable
arpi_esp
parents:
88
diff
changeset
|
3181 } |
34 | 3182 |
1729 | 3183 static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride, int h){ |
3184 return 0; | |
3185 } | |
3186 | |
3187 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type){ | |
3188 int i; | |
2967 | 3189 |
1729 | 3190 memset(cmp, 0, sizeof(void*)*5); |
2967 | 3191 |
1729 | 3192 for(i=0; i<5; i++){ |
3193 switch(type&0xFF){ | |
3194 case FF_CMP_SAD: | |
3195 cmp[i]= c->sad[i]; | |
3196 break; | |
3197 case FF_CMP_SATD: | |
3198 cmp[i]= c->hadamard8_diff[i]; | |
3199 break; | |
3200 case FF_CMP_SSE: | |
3201 cmp[i]= c->sse[i]; | |
3202 break; | |
3203 case FF_CMP_DCT: | |
3204 cmp[i]= c->dct_sad[i]; | |
3205 break; | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3206 case FF_CMP_DCT264: |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3207 cmp[i]= c->dct264_sad[i]; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3208 break; |
2382 | 3209 case FF_CMP_DCTMAX: |
3210 cmp[i]= c->dct_max[i]; | |
3211 break; | |
1729 | 3212 case FF_CMP_PSNR: |
3213 cmp[i]= c->quant_psnr[i]; | |
3214 break; | |
3215 case FF_CMP_BIT: | |
3216 cmp[i]= c->bit[i]; | |
3217 break; | |
3218 case FF_CMP_RD: | |
3219 cmp[i]= c->rd[i]; | |
3220 break; | |
3221 case FF_CMP_VSAD: | |
3222 cmp[i]= c->vsad[i]; | |
3223 break; | |
3224 case FF_CMP_VSSE: | |
3225 cmp[i]= c->vsse[i]; | |
3226 break; | |
3227 case FF_CMP_ZERO: | |
3228 cmp[i]= zero_cmp; | |
3229 break; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3230 case FF_CMP_NSSE: |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3231 cmp[i]= c->nsse[i]; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3232 break; |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
3233 #ifdef CONFIG_SNOW_ENCODER |
2184 | 3234 case FF_CMP_W53: |
3235 cmp[i]= c->w53[i]; | |
3236 break; | |
3237 case FF_CMP_W97: | |
3238 cmp[i]= c->w97[i]; | |
3239 break; | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
3240 #endif |
1729 | 3241 default: |
3242 av_log(NULL, AV_LOG_ERROR,"internal error in cmp function selection\n"); | |
3243 } | |
3244 } | |
3245 } | |
3246 | |
1101 | 3247 /** |
3248 * memset(blocks, 0, sizeof(DCTELEM)*6*64) | |
3249 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
3250 static void clear_blocks_c(DCTELEM *blocks) |
296 | 3251 { |
3252 memset(blocks, 0, sizeof(DCTELEM)*6*64); | |
3253 } | |
3254 | |
866 | 3255 static void add_bytes_c(uint8_t *dst, uint8_t *src, int w){ |
3256 int i; | |
996
ad44196ea483
add/diff_bytes bugfix patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>)
michaelni
parents:
984
diff
changeset
|
3257 for(i=0; i+7<w; i+=8){ |
866 | 3258 dst[i+0] += src[i+0]; |
3259 dst[i+1] += src[i+1]; | |
3260 dst[i+2] += src[i+2]; | |
3261 dst[i+3] += src[i+3]; | |
3262 dst[i+4] += src[i+4]; | |
3263 dst[i+5] += src[i+5]; | |
3264 dst[i+6] += src[i+6]; | |
3265 dst[i+7] += src[i+7]; | |
3266 } | |
3267 for(; i<w; i++) | |
3268 dst[i+0] += src[i+0]; | |
3269 } | |
3270 | |
3271 static void diff_bytes_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ | |
3272 int i; | |
996
ad44196ea483
add/diff_bytes bugfix patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>)
michaelni
parents:
984
diff
changeset
|
3273 for(i=0; i+7<w; i+=8){ |
866 | 3274 dst[i+0] = src1[i+0]-src2[i+0]; |
3275 dst[i+1] = src1[i+1]-src2[i+1]; | |
3276 dst[i+2] = src1[i+2]-src2[i+2]; | |
3277 dst[i+3] = src1[i+3]-src2[i+3]; | |
3278 dst[i+4] = src1[i+4]-src2[i+4]; | |
3279 dst[i+5] = src1[i+5]-src2[i+5]; | |
3280 dst[i+6] = src1[i+6]-src2[i+6]; | |
3281 dst[i+7] = src1[i+7]-src2[i+7]; | |
3282 } | |
3283 for(; i<w; i++) | |
3284 dst[i+0] = src1[i+0]-src2[i+0]; | |
3285 } | |
3286 | |
1527 | 3287 static void sub_hfyu_median_prediction_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){ |
3288 int i; | |
3289 uint8_t l, lt; | |
3290 | |
3291 l= *left; | |
3292 lt= *left_top; | |
3293 | |
3294 for(i=0; i<w; i++){ | |
3295 const int pred= mid_pred(l, src1[i], (l + src1[i] - lt)&0xFF); | |
3296 lt= src1[i]; | |
3297 l= src2[i]; | |
3298 dst[i]= l - pred; | |
2967 | 3299 } |
1527 | 3300 |
3301 *left= l; | |
3302 *left_top= lt; | |
3303 } | |
3304 | |
936 | 3305 #define BUTTERFLY2(o1,o2,i1,i2) \ |
3306 o1= (i1)+(i2);\ | |
3307 o2= (i1)-(i2); | |
3308 | |
3309 #define BUTTERFLY1(x,y) \ | |
3310 {\ | |
3311 int a,b;\ | |
3312 a= x;\ | |
3313 b= y;\ | |
3314 x= a+b;\ | |
3315 y= a-b;\ | |
3316 } | |
3317 | |
4001 | 3318 #define BUTTERFLYA(x,y) (FFABS((x)+(y)) + FFABS((x)-(y))) |
936 | 3319 |
1708 | 3320 static int hadamard8_diff8x8_c(/*MpegEncContext*/ void *s, uint8_t *dst, uint8_t *src, int stride, int h){ |
936 | 3321 int i; |
3322 int temp[64]; | |
3323 int sum=0; | |
2967 | 3324 |
1708 | 3325 assert(h==8); |
936 | 3326 |
3327 for(i=0; i<8; i++){ | |
3328 //FIXME try pointer walks | |
3329 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]); | |
3330 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]); | |
3331 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]); | |
3332 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 | 3333 |
936 | 3334 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3335 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3336 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3337 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3338 |
936 | 3339 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3340 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3341 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3342 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3343 } | |
3344 | |
3345 for(i=0; i<8; i++){ | |
3346 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3347 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3348 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3349 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3350 |
936 | 3351 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3352 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3353 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3354 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
