Mercurial > libavcodec.hg
annotate dsputil.c @ 6390:27a6801cbb27 libavcodec
Fix bug I introduced in r11962.
Thanks to Uoti Urpala for spotting it and proposing the
fix.
author | vitor |
---|---|
date | Sun, 24 Feb 2008 07:47:02 +0000 |
parents | 922774d1fb9c |
children | 7f9e2b5893fc |
rev | line source |
---|---|
0 | 1 /* |
2 * DSP utils | |
429 | 3 * Copyright (c) 2000, 2001 Fabrice Bellard. |
1739
07a484280a82
copyright year update of the files i touched and remembered, things look annoyingly unmaintained otherwise
michael
parents:
1729
diff
changeset
|
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> |
0 | 5 * |
5214 | 6 * gmc & q-pel & 32/64 bit based MC by Michael Niedermayer <michaelni@gmx.at> |
7 * | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
8 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
9 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
10 * FFmpeg is free software; you can redistribute it and/or |
429 | 11 * modify it under the terms of the GNU Lesser General Public |
12 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
13 * version 2.1 of the License, or (at your option) any later version. |
0 | 14 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
15 * FFmpeg is distributed in the hope that it will be useful, |
0 | 16 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
429 | 17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
18 * Lesser General Public License for more details. | |
0 | 19 * |
429 | 20 * You should have received a copy of the GNU Lesser General Public |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
21 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3029
diff
changeset
|
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
0 | 23 */ |
2967 | 24 |
1106 | 25 /** |
26 * @file dsputil.c | |
27 * DSP utils | |
28 */ | |
2967 | 29 |
0 | 30 #include "avcodec.h" |
31 #include "dsputil.h" | |
936 | 32 #include "mpegvideo.h" |
1092 | 33 #include "simple_idct.h" |
1557 | 34 #include "faandct.h" |
5277
7b3fcb7c61ce
Avoid linking with h263.c functions when the relevant codecs
aurel
parents:
5256
diff
changeset
|
35 #include "h263.h" |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
36 #include "snow.h" |
676 | 37 |
2522
e25782262d7d
kill warnings patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2448
diff
changeset
|
38 /* snow.c */ |
e25782262d7d
kill warnings patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2448
diff
changeset
|
39 void ff_spatial_dwt(int *buffer, int width, int height, int stride, int type, int decomposition_count); |
e25782262d7d
kill warnings patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2448
diff
changeset
|
40 |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
41 /* vorbis.c */ |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
42 void vorbis_inverse_coupling(float *mag, float *ang, int blocksize); |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
43 |
5737 | 44 /* flacenc.c */ |
45 void ff_flac_compute_autocorr(const int32_t *data, int len, int lag, double *autoc); | |
46 | |
6384 | 47 /* pngdec.c */ |
48 void ff_add_png_paeth_prediction(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp); | |
49 | |
4176 | 50 uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP] = {0, }; |
4179 | 51 uint32_t ff_squareTbl[512] = {0, }; |
0 | 52 |
6387 | 53 // 0x7f7f7f7f or 0x7f7f7f7f7f7f7f7f or whatever, depending on the cpu's native arithmetic size |
54 #define pb_7f (~0UL/255 * 0x7f) | |
55 #define pb_80 (~0UL/255 * 0x80) | |
6385 | 56 |
1064 | 57 const uint8_t ff_zigzag_direct[64] = { |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
58 0, 1, 8, 16, 9, 2, 3, 10, |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
59 17, 24, 32, 25, 18, 11, 4, 5, |
34 | 60 12, 19, 26, 33, 40, 48, 41, 34, |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
61 27, 20, 13, 6, 7, 14, 21, 28, |
34 | 62 35, 42, 49, 56, 57, 50, 43, 36, |
63 29, 22, 15, 23, 30, 37, 44, 51, | |
64 58, 59, 52, 45, 38, 31, 39, 46, | |
65 53, 60, 61, 54, 47, 55, 62, 63 | |
66 }; | |
67 | |
1567 | 68 /* Specific zigzag scan for 248 idct. NOTE that unlike the |
69 specification, we interleave the fields */ | |
70 const uint8_t ff_zigzag248_direct[64] = { | |
71 0, 8, 1, 9, 16, 24, 2, 10, | |
72 17, 25, 32, 40, 48, 56, 33, 41, | |
73 18, 26, 3, 11, 4, 12, 19, 27, | |
74 34, 42, 49, 57, 50, 58, 35, 43, | |
75 20, 28, 5, 13, 6, 14, 21, 29, | |
76 36, 44, 51, 59, 52, 60, 37, 45, | |
77 22, 30, 7, 15, 23, 31, 38, 46, | |
78 53, 61, 54, 62, 39, 47, 55, 63, | |
79 }; | |
80 | |
220 | 81 /* not permutated inverse zigzag_direct + 1 for MMX quantizer */ |
4197 | 82 DECLARE_ALIGNED_8(uint16_t, inv_zigzag_direct16[64]) = {0, }; |
220 | 83 |
1064 | 84 const uint8_t ff_alternate_horizontal_scan[64] = { |
2967 | 85 0, 1, 2, 3, 8, 9, 16, 17, |
34 | 86 10, 11, 4, 5, 6, 7, 15, 14, |
2967 | 87 13, 12, 19, 18, 24, 25, 32, 33, |
34 | 88 26, 27, 20, 21, 22, 23, 28, 29, |
2967 | 89 30, 31, 34, 35, 40, 41, 48, 49, |
34 | 90 42, 43, 36, 37, 38, 39, 44, 45, |
2967 | 91 46, 47, 50, 51, 56, 57, 58, 59, |
34 | 92 52, 53, 54, 55, 60, 61, 62, 63, |
93 }; | |
94 | |
1064 | 95 const uint8_t ff_alternate_vertical_scan[64] = { |
2967 | 96 0, 8, 16, 24, 1, 9, 2, 10, |
34 | 97 17, 25, 32, 40, 48, 56, 57, 49, |
2967 | 98 41, 33, 26, 18, 3, 11, 4, 12, |
34 | 99 19, 27, 34, 42, 50, 58, 35, 43, |
2967 | 100 51, 59, 20, 28, 5, 13, 6, 14, |
34 | 101 21, 29, 36, 44, 52, 60, 37, 45, |
2967 | 102 53, 61, 22, 30, 7, 15, 23, 31, |
34 | 103 38, 46, 54, 62, 39, 47, 55, 63, |
104 }; | |
105 | |
220 | 106 /* a*inverse[b]>>32 == a/b for all 0<=a<=65536 && 2<=b<=255 */ |
4174 | 107 const uint32_t ff_inverse[256]={ |
2967 | 108 0, 4294967295U,2147483648U,1431655766, 1073741824, 858993460, 715827883, 613566757, |
109 536870912, 477218589, 429496730, 390451573, 357913942, 330382100, 306783379, 286331154, | |
110 268435456, 252645136, 238609295, 226050911, 214748365, 204522253, 195225787, 186737709, | |
111 178956971, 171798692, 165191050, 159072863, 153391690, 148102321, 143165577, 138547333, | |
112 134217728, 130150525, 126322568, 122713352, 119304648, 116080198, 113025456, 110127367, | |
113 107374183, 104755300, 102261127, 99882961, 97612894, 95443718, 93368855, 91382283, | |
114 89478486, 87652394, 85899346, 84215046, 82595525, 81037119, 79536432, 78090315, | |
115 76695845, 75350304, 74051161, 72796056, 71582789, 70409300, 69273667, 68174085, | |
116 67108864, 66076420, 65075263, 64103990, 63161284, 62245903, 61356676, 60492498, | |
117 59652324, 58835169, 58040099, 57266231, 56512728, 55778797, 55063684, 54366675, | |
118 53687092, 53024288, 52377650, 51746594, 51130564, 50529028, 49941481, 49367441, | |
119 48806447, 48258060, 47721859, 47197443, 46684428, 46182445, 45691142, 45210183, | |
120 44739243, 44278014, 43826197, 43383509, 42949673, 42524429, 42107523, 41698712, | |
121 41297763, 40904451, 40518560, 40139882, 39768216, 39403370, 39045158, 38693400, | |
122 38347923, 38008561, 37675152, 37347542, 37025581, 36709123, 36398028, 36092163, | |
123 35791395, 35495598, 35204650, 34918434, 34636834, 34359739, 34087043, 33818641, | |
124 33554432, 33294321, 33038210, 32786010, 32537632, 32292988, 32051995, 31814573, | |
125 31580642, 31350127, 31122952, 30899046, 30678338, 30460761, 30246249, 30034737, | |
126 29826162, 29620465, 29417585, 29217465, 29020050, 28825284, 28633116, 28443493, | |
127 28256364, 28071682, 27889399, 27709467, 27531842, 27356480, 27183338, 27012373, | |
128 26843546, 26676816, 26512144, 26349493, 26188825, 26030105, 25873297, 25718368, | |
129 25565282, 25414008, 25264514, 25116768, 24970741, 24826401, 24683721, 24542671, | |
130 24403224, 24265352, 24129030, 23994231, 23860930, 23729102, 23598722, 23469767, | |
131 23342214, 23216040, 23091223, 22967740, 22845571, 22724695, 22605092, 22486740, | |
132 22369622, 22253717, 22139007, 22025474, 21913099, 21801865, 21691755, 21582751, | |
133 21474837, 21367997, 21262215, 21157475, 21053762, 20951060, 20849356, 20748635, | |
134 20648882, 20550083, 20452226, 20355296, 20259280, 20164166, 20069941, 19976593, | |
135 19884108, 19792477, 19701685, 19611723, 19522579, 19434242, 19346700, 19259944, | |
136 19173962, 19088744, 19004281, 18920561, 18837576, 18755316, 18673771, 18592933, | |
137 18512791, 18433337, 18354562, 18276457, 18199014, 18122225, 18046082, 17970575, | |
138 17895698, 17821442, 17747799, 17674763, 17602325, 17530479, 17459217, 17388532, | |
220 | 139 17318417, 17248865, 17179870, 17111424, 17043522, 16976156, 16909321, 16843010, |
140 }; | |
141 | |
1092 | 142 /* Input permutation for the simple_idct_mmx */ |
143 static const uint8_t simple_mmx_permutation[64]={ | |
2979 | 144 0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, |
145 0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, | |
146 0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, | |
147 0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, | |
148 0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, | |
149 0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, | |
150 0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, | |
151 0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F, | |
1092 | 152 }; |
153 | |
1064 | 154 static int pix_sum_c(uint8_t * pix, int line_size) |
612 | 155 { |
156 int s, i, j; | |
157 | |
158 s = 0; | |
159 for (i = 0; i < 16; i++) { | |
2979 | 160 for (j = 0; j < 16; j += 8) { |
161 s += pix[0]; | |
162 s += pix[1]; | |
163 s += pix[2]; | |
164 s += pix[3]; | |
165 s += pix[4]; | |
166 s += pix[5]; | |
167 s += pix[6]; | |
168 s += pix[7]; | |
169 pix += 8; | |
170 } | |
171 pix += line_size - 16; | |
612 | 172 } |
173 return s; | |
174 } | |
175 | |
1064 | 176 static int pix_norm1_c(uint8_t * pix, int line_size) |
612 | 177 { |
178 int s, i, j; | |
4179 | 179 uint32_t *sq = ff_squareTbl + 256; |
612 | 180 |
181 s = 0; | |
182 for (i = 0; i < 16; i++) { | |
2979 | 183 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
|
184 #if 0 |
2979 | 185 s += sq[pix[0]]; |
186 s += sq[pix[1]]; | |
187 s += sq[pix[2]]; | |
188 s += sq[pix[3]]; | |
189 s += sq[pix[4]]; | |
190 s += sq[pix[5]]; | |
191 s += sq[pix[6]]; | |
192 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
|
193 #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
|
194 #if LONG_MAX > 2147483647 |
2979 | 195 register uint64_t x=*(uint64_t*)pix; |
196 s += sq[x&0xff]; | |
197 s += sq[(x>>8)&0xff]; | |
198 s += sq[(x>>16)&0xff]; | |
199 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
|
200 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
|
201 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
|
202 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
|
203 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
|
204 #else |
2979 | 205 register uint32_t x=*(uint32_t*)pix; |
206 s += sq[x&0xff]; | |
207 s += sq[(x>>8)&0xff]; | |
208 s += sq[(x>>16)&0xff]; | |
209 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
|
210 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
|
211 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
|
212 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
|
213 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
|
214 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
|
215 #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
|
216 #endif |
2979 | 217 pix += 8; |
218 } | |
219 pix += line_size - 16; | |
612 | 220 } |
221 return s; | |
222 } | |
223 | |
6241 | 224 static void bswap_buf(uint32_t *dst, const uint32_t *src, int w){ |
1273 | 225 int i; |
2967 | 226 |
1273 | 227 for(i=0; i+8<=w; i+=8){ |
228 dst[i+0]= bswap_32(src[i+0]); | |
229 dst[i+1]= bswap_32(src[i+1]); | |
230 dst[i+2]= bswap_32(src[i+2]); | |
231 dst[i+3]= bswap_32(src[i+3]); | |
232 dst[i+4]= bswap_32(src[i+4]); | |
233 dst[i+5]= bswap_32(src[i+5]); | |
234 dst[i+6]= bswap_32(src[i+6]); | |
235 dst[i+7]= bswap_32(src[i+7]); | |
236 } | |
237 for(;i<w; i++){ | |
238 dst[i+0]= bswap_32(src[i+0]); | |
239 } | |
240 } | |
612 | 241 |
2184 | 242 static int sse4_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
243 { | |
244 int s, i; | |
4179 | 245 uint32_t *sq = ff_squareTbl + 256; |
2184 | 246 |
247 s = 0; | |
248 for (i = 0; i < h; i++) { | |
249 s += sq[pix1[0] - pix2[0]]; | |
250 s += sq[pix1[1] - pix2[1]]; | |
251 s += sq[pix1[2] - pix2[2]]; | |
252 s += sq[pix1[3] - pix2[3]]; | |
253 pix1 += line_size; | |
254 pix2 += line_size; | |
255 } | |
256 return s; | |
257 } | |
258 | |
1708 | 259 static int sse8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
936 | 260 { |
261 int s, i; | |
4179 | 262 uint32_t *sq = ff_squareTbl + 256; |
936 | 263 |
264 s = 0; | |
1708 | 265 for (i = 0; i < h; i++) { |
936 | 266 s += sq[pix1[0] - pix2[0]]; |
267 s += sq[pix1[1] - pix2[1]]; | |
268 s += sq[pix1[2] - pix2[2]]; | |
269 s += sq[pix1[3] - pix2[3]]; | |
270 s += sq[pix1[4] - pix2[4]]; | |
271 s += sq[pix1[5] - pix2[5]]; | |
272 s += sq[pix1[6] - pix2[6]]; | |
273 s += sq[pix1[7] - pix2[7]]; | |
274 pix1 += line_size; | |
275 pix2 += line_size; | |
276 } | |
277 return s; | |
278 } | |
279 | |
1708 | 280 static int sse16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
884 | 281 { |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
282 int s, i; |
4179 | 283 uint32_t *sq = ff_squareTbl + 256; |
884 | 284 |
285 s = 0; | |
1708 | 286 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
|
287 s += sq[pix1[ 0] - pix2[ 0]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
288 s += sq[pix1[ 1] - pix2[ 1]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
289 s += sq[pix1[ 2] - pix2[ 2]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
290 s += sq[pix1[ 3] - pix2[ 3]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
291 s += sq[pix1[ 4] - pix2[ 4]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
292 s += sq[pix1[ 5] - pix2[ 5]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
293 s += sq[pix1[ 6] - pix2[ 6]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
294 s += sq[pix1[ 7] - pix2[ 7]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
295 s += sq[pix1[ 8] - pix2[ 8]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
296 s += sq[pix1[ 9] - pix2[ 9]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
297 s += sq[pix1[10] - pix2[10]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
298 s += sq[pix1[11] - pix2[11]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
299 s += sq[pix1[12] - pix2[12]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
300 s += sq[pix1[13] - pix2[13]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
301 s += sq[pix1[14] - pix2[14]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
302 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
|
303 |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
304 pix1 += line_size; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
305 pix2 += line_size; |
884 | 306 } |
307 return s; | |
308 } | |
309 | |
2184 | 310 |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
311 #ifdef CONFIG_SNOW_ENCODER //dwt is in snow.c |
2184 | 312 static inline int w_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int w, int h, int type){ |
313 int s, i, j; | |
314 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
|
315 int tmp[32*32]; |
2184 | 316 int level, ori; |
2967 | 317 static const int scale[2][2][4][4]={ |
2184 | 318 { |
319 { | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
320 // 9/7 8x8 dec=3 |
2184 | 321 {268, 239, 239, 213}, |
322 { 0, 224, 224, 152}, | |
323 { 0, 135, 135, 110}, | |
324 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
325 // 9/7 16x16 or 32x32 dec=4 |
2184 | 326 {344, 310, 310, 280}, |
327 { 0, 320, 320, 228}, | |
328 { 0, 175, 175, 136}, | |
329 { 0, 129, 129, 102}, | |
330 } | |
331 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
332 { |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
333 // 5/3 8x8 dec=3 |
2184 | 334 {275, 245, 245, 218}, |
335 { 0, 230, 230, 156}, | |
336 { 0, 138, 138, 113}, | |
337 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
338 // 5/3 16x16 or 32x32 dec=4 |
2184 | 339 {352, 317, 317, 286}, |
340 { 0, 328, 328, 233}, | |
341 { 0, 180, 180, 140}, | |
342 { 0, 132, 132, 105}, | |
343 } | |
344 } | |
345 }; | |
346 | |
347 for (i = 0; i < h; i++) { | |
348 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
|
349 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
|
350 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
|
351 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
|
352 tmp[32*i+j+3] = (pix1[j+3] - pix2[j+3])<<4; |
2184 | 353 } |
354 pix1 += line_size; | |
355 pix2 += line_size; | |
356 } | |
2639 | 357 |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
358 ff_spatial_dwt(tmp, w, h, 32, type, dec_count); |
2184 | 359 |
360 s=0; | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
361 assert(w==h); |
2184 | 362 for(level=0; level<dec_count; level++){ |
363 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
|
364 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
|
365 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
|
366 int stride= 32<<(dec_count-level); |
2184 | 367 int sy= (ori&2) ? stride>>1 : 0; |
2967 | 368 |
2184 | 369 for(i=0; i<size; i++){ |
370 for(j=0; j<size; j++){ | |
371 int v= tmp[sx + sy + i*stride + j] * scale[type][dec_count-3][level][ori]; | |
4001 | 372 s += FFABS(v); |
2184 | 373 } |
374 } | |
375 } | |
376 } | |
2967 | 377 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
|
378 return s>>9; |
2184 | 379 } |
380 | |
381 static int w53_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
382 return w_c(v, pix1, pix2, line_size, 8, h, 1); | |
383 } | |
384 | |
385 static int w97_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
386 return w_c(v, pix1, pix2, line_size, 8, h, 0); | |
387 } | |
388 | |
389 static int w53_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
390 return w_c(v, pix1, pix2, line_size, 16, h, 1); | |
391 } | |
392 | |
393 static int w97_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
394 return w_c(v, pix1, pix2, line_size, 16, h, 0); | |
395 } | |
396 | |
4197 | 397 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
|
398 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
|
399 } |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
400 |
4197 | 401 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
|
402 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
|
403 } |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
404 #endif |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
405 |
1064 | 406 static void get_pixels_c(DCTELEM *restrict block, const uint8_t *pixels, int line_size) |
0 | 407 { |
408 int i; | |
409 | |
410 /* read the pixels */ | |
411 for(i=0;i<8;i++) { | |
516 | 412 block[0] = pixels[0]; |
413 block[1] = pixels[1]; | |
414 block[2] = pixels[2]; | |
415 block[3] = pixels[3]; | |
416 block[4] = pixels[4]; | |
417 block[5] = pixels[5]; | |
418 block[6] = pixels[6]; | |
419 block[7] = pixels[7]; | |
420 pixels += line_size; | |
421 block += 8; | |
0 | 422 } |
423 } | |
424 | |
1064 | 425 static void diff_pixels_c(DCTELEM *restrict block, const uint8_t *s1, |
2979 | 426 const uint8_t *s2, int stride){ |
324 | 427 int i; |
428 | |
429 /* read the pixels */ | |
430 for(i=0;i<8;i++) { | |
516 | 431 block[0] = s1[0] - s2[0]; |
432 block[1] = s1[1] - s2[1]; | |
433 block[2] = s1[2] - s2[2]; | |
434 block[3] = s1[3] - s2[3]; | |
435 block[4] = s1[4] - s2[4]; | |
436 block[5] = s1[5] - s2[5]; | |
437 block[6] = s1[6] - s2[6]; | |
438 block[7] = s1[7] - s2[7]; | |
324 | 439 s1 += stride; |
440 s2 += stride; | |
516 | 441 block += 8; |
324 | 442 } |
443 } | |
444 | |
445 | |
1064 | 446 static void put_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 447 int line_size) |
0 | 448 { |
449 int i; | |
4176 | 450 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 451 |
0 | 452 /* read the pixels */ |
453 for(i=0;i<8;i++) { | |
516 | 454 pixels[0] = cm[block[0]]; |
455 pixels[1] = cm[block[1]]; | |
456 pixels[2] = cm[block[2]]; | |
457 pixels[3] = cm[block[3]]; | |
458 pixels[4] = cm[block[4]]; | |
459 pixels[5] = cm[block[5]]; | |
460 pixels[6] = cm[block[6]]; | |
461 pixels[7] = cm[block[7]]; | |
462 | |
463 pixels += line_size; | |
464 block += 8; | |
0 | 465 } |
466 } | |
467 | |
2256 | 468 static void put_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 469 int line_size) |
2256 | 470 { |
471 int i; | |
4176 | 472 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 473 |
2256 | 474 /* read the pixels */ |
475 for(i=0;i<4;i++) { | |
476 pixels[0] = cm[block[0]]; | |
477 pixels[1] = cm[block[1]]; | |
478 pixels[2] = cm[block[2]]; | |
479 pixels[3] = cm[block[3]]; | |
480 | |
481 pixels += line_size; | |
482 block += 8; | |
483 } | |
484 } | |
485 | |
2257 | 486 static void put_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 487 int line_size) |
2257 | 488 { |
489 int i; | |
4176 | 490 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 491 |
2257 | 492 /* read the pixels */ |
493 for(i=0;i<2;i++) { | |
494 pixels[0] = cm[block[0]]; | |
495 pixels[1] = cm[block[1]]; | |
496 | |
497 pixels += line_size; | |
498 block += 8; | |
499 } | |
500 } | |
501 | |
2967 | 502 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
|
503 uint8_t *restrict pixels, |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
504 int line_size) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
505 { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
506 int i, j; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