3355 | |
2967 | 3356 sum += |
936 | 3357 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3358 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3359 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3360 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3361 } | |
3362 #if 0 | |
3363 static int maxi=0; | |
3364 if(sum>maxi){ | |
3365 maxi=sum; | |
3366 printf("MAX:%d\n", maxi); | |
3367 } | |
3368 #endif | |
3369 return sum; | |
3370 } | |
3371 | |
1729 | 3372 static int hadamard8_intra8x8_c(/*MpegEncContext*/ void *s, uint8_t *src, uint8_t *dummy, int stride, int h){ |
936 | 3373 int i; |
3374 int temp[64]; | |
3375 int sum=0; | |
2967 | 3376 |
1729 | 3377 assert(h==8); |
2967 | 3378 |
936 | 3379 for(i=0; i<8; i++){ |
3380 //FIXME try pointer walks | |
1729 | 3381 BUTTERFLY2(temp[8*i+0], temp[8*i+1], src[stride*i+0],src[stride*i+1]); |
3382 BUTTERFLY2(temp[8*i+2], temp[8*i+3], src[stride*i+2],src[stride*i+3]); | |
3383 BUTTERFLY2(temp[8*i+4], temp[8*i+5], src[stride*i+4],src[stride*i+5]); | |
3384 BUTTERFLY2(temp[8*i+6], temp[8*i+7], src[stride*i+6],src[stride*i+7]); | |
2967 | 3385 |
936 | 3386 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3387 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3388 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3389 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3390 |
936 | 3391 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3392 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3393 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3394 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3395 } | |
3396 | |
3397 for(i=0; i<8; i++){ | |
3398 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3399 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3400 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3401 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3402 |
936 | 3403 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3404 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3405 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3406 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
2967 | 3407 |
3408 sum += | |
936 | 3409 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3410 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3411 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3412 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3413 } | |
2967 | 3414 |
4001 | 3415 sum -= FFABS(temp[8*0] + temp[8*4]); // -mean |
2967 | 3416 |
936 | 3417 return sum; |
3418 } | |
3419 | |
1708 | 3420 static int dct_sad8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3421 MpegEncContext * const s= (MpegEncContext *)c; |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
3422 DECLARE_ALIGNED_16(uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
1016 | 3423 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
2967 | 3424 |
1708 | 3425 assert(h==8); |
936 | 3426 |
3427 s->dsp.diff_pixels(temp, src1, src2, stride); | |
1092 | 3428 s->dsp.fdct(temp); |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
3429 return s->dsp.sum_abs_dctelem(temp); |
936 | 3430 } |
3431 | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3432 #ifdef CONFIG_GPL |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3433 #define DCT8_1D {\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3434 const int s07 = SRC(0) + SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3435 const int s16 = SRC(1) + SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3436 const int s25 = SRC(2) + SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3437 const int s34 = SRC(3) + SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3438 const int a0 = s07 + s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3439 const int a1 = s16 + s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3440 const int a2 = s07 - s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3441 const int a3 = s16 - s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3442 const int d07 = SRC(0) - SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3443 const int d16 = SRC(1) - SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3444 const int d25 = SRC(2) - SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3445 const int d34 = SRC(3) - SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3446 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
|
3447 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
|
3448 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
|
3449 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
|
3450 DST(0, a0 + a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3451 DST(1, a4 + (a7>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3452 DST(2, a2 + (a3>>1)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3453 DST(3, a5 + (a6>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3454 DST(4, a0 - a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3455 DST(5, a6 - (a5>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3456 DST(6, (a2>>1) - a3 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3457 DST(7, (a4>>2) - a7 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3458 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3459 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3460 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
|
3461 MpegEncContext * const s= (MpegEncContext *)c; |
5256 | 3462 DCTELEM dct[8][8]; |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3463 int i; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3464 int sum=0; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3465 |
5256 | 3466 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
|
3467 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3468 #define SRC(x) dct[i][x] |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3469 #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
|
3470 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3471 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3472 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3473 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3474 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3475 #define SRC(x) dct[x][i] |
4001 | 3476 #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
|
3477 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3478 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3479 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3480 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3481 return sum; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3482 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3483 #endif |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3484 |
2382 | 3485 static int dct_max8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
3486 MpegEncContext * const s= (MpegEncContext *)c; | |
3089 | 3487 DECLARE_ALIGNED_8(uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
2382 | 3488 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3489 int sum=0, i; | |
2967 | 3490 |
2382 | 3491 assert(h==8); |
3492 | |
3493 s->dsp.diff_pixels(temp, src1, src2, stride); | |
3494 s->dsp.fdct(temp); | |
3495 | |
3496 for(i=0; i<64; i++) | |
4001 | 3497 sum= FFMAX(sum, FFABS(temp[i])); |
2967 | 3498 |
2382 | 3499 return sum; |
3500 } | |
3501 | |
1708 | 3502 static int quant_psnr8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3503 MpegEncContext * const s= (MpegEncContext *)c; |
3089 | 3504 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64*2/8]); |
1016 | 3505 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3506 DCTELEM * const bak = ((DCTELEM*)aligned_temp)+64; | |
936 | 3507 int sum=0, i; |
3508 | |
1708 | 3509 assert(h==8); |
936 | 3510 s->mb_intra=0; |
2967 | 3511 |
936 | 3512 s->dsp.diff_pixels(temp, src1, src2, stride); |
2967 | 3513 |
936 | 3514 memcpy(bak, temp, 64*sizeof(DCTELEM)); |
2967 | 3515 |
1013 | 3516 s->block_last_index[0/*FIXME*/]= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
1689 | 3517 s->dct_unquantize_inter(s, temp, 0, s->qscale); |
6001 | 3518 ff_simple_idct(temp); //FIXME |
2967 | 3519 |
936 | 3520 for(i=0; i<64; i++) |
3521 sum+= (temp[i]-bak[i])*(temp[i]-bak[i]); | |
2967 | 3522 |
936 | 3523 return sum; |
3524 } | |
3525 | |
1708 | 3526 static int rd8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3527 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3528 const uint8_t *scantable= s->intra_scantable.permutated; |
3089 | 3529 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
3530 DECLARE_ALIGNED_8 (uint64_t, aligned_bak[stride]); | |
1016 | 3531 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3532 uint8_t * const bak= (uint8_t*)aligned_bak; | |
1007 | 3533 int i, last, run, bits, level, distoration, start_i; |
3534 const int esc_length= s->ac_esc_length; | |
3535 uint8_t * length; | |
3536 uint8_t * last_length; | |
2967 | 3537 |
1708 | 3538 assert(h==8); |
3539 | |
1007 | 3540 for(i=0; i<8; i++){ |
3541 ((uint32_t*)(bak + i*stride))[0]= ((uint32_t*)(src2 + i*stride))[0]; | |
3542 ((uint32_t*)(bak + i*stride))[1]= ((uint32_t*)(src2 + i*stride))[1]; | |
3543 } | |
3544 | |
3545 s->dsp.diff_pixels(temp, src1, src2, stride); | |
3546 | |
1013 | 3547 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3548 | |
3549 bits=0; | |
2967 | 3550 |
1013 | 3551 if (s->mb_intra) { |
2967 | 3552 start_i = 1; |
1013 | 3553 length = s->intra_ac_vlc_length; |
3554 last_length= s->intra_ac_vlc_last_length; | |
3555 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma | |
3556 } else { | |
3557 start_i = 0; | |
3558 length = s->inter_ac_vlc_length; | |
3559 last_length= s->inter_ac_vlc_last_length; | |
3560 } | |
2967 | 3561 |
1013 | 3562 if(last>=start_i){ |
1007 | 3563 run=0; |
3564 for(i=start_i; i<last; i++){ | |
3565 int j= scantable[i]; | |
3566 level= temp[j]; | |
2967 | 3567 |
1007 | 3568 if(level){ |
3569 level+=64; | |
3570 if((level&(~127)) == 0){ | |
3571 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3572 }else | |
3573 bits+= esc_length; | |
3574 run=0; | |
3575 }else | |
3576 run++; | |
3577 } | |
3578 i= scantable[last]; | |
2967 | 3579 |
1011 | 3580 level= temp[i] + 64; |
3581 | |
3582 assert(level - 64); | |
2967 | 3583 |
1007 | 3584 if((level&(~127)) == 0){ |
3585 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3586 }else | |
3587 bits+= esc_length; | |
2967 | 3588 |
1013 | 3589 } |
3590 | |
3591 if(last>=0){ | |
1689 | 3592 if(s->mb_intra) |
3593 s->dct_unquantize_intra(s, temp, 0, s->qscale); | |
3594 else | |
3595 s->dct_unquantize_inter(s, temp, 0, s->qscale); | |
1007 | 3596 } |
2967 | 3597 |
1092 | 3598 s->dsp.idct_add(bak, stride, temp); |
2967 | 3599 |
1708 | 3600 distoration= s->dsp.sse[1](NULL, bak, src1, stride, 8); |
1007 | 3601 |
1013 | 3602 return distoration + ((bits*s->qscale*s->qscale*109 + 64)>>7); |
1007 | 3603 } |
3604 | |
1708 | 3605 static int bit8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3606 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3607 const uint8_t *scantable= s->intra_scantable.permutated; |
3089 | 3608 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
1016 | 3609 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
1007 | 3610 int i, last, run, bits, level, start_i; |
3611 const int esc_length= s->ac_esc_length; | |
3612 uint8_t * length; | |
3613 uint8_t * last_length; | |
1708 | 3614 |
3615 assert(h==8); | |
2967 | 3616 |
1013 | 3617 s->dsp.diff_pixels(temp, src1, src2, stride); |
1007 | 3618 |
1013 | 3619 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3620 | |
3621 bits=0; | |
2967 | 3622 |
1007 | 3623 if (s->mb_intra) { |
2967 | 3624 start_i = 1; |
1007 | 3625 length = s->intra_ac_vlc_length; |
3626 last_length= s->intra_ac_vlc_last_length; | |
1013 | 3627 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma |
1007 | 3628 } else { |
3629 start_i = 0; | |
3630 length = s->inter_ac_vlc_length; | |
3631 last_length= s->inter_ac_vlc_last_length; | |
3632 } | |
2967 | 3633 |
1013 | 3634 if(last>=start_i){ |