507 |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
508 for (i = 0; i < 8; i++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
509 for (j = 0; j < 8; j++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
510 if (*block < -128) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
511 *pixels = 0; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
512 else if (*block > 127) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
513 *pixels = 255; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
514 else |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
515 *pixels = (uint8_t)(*block + 128); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
516 block++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
517 pixels++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
518 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
519 pixels += (line_size - 8); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
520 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
521 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
522 |
1064 | 523 static void add_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
516 | 524 int line_size) |
0 | 525 { |
526 int i; | |
4176 | 527 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 528 |
0 | 529 /* read the pixels */ |
530 for(i=0;i<8;i++) { | |
516 | 531 pixels[0] = cm[pixels[0] + block[0]]; |
532 pixels[1] = cm[pixels[1] + block[1]]; | |
533 pixels[2] = cm[pixels[2] + block[2]]; | |
534 pixels[3] = cm[pixels[3] + block[3]]; | |
535 pixels[4] = cm[pixels[4] + block[4]]; | |
536 pixels[5] = cm[pixels[5] + block[5]]; | |
537 pixels[6] = cm[pixels[6] + block[6]]; | |
538 pixels[7] = cm[pixels[7] + block[7]]; | |
539 pixels += line_size; | |
540 block += 8; | |
0 | 541 } |
542 } | |
2256 | 543 |
544 static void add_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, | |
545 int line_size) | |
546 { | |
547 int i; | |
4176 | 548 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 549 |
2256 | 550 /* read the pixels */ |
551 for(i=0;i<4;i++) { | |
552 pixels[0] = cm[pixels[0] + block[0]]; | |
553 pixels[1] = cm[pixels[1] + block[1]]; | |
554 pixels[2] = cm[pixels[2] + block[2]]; | |
555 pixels[3] = cm[pixels[3] + block[3]]; | |
556 pixels += line_size; | |
557 block += 8; | |
558 } | |
559 } | |
2257 | 560 |
561 static void add_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, | |
562 int line_size) | |
563 { | |
564 int i; | |
4176 | 565 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 566 |
2257 | 567 /* read the pixels */ |
568 for(i=0;i<2;i++) { | |
569 pixels[0] = cm[pixels[0] + block[0]]; | |
570 pixels[1] = cm[pixels[1] + block[1]]; | |
571 pixels += line_size; | |
572 block += 8; | |
573 } | |
574 } | |
2763 | 575 |
576 static void add_pixels8_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
577 { | |
578 int i; | |
579 for(i=0;i<8;i++) { | |
580 pixels[0] += block[0]; | |
581 pixels[1] += block[1]; | |
582 pixels[2] += block[2]; | |
583 pixels[3] += block[3]; | |
584 pixels[4] += block[4]; | |
585 pixels[5] += block[5]; | |
586 pixels[6] += block[6]; | |
587 pixels[7] += block[7]; | |
588 pixels += line_size; | |
589 block += 8; | |
590 } | |
591 } | |
592 | |
593 static void add_pixels4_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
594 { | |
595 int i; | |
596 for(i=0;i<4;i++) { | |
597 pixels[0] += block[0]; | |
598 pixels[1] += block[1]; | |
599 pixels[2] += block[2]; | |
600 pixels[3] += block[3]; | |
601 pixels += line_size; | |
602 block += 4; | |
603 } | |
604 } | |
605 | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
606 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
|
607 { |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
608 int sum=0, i; |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
609 for(i=0; i<64; i++) |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
610 sum+= FFABS(block[i]); |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
611 return sum; |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
612 } |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
613 |
385 | 614 #if 0 |
615 | |
616 #define PIXOP2(OPNAME, OP) \ | |
651 | 617 static void OPNAME ## _pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 618 {\ |
619 int i;\ | |
620 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
|
621 OP(*((uint64_t*)block), AV_RN64(pixels));\ |
385 | 622 pixels+=line_size;\ |
623 block +=line_size;\ | |
624 }\ | |
625 }\ | |
626 \ | |
859 | 627 static void OPNAME ## _no_rnd_pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 628 {\ |
629 int i;\ | |
630 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
|
631 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
|
632 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 633 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
634 pixels+=line_size;\ | |
635 block +=line_size;\ | |
636 }\ | |
637 }\ | |
638 \ | |
859 | 639 static void OPNAME ## _pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 640 {\ |
641 int i;\ | |
642 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
|
643 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
|
644 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 645 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
646 pixels+=line_size;\ | |
647 block +=line_size;\ | |
648 }\ | |
649 }\ | |
650 \ | |
859 | 651 static void OPNAME ## _no_rnd_pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 652 {\ |
653 int i;\ | |
654 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
|
655 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
|
656 const uint64_t b= AV_RN64(pixels+line_size);\ |
385 | 657 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
658 pixels+=line_size;\ | |
659 block +=line_size;\ | |
660 }\ | |
661 }\ | |
662 \ | |
859 | 663 static void OPNAME ## _pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 664 {\ |
665 int i;\ | |
666 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
|
667 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
|
668 const uint64_t b= AV_RN64(pixels+line_size);\ |
385 | 669 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
670 pixels+=line_size;\ | |
671 block +=line_size;\ | |
672 }\ | |
673 }\ | |
674 \ | |
859 | 675 static void OPNAME ## _pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 676 {\ |
677 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
678 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
|
679 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 680 uint64_t l0= (a&0x0303030303030303ULL)\ |
681 + (b&0x0303030303030303ULL)\ | |
682 + 0x0202020202020202ULL;\ | |
683 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
684 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
685 uint64_t l1,h1;\ | |
686 \ | |
687 pixels+=line_size;\ | |
688 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
|
689 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
|
690 uint64_t b= AV_RN64(pixels+1);\ |
385 | 691 l1= (a&0x0303030303030303ULL)\ |
692 + (b&0x0303030303030303ULL);\ | |
693 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
694 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
695 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
696 pixels+=line_size;\ | |
697 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
698 a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
699 b= AV_RN64(pixels+1);\ |
385 | 700 l0= (a&0x0303030303030303ULL)\ |
701 + (b&0x0303030303030303ULL)\ | |
702 + 0x0202020202020202ULL;\ | |
703 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
704 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
705 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
706 pixels+=line_size;\ | |
707 block +=line_size;\ | |
708 }\ | |
709 }\ | |
710 \ | |
859 | 711 static void OPNAME ## _no_rnd_pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 712 {\ |
713 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
714 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
|
715 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 716 uint64_t l0= (a&0x0303030303030303ULL)\ |
717 + (b&0x0303030303030303ULL)\ | |
718 + 0x0101010101010101ULL;\ | |
719 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
720 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
721 uint64_t l1,h1;\ | |
722 \ | |
723 pixels+=line_size;\ | |
724 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
|
725 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
|
726 uint64_t b= AV_RN64(pixels+1);\ |
385 | 727 l1= (a&0x0303030303030303ULL)\ |
728 + (b&0x0303030303030303ULL);\ | |
729 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
730 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
731 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
732 pixels+=line_size;\ | |
733 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
734 a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
735 b= AV_RN64(pixels+1);\ |
385 | 736 l0= (a&0x0303030303030303ULL)\ |
737 + (b&0x0303030303030303ULL)\ | |
738 + 0x0101010101010101ULL;\ | |
739 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
740 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
741 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
742 pixels+=line_size;\ | |
743 block +=line_size;\ | |
744 }\ | |
745 }\ | |
746 \ | |
859 | 747 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels_c , 8)\ |
748 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels_x2_c , 8)\ | |
749 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels_y2_c , 8)\ | |
750 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels_xy2_c, 8)\ | |
751 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels_x2_c , 8)\ | |
752 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels_y2_c , 8)\ | |
753 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels_xy2_c, 8) | |
385 | 754 |
755 #define op_avg(a, b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEFEFEFEFEULL)>>1) ) | |
756 #else // 64 bit variant | |
757 | |
758 #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
|
759 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
|
760 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
|
761 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
|
762 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
|
763 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
|
764 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
|
765 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
766 }\ |
1168 | 767 static void OPNAME ## _pixels4_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
768 int i;\ | |
769 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
|
770 OP(*((uint32_t*)(block )), AV_RN32(pixels ));\ |
1168 | 771 pixels+=line_size;\ |
772 block +=line_size;\ | |
773 }\ | |
774 }\ | |
859 | 775 static void OPNAME ## _pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
385 | 776 int i;\ |
777 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
|
778 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
|
779 OP(*((uint32_t*)(block+4)), AV_RN32(pixels+4));\ |
385 | 780 pixels+=line_size;\ |
781 block +=line_size;\ | |
782 }\ | |
783 }\ | |
859 | 784 static inline void OPNAME ## _no_rnd_pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
785 OPNAME ## _pixels8_c(block, pixels, line_size, h);\ | |
651 | 786 }\ |
385 | 787 \ |
651 | 788 static inline void OPNAME ## _no_rnd_pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
789 int src_stride1, int src_stride2, int h){\ | |
385 | 790 int i;\ |
791 for(i=0; i<h; i++){\ | |
651 | 792 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
793 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
|
794 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 795 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
|
796 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
|
797 b= AV_RN32(&src2[i*src_stride2+4]);\ |
1264 | 798 OP(*((uint32_t*)&dst[i*dst_stride+4]), no_rnd_avg32(a, b));\ |
385 | 799 }\ |
800 }\ | |
801 \ | |
651 | 802 static inline void OPNAME ## _pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
803 int src_stride1, int src_stride2, int h){\ | |
385 | 804 int i;\ |
805 for(i=0; i<h; i++){\ | |
651 | 806 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
807 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
|
808 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 809 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
|
810 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
|
811 b= AV_RN32(&src2[i*src_stride2+4]);\ |
1264 | 812 OP(*((uint32_t*)&dst[i*dst_stride+4]), rnd_avg32(a, b));\ |
385 | 813 }\ |
814 }\ | |
815 \ | |
1168 | 816 static inline void OPNAME ## _pixels4_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
817 int src_stride1, int src_stride2, int h){\ | |
818 int i;\ | |
819 for(i=0; i<h; i++){\ | |
820 uint32_t a,b;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
821 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
|
822 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 823 OP(*((uint32_t*)&dst[i*dst_stride ]), rnd_avg32(a, b));\ |
1168 | 824 }\ |
825 }\ | |
826 \ | |
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 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
|
828 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
|
829 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
|
830 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
|
831 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
832 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
|
833 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
|
834 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
|
835 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
836 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
837 \ |
651 | 838 static inline void OPNAME ## _pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
839 int src_stride1, int src_stride2, int h){\ | |
840 OPNAME ## _pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
841 OPNAME ## _pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
842 }\ | |
843 \ | |
844 static inline void OPNAME ## _no_rnd_pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ | |
845 int src_stride1, int src_stride2, int h){\ | |
846 OPNAME ## _no_rnd_pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
847 OPNAME ## _no_rnd_pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
848 }\ | |
849 \ | |
859 | 850 static inline void OPNAME ## _no_rnd_pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 851 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
852 }\ | |
853 \ | |
859 | 854 static inline void OPNAME ## _pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 855 OPNAME ## _pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
856 }\ | |
857 \ | |
859 | 858 static inline void OPNAME ## _no_rnd_pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 859 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
860 }\ | |
861 \ | |
859 | 862 static inline void OPNAME ## _pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 863 OPNAME ## _pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
385 | 864 }\ |
865 \ | |
651 | 866 static inline void OPNAME ## _pixels8_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
867 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
868 int i;\ | |
869 for(i=0; i<h; i++){\ | |
870 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
|
871 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
|
872 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
|
873 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
|
874 d= AV_RN32(&src4[i*src_stride4]);\ |
651 | 875 l0= (a&0x03030303UL)\ |
876 + (b&0x03030303UL)\ | |
877 + 0x02020202UL;\ | |
878 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
879 + ((b&0xFCFCFCFCUL)>>2);\ | |
880 l1= (c&0x03030303UL)\ | |
881 + (d&0x03030303UL);\ | |
882 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
883 + ((d&0xFCFCFCFCUL)>>2);\ | |
884 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
|
885 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
|
886 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
|
887 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
|
888 d= AV_RN32(&src4[i*src_stride4+4]);\ |
651 | 889 l0= (a&0x03030303UL)\ |
890 + (b&0x03030303UL)\ | |
891 + 0x02020202UL;\ | |
892 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
893 + ((b&0xFCFCFCFCUL)>>2);\ | |
894 l1= (c&0x03030303UL)\ | |
895 + (d&0x03030303UL);\ | |
896 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
897 + ((d&0xFCFCFCFCUL)>>2);\ | |
898 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
899 }\ | |
900 }\ | |
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
|
901 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
902 static inline void OPNAME ## _pixels4_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
|
903 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
|
904 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
905 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
906 static inline void OPNAME ## _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
|
907 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
|
908 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
909 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
910 static inline void OPNAME ## _pixels2_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
|
911 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
|
912 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
913 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
914 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
|
915 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
|
916 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
917 \ |
651 | 918 static inline void OPNAME ## _no_rnd_pixels8_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
919 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
385 | 920 int i;\ |
921 for(i=0; i<h; i++){\ | |
651 | 922 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
|
923 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
|
924 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
|
925 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
|
926 d= AV_RN32(&src4[i*src_stride4]);\ |
651 | 927 l0= (a&0x03030303UL)\ |
928 + (b&0x03030303UL)\ | |
929 + 0x01010101UL;\ | |
930 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
931 + ((b&0xFCFCFCFCUL)>>2);\ | |
932 l1= (c&0x03030303UL)\ | |
933 + (d&0x03030303UL);\ | |
934 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
935 + ((d&0xFCFCFCFCUL)>>2);\ | |
936 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
|
937 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
|
938 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
|
939 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
|
940 d= AV_RN32(&src4[i*src_stride4+4]);\ |
651 | 941 l0= (a&0x03030303UL)\ |
942 + (b&0x03030303UL)\ | |
943 + 0x01010101UL;\ | |
944 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
945 + ((b&0xFCFCFCFCUL)>>2);\ | |
946 l1= (c&0x03030303UL)\ | |
947 + (d&0x03030303UL);\ | |
948 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
949 + ((d&0xFCFCFCFCUL)>>2);\ | |
950 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
385 | 951 }\ |
952 }\ | |
651 | 953 static inline void OPNAME ## _pixels16_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
954 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
955 OPNAME ## _pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
956 OPNAME ## _pixels8_l4(dst+8, src1+8, src2+8, src3+8, src4+8, dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
957 }\ | |
958 static inline void OPNAME ## _no_rnd_pixels16_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ | |
959 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
960 OPNAME ## _no_rnd_pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
961 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);\ | |
962 }\ | |
385 | 963 \ |
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
|
964 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
|
965 {\ |
85b71f9f7450
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 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
|
967 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
|
968 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
|
969 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
|
970 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
|
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 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
|
973 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
|
974 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
|
975 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
|
976 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
|
977 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
|
978 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
979 block[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
|
980 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
|
981 \ |
85b71f9f7450
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 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
|
983 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
|
984 \ |
85b71f9f7450
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 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
|
986 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
|
987 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
|
988 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
|