1007 | 3635 run=0; |
3636 for(i=start_i; i<last; i++){ | |
3637 int j= scantable[i]; | |
3638 level= temp[j]; | |
2967 | 3639 |
1007 | 3640 if(level){ |
3641 level+=64; | |
3642 if((level&(~127)) == 0){ | |
3643 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3644 }else | |
3645 bits+= esc_length; | |
3646 run=0; | |
3647 }else | |
3648 run++; | |
3649 } | |
3650 i= scantable[last]; | |
2967 | 3651 |
1013 | 3652 level= temp[i] + 64; |
2967 | 3653 |
1013 | 3654 assert(level - 64); |
2967 | 3655 |
1007 | 3656 if((level&(~127)) == 0){ |
3657 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3658 }else | |
3659 bits+= esc_length; | |
3660 } | |
3661 | |
3662 return bits; | |
3663 } | |
3664 | |
1729 | 3665 static int vsad_intra16_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ |
3666 int score=0; | |
3667 int x,y; | |
2967 | 3668 |
1729 | 3669 for(y=1; y<h; y++){ |
3670 for(x=0; x<16; x+=4){ | |
4001 | 3671 score+= FFABS(s[x ] - s[x +stride]) + FFABS(s[x+1] - s[x+1+stride]) |
3672 +FFABS(s[x+2] - s[x+2+stride]) + FFABS(s[x+3] - s[x+3+stride]); | |
1729 | 3673 } |
3674 s+= stride; | |
3675 } | |
2967 | 3676 |
1729 | 3677 return score; |
3678 } | |
3679 | |
3680 static int vsad16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3681 int score=0; | |
3682 int x,y; | |
2967 | 3683 |
1729 | 3684 for(y=1; y<h; y++){ |
3685 for(x=0; x<16; x++){ | |
4001 | 3686 score+= FFABS(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); |
1729 | 3687 } |
3688 s1+= stride; | |
3689 s2+= stride; | |
3690 } | |
2967 | 3691 |
1729 | 3692 return score; |
3693 } | |
3694 | |
3695 #define SQ(a) ((a)*(a)) | |
3696 static int vsse_intra16_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ | |
3697 int score=0; | |
3698 int x,y; | |
2967 | 3699 |
1729 | 3700 for(y=1; y<h; y++){ |
3701 for(x=0; x<16; x+=4){ | |
2967 | 3702 score+= SQ(s[x ] - s[x +stride]) + SQ(s[x+1] - s[x+1+stride]) |
1729 | 3703 +SQ(s[x+2] - s[x+2+stride]) + SQ(s[x+3] - s[x+3+stride]); |
3704 } | |
3705 s+= stride; | |
3706 } | |
2967 | 3707 |
1729 | 3708 return score; |
3709 } | |
3710 | |
3711 static int vsse16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3712 int score=0; | |
3713 int x,y; | |
2967 | 3714 |
1729 | 3715 for(y=1; y<h; y++){ |
3716 for(x=0; x<16; x++){ | |
3717 score+= SQ(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); | |
3718 } | |
3719 s1+= stride; | |
3720 s2+= stride; | |
3721 } | |
2967 | 3722 |
1729 | 3723 return score; |
3724 } | |
3725 | |
5255 | 3726 static int ssd_int8_vs_int16_c(const int8_t *pix1, const int16_t *pix2, |
3727 int size){ | |
4749 | 3728 int score=0; |
3729 int i; | |
3730 for(i=0; i<size; i++) | |
3731 score += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]); | |
3732 return score; | |
3733 } | |
3734 | |
1708 | 3735 WARPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c) |
1729 | 3736 WARPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c) |
1708 | 3737 WARPER8_16_SQ(dct_sad8x8_c, dct_sad16_c) |
3013 | 3738 #ifdef CONFIG_GPL |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3739 WARPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c) |
3013 | 3740 #endif |
2382 | 3741 WARPER8_16_SQ(dct_max8x8_c, dct_max16_c) |
1708 | 3742 WARPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c) |
3743 WARPER8_16_SQ(rd8x8_c, rd16_c) | |
3744 WARPER8_16_SQ(bit8x8_c, bit16_c) | |
936 | 3745 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3746 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
|
3747 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3748 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3749 dst[i] *= src[i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3750 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3751 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3752 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
|
3753 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3754 src1 += len-1; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3755 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3756 dst[i] = src0[i] * src1[-i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3757 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3758 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3759 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
|
3760 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3761 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3762 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
|
3763 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3764 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3765 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
|
3766 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3767 for(i=0; i<len; i++) { |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3768 int_fast32_t tmp = ((int32_t*)src)[i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3769 if(tmp & 0xf0000){ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3770 tmp = (0x43c0ffff - tmp)>>31; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3771 // 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
|
3772 // if(tmp > 0x43c0ffff) tmp = 0xFFFF; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3773 // else tmp = 0; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3774 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3775 dst[i] = tmp - 0x8000; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3776 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3777 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3778 |
5887 | 3779 #define W0 2048 |
3780 #define W1 2841 /* 2048*sqrt (2)*cos (1*pi/16) */ | |
3781 #define W2 2676 /* 2048*sqrt (2)*cos (2*pi/16) */ | |
3782 #define W3 2408 /* 2048*sqrt (2)*cos (3*pi/16) */ | |
3783 #define W4 2048 /* 2048*sqrt (2)*cos (4*pi/16) */ | |
3784 #define W5 1609 /* 2048*sqrt (2)*cos (5*pi/16) */ | |
3785 #define W6 1108 /* 2048*sqrt (2)*cos (6*pi/16) */ | |
3786 #define W7 565 /* 2048*sqrt (2)*cos (7*pi/16) */ | |
3787 | |
3788 static void wmv2_idct_row(short * b) | |
3789 { | |
3790 int s1,s2; | |
3791 int a0,a1,a2,a3,a4,a5,a6,a7; | |
3792 /*step 1*/ | |
3793 a1 = W1*b[1]+W7*b[7]; | |
3794 a7 = W7*b[1]-W1*b[7]; | |
3795 a5 = W5*b[5]+W3*b[3]; | |
3796 a3 = W3*b[5]-W5*b[3]; | |
3797 a2 = W2*b[2]+W6*b[6]; | |
3798 a6 = W6*b[2]-W2*b[6]; | |
3799 a0 = W0*b[0]+W0*b[4]; | |
3800 a4 = W0*b[0]-W0*b[4]; | |
3801 /*step 2*/ | |
3802 s1 = (181*(a1-a5+a7-a3)+128)>>8;//1,3,5,7, | |
3803 s2 = (181*(a1-a5-a7+a3)+128)>>8; | |
3804 /*step 3*/ | |
3805 b[0] = (a0+a2+a1+a5 + (1<<7))>>8; | |
3806 b[1] = (a4+a6 +s1 + (1<<7))>>8; | |
3807 b[2] = (a4-a6 +s2 + (1<<7))>>8; | |
3808 b[3] = (a0-a2+a7+a3 + (1<<7))>>8; | |
3809 b[4] = (a0-a2-a7-a3 + (1<<7))>>8; | |
3810 b[5] = (a4-a6 -s2 + (1<<7))>>8; | |
3811 b[6] = (a4+a6 -s1 + (1<<7))>>8; | |
3812 b[7] = (a0+a2-a1-a5 + (1<<7))>>8; | |
3813 } | |
3814 static void wmv2_idct_col(short * b) | |
3815 { | |
3816 int s1,s2; | |
3817 int a0,a1,a2,a3,a4,a5,a6,a7; | |
3818 /*step 1, with extended precision*/ | |
3819 a1 = (W1*b[8*1]+W7*b[8*7] + 4)>>3; | |
3820 a7 = (W7*b[8*1]-W1*b[8*7] + 4)>>3; | |
3821 a5 = (W5*b[8*5]+W3*b[8*3] + 4)>>3; | |
3822 a3 = (W3*b[8*5]-W5*b[8*3] + 4)>>3; | |
3823 a2 = (W2*b[8*2]+W6*b[8*6] + 4)>>3; | |
3824 a6 = (W6*b[8*2]-W2*b[8*6] + 4)>>3; | |
3825 a0 = (W0*b[8*0]+W0*b[8*4] )>>3; | |
3826 a4 = (W0*b[8*0]-W0*b[8*4] )>>3; | |
3827 /*step 2*/ | |
3828 s1 = (181*(a1-a5+a7-a3)+128)>>8; | |
3829 s2 = (181*(a1-a5-a7+a3)+128)>>8; | |
3830 /*step 3*/ | |
3831 b[8*0] = (a0+a2+a1+a5 + (1<<13))>>14; | |
3832 b[8*1] = (a4+a6 +s1 + (1<<13))>>14; | |
3833 b[8*2] = (a4-a6 +s2 + (1<<13))>>14; | |
3834 b[8*3] = (a0-a2+a7+a3 + (1<<13))>>14; | |
3835 | |
3836 b[8*4] = (a0-a2-a7-a3 + (1<<13))>>14; | |
3837 b[8*5] = (a4-a6 -s2 + (1<<13))>>14; | |
3838 b[8*6] = (a4+a6 -s1 + (1<<13))>>14; | |
3839 b[8*7] = (a0+a2-a1-a5 + (1<<13))>>14; | |
3840 } | |
3841 void ff_wmv2_idct_c(short * block){ | |
3842 int i; | |
3843 | |
3844 for(i=0;i<64;i+=8){ | |
3845 wmv2_idct_row(block+i); | |
3846 } | |
3847 for(i=0;i<8;i++){ | |
3848 wmv2_idct_col(block+i); | |
3849 } | |
3850 } | |
1092 | 3851 /* XXX: those functions should be suppressed ASAP when all IDCTs are |
3852 converted */ | |
5887 | 3853 static void ff_wmv2_idct_put_c(uint8_t *dest, int line_size, DCTELEM *block) |
3854 { | |
3855 ff_wmv2_idct_c(block); | |
3856 put_pixels_clamped_c(block, dest, line_size); | |
3857 } | |
3858 static void ff_wmv2_idct_add_c(uint8_t *dest, int line_size, DCTELEM *block) | |
3859 { | |
3860 ff_wmv2_idct_c(block); | |
3861 add_pixels_clamped_c(block, dest, line_size); | |
3862 } | |
1092 | 3863 static void ff_jref_idct_put(uint8_t *dest, int line_size, DCTELEM *block) |
3864 { | |
3865 j_rev_dct (block); | |
3866 put_pixels_clamped_c(block, dest, line_size); | |
3867 } | |
3868 static void ff_jref_idct_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3869 { | |
3870 j_rev_dct (block); | |
3871 add_pixels_clamped_c(block, dest, line_size); | |
3872 } | |
3873 | |
2256 | 3874 static void ff_jref_idct4_put(uint8_t *dest, int line_size, DCTELEM *block) |
3875 { | |
3876 j_rev_dct4 (block); | |
3877 put_pixels_clamped4_c(block, dest, line_size); | |
3878 } | |
3879 static void ff_jref_idct4_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3880 { | |
3881 j_rev_dct4 (block); | |
3882 add_pixels_clamped4_c(block, dest, line_size); | |
3883 } | |
3884 | |
2257 | 3885 static void ff_jref_idct2_put(uint8_t *dest, int line_size, DCTELEM *block) |
3886 { | |
3887 j_rev_dct2 (block); | |
3888 put_pixels_clamped2_c(block, dest, line_size); | |
3889 } | |
3890 static void ff_jref_idct2_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3891 { | |
3892 j_rev_dct2 (block); | |
3893 add_pixels_clamped2_c(block, dest, line_size); | |
3894 } | |
3895 | |
2259 | 3896 static void ff_jref_idct1_put(uint8_t *dest, int line_size, DCTELEM *block) |
3897 { | |
4176 | 3898 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2259 | 3899 |
3900 dest[0] = cm[(block[0] + 4)>>3]; | |
3901 } | |
3902 static void ff_jref_idct1_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3903 { | |
4176 | 3904 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2259 | 3905 |
3906 dest[0] = cm[dest[0] + ((block[0] + 4)>>3)]; | |
3907 } | |
3908 | |
5143 | 3909 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
|
3910 |
1201 | 3911 /* init static data */ |
4197 | 3912 void dsputil_static_init(void) |
0 | 3913 { |
751 | 3914 int i; |
0 | 3915 |
4176 | 3916 for(i=0;i<256;i++) ff_cropTbl[i + MAX_NEG_CROP] = i; |
1201 | 3917 for(i=0;i<MAX_NEG_CROP;i++) { |
4176 | 3918 ff_cropTbl[i] = 0; |
3919 ff_cropTbl[i + MAX_NEG_CROP + 256] = 255; | |
1201 | 3920 } |
2967 | 3921 |
1201 | 3922 for(i=0;i<512;i++) { |
4179 | 3923 ff_squareTbl[i] = (i - 256) * (i - 256); |
1201 | 3924 } |
2967 | 3925 |
4197 | 3926 for(i=0; i<64; i++) inv_zigzag_direct16[ff_zigzag_direct[i]]= i+1; |
1201 | 3927 } |
0 | 3928 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3929 int ff_check_alignment(void){ |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3930 static int did_fail=0; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3931 DECLARE_ALIGNED_16(int, aligned); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3932 |
5150 | 3933 if((long)&aligned & 15){ |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3934 if(!did_fail){ |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3935 #if defined(HAVE_MMX) || defined(HAVE_ALTIVEC) |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3936 av_log(NULL, AV_LOG_ERROR, |
4292 | 3937 "Compiler did not align stack variables. Libavcodec has been miscompiled\n" |
3938 "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
|
3939 "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
|
3940 "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
|
3941 #endif |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3942 did_fail=1; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3943 } |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3944 return -1; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3945 } |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3946 return 0; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3947 } |