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 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
|
991 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
|
992 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
|
993 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
|
994 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
995 }\ |
85b71f9f7450
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 \ |
85b71f9f7450
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 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
|
998 {\ |
85b71f9f7450
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 int i;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1000 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
|
1001 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
|
1002 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
|
1003 + (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
|
1004 + 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
|
1005 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
|
1006 + ((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
|
1007 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
|
1008 \ |
85b71f9f7450
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 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
|
1010 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
|
1011 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
|
1012 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
|
1013 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
|
1014 + (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
|
1015 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
|
1016 + ((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
|
1017 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
|
1018 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
|
1019 block +=line_size;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1020 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1021 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
|
1022 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
|
1023 + (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
|
1024 + 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
|
1025 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
|
1026 + ((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
|
1027 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
|
1028 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
|
1029 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
|
1030 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1031 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1032 \ |
859 | 1033 static inline void OPNAME ## _pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1034 {\ |
1035 int j;\ | |
1036 for(j=0; j<2; j++){\ | |
1037 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1038 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
|
1039 const uint32_t b= AV_RN32(pixels+1);\ |
385 | 1040 uint32_t l0= (a&0x03030303UL)\ |
1041 + (b&0x03030303UL)\ | |
1042 + 0x02020202UL;\ | |
1043 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1044 + ((b&0xFCFCFCFCUL)>>2);\ | |
1045 uint32_t l1,h1;\ | |
1046 \ | |
1047 pixels+=line_size;\ | |
1048 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
|
1049 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
|
1050 uint32_t b= AV_RN32(pixels+1);\ |
385 | 1051 l1= (a&0x03030303UL)\ |
1052 + (b&0x03030303UL);\ | |
1053 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1054 + ((b&0xFCFCFCFCUL)>>2);\ | |
1055 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1056 pixels+=line_size;\ | |
1057 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1058 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1059 b= AV_RN32(pixels+1);\ |
385 | 1060 l0= (a&0x03030303UL)\ |
1061 + (b&0x03030303UL)\ | |
1062 + 0x02020202UL;\ | |
1063 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1064 + ((b&0xFCFCFCFCUL)>>2);\ | |
1065 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1066 pixels+=line_size;\ | |
1067 block +=line_size;\ | |
1068 }\ | |
1069 pixels+=4-line_size*(h+1);\ | |
1070 block +=4-line_size*h;\ | |
1071 }\ | |
1072 }\ | |
1073 \ | |
859 | 1074 static inline void OPNAME ## _no_rnd_pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1075 {\ |
1076 int j;\ | |
1077 for(j=0; j<2; j++){\ | |
1078 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1079 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
|
1080 const uint32_t b= AV_RN32(pixels+1);\ |
385 | 1081 uint32_t l0= (a&0x03030303UL)\ |
1082 + (b&0x03030303UL)\ | |
1083 + 0x01010101UL;\ | |
1084 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1085 + ((b&0xFCFCFCFCUL)>>2);\ | |
1086 uint32_t l1,h1;\ | |
1087 \ | |
1088 pixels+=line_size;\ | |
1089 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
|
1090 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
|
1091 uint32_t b= AV_RN32(pixels+1);\ |
385 | 1092 l1= (a&0x03030303UL)\ |
1093 + (b&0x03030303UL);\ | |
1094 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1095 + ((b&0xFCFCFCFCUL)>>2);\ | |
1096 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1097 pixels+=line_size;\ | |
1098 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1099 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1100 b= AV_RN32(pixels+1);\ |
385 | 1101 l0= (a&0x03030303UL)\ |
1102 + (b&0x03030303UL)\ | |
1103 + 0x01010101UL;\ | |
1104 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1105 + ((b&0xFCFCFCFCUL)>>2);\ | |
1106 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1107 pixels+=line_size;\ | |
1108 block +=line_size;\ | |
1109 }\ | |
1110 pixels+=4-line_size*(h+1);\ | |
1111 block +=4-line_size*h;\ | |
1112 }\ | |
1113 }\ | |
1114 \ | |
859 | 1115 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels8_c , 8)\ |
1116 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels8_x2_c , 8)\ | |
1117 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels8_y2_c , 8)\ | |
1118 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels8_xy2_c, 8)\ | |
1119 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_c , OPNAME ## _pixels8_c , 8)\ | |
1120 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels8_x2_c , 8)\ | |
1121 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels8_y2_c , 8)\ | |
1122 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels8_xy2_c, 8)\ | |
651 | 1123 |
1264 | 1124 #define op_avg(a, b) a = rnd_avg32(a, b) |
385 | 1125 #endif |
1126 #define op_put(a, b) a = b | |
1127 | |
1128 PIXOP2(avg, op_avg) | |
1129 PIXOP2(put, op_put) | |
1130 #undef op_avg | |
1131 #undef op_put | |
1132 | |
0 | 1133 #define avg2(a,b) ((a+b+1)>>1) |
1134 #define avg4(a,b,c,d) ((a+b+c+d+2)>>2) | |
1135 | |
1864 | 1136 static void put_no_rnd_pixels16_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ |
1137 put_no_rnd_pixels16_l2(dst, a, b, stride, stride, stride, h); | |
1138 } | |
1139 | |
1140 static void put_no_rnd_pixels8_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ | |
1141 put_no_rnd_pixels8_l2(dst, a, b, stride, stride, stride, h); | |
1142 } | |
753 | 1143 |
1064 | 1144 static void gmc1_c(uint8_t *dst, uint8_t *src, int stride, int h, int x16, int y16, int rounder) |
255 | 1145 { |
1146 const int A=(16-x16)*(16-y16); | |
1147 const int B=( x16)*(16-y16); | |
1148 const int C=(16-x16)*( y16); | |
1149 const int D=( x16)*( y16); | |
1150 int i; | |
1151 | |
1152 for(i=0; i<h; i++) | |
1153 { | |
651 | 1154 dst[0]= (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + rounder)>>8; |
1155 dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8; | |
1156 dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8; | |
1157 dst[3]= (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + rounder)>>8; | |
1158 dst[4]= (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + rounder)>>8; | |
1159 dst[5]= (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + rounder)>>8; | |
1160 dst[6]= (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + rounder)>>8; | |
1161 dst[7]= (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + rounder)>>8; | |
1162 dst+= stride; | |
1163 src+= stride; | |
255 | 1164 } |
1165 } | |
1166 | |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
1167 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, |
753 | 1168 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height) |
1169 { | |
1170 int y, vx, vy; | |
1171 const int s= 1<<shift; | |
2967 | 1172 |
753 | 1173 width--; |
1174 height--; | |
1175 | |
1176 for(y=0; y<h; y++){ | |
1177 int x; | |
1178 | |
1179 vx= ox; | |
1180 vy= oy; | |
1181 for(x=0; x<8; x++){ //XXX FIXME optimize | |
1182 int src_x, src_y, frac_x, frac_y, index; | |
1183 | |
1184 src_x= vx>>16; | |
1185 src_y= vy>>16; | |
1186 frac_x= src_x&(s-1); | |
1187 frac_y= src_y&(s-1); | |
1188 src_x>>=shift; | |
1189 src_y>>=shift; | |
2967 | 1190 |
753 | 1191 if((unsigned)src_x < width){ |
1192 if((unsigned)src_y < height){ | |
1193 index= src_x + src_y*stride; | |
1194 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) | |
1195 + src[index +1]* frac_x )*(s-frac_y) | |
1196 + ( src[index+stride ]*(s-frac_x) | |
1197 + src[index+stride+1]* frac_x )* frac_y | |
1198 + r)>>(shift*2); | |
1199 }else{ | |
4594 | 1200 index= src_x + av_clip(src_y, 0, height)*stride; |
2967 | 1201 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) |
753 | 1202 + src[index +1]* frac_x )*s |
1203 + r)>>(shift*2); | |
1204 } | |
1205 }else{ | |
1206 if((unsigned)src_y < height){ | |
4594 | 1207 index= av_clip(src_x, 0, width) + src_y*stride; |
2967 | 1208 dst[y*stride + x]= ( ( src[index ]*(s-frac_y) |
753 | 1209 + src[index+stride ]* frac_y )*s |
1210 + r)>>(shift*2); | |
1211 }else{ | |
4594 | 1212 index= av_clip(src_x, 0, width) + av_clip(src_y, 0, height)*stride; |
753 | 1213 dst[y*stride + x]= src[index ]; |
1214 } | |
1215 } | |
2967 | 1216 |
753 | 1217 vx+= dxx; |
1218 vy+= dyx; | |
1219 } | |
1220 ox += dxy; | |
1221 oy += dyy; | |
1222 } | |
1223 } | |
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
|
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 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
|
1226 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
|
1227 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
|
1228 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
|
1229 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
|
1230 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
|
1231 } |
85b71f9f7450
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 |
85b71f9f7450
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 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
|
1235 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
|
1236 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
|
1237 for (j=0; j < width; j++) { |
2979 | 1238 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
|
1239 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1240 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
|
1241 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
|
1242 } |
85b71f9f7450
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 |
85b71f9f7450
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 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
|
1246 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
|
1247 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
|
1248 for (j=0; j < width; j++) { |
2979 | 1249 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
|
1250 } |
85b71f9f7450
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 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
|
1252 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
|
1253 } |
85b71f9f7450
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 } |
2967 | 1255 |
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
|
1256 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
|
1257 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
|
1258 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
|
1259 for (j=0; j < width; j++) { |
2979 | 1260 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
|
1261 } |
85b71f9f7450
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 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
|
1263 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
|
1264 } |
85b71f9f7450
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 } |
2967 | 1266 |
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
|
1267 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
|
1268 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
|
1269 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
|
1270 for (j=0; j < width; j++) { |
2979 | 1271 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
|
1272 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1273 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
|
1274 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
|
1275 } |
85b71f9f7450
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 |
85b71f9f7450
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 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
|
1279 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
|
1280 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
|
1281 for (j=0; j < width; j++) { |
2979 | 1282 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
|
1283 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1284 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
|
1285 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
|
1286 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1287 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1288 |
85b71f9f7450
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 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
|
1290 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
|
1291 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
|
1292 for (j=0; j < width; j++) { |
2979 | 1293 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
|
1294 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1295 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
|
1296 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
|
1297 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1298 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1299 |
85b71f9f7450
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 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
|
1301 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
|
1302 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
|
1303 for (j=0; j < width; j++) { |
2979 | 1304 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
|
1305 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1306 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
|
1307 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
|
1308 } |
85b71f9f7450
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 |
85b71f9f7450
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 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
|
1312 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
|
1313 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
|
1314 for (j=0; j < width; j++) { |
2979 | 1315 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
|
1316 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1317 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
|
1318 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
|
1319 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1320 } |
1319 | 1321 |
1322 static inline void avg_tpel_pixels_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1323 switch(width){ | |
1324 case 2: avg_pixels2_c (dst, src, stride, height); break; | |
1325 case 4: avg_pixels4_c (dst, src, stride, height); break; | |
1326 case 8: avg_pixels8_c (dst, src, stride, height); break; | |
1327 case 16:avg_pixels16_c(dst, src, stride, height); break; | |
1328 } | |
1329 } | |
1330 | |
1331 static inline void avg_tpel_pixels_mc10_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1332 int i,j; | |
1333 for (i=0; i < height; i++) { | |
1334 for (j=0; j < width; j++) { | |
2979 | 1335 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1336 } |
1337 src += stride; | |
1338 dst += stride; | |
1339 } | |
1340 } | |
1341 | |
1342 static inline void avg_tpel_pixels_mc20_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1343 int i,j; | |
1344 for (i=0; i < height; i++) { | |
1345 for (j=0; j < width; j++) { | |
2979 | 1346 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1347 } |
1348 src += stride; | |
1349 dst += stride; | |
1350 } | |
1351 } | |
2967 | 1352 |
1319 | 1353 static inline void avg_tpel_pixels_mc01_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1354 int i,j; | |
1355 for (i=0; i < height; i++) { | |
1356 for (j=0; j < width; j++) { | |
2979 | 1357 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1358 } |
1359 src += stride; | |
1360 dst += stride; | |
1361 } | |
1362 } | |
2967 | 1363 |
1319 | 1364 static inline void avg_tpel_pixels_mc11_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1365 int i,j; | |
1366 for (i=0; i < height; i++) { | |
1367 for (j=0; j < width; j++) { | |
2979 | 1368 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 | 1369 } |
1370 src += stride; | |
1371 dst += stride; | |
1372 } | |
1373 } | |
1374 | |
1375 static inline void avg_tpel_pixels_mc12_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1376 int i,j; | |
1377 for (i=0; i < height; i++) { | |
1378 for (j=0; j < width; j++) { | |
2979 | 1379 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 | 1380 } |
1381 src += stride; | |
1382 dst += stride; | |
1383 } | |
1384 } | |
1385 | |
1386 static inline void avg_tpel_pixels_mc02_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1387 int i,j; | |
1388 for (i=0; i < height; i++) { | |
1389 for (j=0; j < width; j++) { | |
2979 | 1390 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1391 } |
1392 src += stride; | |
1393 dst += stride; | |
1394 } | |
1395 } | |
1396 | |
1397 static inline void avg_tpel_pixels_mc21_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1398 int i,j; | |
1399 for (i=0; i < height; i++) { | |
1400 for (j=0; j < width; j++) { | |
2979 | 1401 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 | 1402 } |
1403 src += stride; | |
1404 dst += stride; | |
1405 } | |
1406 } | |
1407 | |
1408 static inline void avg_tpel_pixels_mc22_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1409 int i,j; | |
1410 for (i=0; i < height; i++) { | |
1411 for (j=0; j < width; j++) { | |
2979 | 1412 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 | 1413 } |
1414 src += stride; | |
1415 dst += stride; | |
1416 } | |
1417 } | |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1418 #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
|
1419 #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
|
1420 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
|
1421 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
|
1422 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
|
1423 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
|
1424 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
|
1425 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
|
1426 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
|
1427 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
|
1428 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
|
1429 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
|
1430 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
|
1431 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
|
1432 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
|
1433 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
|
1434 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
|
1435 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
|
1436 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
|
1437 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
|
1438 #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
|
1439 |
1168 | 1440 #define H264_CHROMA_MC(OPNAME, OP)\ |
1441 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){\ | |
1442 const int A=(8-x)*(8-y);\ | |
1443 const int B=( x)*(8-y);\ | |
1444 const int C=(8-x)*( y);\ | |
1445 const int D=( x)*( y);\ | |
1446 int i;\ | |
1447 \ | |
1448 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1449 \ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1450 if(D){\ |
6054 | 1451 for(i=0; i<h; i++){\ |
6053 | 1452 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1453 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1454 dst+= stride;\ | |
1455 src+= stride;\ | |
1456 }\ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1457 }else{\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1458 const int E= B+C;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1459 const int step= C ? stride : 1;\ |
6054 | 1460 for(i=0; i<h; i++){\ |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1461 OP(dst[0], (A*src[0] + E*src[step+0]));\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1462 OP(dst[1], (A*src[1] + E*src[step+1]));\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1463 dst+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1464 src+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1465 }\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1466 }\ |
1168 | 1467 }\ |
1468 \ | |
1469 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){\ | |
1470 const int A=(8-x)*(8-y);\ | |
1471 const int B=( x)*(8-y);\ | |