861 | 3948 |
1201 | 3949 void dsputil_init(DSPContext* c, AVCodecContext *avctx) |
3950 { | |
3951 int i; | |
0 | 3952 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3953 ff_check_alignment(); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3954 |
1092 | 3955 #ifdef CONFIG_ENCODERS |
1567 | 3956 if(avctx->dct_algo==FF_DCT_FASTINT) { |
1092 | 3957 c->fdct = fdct_ifast; |
2979 | 3958 c->fdct248 = fdct_ifast248; |
2967 | 3959 } |
1567 | 3960 else if(avctx->dct_algo==FF_DCT_FAAN) { |
1557 | 3961 c->fdct = ff_faandct; |
2979 | 3962 c->fdct248 = ff_faandct248; |
2967 | 3963 } |
1567 | 3964 else { |
1092 | 3965 c->fdct = ff_jpeg_fdct_islow; //slow/accurate/default |
2979 | 3966 c->fdct248 = ff_fdct248_islow; |
1567 | 3967 } |
1092 | 3968 #endif //CONFIG_ENCODERS |
3969 | |
2256 | 3970 if(avctx->lowres==1){ |
5064 | 3971 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
|
3972 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
|
3973 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
|
3974 }else{ |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
3975 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
|
3976 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
|
3977 } |
2256 | 3978 c->idct = j_rev_dct4; |
1092 | 3979 c->idct_permutation_type= FF_NO_IDCT_PERM; |
2257 | 3980 }else if(avctx->lowres==2){ |
3981 c->idct_put= ff_jref_idct2_put; | |
3982 c->idct_add= ff_jref_idct2_add; | |
3983 c->idct = j_rev_dct2; | |
3984 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2259 | 3985 }else if(avctx->lowres==3){ |
3986 c->idct_put= ff_jref_idct1_put; | |
3987 c->idct_add= ff_jref_idct1_add; | |
3988 c->idct = j_rev_dct1; | |
3989 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2256 | 3990 }else{ |
3991 if(avctx->idct_algo==FF_IDCT_INT){ | |
3992 c->idct_put= ff_jref_idct_put; | |
3993 c->idct_add= ff_jref_idct_add; | |
3994 c->idct = j_rev_dct; | |
3995 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM; | |
5007 | 3996 }else if((ENABLE_VP3_DECODER || ENABLE_VP5_DECODER || ENABLE_VP6_DECODER || ENABLE_THEORA_DECODER ) && |
3997 avctx->idct_algo==FF_IDCT_VP3){ | |
2693 | 3998 c->idct_put= ff_vp3_idct_put_c; |
3999 c->idct_add= ff_vp3_idct_add_c; | |
4000 c->idct = ff_vp3_idct_c; | |
4001 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
5887 | 4002 }else if(avctx->idct_algo==FF_IDCT_WMV2){ |
4003 c->idct_put= ff_wmv2_idct_put_c; | |
4004 c->idct_add= ff_wmv2_idct_add_c; | |
4005 c->idct = ff_wmv2_idct_c; | |
4006 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2256 | 4007 }else{ //accurate/default |
6001 | 4008 c->idct_put= ff_simple_idct_put; |
4009 c->idct_add= ff_simple_idct_add; | |
4010 c->idct = ff_simple_idct; | |
2256 | 4011 c->idct_permutation_type= FF_NO_IDCT_PERM; |
4012 } | |
1092 | 4013 } |
4014 | |
5064 | 4015 if (ENABLE_H264_DECODER) { |
5065 | 4016 c->h264_idct_add= ff_h264_idct_add_c; |
4017 c->h264_idct8_add= ff_h264_idct8_add_c; | |
4018 c->h264_idct_dc_add= ff_h264_idct_dc_add_c; | |
4019 c->h264_idct8_dc_add= ff_h264_idct8_dc_add_c; | |
5064 | 4020 } |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4021 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4022 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
|
4023 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
|
4024 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
|
4025 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
|
4026 c->add_pixels_clamped = add_pixels_clamped_c; |
2763 | 4027 c->add_pixels8 = add_pixels8_c; |
4028 c->add_pixels4 = add_pixels4_c; | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
4029 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
|
4030 c->gmc1 = gmc1_c; |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
4031 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
|
4032 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
|
4033 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
|
4034 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
|
4035 |
859 | 4036 /* TODO [0] 16 [1] 8 */ |
1708 | 4037 c->pix_abs[0][0] = pix_abs16_c; |
4038 c->pix_abs[0][1] = pix_abs16_x2_c; | |
4039 c->pix_abs[0][2] = pix_abs16_y2_c; | |
4040 c->pix_abs[0][3] = pix_abs16_xy2_c; | |
4041 c->pix_abs[1][0] = pix_abs8_c; | |
4042 c->pix_abs[1][1] = pix_abs8_x2_c; | |
4043 c->pix_abs[1][2] = pix_abs8_y2_c; | |
4044 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
|
4045 |
859 | 4046 #define dspfunc(PFX, IDX, NUM) \ |
4047 c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## NUM ## _c; \ | |
4048 c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## NUM ## _x2_c; \ | |
4049 c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## NUM ## _y2_c; \ | |
4050 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
|
4051 |
859 | 4052 dspfunc(put, 0, 16); |
4053 dspfunc(put_no_rnd, 0, 16); | |
4054 dspfunc(put, 1, 8); | |
4055 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
|
4056 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
|
4057 dspfunc(put, 3, 2); |
0 | 4058 |
859 | 4059 dspfunc(avg, 0, 16); |
4060 dspfunc(avg_no_rnd, 0, 16); | |
4061 dspfunc(avg, 1, 8); | |
4062 dspfunc(avg_no_rnd, 1, 8); | |
1319 | 4063 dspfunc(avg, 2, 4); |
4064 dspfunc(avg, 3, 2); | |
859 | 4065 #undef dspfunc |
857 | 4066 |
1864 | 4067 c->put_no_rnd_pixels_l2[0]= put_no_rnd_pixels16_l2_c; |
4068 c->put_no_rnd_pixels_l2[1]= put_no_rnd_pixels8_l2_c; | |
4069 | |
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
|
4070 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
|
4071 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
|
4072 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
|
4073 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
|
4074 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
|
4075 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
|
4076 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
|
4077 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
|
4078 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
|
4079 |
1319 | 4080 c->avg_tpel_pixels_tab[ 0] = avg_tpel_pixels_mc00_c; |
4081 c->avg_tpel_pixels_tab[ 1] = avg_tpel_pixels_mc10_c; | |
4082 c->avg_tpel_pixels_tab[ 2] = avg_tpel_pixels_mc20_c; | |
4083 c->avg_tpel_pixels_tab[ 4] = avg_tpel_pixels_mc01_c; | |
4084 c->avg_tpel_pixels_tab[ 5] = avg_tpel_pixels_mc11_c; | |
4085 c->avg_tpel_pixels_tab[ 6] = avg_tpel_pixels_mc21_c; | |
4086 c->avg_tpel_pixels_tab[ 8] = avg_tpel_pixels_mc02_c; | |
4087 c->avg_tpel_pixels_tab[ 9] = avg_tpel_pixels_mc12_c; | |
4088 c->avg_tpel_pixels_tab[10] = avg_tpel_pixels_mc22_c; | |
4089 | |
859 | 4090 #define dspfunc(PFX, IDX, NUM) \ |
4091 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_c; \ | |
4092 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_c; \ | |
4093 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_c; \ | |
4094 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_c; \ | |
4095 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_c; \ | |
4096 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_c; \ | |
4097 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_c; \ | |
4098 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_c; \ | |
4099 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_c; \ | |
4100 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_c; \ | |
4101 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_c; \ | |
4102 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_c; \ | |
4103 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_c; \ | |
4104 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_c; \ | |
4105 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_c; \ | |
4106 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_c | |
857 | 4107 |
859 | 4108 dspfunc(put_qpel, 0, 16); |
4109 dspfunc(put_no_rnd_qpel, 0, 16); | |
4110 | |
4111 dspfunc(avg_qpel, 0, 16); | |
4112 /* dspfunc(avg_no_rnd_qpel, 0, 16); */ | |
857 | 4113 |
859 | 4114 dspfunc(put_qpel, 1, 8); |
4115 dspfunc(put_no_rnd_qpel, 1, 8); | |
4116 | |
4117 dspfunc(avg_qpel, 1, 8); | |
4118 /* dspfunc(avg_no_rnd_qpel, 1, 8); */ | |
1168 | 4119 |
4120 dspfunc(put_h264_qpel, 0, 16); | |
4121 dspfunc(put_h264_qpel, 1, 8); | |
4122 dspfunc(put_h264_qpel, 2, 4); | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
4123 dspfunc(put_h264_qpel, 3, 2); |
1168 | 4124 dspfunc(avg_h264_qpel, 0, 16); |
4125 dspfunc(avg_h264_qpel, 1, 8); | |
4126 dspfunc(avg_h264_qpel, 2, 4); | |
4127 | |
859 | 4128 #undef dspfunc |
1168 | 4129 c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_c; |
4130 c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_c; | |
4131 c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_c; | |
4132 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_c; | |
4133 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_c; | |
4134 c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_c; | |
3663 | 4135 c->put_no_rnd_h264_chroma_pixels_tab[0]= put_no_rnd_h264_chroma_mc8_c; |
857 | 4136 |
2415 | 4137 c->weight_h264_pixels_tab[0]= weight_h264_pixels16x16_c; |
4138 c->weight_h264_pixels_tab[1]= weight_h264_pixels16x8_c; | |
4139 c->weight_h264_pixels_tab[2]= weight_h264_pixels8x16_c; | |
4140 c->weight_h264_pixels_tab[3]= weight_h264_pixels8x8_c; | |
4141 c->weight_h264_pixels_tab[4]= weight_h264_pixels8x4_c; | |
4142 c->weight_h264_pixels_tab[5]= weight_h264_pixels4x8_c; | |
4143 c->weight_h264_pixels_tab[6]= weight_h264_pixels4x4_c; | |
4144 c->weight_h264_pixels_tab[7]= weight_h264_pixels4x2_c; | |
4145 c->weight_h264_pixels_tab[8]= weight_h264_pixels2x4_c; | |
4146 c->weight_h264_pixels_tab[9]= weight_h264_pixels2x2_c; | |
4147 c->biweight_h264_pixels_tab[0]= biweight_h264_pixels16x16_c; | |
4148 c->biweight_h264_pixels_tab[1]= biweight_h264_pixels16x8_c; | |
4149 c->biweight_h264_pixels_tab[2]= biweight_h264_pixels8x16_c; | |
4150 c->biweight_h264_pixels_tab[3]= biweight_h264_pixels8x8_c; | |
4151 c->biweight_h264_pixels_tab[4]= biweight_h264_pixels8x4_c; | |
4152 c->biweight_h264_pixels_tab[5]= biweight_h264_pixels4x8_c; | |
4153 c->biweight_h264_pixels_tab[6]= biweight_h264_pixels4x4_c; | |
4154 c->biweight_h264_pixels_tab[7]= biweight_h264_pixels4x2_c; | |
4155 c->biweight_h264_pixels_tab[8]= biweight_h264_pixels2x4_c; | |
4156 c->biweight_h264_pixels_tab[9]= biweight_h264_pixels2x2_c; | |
4157 | |
3432 | 4158 #ifdef CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4159 ff_cavsdsp_init(c,avctx); |
3432 | 4160 #endif |
3526 | 4161 #if defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
4162 ff_vc1dsp_init(c,avctx); | |
4163 #endif | |
5887 | 4164 #if defined(CONFIG_WMV2_DECODER) || defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
4165 ff_intrax8dsp_init(c,avctx); | |
4166 #endif | |
4296 | 4167 #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
|
4168 ff_h264dspenc_init(c,avctx); |
4296 | 4169 #endif |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4170 |
936 | 4171 c->put_mspel_pixels_tab[0]= put_mspel8_mc00_c; |
4172 c->put_mspel_pixels_tab[1]= put_mspel8_mc10_c; | |
4173 c->put_mspel_pixels_tab[2]= put_mspel8_mc20_c; | |
4174 c->put_mspel_pixels_tab[3]= put_mspel8_mc30_c; | |
4175 c->put_mspel_pixels_tab[4]= put_mspel8_mc02_c; | |
4176 c->put_mspel_pixels_tab[5]= put_mspel8_mc12_c; | |
4177 c->put_mspel_pixels_tab[6]= put_mspel8_mc22_c; | |
4178 c->put_mspel_pixels_tab[7]= put_mspel8_mc32_c; | |
2967 | 4179 |
1708 | 4180 #define SET_CMP_FUNC(name) \ |
4181 c->name[0]= name ## 16_c;\ | |
4182 c->name[1]= name ## 8x8_c; | |
2967 | 4183 |
1708 | 4184 SET_CMP_FUNC(hadamard8_diff) |
1729 | 4185 c->hadamard8_diff[4]= hadamard8_intra16_c; |
1708 | 4186 SET_CMP_FUNC(dct_sad) |
2382 | 4187 SET_CMP_FUNC(dct_max) |
3013 | 4188 #ifdef CONFIG_GPL |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
4189 SET_CMP_FUNC(dct264_sad) |
3013 | 4190 #endif |
1708 | 4191 c->sad[0]= pix_abs16_c; |
4192 c->sad[1]= pix_abs8_c; | |