1472 const int C=(8-x)*( y);\ | |
1473 const int D=( x)*( y);\ | |
1474 int i;\ | |
1475 \ | |
1476 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1477 \ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1478 if(D){\ |
6054 | 1479 for(i=0; i<h; i++){\ |
6053 | 1480 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1481 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1482 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1483 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1484 dst+= stride;\ | |
1485 src+= stride;\ | |
1486 }\ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1487 }else{\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1488 const int E= B+C;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1489 const int step= C ? stride : 1;\ |
6054 | 1490 for(i=0; i<h; i++){\ |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1491 OP(dst[0], (A*src[0] + E*src[step+0]));\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1492 OP(dst[1], (A*src[1] + E*src[step+1]));\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1493 OP(dst[2], (A*src[2] + E*src[step+2]));\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1494 OP(dst[3], (A*src[3] + E*src[step+3]));\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1495 dst+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1496 src+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1497 }\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1498 }\ |
1168 | 1499 }\ |
1500 \ | |
1501 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){\ | |
1502 const int A=(8-x)*(8-y);\ | |
1503 const int B=( x)*(8-y);\ | |
1504 const int C=(8-x)*( y);\ | |
1505 const int D=( x)*( y);\ | |
1506 int i;\ | |
1507 \ | |
1508 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1509 \ | |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1510 if(D){\ |
6054 | 1511 for(i=0; i<h; i++){\ |
6053 | 1512 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1513 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1514 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1515 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1516 OP(dst[4], (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5]));\ | |
1517 OP(dst[5], (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6]));\ | |
1518 OP(dst[6], (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7]));\ | |
1519 OP(dst[7], (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8]));\ | |
1520 dst+= stride;\ | |
1521 src+= stride;\ | |
1522 }\ | |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1523 }else{\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1524 const int E= B+C;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1525 const int step= C ? stride : 1;\ |
6054 | 1526 for(i=0; i<h; i++){\ |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1527 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
|
1528 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
|
1529 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
|
1530 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
|
1531 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
|
1532 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
|
1533 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
|
1534 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
|
1535 dst+= stride;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1536 src+= stride;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1537 }\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1538 }\ |
1168 | 1539 } |
1540 | |
1541 #define op_avg(a, b) a = (((a)+(((b) + 32)>>6)+1)>>1) | |
1542 #define op_put(a, b) a = (((b) + 32)>>6) | |
1543 | |
1544 H264_CHROMA_MC(put_ , op_put) | |
1545 H264_CHROMA_MC(avg_ , op_avg) | |
1546 #undef op_avg | |
1547 #undef op_put | |
1548 | |
3663 | 1549 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){ |
1550 const int A=(8-x)*(8-y); | |
1551 const int B=( x)*(8-y); | |
1552 const int C=(8-x)*( y); | |
1553 const int D=( x)*( y); | |
1554 int i; | |
1555 | |
1556 assert(x<8 && y<8 && x>=0 && y>=0); | |
1557 | |
1558 for(i=0; i<h; i++) | |
1559 { | |
1560 dst[0] = (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6; | |
1561 dst[1] = (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6; | |
1562 dst[2] = (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6; | |
1563 dst[3] = (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6; | |
1564 dst[4] = (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6; | |
1565 dst[5] = (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6; | |
1566 dst[6] = (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6; | |
1567 dst[7] = (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6; | |
1568 dst+= stride; | |
1569 src+= stride; | |
1570 } | |
1571 } | |
1572 | |
651 | 1573 #define QPEL_MC(r, OPNAME, RND, OP) \ |
1064 | 1574 static void OPNAME ## mpeg4_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
4176 | 1575 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1576 int i;\ |
1577 for(i=0; i<h; i++)\ | |
1578 {\ | |
1579 OP(dst[0], (src[0]+src[1])*20 - (src[0]+src[2])*6 + (src[1]+src[3])*3 - (src[2]+src[4]));\ | |
1580 OP(dst[1], (src[1]+src[2])*20 - (src[0]+src[3])*6 + (src[0]+src[4])*3 - (src[1]+src[5]));\ | |
1581 OP(dst[2], (src[2]+src[3])*20 - (src[1]+src[4])*6 + (src[0]+src[5])*3 - (src[0]+src[6]));\ | |
1582 OP(dst[3], (src[3]+src[4])*20 - (src[2]+src[5])*6 + (src[1]+src[6])*3 - (src[0]+src[7]));\ | |
1583 OP(dst[4], (src[4]+src[5])*20 - (src[3]+src[6])*6 + (src[2]+src[7])*3 - (src[1]+src[8]));\ | |
1584 OP(dst[5], (src[5]+src[6])*20 - (src[4]+src[7])*6 + (src[3]+src[8])*3 - (src[2]+src[8]));\ | |
1585 OP(dst[6], (src[6]+src[7])*20 - (src[5]+src[8])*6 + (src[4]+src[8])*3 - (src[3]+src[7]));\ | |
1586 OP(dst[7], (src[7]+src[8])*20 - (src[6]+src[8])*6 + (src[5]+src[7])*3 - (src[4]+src[6]));\ | |
1587 dst+=dstStride;\ | |
1588 src+=srcStride;\ | |
1589 }\ | |
1590 }\ | |
1591 \ | |
1064 | 1592 static void OPNAME ## mpeg4_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
984 | 1593 const int w=8;\ |
4176 | 1594 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1595 int i;\ |
1596 for(i=0; i<w; i++)\ | |
1597 {\ | |
1598 const int src0= src[0*srcStride];\ | |
1599 const int src1= src[1*srcStride];\ | |
1600 const int src2= src[2*srcStride];\ | |
1601 const int src3= src[3*srcStride];\ | |
1602 const int src4= src[4*srcStride];\ | |
1603 const int src5= src[5*srcStride];\ | |
1604 const int src6= src[6*srcStride];\ | |
1605 const int src7= src[7*srcStride];\ | |
1606 const int src8= src[8*srcStride];\ | |
1607 OP(dst[0*dstStride], (src0+src1)*20 - (src0+src2)*6 + (src1+src3)*3 - (src2+src4));\ | |
1608 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*6 + (src0+src4)*3 - (src1+src5));\ | |
1609 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*6 + (src0+src5)*3 - (src0+src6));\ | |
1610 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*6 + (src1+src6)*3 - (src0+src7));\ | |
1611 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*6 + (src2+src7)*3 - (src1+src8));\ | |
1612 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*6 + (src3+src8)*3 - (src2+src8));\ | |
1613 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*6 + (src4+src8)*3 - (src3+src7));\ | |
1614 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src8)*6 + (src5+src7)*3 - (src4+src6));\ | |
1615 dst++;\ | |
1616 src++;\ | |
1617 }\ | |
1618 }\ | |
1619 \ | |
1064 | 1620 static void OPNAME ## mpeg4_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
4176 | 1621 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1622 int i;\ |
954 | 1623 \ |
651 | 1624 for(i=0; i<h; i++)\ |
1625 {\ | |
1626 OP(dst[ 0], (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]));\ | |
1627 OP(dst[ 1], (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]));\ | |
1628 OP(dst[ 2], (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]));\ | |
1629 OP(dst[ 3], (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]));\ | |
1630 OP(dst[ 4], (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]));\ | |
1631 OP(dst[ 5], (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]));\ | |
1632 OP(dst[ 6], (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]));\ | |
1633 OP(dst[ 7], (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]));\ | |
1634 OP(dst[ 8], (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]));\ | |
1635 OP(dst[ 9], (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]));\ | |
1636 OP(dst[10], (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]));\ | |
1637 OP(dst[11], (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]));\ | |
1638 OP(dst[12], (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]));\ | |
1639 OP(dst[13], (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]));\ | |
1640 OP(dst[14], (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]));\ | |
1641 OP(dst[15], (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]));\ | |
1642 dst+=dstStride;\ | |
1643 src+=srcStride;\ | |
1644 }\ | |
255 | 1645 }\ |
1646 \ | |
1064 | 1647 static void OPNAME ## mpeg4_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
4176 | 1648 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1649 int i;\ |
954 | 1650 const int w=16;\ |
651 | 1651 for(i=0; i<w; i++)\ |
1652 {\ | |
1653 const int src0= src[0*srcStride];\ | |
1654 const int src1= src[1*srcStride];\ | |
1655 const int src2= src[2*srcStride];\ | |
1656 const int src3= src[3*srcStride];\ | |
1657 const int src4= src[4*srcStride];\ | |
1658 const int src5= src[5*srcStride];\ | |
1659 const int src6= src[6*srcStride];\ | |
1660 const int src7= src[7*srcStride];\ | |
1661 const int src8= src[8*srcStride];\ | |
1662 const int src9= src[9*srcStride];\ | |
1663 const int src10= src[10*srcStride];\ | |
1664 const int src11= src[11*srcStride];\ | |
1665 const int src12= src[12*srcStride];\ | |
1666 const int src13= src[13*srcStride];\ | |
1667 const int src14= src[14*srcStride];\ | |
1668 const int src15= src[15*srcStride];\ | |
1669 const int src16= src[16*srcStride];\ | |
1670 OP(dst[ 0*dstStride], (src0 +src1 )*20 - (src0 +src2 )*6 + (src1 +src3 )*3 - (src2 +src4 ));\ | |
1671 OP(dst[ 1*dstStride], (src1 +src2 )*20 - (src0 +src3 )*6 + (src0 +src4 )*3 - (src1 +src5 ));\ | |
1672 OP(dst[ 2*dstStride], (src2 +src3 )*20 - (src1 +src4 )*6 + (src0 +src5 )*3 - (src0 +src6 ));\ | |
1673 OP(dst[ 3*dstStride], (src3 +src4 )*20 - (src2 +src5 )*6 + (src1 +src6 )*3 - (src0 +src7 ));\ | |
1674 OP(dst[ 4*dstStride], (src4 +src5 )*20 - (src3 +src6 )*6 + (src2 +src7 )*3 - (src1 +src8 ));\ | |
1675 OP(dst[ 5*dstStride], (src5 +src6 )*20 - (src4 +src7 )*6 + (src3 +src8 )*3 - (src2 +src9 ));\ | |
1676 OP(dst[ 6*dstStride], (src6 +src7 )*20 - (src5 +src8 )*6 + (src4 +src9 )*3 - (src3 +src10));\ | |
1677 OP(dst[ 7*dstStride], (src7 +src8 )*20 - (src6 +src9 )*6 + (src5 +src10)*3 - (src4 +src11));\ | |
1678 OP(dst[ 8*dstStride], (src8 +src9 )*20 - (src7 +src10)*6 + (src6 +src11)*3 - (src5 +src12));\ | |
1679 OP(dst[ 9*dstStride], (src9 +src10)*20 - (src8 +src11)*6 + (src7 +src12)*3 - (src6 +src13));\ | |
1680 OP(dst[10*dstStride], (src10+src11)*20 - (src9 +src12)*6 + (src8 +src13)*3 - (src7 +src14));\ | |
1681 OP(dst[11*dstStride], (src11+src12)*20 - (src10+src13)*6 + (src9 +src14)*3 - (src8 +src15));\ | |
1682 OP(dst[12*dstStride], (src12+src13)*20 - (src11+src14)*6 + (src10+src15)*3 - (src9 +src16));\ | |
1683 OP(dst[13*dstStride], (src13+src14)*20 - (src12+src15)*6 + (src11+src16)*3 - (src10+src16));\ | |
1684 OP(dst[14*dstStride], (src14+src15)*20 - (src13+src16)*6 + (src12+src16)*3 - (src11+src15));\ | |
1685 OP(dst[15*dstStride], (src15+src16)*20 - (src14+src16)*6 + (src13+src15)*3 - (src12+src14));\ | |
1686 dst++;\ | |
1687 src++;\ | |
1688 }\ | |
255 | 1689 }\ |
1690 \ | |
1064 | 1691 static void OPNAME ## qpel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1692 OPNAME ## pixels8_c(dst, src, stride, 8);\ |
255 | 1693 }\ |
1694 \ | |
1064 | 1695 static void OPNAME ## qpel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
1696 uint8_t half[64];\ | |
651 | 1697 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1698 OPNAME ## pixels8_l2(dst, src, half, stride, stride, 8, 8);\ | |
1699 }\ | |
1700 \ | |
1064 | 1701 static void OPNAME ## qpel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 1702 OPNAME ## mpeg4_qpel8_h_lowpass(dst, src, stride, stride, 8);\ |
255 | 1703 }\ |
1704 \ | |
1064 | 1705 static void OPNAME ## qpel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
1706 uint8_t half[64];\ | |
651 | 1707 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1708 OPNAME ## pixels8_l2(dst, src+1, half, stride, stride, 8, 8);\ | |
1709 }\ | |
1710 \ | |
1064 | 1711 static void OPNAME ## qpel8_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
1712 uint8_t full[16*9];\ | |
1713 uint8_t half[64];\ | |
651 | 1714 copy_block9(full, src, 16, stride, 9);\ |
984 | 1715 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1716 OPNAME ## pixels8_l2(dst, full, half, stride, 16, 8, 8);\ |
1717 }\ | |
1718 \ | |
1064 | 1719 static void OPNAME ## qpel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
1720 uint8_t full[16*9];\ | |
651 | 1721 copy_block9(full, src, 16, stride, 9);\ |
984 | 1722 OPNAME ## mpeg4_qpel8_v_lowpass(dst, full, stride, 16);\ |
255 | 1723 }\ |
1724 \ | |
1064 | 1725 static void OPNAME ## qpel8_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
1726 uint8_t full[16*9];\ | |
1727 uint8_t half[64];\ | |
651 | 1728 copy_block9(full, src, 16, stride, 9);\ |
984 | 1729 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1730 OPNAME ## pixels8_l2(dst, full+16, half, stride, 16, 8, 8);\ |
1731 }\ | |
1064 | 1732 void ff_ ## OPNAME ## qpel8_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1733 uint8_t full[16*9];\ | |
1734 uint8_t halfH[72];\ | |
1735 uint8_t halfV[64];\ | |
1736 uint8_t halfHV[64];\ | |
651 | 1737 copy_block9(full, src, 16, stride, 9);\ |
1738 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1739 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1740 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1741 OPNAME ## pixels8_l4(dst, full, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1742 }\ |
1064 | 1743 static void OPNAME ## qpel8_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
1744 uint8_t full[16*9];\ | |
1745 uint8_t halfH[72];\ | |
1746 uint8_t halfHV[64];\ | |
984 | 1747 copy_block9(full, src, 16, stride, 9);\ |
1748 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1749 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1750 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1751 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1752 }\ | |
1064 | 1753 void ff_ ## OPNAME ## qpel8_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1754 uint8_t full[16*9];\ | |
1755 uint8_t halfH[72];\ | |
1756 uint8_t halfV[64];\ | |
1757 uint8_t halfHV[64];\ | |
651 | 1758 copy_block9(full, src, 16, stride, 9);\ |
1759 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1760 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1761 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1762 OPNAME ## pixels8_l4(dst, full+1, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1763 }\ |
1064 | 1764 static void OPNAME ## qpel8_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
1765 uint8_t full[16*9];\ | |
1766 uint8_t halfH[72];\ | |
1767 uint8_t halfHV[64];\ | |
984 | 1768 copy_block9(full, src, 16, stride, 9);\ |
1769 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1770 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1771 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1772 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1773 }\ | |
1064 | 1774 void ff_ ## OPNAME ## qpel8_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1775 uint8_t full[16*9];\ | |
1776 uint8_t halfH[72];\ | |
1777 uint8_t halfV[64];\ | |
1778 uint8_t halfHV[64];\ | |
651 | 1779 copy_block9(full, src, 16, stride, 9);\ |
1780 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1781 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1782 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1783 OPNAME ## pixels8_l4(dst, full+16, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
1784 }\ | |
1064 | 1785 static void OPNAME ## qpel8_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
1786 uint8_t full[16*9];\ | |
1787 uint8_t halfH[72];\ | |
1788 uint8_t halfHV[64];\ | |
984 | 1789 copy_block9(full, src, 16, stride, 9);\ |
1790 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1791 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1792 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1793 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1794 }\ | |
1064 | 1795 void ff_ ## OPNAME ## qpel8_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1796 uint8_t full[16*9];\ | |
1797 uint8_t halfH[72];\ | |
1798 uint8_t halfV[64];\ | |
1799 uint8_t halfHV[64];\ | |
651 | 1800 copy_block9(full, src, 16, stride, 9);\ |
1801 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full , 8, 16, 9);\ | |
984 | 1802 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1803 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1804 OPNAME ## pixels8_l4(dst, full+17, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1805 }\ |
1064 | 1806 static void OPNAME ## qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
1807 uint8_t full[16*9];\ | |
1808 uint8_t halfH[72];\ | |
1809 uint8_t halfHV[64];\ | |
984 | 1810 copy_block9(full, src, 16, stride, 9);\ |
1811 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1812 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1813 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1814 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1815 }\ | |
1064 | 1816 static void OPNAME ## qpel8_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
1817 uint8_t halfH[72];\ | |
1818 uint8_t halfHV[64];\ | |
651 | 1819 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1820 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1821 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ |
1822 }\ | |
1064 | 1823 static void OPNAME ## qpel8_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
1824 uint8_t halfH[72];\ | |
1825 uint8_t halfHV[64];\ | |
651 | 1826 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1827 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1828 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ |
1829 }\ | |
1064 | 1830 void ff_ ## OPNAME ## qpel8_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1831 uint8_t full[16*9];\ | |
1832 uint8_t halfH[72];\ | |
1833 uint8_t halfV[64];\ | |
1834 uint8_t halfHV[64];\ | |
651 | 1835 copy_block9(full, src, 16, stride, 9);\ |
1836 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1837 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1838 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1839 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
255 | 1840 }\ |
1064 | 1841 static void OPNAME ## qpel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
1842 uint8_t full[16*9];\ | |
1843 uint8_t halfH[72];\ | |
984 | 1844 copy_block9(full, src, 16, stride, 9);\ |
1845 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1846 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1847 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1848 }\ | |
1064 | 1849 void ff_ ## OPNAME ## qpel8_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1850 uint8_t full[16*9];\ | |
1851 uint8_t halfH[72];\ | |
1852 uint8_t halfV[64];\ | |
1853 uint8_t halfHV[64];\ | |
651 | 1854 copy_block9(full, src, 16, stride, 9);\ |
1855 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1856 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1857 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1858 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
1859 }\ | |
1064 | 1860 static void OPNAME ## qpel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
1861 uint8_t full[16*9];\ | |
1862 uint8_t halfH[72];\ | |
984 | 1863 copy_block9(full, src, 16, stride, 9);\ |
1864 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1865 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1866 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1867 }\ | |
1064 | 1868 static void OPNAME ## qpel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
1869 uint8_t halfH[72];\ | |
651 | 1870 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1871 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ |
651 | 1872 }\ |
1064 | 1873 static void OPNAME ## qpel16_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1874 OPNAME ## pixels16_c(dst, src, stride, 16);\ |
255 | 1875 }\ |
651 | 1876 \ |
1064 | 1877 static void OPNAME ## qpel16_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
1878 uint8_t half[256];\ | |
651 | 1879 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
1880 OPNAME ## pixels16_l2(dst, src, half, stride, stride, 16, 16);\ | |
1881 }\ | |
1882 \ | |
1064 | 1883 static void OPNAME ## qpel16_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 1884 OPNAME ## mpeg4_qpel16_h_lowpass(dst, src, stride, stride, 16);\ |
1885 }\ | |
1886 \ | |
1064 | 1887 static void OPNAME ## qpel16_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
1888 uint8_t half[256];\ | |
651 | 1889 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
1890 OPNAME ## pixels16_l2(dst, src+1, half, stride, stride, 16, 16);\ | |
1891 }\ | |
1892 \ | |
1064 | 1893 static void OPNAME ## qpel16_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
1894 uint8_t full[24*17];\ | |
1895 uint8_t half[256];\ | |
651 | 1896 copy_block17(full, src, 24, stride, 17);\ |
954 | 1897 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 1898 OPNAME ## pixels16_l2(dst, full, half, stride, 24, 16, 16);\ |
255 | 1899 }\ |
651 | 1900 \ |
1064 | 1901 static void OPNAME ## qpel16_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
1902 uint8_t full[24*17];\ | |
651 | 1903 copy_block17(full, src, 24, stride, 17);\ |
954 | 1904 OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24);\ |
651 | 1905 }\ |
1906 \ | |
1064 | 1907 static void OPNAME ## qpel16_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
1908 uint8_t full[24*17];\ | |
1909 uint8_t half[256];\ | |
651 | 1910 copy_block17(full, src, 24, stride, 17);\ |
954 | 1911 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 1912 OPNAME ## pixels16_l2(dst, full+24, half, stride, 24, 16, 16);\ |
255 | 1913 }\ |
1064 | 1914 void ff_ ## OPNAME ## qpel16_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1915 uint8_t full[24*17];\ | |
1916 uint8_t halfH[272];\ | |
1917 uint8_t halfV[256];\ | |
1918 uint8_t halfHV[256];\ | |
651 | 1919 copy_block17(full, src, 24, stride, 17);\ |
1920 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1921 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
1922 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1923 OPNAME ## pixels16_l4(dst, full, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
1924 }\ | |
1064 | 1925 static void OPNAME ## qpel16_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
1926 uint8_t full[24*17];\ | |
1927 uint8_t halfH[272];\ | |
1928 uint8_t halfHV[256];\ | |
984 | 1929 copy_block17(full, src, 24, stride, 17);\ |
1930 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1931 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
1932 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1933 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
1934 }\ | |
1064 | 1935 void ff_ ## OPNAME ## qpel16_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1936 uint8_t full[24*17];\ | |
1937 uint8_t halfH[272];\ | |
1938 uint8_t halfV[256];\ | |
1939 uint8_t halfHV[256];\ | |
651 | 1940 copy_block17(full, src, 24, stride, 17);\ |
1941 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1942 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
1943 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1944 OPNAME ## pixels16_l4(dst, full+1, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
1945 }\ | |
1064 | 1946 static void OPNAME ## qpel16_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
1947 uint8_t full[24*17];\ | |
1948 uint8_t halfH[272];\ | |
1949 uint8_t halfHV[256];\ | |
984 | 1950 copy_block17(full, src, 24, stride, 17);\ |
1951 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1952 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
1953 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1954 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
1955 }\ | |
1064 | 1956 void ff_ ## OPNAME ## qpel16_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1957 uint8_t full[24*17];\ | |
1958 uint8_t halfH[272];\ | |
1959 uint8_t halfV[256];\ | |
1960 uint8_t halfHV[256];\ | |
651 | 1961 copy_block17(full, src, 24, stride, 17);\ |
1962 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1963 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
1964 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1965 OPNAME ## pixels16_l4(dst, full+24, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
255 | 1966 }\ |
1064 | 1967 static void OPNAME ## qpel16_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
1968 uint8_t full[24*17];\ | |
1969 uint8_t halfH[272];\ | |
1970 uint8_t halfHV[256];\ | |
984 | 1971 copy_block17(full, src, 24, stride, 17);\ |
1972 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1973 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
1974 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1975 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
1976 }\ | |
1064 | 1977 void ff_ ## OPNAME ## qpel16_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1978 uint8_t full[24*17];\ | |
1979 uint8_t halfH[272];\ | |
1980 uint8_t halfV[256];\ | |
1981 uint8_t halfHV[256];\ | |
651 | 1982 copy_block17(full, src, 24, stride, 17);\ |
1983 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full , 16, 24, 17);\ | |
954 | 1984 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
1985 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1986 OPNAME ## pixels16_l4(dst, full+25, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
1987 }\ | |
1064 | 1988 static void OPNAME ## qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
1989 uint8_t full[24*17];\ | |
1990 uint8_t halfH[272];\ | |
1991 uint8_t halfHV[256];\ | |
984 | 1992 copy_block17(full, src, 24, stride, 17);\ |
1993 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1994 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
1995 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1996 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
1997 }\ | |
1064 | 1998 static void OPNAME ## qpel16_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
1999 uint8_t halfH[272];\ | |
2000 uint8_t halfHV[256];\ | |
651 | 2001 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2002 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 2003 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ |
255 | 2004 }\ |
1064 | 2005 static void OPNAME ## qpel16_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
2006 uint8_t halfH[272];\ | |
2007 uint8_t halfHV[256];\ | |
651 | 2008 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2009 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 2010 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ |
2011 }\ | |
1064 | 2012 void ff_ ## OPNAME ## qpel16_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2013 uint8_t full[24*17];\ | |
2014 uint8_t halfH[272];\ | |
2015 uint8_t halfV[256];\ | |
2016 uint8_t halfHV[256];\ | |
651 | 2017 copy_block17(full, src, 24, stride, 17);\ |
2018 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2019 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
2020 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2021 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
255 | 2022 }\ |
1064 | 2023 static void OPNAME ## qpel16_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
2024 uint8_t full[24*17];\ | |
2025 uint8_t halfH[272];\ | |
984 | 2026 copy_block17(full, src, 24, stride, 17);\ |
2027 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2028 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
2029 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2030 }\ | |
1064 | 2031 void ff_ ## OPNAME ## qpel16_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2032 uint8_t full[24*17];\ | |
2033 uint8_t halfH[272];\ | |
2034 uint8_t halfV[256];\ | |
2035 uint8_t halfHV[256];\ | |
651 | 2036 copy_block17(full, src, 24, stride, 17);\ |
2037 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2038 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
2039 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2040 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
2041 }\ | |
1064 | 2042 static void OPNAME ## qpel16_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
2043 uint8_t full[24*17];\ | |
2044 uint8_t halfH[272];\ | |
984 | 2045 copy_block17(full, src, 24, stride, 17);\ |
2046 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2047 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2048 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2049 }\ | |
1064 | 2050 static void OPNAME ## qpel16_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
2051 uint8_t halfH[272];\ | |
651 | 2052 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2053 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ |
859 | 2054 } |
255 | 2055 |
651 | 2056 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) |
2057 #define op_avg_no_rnd(a, b) a = (((a)+cm[((b) + 15)>>5])>>1) | |
2058 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2059 #define op_put_no_rnd(a, b) a = cm[((b) + 15)>>5] | |
2060 | |
2061 QPEL_MC(0, put_ , _ , op_put) | |
2062 QPEL_MC(1, put_no_rnd_, _no_rnd_, op_put_no_rnd) | |
2063 QPEL_MC(0, avg_ , _ , op_avg) | |
2064 //QPEL_MC(1, avg_no_rnd , _ , op_avg) | |
2065 #undef op_avg | |
2066 #undef op_avg_no_rnd | |
2067 #undef op_put | |
2068 #undef op_put_no_rnd | |
255 | 2069 |
1168 | 2070 #if 1 |
2071 #define H264_LOWPASS(OPNAME, OP, OP2) \ | |
5151 | 2072 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
|
2073 const int h=2;\ |
4176 | 2074 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
|
2075 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2076 for(i=0; i<h; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2077 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2078 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
|
2079 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
|
2080 dst+=dstStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2081 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2082 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2083 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2084 \ |
5151 | 2085 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
|
2086 const int w=2;\ |
4176 | 2087 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
|
2088 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2089 for(i=0; i<w; i++)\ |
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 const int srcB= src[-2*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2092 const int srcA= src[-1*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2093 const int src0= src[0 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2094 const int src1= src[1 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2095 const int src2= src[2 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2096 const int src3= src[3 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2097 const int src4= src[4 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2098 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
|
2099 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
|
2100 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2101 src++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2102 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2103 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2104 \ |
5151 | 2105 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
|
2106 const int h=2;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2107 const int w=2;\ |
4176 | 2108 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
|
2109 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2110 src -= 2*srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2111 for(i=0; i<h+5; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2112 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2113 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
|
2114 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
|
2115 tmp+=tmpStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2116 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2117 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2118 tmp -= tmpStride*(h+5-2);\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2119 for(i=0; i<w; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2120 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2121 const int tmpB= tmp[-2*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2122 const int tmpA= tmp[-1*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2123 const int tmp0= tmp[0 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2124 const int tmp1= tmp[1 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2125 const int tmp2= tmp[2 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2126 const int tmp3= tmp[3 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2127 const int tmp4= tmp[4 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2128 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
|
2129 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
|
2130 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2131 tmp++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2132 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2133 }\ |
1168 | 2134 static void OPNAME ## h264_qpel4_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
2135 const int h=4;\ | |
4176 | 2136 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2137 int i;\ |
2138 for(i=0; i<h; i++)\ | |
2139 {\ | |
2140 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]));\ | |
2141 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]));\ | |
2142 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]));\ | |
2143 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]));\ | |
2144 dst+=dstStride;\ | |
2145 src+=srcStride;\ | |
2146 }\ | |
2147 }\ | |
2148 \ | |
2149 static void OPNAME ## h264_qpel4_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2150 const int w=4;\ | |
4176 | 2151 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2152 int i;\ |
2153 for(i=0; i<w; i++)\ | |
2154 {\ | |
2155 const int srcB= src[-2*srcStride];\ | |
2156 const int srcA= src[-1*srcStride];\ | |
2157 const int src0= src[0 *srcStride];\ | |
2158 const int src1= src[1 *srcStride];\ | |
2159 const int src2= src[2 *srcStride];\ | |
2160 const int src3= src[3 *srcStride];\ | |
2161 const int src4= src[4 *srcStride];\ | |
2162 const int src5= src[5 *srcStride];\ | |
2163 const int src6= src[6 *srcStride];\ | |
2164 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2165 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2166 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2167 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2168 dst++;\ | |
2169 src++;\ | |
2170 }\ | |
2171 }\ | |
2172 \ | |
2173 static void OPNAME ## h264_qpel4_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2174 const int h=4;\ | |
2175 const int w=4;\ | |
4176 | 2176 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2177 int i;\ |
2178 src -= 2*srcStride;\ | |
2179 for(i=0; i<h+5; i++)\ | |
2180 {\ | |
2181 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]);\ | |
2182 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]);\ | |
2183 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]);\ | |
2184 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]);\ | |
2185 tmp+=tmpStride;\ | |
2186 src+=srcStride;\ | |
2187 }\ | |
2188 tmp -= tmpStride*(h+5-2);\ | |
2189 for(i=0; i<w; i++)\ | |
2190 {\ | |
2191 const int tmpB= tmp[-2*tmpStride];\ | |
2192 const int tmpA= tmp[-1*tmpStride];\ | |
2193 const int tmp0= tmp[0 *tmpStride];\ | |
2194 const int tmp1= tmp[1 *tmpStride];\ | |
2195 const int tmp2= tmp[2 *tmpStride];\ | |
2196 const int tmp3= tmp[3 *tmpStride];\ | |
2197 const int tmp4= tmp[4 *tmpStride];\ | |
2198 const int tmp5= tmp[5 *tmpStride];\ | |
2199 const int tmp6= tmp[6 *tmpStride];\ | |
2200 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2201 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2202 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2203 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2204 dst++;\ | |
2205 tmp++;\ | |
2206 }\ | |
2207 }\ | |
2208 \ | |
2209 static void OPNAME ## h264_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2210 const int h=8;\ | |
4176 | 2211 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2212 int i;\ |
2213 for(i=0; i<h; i++)\ | |
2214 {\ | |
2215 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]));\ | |
2216 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]));\ | |
2217 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]));\ | |
2218 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]));\ | |
2219 OP(dst[4], (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]));\ | |
2220 OP(dst[5], (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]));\ | |
2221 OP(dst[6], (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]));\ | |
2222 OP(dst[7], (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]));\ | |
2223 dst+=dstStride;\ | |
2224 src+=srcStride;\ | |
2225 }\ | |
2226 }\ | |
2227 \ | |
2228 static void OPNAME ## h264_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2229 const int w=8;\ | |
4176 | 2230 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2231 int i;\ |
2232 for(i=0; i<w; i++)\ | |
2233 {\ | |
2234 const int srcB= src[-2*srcStride];\ | |
2235 const int srcA= src[-1*srcStride];\ | |
2236 const int src0= src[0 *srcStride];\ | |
2237 const int src1= src[1 *srcStride];\ | |
2238 const int src2= src[2 *srcStride];\ | |
2239 const int src3= src[3 *srcStride];\ | |
2240 const int src4= src[4 *srcStride];\ | |
2241 const int src5= src[5 *srcStride];\ | |
2242 const int src6= src[6 *srcStride];\ | |
2243 const int src7= src[7 *srcStride];\ | |
2244 const int src8= src[8 *srcStride];\ | |
2245 const int src9= src[9 *srcStride];\ | |
2246 const int src10=src[10*srcStride];\ | |
2247 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2248 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2249 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2250 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2251 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*5 + (src2+src7));\ | |
2252 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*5 + (src3+src8));\ | |
2253 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*5 + (src4+src9));\ | |
2254 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src9)*5 + (src5+src10));\ | |
2255 dst++;\ | |
2256 src++;\ | |
2257 }\ | |
2258 }\ | |
2259 \ | |
2260 static void OPNAME ## h264_qpel8_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2261 const int h=8;\ | |
2262 const int w=8;\ | |
4176 | 2263 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2264 int i;\ |
2265 src -= 2*srcStride;\ | |
2266 for(i=0; i<h+5; i++)\ | |
2267 {\ | |
2268 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]);\ | |
2269 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]);\ | |
2270 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]);\ | |
2271 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]);\ | |
2272 tmp[4]= (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]);\ | |
2273 tmp[5]= (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]);\ | |
2274 tmp[6]= (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]);\ | |
2275 tmp[7]= (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]);\ | |
2276 tmp+=tmpStride;\ | |
2277 src+=srcStride;\ | |
2278 }\ | |
2279 tmp -= tmpStride*(h+5-2);\ | |
2280 for(i=0; i<w; i++)\ | |
2281 {\ | |
2282 const int tmpB= tmp[-2*tmpStride];\ | |
2283 const int tmpA= tmp[-1*tmpStride];\ | |
2284 const int tmp0= tmp[0 *tmpStride];\ | |
2285 const int tmp1= tmp[1 *tmpStride];\ | |
2286 const int tmp2= tmp[2 *tmpStride];\ | |
2287 const int tmp3= tmp[3 *tmpStride];\ | |
2288 const int tmp4= tmp[4 *tmpStride];\ | |
2289 const int tmp5= tmp[5 *tmpStride];\ | |
2290 const int tmp6= tmp[6 *tmpStride];\ | |
2291 const int tmp7= tmp[7 *tmpStride];\ | |
2292 const int tmp8= tmp[8 *tmpStride];\ | |
2293 const int tmp9= tmp[9 *tmpStride];\ | |
2294 const int tmp10=tmp[10*tmpStride];\ | |
2295 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2296 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2297 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2298 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2299 OP2(dst[4*dstStride], (tmp4+tmp5)*20 - (tmp3+tmp6)*5 + (tmp2+tmp7));\ | |
2300 OP2(dst[5*dstStride], (tmp5+tmp6)*20 - (tmp4+tmp7)*5 + (tmp3+tmp8));\ | |
2301 OP2(dst[6*dstStride], (tmp6+tmp7)*20 - (tmp5+tmp8)*5 + (tmp4+tmp9));\ | |
2302 OP2(dst[7*dstStride], (tmp7+tmp8)*20 - (tmp6+tmp9)*5 + (tmp5+tmp10));\ | |
2303 dst++;\ | |
2304 tmp++;\ | |
2305 }\ | |
2306 }\ | |
2307 \ | |
2308 static void OPNAME ## h264_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2309 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2310 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2311 src += 8*srcStride;\ | |
2312 dst += 8*dstStride;\ | |
2313 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2314 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2315 }\ | |
2316 \ | |
2317 static void OPNAME ## h264_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2318 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2319 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2320 src += 8*srcStride;\ | |
2321 dst += 8*dstStride;\ | |
2322 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2323 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2324 }\ | |
2325 \ | |
2326 static void OPNAME ## h264_qpel16_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2327 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2328 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2329 src += 8*srcStride;\ | |
2330 dst += 8*dstStride;\ | |
2331 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2332 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2333 }\ | |
2334 | |
2335 #define H264_MC(OPNAME, SIZE) \ | |
2336 static void OPNAME ## h264_qpel ## SIZE ## _mc00_c (uint8_t *dst, uint8_t *src, int stride){\ | |
2337 OPNAME ## pixels ## SIZE ## _c(dst, src, stride, SIZE);\ | |
2338 }\ | |
2339 \ | |
2340 static void OPNAME ## h264_qpel ## SIZE ## _mc10_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2341 uint8_t half[SIZE*SIZE];\ | |
2342 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2343 OPNAME ## pixels ## SIZE ## _l2(dst, src, half, stride, stride, SIZE, SIZE);\ | |