4193 c->sse[0]= sse16_c; | |
4194 c->sse[1]= sse8_c; | |
2184 | 4195 c->sse[2]= sse4_c; |
1708 | 4196 SET_CMP_FUNC(quant_psnr) |
4197 SET_CMP_FUNC(rd) | |
4198 SET_CMP_FUNC(bit) | |
1729 | 4199 c->vsad[0]= vsad16_c; |
4200 c->vsad[4]= vsad_intra16_c; | |
4201 c->vsse[0]= vsse16_c; | |
4202 c->vsse[4]= vsse_intra16_c; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4203 c->nsse[0]= nsse16_c; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4204 c->nsse[1]= nsse8_c; |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
4205 #ifdef CONFIG_SNOW_ENCODER |
2184 | 4206 c->w53[0]= w53_16_c; |
4207 c->w53[1]= w53_8_c; | |
4208 c->w97[0]= w97_16_c; | |
4209 c->w97[1]= w97_8_c; | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
4210 #endif |
2184 | 4211 |
4749 | 4212 c->ssd_int8_vs_int16 = ssd_int8_vs_int16_c; |
4213 | |
866 | 4214 c->add_bytes= add_bytes_c; |
4215 c->diff_bytes= diff_bytes_c; | |
1527 | 4216 c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_c; |
1273 | 4217 c->bswap_buf= bswap_buf; |
2633 | 4218 |
4219 c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_c; | |
4220 c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_c; | |
4221 c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_c; | |
4222 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
|
4223 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
|
4224 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
|
4225 c->h264_loop_filter_strength= NULL; |
2967 | 4226 |
5277
7b3fcb7c61ce
Avoid linking with h263.c functions when the relevant codecs
aurel
parents:
5256
diff
changeset
|
4227 if (ENABLE_ANY_H263) { |
5278 | 4228 c->h263_h_loop_filter= h263_h_loop_filter_c; |
4229 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
|
4230 } |
2967 | 4231 |
2045 | 4232 c->h261_loop_filter= h261_loop_filter_c; |
2967 | 4233 |
1784 | 4234 c->try_8x8basis= try_8x8basis_c; |
4235 c->add_8x8basis= add_8x8basis_c; | |
866 | 4236 |
4598
5111fceeb971
The Snow DEcoding routines should be under CONFIG_SNOW_DECODER.
diego
parents:
4594
diff
changeset
|
4237 #ifdef CONFIG_SNOW_DECODER |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4238 c->vertical_compose97i = ff_snow_vertical_compose97i; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4239 c->horizontal_compose97i = ff_snow_horizontal_compose97i; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4240 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
|
4241 #endif |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4242 |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4243 #ifdef CONFIG_VORBIS_DECODER |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4244 c->vorbis_inverse_coupling = vorbis_inverse_coupling; |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4245 #endif |
5737 | 4246 #ifdef CONFIG_FLAC_ENCODER |
4247 c->flac_compute_autocorr = ff_flac_compute_autocorr; | |
4248 #endif | |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4249 c->vector_fmul = vector_fmul_c; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4250 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
|
4251 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
|
4252 c->float_to_int16 = ff_float_to_int16_c; |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4253 |
3245 | 4254 c->shrink[0]= ff_img_copy_plane; |
4255 c->shrink[1]= ff_shrink22; | |
4256 c->shrink[2]= ff_shrink44; | |
4257 c->shrink[3]= ff_shrink88; | |
4258 | |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4259 c->prefetch= just_return; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4260 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4261 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
|
4262 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
|
4263 |
5146
5db8d038eb1f
use if() instead of #ifdef and ensure all possible optimisations are enabled
aurel
parents:
5143
diff
changeset
|
4264 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
|
4265 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
|
4266 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
|
4267 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
|
4268 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
|
4269 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
|
4270 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
|
4271 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
|
4272 if (ENABLE_BFIN) dsputil_init_bfin (c, avctx); |
1092 | 4273 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4274 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
|
4275 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
|
4276 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
|
4277 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
|
4278 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
|
4279 } |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4280 |
1092 | 4281 switch(c->idct_permutation_type){ |
4282 case FF_NO_IDCT_PERM: | |
4283 for(i=0; i<64; i++) | |
4284 c->idct_permutation[i]= i; | |
4285 break; | |
4286 case FF_LIBMPEG2_IDCT_PERM: | |
4287 for(i=0; i<64; i++) | |
4288 c->idct_permutation[i]= (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2); | |
4289 break; | |
4290 case FF_SIMPLE_IDCT_PERM: | |
4291 for(i=0; i<64; i++) | |
4292 c->idct_permutation[i]= simple_mmx_permutation[i]; | |
4293 break; | |
4294 case FF_TRANSPOSE_IDCT_PERM: | |
4295 for(i=0; i<64; i++) | |
4296 c->idct_permutation[i]= ((i&7)<<3) | (i>>3); | |
4297 break; | |
2696
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4298 case FF_PARTTRANS_IDCT_PERM: |
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4299 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
|
4300 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
|
4301 break; |
1092 | 4302 default: |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1571
diff
changeset
|
4303 av_log(avctx, AV_LOG_ERROR, "Internal error, IDCT permutation not set\n"); |
1092 | 4304 } |
0 | 4305 } |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
220
diff
changeset
|
4306 |