2344 }\ | |
2345 \ | |
2346 static void OPNAME ## h264_qpel ## SIZE ## _mc20_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2347 OPNAME ## h264_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride);\ | |
2348 }\ | |
2349 \ | |
2350 static void OPNAME ## h264_qpel ## SIZE ## _mc30_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2351 uint8_t half[SIZE*SIZE];\ | |
2352 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2353 OPNAME ## pixels ## SIZE ## _l2(dst, src+1, half, stride, stride, SIZE, SIZE);\ | |
2354 }\ | |
2355 \ | |
2356 static void OPNAME ## h264_qpel ## SIZE ## _mc01_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2357 uint8_t full[SIZE*(SIZE+5)];\ | |
2358 uint8_t * const full_mid= full + SIZE*2;\ | |
2359 uint8_t half[SIZE*SIZE];\ | |
2360 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2361 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2362 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid, half, stride, SIZE, SIZE, SIZE);\ | |
2363 }\ | |
2364 \ | |
2365 static void OPNAME ## h264_qpel ## SIZE ## _mc02_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 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2369 OPNAME ## h264_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE);\ | |
2370 }\ | |
2371 \ | |
2372 static void OPNAME ## h264_qpel ## SIZE ## _mc03_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2373 uint8_t full[SIZE*(SIZE+5)];\ | |
2374 uint8_t * const full_mid= full + SIZE*2;\ | |
2375 uint8_t half[SIZE*SIZE];\ | |
2376 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2377 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2378 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid+SIZE, half, stride, SIZE, SIZE, SIZE);\ | |
2379 }\ | |
2380 \ | |
2381 static void OPNAME ## h264_qpel ## SIZE ## _mc11_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2382 uint8_t full[SIZE*(SIZE+5)];\ | |
2383 uint8_t * const full_mid= full + SIZE*2;\ | |
2384 uint8_t halfH[SIZE*SIZE];\ | |
2385 uint8_t halfV[SIZE*SIZE];\ | |
2386 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2387 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2388 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2389 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2390 }\ | |
2391 \ | |
2392 static void OPNAME ## h264_qpel ## SIZE ## _mc31_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2393 uint8_t full[SIZE*(SIZE+5)];\ | |
2394 uint8_t * const full_mid= full + SIZE*2;\ | |
2395 uint8_t halfH[SIZE*SIZE];\ | |
2396 uint8_t halfV[SIZE*SIZE];\ | |
2397 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2398 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2399 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2400 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2401 }\ | |
2402 \ | |
2403 static void OPNAME ## h264_qpel ## SIZE ## _mc13_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2404 uint8_t full[SIZE*(SIZE+5)];\ | |
2405 uint8_t * const full_mid= full + SIZE*2;\ | |
2406 uint8_t halfH[SIZE*SIZE];\ | |
2407 uint8_t halfV[SIZE*SIZE];\ | |
2408 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2409 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2410 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2411 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2412 }\ | |
2413 \ | |
2414 static void OPNAME ## h264_qpel ## SIZE ## _mc33_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2415 uint8_t full[SIZE*(SIZE+5)];\ | |
2416 uint8_t * const full_mid= full + SIZE*2;\ | |
2417 uint8_t halfH[SIZE*SIZE];\ | |
2418 uint8_t halfV[SIZE*SIZE];\ | |
2419 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2420 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2421 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2422 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2423 }\ | |
2424 \ | |
2425 static void OPNAME ## h264_qpel ## SIZE ## _mc22_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2426 int16_t tmp[SIZE*(SIZE+5)];\ | |
2427 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass(dst, tmp, src, stride, SIZE, stride);\ | |
2428 }\ | |
2429 \ | |
2430 static void OPNAME ## h264_qpel ## SIZE ## _mc21_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2431 int16_t tmp[SIZE*(SIZE+5)];\ | |
2432 uint8_t halfH[SIZE*SIZE];\ | |
2433 uint8_t halfHV[SIZE*SIZE];\ | |
2434 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2435 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2436 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2437 }\ | |
2438 \ | |
2439 static void OPNAME ## h264_qpel ## SIZE ## _mc23_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2440 int16_t tmp[SIZE*(SIZE+5)];\ | |
2441 uint8_t halfH[SIZE*SIZE];\ | |
2442 uint8_t halfHV[SIZE*SIZE];\ | |
2443 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2444 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2445 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2446 }\ | |
2447 \ | |
2448 static void OPNAME ## h264_qpel ## SIZE ## _mc12_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2449 uint8_t full[SIZE*(SIZE+5)];\ | |
2450 uint8_t * const full_mid= full + SIZE*2;\ | |
2451 int16_t tmp[SIZE*(SIZE+5)];\ | |
2452 uint8_t halfV[SIZE*SIZE];\ | |
2453 uint8_t halfHV[SIZE*SIZE];\ | |
2454 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2455 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2456 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2457 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2458 }\ | |
2459 \ | |
2460 static void OPNAME ## h264_qpel ## SIZE ## _mc32_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2461 uint8_t full[SIZE*(SIZE+5)];\ | |
2462 uint8_t * const full_mid= full + SIZE*2;\ | |
2463 int16_t tmp[SIZE*(SIZE+5)];\ | |
2464 uint8_t halfV[SIZE*SIZE];\ | |
2465 uint8_t halfHV[SIZE*SIZE];\ | |
2466 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2467 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2468 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2469 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2470 }\ | |
2471 | |
2472 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) | |
2473 //#define op_avg2(a, b) a = (((a)*w1+cm[((b) + 16)>>5]*w2 + o + 64)>>7) | |
2474 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2475 #define op2_avg(a, b) a = (((a)+cm[((b) + 512)>>10]+1)>>1) | |
2476 #define op2_put(a, b) a = cm[((b) + 512)>>10] | |
2477 | |
2478 H264_LOWPASS(put_ , op_put, op2_put) | |
2479 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
|
2480 H264_MC(put_, 2) |
1168 | 2481 H264_MC(put_, 4) |
2482 H264_MC(put_, 8) | |
2483 H264_MC(put_, 16) | |
2484 H264_MC(avg_, 4) | |
2485 H264_MC(avg_, 8) | |
2486 H264_MC(avg_, 16) | |
2487 | |
2488 #undef op_avg | |
2489 #undef op_put | |
2490 #undef op2_avg | |
2491 #undef op2_put | |
2492 #endif | |
2493 | |
4594 | 2494 #define op_scale1(x) block[x] = av_clip_uint8( (block[x]*weight + offset) >> log2_denom ) |
2495 #define op_scale2(x) dst[x] = av_clip_uint8( (src[x]*weights + dst[x]*weightd + offset) >> (log2_denom+1)) | |
2415 | 2496 #define H264_WEIGHT(W,H) \ |
2497 static void weight_h264_pixels ## W ## x ## H ## _c(uint8_t *block, int stride, int log2_denom, int weight, int offset){ \ | |
3029 | 2498 int y; \ |
2415 | 2499 offset <<= log2_denom; \ |
2500 if(log2_denom) offset += 1<<(log2_denom-1); \ | |
2501 for(y=0; y<H; y++, block += stride){ \ | |
2502 op_scale1(0); \ | |
2503 op_scale1(1); \ | |
2504 if(W==2) continue; \ | |
2505 op_scale1(2); \ | |
2506 op_scale1(3); \ | |
2507 if(W==4) continue; \ | |
2508 op_scale1(4); \ | |
2509 op_scale1(5); \ | |
2510 op_scale1(6); \ | |
2511 op_scale1(7); \ | |
2512 if(W==8) continue; \ | |
2513 op_scale1(8); \ | |
2514 op_scale1(9); \ | |
2515 op_scale1(10); \ | |
2516 op_scale1(11); \ | |
2517 op_scale1(12); \ | |
2518 op_scale1(13); \ | |
2519 op_scale1(14); \ | |
2520 op_scale1(15); \ | |
2521 } \ | |
2522 } \ | |
3029 | 2523 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){ \ |
2524 int y; \ | |
2525 offset = ((offset + 1) | 1) << log2_denom; \ | |
2415 | 2526 for(y=0; y<H; y++, dst += stride, src += stride){ \ |
2527 op_scale2(0); \ | |
2528 op_scale2(1); \ | |
2529 if(W==2) continue; \ | |
2530 op_scale2(2); \ | |
2531 op_scale2(3); \ | |
2532 if(W==4) continue; \ | |
2533 op_scale2(4); \ | |
2534 op_scale2(5); \ | |
2535 op_scale2(6); \ | |
2536 op_scale2(7); \ | |
2537 if(W==8) continue; \ | |
2538 op_scale2(8); \ | |
2539 op_scale2(9); \ | |
2540 op_scale2(10); \ | |
2541 op_scale2(11); \ | |
2542 op_scale2(12); \ | |
2543 op_scale2(13); \ | |
2544 op_scale2(14); \ | |
2545 op_scale2(15); \ | |
2546 } \ | |
2547 } | |
2548 | |
2549 H264_WEIGHT(16,16) | |
2550 H264_WEIGHT(16,8) | |
2551 H264_WEIGHT(8,16) | |
2552 H264_WEIGHT(8,8) | |
2553 H264_WEIGHT(8,4) | |
2554 H264_WEIGHT(4,8) | |
2555 H264_WEIGHT(4,4) | |
2556 H264_WEIGHT(4,2) | |
2557 H264_WEIGHT(2,4) | |
2558 H264_WEIGHT(2,2) | |
2559 | |
2560 #undef op_scale1 | |
2561 #undef op_scale2 | |
2562 #undef H264_WEIGHT | |
2563 | |
936 | 2564 static void wmv2_mspel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){ |
4176 | 2565 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
936 | 2566 int i; |
2567 | |
2568 for(i=0; i<h; i++){ | |
2569 dst[0]= cm[(9*(src[0] + src[1]) - (src[-1] + src[2]) + 8)>>4]; | |
2570 dst[1]= cm[(9*(src[1] + src[2]) - (src[ 0] + src[3]) + 8)>>4]; | |
2571 dst[2]= cm[(9*(src[2] + src[3]) - (src[ 1] + src[4]) + 8)>>4]; | |
2572 dst[3]= cm[(9*(src[3] + src[4]) - (src[ 2] + src[5]) + 8)>>4]; | |
2573 dst[4]= cm[(9*(src[4] + src[5]) - (src[ 3] + src[6]) + 8)>>4]; | |
2574 dst[5]= cm[(9*(src[5] + src[6]) - (src[ 4] + src[7]) + 8)>>4]; | |
2575 dst[6]= cm[(9*(src[6] + src[7]) - (src[ 5] + src[8]) + 8)>>4]; | |
2576 dst[7]= cm[(9*(src[7] + src[8]) - (src[ 6] + src[9]) + 8)>>4]; | |
2577 dst+=dstStride; | |
2967 | 2578 src+=srcStride; |
936 | 2579 } |
2580 } | |
2581 | |
3432 | 2582 #ifdef CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2583 /* AVS specific */ |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2584 void ff_cavsdsp_init(DSPContext* c, AVCodecContext *avctx); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2585 |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2586 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
|
2587 put_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2588 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2589 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
|
2590 avg_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2591 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2592 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
|
2593 put_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2594 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2595 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
|
2596 avg_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2597 } |
3432 | 2598 #endif /* CONFIG_CAVS_DECODER */ |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2599 |
3526 | 2600 #if defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
2601 /* VC-1 specific */ | |
2602 void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx); | |
2603 | |
2604 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, uint8_t *src, int stride, int rnd) { | |
2605 put_pixels8_c(dst, src, stride, 8); | |
2606 } | |
2607 #endif /* CONFIG_VC1_DECODER||CONFIG_WMV3_DECODER */ | |
2608 | |
5887 | 2609 void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx); |
5899 | 2610 |
4296 | 2611 /* H264 specific */ |
5411
362aec4ef932
Take care of some renames (Doxygen and function name) after the previous pure rename patch.
takis
parents:
5394
diff
changeset
|
2612 void ff_h264dspenc_init(DSPContext* c, AVCodecContext *avctx); |
4296 | 2613 |
936 | 2614 static void wmv2_mspel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int w){ |
4176 | 2615 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
936 | 2616 int i; |
2617 | |
2618 for(i=0; i<w; i++){ | |
2619 const int src_1= src[ -srcStride]; | |
2620 const int src0 = src[0 ]; | |
2621 const int src1 = src[ srcStride]; | |
2622 const int src2 = src[2*srcStride]; | |
2623 const int src3 = src[3*srcStride]; | |
2624 const int src4 = src[4*srcStride]; | |
2625 const int src5 = src[5*srcStride]; | |
2626 const int src6 = src[6*srcStride]; | |
2627 const int src7 = src[7*srcStride]; | |
2628 const int src8 = src[8*srcStride]; | |
2629 const int src9 = src[9*srcStride]; | |
2630 dst[0*dstStride]= cm[(9*(src0 + src1) - (src_1 + src2) + 8)>>4]; | |
2631 dst[1*dstStride]= cm[(9*(src1 + src2) - (src0 + src3) + 8)>>4]; | |
2632 dst[2*dstStride]= cm[(9*(src2 + src3) - (src1 + src4) + 8)>>4]; | |
2633 dst[3*dstStride]= cm[(9*(src3 + src4) - (src2 + src5) + 8)>>4]; | |
2634 dst[4*dstStride]= cm[(9*(src4 + src5) - (src3 + src6) + 8)>>4]; | |
2635 dst[5*dstStride]= cm[(9*(src5 + src6) - (src4 + src7) + 8)>>4]; | |
2636 dst[6*dstStride]= cm[(9*(src6 + src7) - (src5 + src8) + 8)>>4]; | |
2637 dst[7*dstStride]= cm[(9*(src7 + src8) - (src6 + src9) + 8)>>4]; | |
2638 src++; | |
2639 dst++; | |
2640 } | |
2641 } | |
2642 | |
2643 static void put_mspel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){ | |
2644 put_pixels8_c(dst, src, stride, 8); | |
2645 } | |
2646 | |
2647 static void put_mspel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){ | |
2648 uint8_t half[64]; | |
2649 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2650 put_pixels8_l2(dst, src, half, stride, stride, 8, 8); | |
2651 } | |
2652 | |
2653 static void put_mspel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){ | |
2654 wmv2_mspel8_h_lowpass(dst, src, stride, stride, 8); | |
2655 } | |
2656 | |
2657 static void put_mspel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){ | |
2658 uint8_t half[64]; | |
2659 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2660 put_pixels8_l2(dst, src+1, half, stride, stride, 8, 8); | |
2661 } | |
2662 | |
2663 static void put_mspel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){ | |
2664 wmv2_mspel8_v_lowpass(dst, src, stride, stride, 8); | |
2665 } | |
2666 | |
2667 static void put_mspel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){ | |
2668 uint8_t halfH[88]; | |
2669 uint8_t halfV[64]; | |
2670 uint8_t halfHV[64]; | |
2671 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2672 wmv2_mspel8_v_lowpass(halfV, src, 8, stride, 8); | |
2673 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2674 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2675 } | |
2676 static void put_mspel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){ | |
2677 uint8_t halfH[88]; | |
2678 uint8_t halfV[64]; | |
2679 uint8_t halfHV[64]; | |
2680 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2681 wmv2_mspel8_v_lowpass(halfV, src+1, 8, stride, 8); | |
2682 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2683 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2684 } | |
2685 static void put_mspel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){ | |
2686 uint8_t halfH[88]; | |
2687 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2688 wmv2_mspel8_v_lowpass(dst, halfH+8, stride, 8, 8); | |
2689 } | |
2690 | |
1644 | 2691 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
|
2692 if(ENABLE_ANY_H263) { |
1644 | 2693 int x; |
2694 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2695 |
1644 | 2696 for(x=0; x<8; x++){ |
2697 int d1, d2, ad1; | |
2698 int p0= src[x-2*stride]; | |
2699 int p1= src[x-1*stride]; | |
2700 int p2= src[x+0*stride]; | |
2701 int p3= src[x+1*stride]; | |
2702 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2703 | |
2704 if (d<-2*strength) d1= 0; | |
2705 else if(d<- strength) d1=-2*strength - d; | |
2706 else if(d< strength) d1= d; | |
2707 else if(d< 2*strength) d1= 2*strength - d; | |
2708 else d1= 0; | |
2967 | 2709 |
1644 | 2710 p1 += d1; |
2711 p2 -= d1; | |
2712 if(p1&256) p1= ~(p1>>31); | |
2713 if(p2&256) p2= ~(p2>>31); | |
2967 | 2714 |
1644 | 2715 src[x-1*stride] = p1; |
2716 src[x+0*stride] = p2; | |
2717 | |
4001 | 2718 ad1= FFABS(d1)>>1; |
2967 | 2719 |
4594 | 2720 d2= av_clip((p0-p3)/4, -ad1, ad1); |
2967 | 2721 |
1644 | 2722 src[x-2*stride] = p0 - d2; |
2723 src[x+ stride] = p3 + d2; | |
2724 } | |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2725 } |
1644 | 2726 } |
2727 | |
2728 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
|
2729 if(ENABLE_ANY_H263) { |
1644 | 2730 int y; |
2731 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2732 |
1644 | 2733 for(y=0; y<8; y++){ |
2734 int d1, d2, ad1; | |
2735 int p0= src[y*stride-2]; | |
2736 int p1= src[y*stride-1]; | |
2737 int p2= src[y*stride+0]; | |
2738 int p3= src[y*stride+1]; | |
2739 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2740 | |
2741 if (d<-2*strength) d1= 0; | |
2742 else if(d<- strength) d1=-2*strength - d; | |
2743 else if(d< strength) d1= d; | |
2744 else if(d< 2*strength) d1= 2*strength - d; | |
2745 else d1= 0; | |
2967 | 2746 |
1644 | 2747 p1 += d1; |
2748 p2 -= d1; | |
2749 if(p1&256) p1= ~(p1>>31); | |
2750 if(p2&256) p2= ~(p2>>31); | |
2967 | 2751 |
1644 | 2752 src[y*stride-1] = p1; |
2753 src[y*stride+0] = p2; | |
2754 | |
4001 | 2755 ad1= FFABS(d1)>>1; |
2967 | 2756 |
4594 | 2757 d2= av_clip((p0-p3)/4, -ad1, ad1); |
2967 | 2758 |
1644 | 2759 src[y*stride-2] = p0 - d2; |
2760 src[y*stride+1] = p3 + d2; | |
2761 } | |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2762 } |
1644 | 2763 } |
936 | 2764 |
2045 | 2765 static void h261_loop_filter_c(uint8_t *src, int stride){ |
2766 int x,y,xy,yz; | |
2767 int temp[64]; | |
2768 | |
2769 for(x=0; x<8; x++){ | |
2770 temp[x ] = 4*src[x ]; | |
2771 temp[x + 7*8] = 4*src[x + 7*stride]; | |
2772 } | |
2773 for(y=1; y<7; y++){ | |
2774 for(x=0; x<8; x++){ | |
2775 xy = y * stride + x; | |
2776 yz = y * 8 + x; | |
2777 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
|
2778 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2779 } |
2967 | 2780 |
2045 | 2781 for(y=0; y<8; y++){ |
2782 src[ y*stride] = (temp[ y*8] + 2)>>2; | |
2783 src[7+y*stride] = (temp[7+y*8] + 2)>>2; | |
2784 for(x=1; x<7; x++){ | |
2785 xy = y * stride + x; | |
2786 yz = y * 8 + x; | |
2787 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
|
2788 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2789 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2790 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2791 |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2792 static inline void h264_loop_filter_luma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0) |
2633 | 2793 { |
2794 int i, d; | |
2795 for( i = 0; i < 4; i++ ) { | |
2796 if( tc0[i] < 0 ) { | |
2797 pix += 4*ystride; | |
2798 continue; | |
2799 } | |
2800 for( d = 0; d < 4; d++ ) { | |
2801 const int p0 = pix[-1*xstride]; | |
2802 const int p1 = pix[-2*xstride]; | |
2803 const int p2 = pix[-3*xstride]; | |
2804 const int q0 = pix[0]; | |
2805 const int q1 = pix[1*xstride]; | |
2806 const int q2 = pix[2*xstride]; | |
2967 | 2807 |
4001 | 2808 if( FFABS( p0 - q0 ) < alpha && |
2809 FFABS( p1 - p0 ) < beta && | |
2810 FFABS( q1 - q0 ) < beta ) { | |
2967 | 2811 |
2633 | 2812 int tc = tc0[i]; |
2813 int i_delta; | |
2967 | 2814 |
4001 | 2815 if( FFABS( p2 - p0 ) < beta ) { |
4594 | 2816 pix[-2*xstride] = p1 + av_clip( (( p2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - p1, -tc0[i], tc0[i] ); |
2633 | 2817 tc++; |
2818 } | |
4001 | 2819 if( FFABS( q2 - q0 ) < beta ) { |
4594 | 2820 pix[ xstride] = q1 + av_clip( (( q2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - q1, -tc0[i], tc0[i] ); |
2633 | 2821 tc++; |
2822 } | |
2967 | 2823 |
4594 | 2824 i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
2825 pix[-xstride] = av_clip_uint8( p0 + i_delta ); /* p0' */ | |
2826 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */ | |
2633 | 2827 } |
2828 pix += ystride; | |
2829 } | |
2830 } | |
2831 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2832 static void h264_v_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2833 { |
2834 h264_loop_filter_luma_c(pix, stride, 1, alpha, beta, tc0); | |
2835 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2836 static void h264_h_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2837 { |
2838 h264_loop_filter_luma_c(pix, 1, stride, alpha, beta, tc0); | |
2839 } | |
2840 | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2841 static inline void h264_loop_filter_chroma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0) |
2633 | 2842 { |
2843 int i, d; | |
2844 for( i = 0; i < 4; i++ ) { | |
2845 const int tc = tc0[i]; | |
2846 if( tc <= 0 ) { | |
2847 pix += 2*ystride; | |
2848 continue; | |
2849 } | |
2850 for( d = 0; d < 2; d++ ) { | |
2851 const int p0 = pix[-1*xstride]; | |
2852 const int p1 = pix[-2*xstride]; | |
2853 const int q0 = pix[0]; | |
2854 const int q1 = pix[1*xstride]; | |
2855 | |
4001 | 2856 if( FFABS( p0 - q0 ) < alpha && |
2857 FFABS( p1 - p0 ) < beta && | |
2858 FFABS( q1 - q0 ) < beta ) { | |
2633 | 2859 |
4594 | 2860 int delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
2861 | |
2862 pix[-xstride] = av_clip_uint8( p0 + delta ); /* p0' */ | |
2863 pix[0] = av_clip_uint8( q0 - delta ); /* q0' */ | |
2633 | 2864 } |
2865 pix += ystride; | |
2866 } | |
2867 } | |
2868 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2869 static void h264_v_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2870 { |
2871 h264_loop_filter_chroma_c(pix, stride, 1, alpha, beta, tc0); | |
2872 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2873 static void h264_h_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2874 { |
2875 h264_loop_filter_chroma_c(pix, 1, stride, alpha, beta, tc0); | |
2876 } | |
2877 | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2878 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
|
2879 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2880 int d; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2881 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
|
2882 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
|
2883 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
|
2884 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
|
2885 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
|
2886 |
4001 | 2887 if( FFABS( p0 - q0 ) < alpha && |
2888 FFABS( p1 - p0 ) < beta && | |
2889 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
|
2890 |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2891 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
|
2892 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
|
2893 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2894 pix += ystride; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2895 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2896 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2897 static void h264_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
|
2898 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2899 h264_loop_filter_chroma_intra_c(pix, 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
|
2900 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2901 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
|
2902 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2903 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
|
2904 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2905 |
1708 | 2906 static inline int pix_abs16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2907 { |
2908 int s, i; | |
2909 | |
2910 s = 0; | |
1708 | 2911 for(i=0;i<h;i++) { |
0 | 2912 s += abs(pix1[0] - pix2[0]); |
2913 s += abs(pix1[1] - pix2[1]); | |
2914 s += abs(pix1[2] - pix2[2]); | |
2915 s += abs(pix1[3] - pix2[3]); | |
2916 s += abs(pix1[4] - pix2[4]); | |
2917 s += abs(pix1[5] - pix2[5]); | |
2918 s += abs(pix1[6] - pix2[6]); | |
2919 s += abs(pix1[7] - pix2[7]); | |
2920 s += abs(pix1[8] - pix2[8]); | |
2921 s += abs(pix1[9] - pix2[9]); | |
2922 s += abs(pix1[10] - pix2[10]); | |
2923 s += abs(pix1[11] - pix2[11]); | |
2924 s += abs(pix1[12] - pix2[12]); | |
2925 s += abs(pix1[13] - pix2[13]); | |
2926 s += abs(pix1[14] - pix2[14]); | |
2927 s += abs(pix1[15] - pix2[15]); | |
2928 pix1 += line_size; | |
2929 pix2 += line_size; | |
2930 } | |
2931 return s; | |
2932 } | |
2933 | |
1708 | 2934 static int pix_abs16_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2935 { |
2936 int s, i; | |
2937 | |
2938 s = 0; | |
1708 | 2939 for(i=0;i<h;i++) { |
0 | 2940 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
2941 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
2942 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
2943 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
2944 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
2945 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
2946 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
2947 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
2948 s += abs(pix1[8] - avg2(pix2[8], pix2[9])); | |
2949 s += abs(pix1[9] - avg2(pix2[9], pix2[10])); | |
2950 s += abs(pix1[10] - avg2(pix2[10], pix2[11])); | |
2951 s += abs(pix1[11] - avg2(pix2[11], pix2[12])); | |
2952 s += abs(pix1[12] - avg2(pix2[12], pix2[13])); | |
2953 s += abs(pix1[13] - avg2(pix2[13], pix2[14])); | |
2954 s += abs(pix1[14] - avg2(pix2[14], pix2[15])); | |
2955 s += abs(pix1[15] - avg2(pix2[15], pix2[16])); | |
2956 pix1 += line_size; | |
2957 pix2 += line_size; | |
2958 } | |
2959 return s; | |
2960 } | |
2961 | |
1708 | 2962 static int pix_abs16_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2963 { |
2964 int s, i; | |
1064 | 2965 uint8_t *pix3 = pix2 + line_size; |
0 | 2966 |
2967 s = 0; | |
1708 | 2968 for(i=0;i<h;i++) { |
0 | 2969 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
2970 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
2971 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
2972 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
2973 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
2974 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
2975 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
2976 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
2977 s += abs(pix1[8] - avg2(pix2[8], pix3[8])); | |
2978 s += abs(pix1[9] - avg2(pix2[9], pix3[9])); | |
2979 s += abs(pix1[10] - avg2(pix2[10], pix3[10])); | |
2980 s += abs(pix1[11] - avg2(pix2[11], pix3[11])); | |
2981 s += abs(pix1[12] - avg2(pix2[12], pix3[12])); | |
2982 s += abs(pix1[13] - avg2(pix2[13], pix3[13])); | |
2983 s += abs(pix1[14] - avg2(pix2[14], pix3[14])); | |
2984 s += abs(pix1[15] - avg2(pix2[15], pix3[15])); | |
2985 pix1 += line_size; | |
2986 pix2 += line_size; | |
2987 pix3 += line_size; | |
2988 } | |
2989 return s; | |
2990 } | |
2991 | |
1708 | 2992 static int pix_abs16_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2993 { |
2994 int s, i; | |
1064 | 2995 uint8_t *pix3 = pix2 + line_size; |
0 | 2996 |
2997 s = 0; | |
1708 | 2998 for(i=0;i<h;i++) { |
0 | 2999 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
3000 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
3001 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
3002 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
3003 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
3004 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
3005 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
3006 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
3007 s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9])); | |
3008 s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10])); | |
3009 s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11])); | |
3010 s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12])); | |
3011 s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13])); | |
3012 s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14])); | |
3013 s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15])); | |
3014 s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16])); | |
3015 pix1 += line_size; | |
3016 pix2 += line_size; | |
3017 pix3 += line_size; | |
3018 } | |
3019 return s; | |
3020 } | |
3021 | |
1708 | 3022 static inline int pix_abs8_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3023 { |
3024 int s, i; | |
3025 | |
3026 s = 0; | |
1708 | 3027 for(i=0;i<h;i++) { |
294 | 3028 s += abs(pix1[0] - pix2[0]); |
3029 s += abs(pix1[1] - pix2[1]); | |
3030 s += abs(pix1[2] - pix2[2]); | |
3031 s += abs(pix1[3] - pix2[3]); | |
3032 s += abs(pix1[4] - pix2[4]); | |
3033 s += abs(pix1[5] - pix2[5]); | |
3034 s += abs(pix1[6] - pix2[6]); | |
3035 s += abs(pix1[7] - pix2[7]); | |
3036 pix1 += line_size; | |
3037 pix2 += line_size; | |
3038 } | |
3039 return s; | |
3040 } | |
3041 | |
1708 | 3042 static int pix_abs8_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3043 { |
3044 int s, i; | |
3045 | |
3046 s = 0; | |
1708 | 3047 for(i=0;i<h;i++) { |
294 | 3048 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
3049 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
3050 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
3051 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
3052 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
3053 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
3054 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
3055 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
3056 pix1 += line_size; | |
3057 pix2 += line_size; | |
3058 } | |
3059 return s; | |
3060 } | |
3061 | |
1708 | 3062 static int pix_abs8_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3063 { |
3064 int s, i; | |
1064 | 3065 uint8_t *pix3 = pix2 + line_size; |
294 | 3066 |
3067 s = 0; | |
1708 | 3068 for(i=0;i<h;i++) { |
294 | 3069 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
3070 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
3071 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
3072 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
3073 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
3074 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
3075 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
3076 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
3077 pix1 += line_size; | |
3078 pix2 += line_size; | |
3079 pix3 += line_size; | |
3080 } | |
3081 return s; | |
3082 } | |
3083 | |
1708 | 3084 static int pix_abs8_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3085 { |
3086 int s, i; | |
1064 | 3087 uint8_t *pix3 = pix2 + line_size; |
294 | 3088 |
3089 s = 0; | |
1708 | 3090 for(i=0;i<h;i++) { |
294 | 3091 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
3092 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
3093 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
3094 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
3095 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
3096 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
3097 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
3098 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
3099 pix1 += line_size; | |
3100 pix2 += line_size; | |
3101 pix3 += line_size; | |
3102 } | |
3103 return s; | |
3104 } | |
3105 | |
2834 | 3106 static int nsse16_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3107 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3108 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3109 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3110 int x,y; |
2066 | 3111 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3112 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3113 for(x=0; x<16; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3114 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3115 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3116 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3117 for(x=0; x<15; x++){ |
4001 | 3118 score2+= FFABS( s1[x ] - s1[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3119 - s1[x+1] + s1[x+1+stride]) |
4001 | 3120 -FFABS( s2[x ] - s2[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3121 - s2[x+1] + s2[x+1+stride]); |
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 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3124 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3125 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3126 } |
2066 | 3127 |
4001 | 3128 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight; |
3129 else return score1 + FFABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3130 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3131 |
2834 | 3132 static int nsse8_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3133 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3134 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3135 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3136 int x,y; |
2967 | 3137 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3138 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3139 for(x=0; x<8; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3140 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3141 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3142 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3143 for(x=0; x<7; x++){ |
4001 | 3144 score2+= FFABS( s1[x ] - s1[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3145 - s1[x+1] + s1[x+1+stride]) |
4001 | 3146 -FFABS( s2[x ] - s2[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3147 - s2[x+1] + s2[x+1+stride]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3148 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3149 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3150 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3151 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3152 } |
2967 | 3153 |
4001 | 3154 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight; |
3155 else return score1 + FFABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3156 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3157 |
1784 | 3158 static int try_8x8basis_c(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale){ |
3159 int i; | |
3160 unsigned int sum=0; | |
3161 | |
3162 for(i=0; i<8*8; i++){ | |
3163 int b= rem[i] + ((basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT)); | |
3164 int w= weight[i]; | |
3165 b>>= RECON_SHIFT; | |
3166 assert(-512<b && b<512); | |
3167 | |
3168 sum += (w*b)*(w*b)>>4; | |
3169 } | |
3170 return sum>>2; | |
3171 } | |
3172 | |
3173 static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale){ | |
3174 int i; | |
3175 | |
3176 for(i=0; i<8*8; i++){ | |
3177 rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT); | |
2967 | 3178 } |
1784 | 3179 } |
3180 | |
1100 | 3181 /** |
3182 * permutes an 8x8 block. | |
1101 | 3183 * @param block the block which will be permuted according to the given permutation vector |
1100 | 3184 * @param permutation the permutation vector |
3185 * @param last the last non zero coefficient in scantable order, used to speed the permutation up | |
2967 | 3186 * @param scantable the used scantable, this is only used to speed the permutation up, the block is not |
1101 | 3187 * (inverse) permutated to scantable order! |
1100 | 3188 */ |
1064 | 3189 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
|
3190 { |
764 | 3191 int i; |
945 | 3192 DCTELEM temp[64]; |
2967 | 3193 |
764 | 3194 if(last<=0) return; |
5129 | 3195 //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
|
3196 |
764 | 3197 for(i=0; i<=last; i++){ |
3198 const int j= scantable[i]; | |
3199 temp[j]= block[j]; | |
3200 block[j]=0; | |
3201 } | |
2967 | 3202 |
764 | 3203 for(i=0; i<=last; i++){ |
3204 const int j= scantable[i]; | |
3205 const int perm_j= permutation[j]; | |
3206 block[perm_j]= temp[j]; | |
3207 } | |
174
ac5075a55488
new IDCT code by Michael Niedermayer (michaelni@gmx.at) - #define SIMPLE_IDCT to enable
arpi_esp
parents:
88
diff
changeset
|
3208 } |
34 | 3209 |
1729 | 3210 static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride, int h){ |
3211 return 0; | |
3212 } | |
3213 | |
3214 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type){ | |
3215 int i; | |
2967 | 3216 |
1729 | 3217 memset(cmp, 0, sizeof(void*)*5); |
2967 | 3218 |
1729 | 3219 for(i=0; i<5; i++){ |
3220 switch(type&0xFF){ | |
3221 case FF_CMP_SAD: | |
3222 cmp[i]= c->sad[i]; | |
3223 break; | |
3224 case FF_CMP_SATD: | |
3225 cmp[i]= c->hadamard8_diff[i]; | |
3226 break; | |
3227 case FF_CMP_SSE: | |
3228 cmp[i]= c->sse[i]; | |
3229 break; | |
3230 case FF_CMP_DCT: | |
3231 cmp[i]= c->dct_sad[i]; | |
3232 break; | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3233 case FF_CMP_DCT264: |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3234 cmp[i]= c->dct264_sad[i]; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3235 break; |
2382 | 3236 case FF_CMP_DCTMAX: |
3237 cmp[i]= c->dct_max[i]; | |
3238 break; | |
1729 | 3239 case FF_CMP_PSNR: |
3240 cmp[i]= c->quant_psnr[i]; | |
3241 break; | |
3242 case FF_CMP_BIT: | |
3243 cmp[i]= c->bit[i]; | |
3244 break; | |
3245 case FF_CMP_RD: | |
3246 cmp[i]= c->rd[i]; | |
3247 break; | |
3248 case FF_CMP_VSAD: | |
3249 cmp[i]= c->vsad[i]; | |
3250 break; | |
3251 case FF_CMP_VSSE: | |
3252 cmp[i]= c->vsse[i]; | |
3253 break; | |
3254 case FF_CMP_ZERO: | |
3255 cmp[i]= zero_cmp; | |
3256 break; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3257 case FF_CMP_NSSE: |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3258 cmp[i]= c->nsse[i]; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3259 break; |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
3260 #ifdef CONFIG_SNOW_ENCODER |
2184 | 3261 case FF_CMP_W53: |
3262 cmp[i]= c->w53[i]; | |
3263 break; | |
3264 case FF_CMP_W97: | |
3265 cmp[i]= c->w97[i]; | |
3266 break; | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
3267 #endif |
1729 | 3268 default: |
3269 av_log(NULL, AV_LOG_ERROR,"internal error in cmp function selection\n"); | |
3270 } | |
3271 } | |
3272 } | |
3273 | |
1101 | 3274 /** |
3275 * memset(blocks, 0, sizeof(DCTELEM)*6*64) | |
3276 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
3277 static void clear_blocks_c(DCTELEM *blocks) |
296 | 3278 { |
3279 memset(blocks, 0, sizeof(DCTELEM)*6*64); | |
3280 } | |
3281 | |
866 | 3282 static void add_bytes_c(uint8_t *dst, uint8_t *src, int w){ |
6385 | 3283 long i; |
3284 for(i=0; i<=w-sizeof(long); i+=sizeof(long)){ | |
3285 long a = *(long*)(src+i); | |
3286 long b = *(long*)(dst+i); | |
3287 *(long*)(dst+i) = ((a&pb_7f) + (b&pb_7f)) ^ ((a^b)&pb_80); | |
866 | 3288 } |
3289 for(; i<w; i++) | |
3290 dst[i+0] += src[i+0]; | |
3291 } | |
3292 | |
6384 | 3293 static void add_bytes_l2_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ |
6385 | 3294 long i; |
6384 | 3295 for(i=0; i<=w-sizeof(long); i+=sizeof(long)){ |
3296 long a = *(long*)(src1+i); | |
3297 long b = *(long*)(src2+i); | |
6385 | 3298 *(long*)(dst+i) = ((a&pb_7f) + (b&pb_7f)) ^ ((a^b)&pb_80); |
6384 | 3299 } |
3300 for(; i<w; i++) | |
3301 dst[i] = src1[i]+src2[i]; | |
3302 } | |
3303 | |
866 | 3304 static void diff_bytes_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ |
6385 | 3305 long i; |
3306 #ifndef HAVE_FAST_UNALIGNED | |
3307 if((long)src2 & (sizeof(long)-1)){ | |
6386 | 3308 for(i=0; i+7<w; i+=8){ |
3309 dst[i+0] = src1[i+0]-src2[i+0]; | |
3310 dst[i+1] = src1[i+1]-src2[i+1]; | |
3311 dst[i+2] = src1[i+2]-src2[i+2]; | |
3312 dst[i+3] = src1[i+3]-src2[i+3]; | |
3313 dst[i+4] = src1[i+4]-src2[i+4]; | |
3314 dst[i+5] = src1[i+5]-src2[i+5]; | |
3315 dst[i+6] = src1[i+6]-src2[i+6]; | |
3316 dst[i+7] = src1[i+7]-src2[i+7]; | |
3317 } | |
6385 | 3318 }else |
3319 #endif | |
3320 for(i=0; i<=w-sizeof(long); i+=sizeof(long)){ | |
3321 long a = *(long*)(src1+i); | |
3322 long b = *(long*)(src2+i); | |
3323 *(long*)(dst+i) = ((a|pb_80) - (b&pb_7f)) ^ ((a^b^pb_80)&pb_80); | |
3324 } | |
866 | 3325 for(; i<w; i++) |
3326 dst[i+0] = src1[i+0]-src2[i+0]; | |
3327 } | |
3328 | |
1527 | 3329 static void sub_hfyu_median_prediction_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){ |
3330 int i; | |
3331 uint8_t l, lt; | |
3332 | |
3333 l= *left; | |
3334 lt= *left_top; | |
3335 | |
3336 for(i=0; i<w; i++){ | |
3337 const int pred= mid_pred(l, src1[i], (l + src1[i] - lt)&0xFF); | |
3338 lt= src1[i]; | |
3339 l= src2[i]; | |
3340 dst[i]= l - pred; | |
2967 | 3341 } |
1527 | 3342 |
3343 *left= l; | |
3344 *left_top= lt; | |
3345 } | |
3346 | |
936 | 3347 #define BUTTERFLY2(o1,o2,i1,i2) \ |
3348 o1= (i1)+(i2);\ | |
3349 o2= (i1)-(i2); | |
3350 | |
3351 #define BUTTERFLY1(x,y) \ | |
3352 {\ | |
3353 int a,b;\ | |
3354 a= x;\ | |
3355 b= y;\ | |
3356 x= a+b;\ | |
3357 y= a-b;\ | |
3358 } | |
3359 | |
4001 | 3360 #define BUTTERFLYA(x,y) (FFABS((x)+(y)) + FFABS((x)-(y))) |
936 | 3361 |
1708 | 3362 static int hadamard8_diff8x8_c(/*MpegEncContext*/ void *s, uint8_t *dst, uint8_t *src, int stride, int h){ |
936 | 3363 int i; |
3364 int temp[64]; | |
3365 int sum=0; | |
2967 | 3366 |
1708 | 3367 assert(h==8); |
936 | 3368 |
3369 for(i=0; i<8; i++){ | |
3370 //FIXME try pointer walks | |
3371 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]); | |
3372 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]); | |
3373 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]); | |
3374 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 | 3375 |
936 | 3376 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3377 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3378 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3379 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3380 |
936 | 3381 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3382 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3383 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3384 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3385 } | |
3386 | |
3387 for(i=0; i<8; i++){ | |
3388 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3389 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3390 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3391 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3392 |
936 | 3393 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3394 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3395 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3396 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
3397 | |
2967 | 3398 sum += |
936 | 3399 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3400 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3401 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3402 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3403 } | |
3404 #if 0 | |
3405 static int maxi=0; | |
3406 if(sum>maxi){ | |
3407 maxi=sum; | |
3408 printf("MAX:%d\n", maxi); | |
3409 } | |
3410 #endif | |
3411 return sum; | |
3412 } | |
3413 | |
1729 | 3414 static int hadamard8_intra8x8_c(/*MpegEncContext*/ void *s, uint8_t *src, uint8_t *dummy, int stride, int h){ |
936 | 3415 int i; |
3416 int temp[64]; | |
3417 int sum=0; | |
2967 | 3418 |
1729 | 3419 assert(h==8); |
2967 | 3420 |
936 | 3421 for(i=0; i<8; i++){ |
3422 //FIXME try pointer walks | |
1729 | 3423 BUTTERFLY2(temp[8*i+0], temp[8*i+1], src[stride*i+0],src[stride*i+1]); |
3424 BUTTERFLY2(temp[8*i+2], temp[8*i+3], src[stride*i+2],src[stride*i+3]); | |
3425 BUTTERFLY2(temp[8*i+4], temp[8*i+5], src[stride*i+4],src[stride*i+5]); | |
3426 BUTTERFLY2(temp[8*i+6], temp[8*i+7], src[stride*i+6],src[stride*i+7]); | |
2967 | 3427 |
936 | 3428 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3429 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3430 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3431 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3432 |
936 | 3433 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3434 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3435 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3436 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3437 } | |
3438 | |
3439 for(i=0; i<8; i++){ | |
3440 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3441 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3442 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3443 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3444 |
936 | 3445 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3446 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3447 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3448 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
2967 | 3449 |
3450 sum += | |
936 | 3451 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3452 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3453 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3454 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3455 } | |
2967 | 3456 |
4001 | 3457 sum -= FFABS(temp[8*0] + temp[8*4]); // -mean |
2967 | 3458 |
936 | 3459 return sum; |
3460 } | |
3461 | |
1708 | 3462 static int dct_sad8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3463 MpegEncContext * const s= (MpegEncContext *)c; |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
3464 DECLARE_ALIGNED_16(uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
1016 | 3465 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
2967 | 3466 |
1708 | 3467 assert(h==8); |
936 | 3468 |
3469 s->dsp.diff_pixels(temp, src1, src2, stride); | |
1092 | 3470 s->dsp.fdct(temp); |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
3471 return s->dsp.sum_abs_dctelem(temp); |
936 | 3472 } |
3473 | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3474 #ifdef CONFIG_GPL |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3475 #define DCT8_1D {\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3476 const int s07 = SRC(0) + SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3477 const int s16 = SRC(1) + SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3478 const int s25 = SRC(2) + SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3479 const int s34 = SRC(3) + SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3480 const int a0 = s07 + s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3481 const int a1 = s16 + s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3482 const int a2 = s07 - s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3483 const int a3 = s16 - s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3484 const int d07 = SRC(0) - SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3485 const int d16 = SRC(1) - SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3486 const int d25 = SRC(2) - SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3487 const int d34 = SRC(3) - SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3488 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
|
3489 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
|
3490 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
|
3491 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
|
3492 DST(0, a0 + a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3493 DST(1, a4 + (a7>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3494 DST(2, a2 + (a3>>1)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3495 DST(3, a5 + (a6>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3496 DST(4, a0 - a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3497 DST(5, a6 - (a5>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3498 DST(6, (a2>>1) - a3 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3499 DST(7, (a4>>2) - a7 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3500 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3501 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3502 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
|
3503 MpegEncContext * const s= (MpegEncContext *)c; |
5256 | 3504 DCTELEM dct[8][8]; |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3505 int i; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3506 int sum=0; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3507 |
5256 | 3508 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
|
3509 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3510 #define SRC(x) dct[i][x] |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3511 #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
|
3512 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3513 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3514 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3515 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3516 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3517 #define SRC(x) dct[x][i] |
4001 | 3518 #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
|
3519 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3520 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3521 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3522 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3523 return sum; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3524 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3525 #endif |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3526 |
2382 | 3527 static int dct_max8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
3528 MpegEncContext * const s= (MpegEncContext *)c; | |
3089 | 3529 DECLARE_ALIGNED_8(uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
2382 | 3530 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3531 int sum=0, i; | |
2967 | 3532 |
2382 | 3533 assert(h==8); |
3534 | |
3535 s->dsp.diff_pixels(temp, src1, src2, stride); | |
3536 s->dsp.fdct(temp); | |
3537 | |
3538 for(i=0; i<64; i++) | |
4001 | 3539 sum= FFMAX(sum, FFABS(temp[i])); |
2967 | 3540 |
2382 | 3541 return sum; |
3542 } | |
3543 | |
1708 | 3544 static int quant_psnr8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3545 MpegEncContext * const s= (MpegEncContext *)c; |
3089 | 3546 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64*2/8]); |
1016 | 3547 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3548 DCTELEM * const bak = ((DCTELEM*)aligned_temp)+64; | |
936 | 3549 int sum=0, i; |
3550 | |
1708 | 3551 assert(h==8); |
936 | 3552 s->mb_intra=0; |
2967 | 3553 |
936 | 3554 s->dsp.diff_pixels(temp, src1, src2, stride); |
2967 | 3555 |
936 | 3556 memcpy(bak, temp, 64*sizeof(DCTELEM)); |
2967 | 3557 |
1013 | 3558 s->block_last_index[0/*FIXME*/]= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
1689 | 3559 s->dct_unquantize_inter(s, temp, 0, s->qscale); |
6001 | 3560 ff_simple_idct(temp); //FIXME |
2967 | 3561 |
936 | 3562 for(i=0; i<64; i++) |
3563 sum+= (temp[i]-bak[i])*(temp[i]-bak[i]); | |
2967 | 3564 |
936 | 3565 return sum; |
3566 } | |
3567 | |
1708 | 3568 static int rd8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3569 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3570 const uint8_t *scantable= s->intra_scantable.permutated; |
3089 | 3571 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
3572 DECLARE_ALIGNED_8 (uint64_t, aligned_bak[stride]); | |
1016 | 3573 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3574 uint8_t * const bak= (uint8_t*)aligned_bak; | |
1007 | 3575 int i, last, run, bits, level, distoration, start_i; |
3576 const int esc_length= s->ac_esc_length; | |
3577 uint8_t * length; | |
3578 uint8_t * last_length; | |
2967 | 3579 |
1708 | 3580 assert(h==8); |
3581 | |
1007 | 3582 for(i=0; i<8; i++){ |
3583 ((uint32_t*)(bak + i*stride))[0]= ((uint32_t*)(src2 + i*stride))[0]; | |
3584 ((uint32_t*)(bak + i*stride))[1]= ((uint32_t*)(src2 + i*stride))[1]; | |
3585 } | |
3586 | |
3587 s->dsp.diff_pixels(temp, src1, src2, stride); | |
3588 | |
1013 | 3589 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3590 | |
3591 bits=0; | |
2967 | 3592 |
1013 | 3593 if (s->mb_intra) { |
2967 | 3594 start_i = 1; |
1013 | 3595 length = s->intra_ac_vlc_length; |
3596 last_length= s->intra_ac_vlc_last_length; | |
3597 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma | |
3598 } else { | |
3599 start_i = 0; | |
3600 length = s->inter_ac_vlc_length; | |
3601 last_length= s->inter_ac_vlc_last_length; | |
3602 } | |
2967 | 3603 |
1013 | 3604 if(last>=start_i){ |
1007 | 3605 run=0; |
3606 for(i=start_i; i<last; i++){ | |
3607 int j= scantable[i]; | |
3608 level= temp[j]; | |
2967 | 3609 |
1007 | 3610 if(level){ |
3611 level+=64; | |
3612 if((level&(~127)) == 0){ | |
3613 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3614 }else | |
3615 bits+= esc_length; | |
3616 run=0; | |
3617 }else | |
3618 run++; | |
3619 } | |
3620 i= scantable[last]; | |
2967 | 3621 |
1011 | 3622 level= temp[i] + 64; |
3623 | |
3624 assert(level - 64); | |
2967 | 3625 |
1007 | 3626 if((level&(~127)) == 0){ |
3627 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3628 }else | |
3629 bits+= esc_length; | |
2967 | 3630 |
1013 | 3631 } |
3632 | |
3633 if(last>=0){ | |
1689 | 3634 if(s->mb_intra) |
3635 s->dct_unquantize_intra(s, temp, 0, s->qscale); | |
3636 else | |
3637 s->dct_unquantize_inter(s, temp, 0, s->qscale); | |
1007 | 3638 } |
2967 | 3639 |
1092 | 3640 s->dsp.idct_add(bak, stride, temp); |
2967 | 3641 |
1708 | 3642 distoration= s->dsp.sse[1](NULL, bak, src1, stride, 8); |
1007 | 3643 |
1013 | 3644 return distoration + ((bits*s->qscale*s->qscale*109 + 64)>>7); |
1007 | 3645 } |
3646 | |
1708 | 3647 static int bit8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3648 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3649 const uint8_t *scantable= s->intra_scantable.permutated; |
3089 | 3650 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
1016 | 3651 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
1007 | 3652 int i, last, run, bits, level, start_i; |
3653 const int esc_length= s->ac_esc_length; | |
3654 uint8_t * length; | |
3655 uint8_t * last_length; | |
1708 | 3656 |
3657 assert(h==8); | |
2967 | 3658 |
1013 | 3659 s->dsp.diff_pixels(temp, src1, src2, stride); |
1007 | 3660 |
1013 | 3661 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3662 | |
3663 bits=0; | |
2967 | 3664 |
1007 | 3665 if (s->mb_intra) { |
2967 | 3666 start_i = 1; |
1007 | 3667 length = s->intra_ac_vlc_length; |
3668 last_length= s->intra_ac_vlc_last_length; | |
1013 | 3669 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma |
1007 | 3670 } else { |
3671 start_i = 0; | |
3672 length = s->inter_ac_vlc_length; | |
3673 last_length= s->inter_ac_vlc_last_length; | |
3674 } | |
2967 | 3675 |
1013 | 3676 if(last>=start_i){ |
1007 | 3677 run=0; |
3678 for(i=start_i; i<last; i++){ | |
3679 int j= scantable[i]; | |
3680 level= temp[j]; | |
2967 | 3681 |
1007 | 3682 if(level){ |
3683 level+=64; | |
3684 if((level&(~127)) == 0){ | |
3685 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3686 }else | |
3687 bits+= esc_length; | |
3688 run=0; | |
3689 }else | |
3690 run++; | |
3691 } | |
3692 i= scantable[last]; | |
2967 | 3693 |
1013 | 3694 level= temp[i] + 64; |
2967 | 3695 |
1013 | 3696 assert(level - 64); |
2967 | 3697 |
1007 | 3698 if((level&(~127)) == 0){ |
3699 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3700 }else | |
3701 bits+= esc_length; | |
3702 } | |
3703 | |
3704 return bits; | |
3705 } | |
3706 | |
1729 | 3707 static int vsad_intra16_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ |
3708 int score=0; | |
3709 int x,y; | |
2967 | 3710 |
1729 | 3711 for(y=1; y<h; y++){ |
3712 for(x=0; x<16; x+=4){ | |
4001 | 3713 score+= FFABS(s[x ] - s[x +stride]) + FFABS(s[x+1] - s[x+1+stride]) |
3714 +FFABS(s[x+2] - s[x+2+stride]) + FFABS(s[x+3] - s[x+3+stride]); | |
1729 | 3715 } |
3716 s+= stride; | |
3717 } | |
2967 | 3718 |
1729 | 3719 return score; |
3720 } | |
3721 | |
3722 static int vsad16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3723 int score=0; | |
3724 int x,y; | |
2967 | 3725 |
1729 | 3726 for(y=1; y<h; y++){ |
3727 for(x=0; x<16; x++){ | |
4001 | 3728 score+= FFABS(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); |
1729 | 3729 } |
3730 s1+= stride; | |
3731 s2+= stride; | |
3732 } | |
2967 | 3733 |
1729 | 3734 return score; |
3735 } | |
3736 | |
3737 #define SQ(a) ((a)*(a)) | |
3738 static int vsse_intra16_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ | |
3739 int score=0; | |
3740 int x,y; | |
2967 | 3741 |
1729 | 3742 for(y=1; y<h; y++){ |
3743 for(x=0; x<16; x+=4){ | |
2967 | 3744 score+= SQ(s[x ] - s[x +stride]) + SQ(s[x+1] - s[x+1+stride]) |
1729 | 3745 +SQ(s[x+2] - s[x+2+stride]) + SQ(s[x+3] - s[x+3+stride]); |
3746 } | |
3747 s+= stride; | |
3748 } | |
2967 | 3749 |
1729 | 3750 return score; |
3751 } | |
3752 | |
3753 static int vsse16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3754 int score=0; | |
3755 int x,y; | |
2967 | 3756 |
1729 | 3757 for(y=1; y<h; y++){ |
3758 for(x=0; x<16; x++){ | |
3759 score+= SQ(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); | |
3760 } | |
3761 s1+= stride; | |
3762 s2+= stride; | |
3763 } | |
2967 | 3764 |
1729 | 3765 return score; |
3766 } | |
3767 | |
5255 | 3768 static int ssd_int8_vs_int16_c(const int8_t *pix1, const int16_t *pix2, |
3769 int size){ | |
4749 | 3770 int score=0; |
3771 int i; | |
3772 for(i=0; i<size; i++) | |
3773 score += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]); | |
3774 return score; | |
3775 } | |
3776 | |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3777 WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c) |
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3778 WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c) |
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3779 WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c) |
3013 | 3780 #ifdef CONFIG_GPL |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3781 WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c) |
3013 | 3782 #endif |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3783 WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c) |
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3784 WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c) |
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3785 WRAPPER8_16_SQ(rd8x8_c, rd16_c) |
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3786 WRAPPER8_16_SQ(bit8x8_c, bit16_c) |
936 | 3787 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3788 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
|
3789 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3790 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3791 dst[i] *= src[i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3792 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3793 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3794 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
|
3795 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3796 src1 += len-1; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3797 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3798 dst[i] = src0[i] * src1[-i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3799 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3800 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3801 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
|
3802 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3803 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3804 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
|
3805 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3806 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3807 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
|
3808 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3809 for(i=0; i<len; i++) { |
6218 | 3810 int_fast32_t tmp = ((const int32_t*)src)[i]; |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3811 if(tmp & 0xf0000){ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3812 tmp = (0x43c0ffff - tmp)>>31; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3813 // 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
|
3814 // if(tmp > 0x43c0ffff) tmp = 0xFFFF; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3815 // else tmp = 0; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3816 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3817 dst[i] = tmp - 0x8000; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3818 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3819 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3820 |
5887 | 3821 #define W0 2048 |
3822 #define W1 2841 /* 2048*sqrt (2)*cos (1*pi/16) */ | |
3823 #define W2 2676 /* 2048*sqrt (2)*cos (2*pi/16) */ | |
3824 #define W3 2408 /* 2048*sqrt (2)*cos (3*pi/16) */ | |
3825 #define W4 2048 /* 2048*sqrt (2)*cos (4*pi/16) */ | |
3826 #define W5 1609 /* 2048*sqrt (2)*cos (5*pi/16) */ | |
3827 #define W6 1108 /* 2048*sqrt (2)*cos (6*pi/16) */ | |
3828 #define W7 565 /* 2048*sqrt (2)*cos (7*pi/16) */ | |
3829 | |
3830 static void wmv2_idct_row(short * b) | |
3831 { | |
3832 int s1,s2; | |
3833 int a0,a1,a2,a3,a4,a5,a6,a7; | |
3834 /*step 1*/ | |
3835 a1 = W1*b[1]+W7*b[7]; | |
3836 a7 = W7*b[1]-W1*b[7]; | |
3837 a5 = W5*b[5]+W3*b[3]; | |
3838 a3 = W3*b[5]-W5*b[3]; | |
3839 a2 = W2*b[2]+W6*b[6]; | |
3840 a6 = W6*b[2]-W2*b[6]; | |
3841 a0 = W0*b[0]+W0*b[4]; | |
3842 a4 = W0*b[0]-W0*b[4]; | |
3843 /*step 2*/ | |
3844 s1 = (181*(a1-a5+a7-a3)+128)>>8;//1,3,5,7, | |
3845 s2 = (181*(a1-a5-a7+a3)+128)>>8; | |
3846 /*step 3*/ | |
3847 b[0] = (a0+a2+a1+a5 + (1<<7))>>8; | |
3848 b[1] = (a4+a6 +s1 + (1<<7))>>8; | |
3849 b[2] = (a4-a6 +s2 + (1<<7))>>8; | |
3850 b[3] = (a0-a2+a7+a3 + (1<<7))>>8; | |
3851 b[4] = (a0-a2-a7-a3 + (1<<7))>>8; | |
3852 b[5] = (a4-a6 -s2 + (1<<7))>>8; | |
3853 b[6] = (a4+a6 -s1 + (1<<7))>>8; | |
3854 b[7] = (a0+a2-a1-a5 + (1<<7))>>8; | |
3855 } | |
3856 static void wmv2_idct_col(short * b) | |
3857 { | |
3858 int s1,s2; | |
3859 int a0,a1,a2,a3,a4,a5,a6,a7; | |
3860 /*step 1, with extended precision*/ | |
3861 a1 = (W1*b[8*1]+W7*b[8*7] + 4)>>3; | |
3862 a7 = (W7*b[8*1]-W1*b[8*7] + 4)>>3; | |
3863 a5 = (W5*b[8*5]+W3*b[8*3] + 4)>>3; | |
3864 a3 = (W3*b[8*5]-W5*b[8*3] + 4)>>3; | |
3865 a2 = (W2*b[8*2]+W6*b[8*6] + 4)>>3; | |
3866 a6 = (W6*b[8*2]-W2*b[8*6] + 4)>>3; | |
3867 a0 = (W0*b[8*0]+W0*b[8*4] )>>3; | |
3868 a4 = (W0*b[8*0]-W0*b[8*4] )>>3; | |
3869 /*step 2*/ | |
3870 s1 = (181*(a1-a5+a7-a3)+128)>>8; | |
3871 s2 = (181*(a1-a5-a7+a3)+128)>>8; | |
3872 /*step 3*/ | |
3873 b[8*0] = (a0+a2+a1+a5 + (1<<13))>>14; | |
3874 b[8*1] = (a4+a6 +s1 + (1<<13))>>14; | |
3875 b[8*2] = (a4-a6 +s2 + (1<<13))>>14; | |
3876 b[8*3] = (a0-a2+a7+a3 + (1<<13))>>14; | |
3877 | |
3878 b[8*4] = (a0-a2-a7-a3 + (1<<13))>>14; | |
3879 b[8*5] = (a4-a6 -s2 + (1<<13))>>14; | |
3880 b[8*6] = (a4+a6 -s1 + (1<<13))>>14; | |
3881 b[8*7] = (a0+a2-a1-a5 + (1<<13))>>14; | |
3882 } | |
3883 void ff_wmv2_idct_c(short * block){ | |
3884 int i; | |
3885 | |
3886 for(i=0;i<64;i+=8){ | |
3887 wmv2_idct_row(block+i); | |
3888 } | |
3889 for(i=0;i<8;i++){ | |
3890 wmv2_idct_col(block+i); | |
3891 } | |
3892 } | |
1092 | 3893 /* XXX: those functions should be suppressed ASAP when all IDCTs are |
3894 converted */ | |
5887 | 3895 static void ff_wmv2_idct_put_c(uint8_t *dest, int line_size, DCTELEM *block) |
3896 { | |
3897 ff_wmv2_idct_c(block); | |
3898 put_pixels_clamped_c(block, dest, line_size); | |
3899 } | |
3900 static void ff_wmv2_idct_add_c(uint8_t *dest, int line_size, DCTELEM *block) | |
3901 { | |
3902 ff_wmv2_idct_c(block); | |
3903 add_pixels_clamped_c(block, dest, line_size); | |
3904 } | |
1092 | 3905 static void ff_jref_idct_put(uint8_t *dest, int line_size, DCTELEM *block) |
3906 { | |
3907 j_rev_dct (block); | |
3908 put_pixels_clamped_c(block, dest, line_size); | |
3909 } | |
3910 static void ff_jref_idct_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3911 { | |
3912 j_rev_dct (block); | |
3913 add_pixels_clamped_c(block, dest, line_size); | |
3914 } | |
3915 | |
2256 | 3916 static void ff_jref_idct4_put(uint8_t *dest, int line_size, DCTELEM *block) |
3917 { | |
3918 j_rev_dct4 (block); | |
3919 put_pixels_clamped4_c(block, dest, line_size); | |
3920 } | |
3921 static void ff_jref_idct4_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3922 { | |
3923 j_rev_dct4 (block); | |
3924 add_pixels_clamped4_c(block, dest, line_size); | |
3925 } | |
3926 | |
2257 | 3927 static void ff_jref_idct2_put(uint8_t *dest, int line_size, DCTELEM *block) |
3928 { | |
3929 j_rev_dct2 (block); | |
3930 put_pixels_clamped2_c(block, dest, line_size); | |
3931 } | |
3932 static void ff_jref_idct2_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3933 { | |
3934 j_rev_dct2 (block); | |
3935 add_pixels_clamped2_c(block, dest, line_size); | |
3936 } | |
3937 | |
2259 | 3938 static void ff_jref_idct1_put(uint8_t *dest, int line_size, DCTELEM *block) |
3939 { | |
4176 | 3940 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2259 | 3941 |
3942 dest[0] = cm[(block[0] + 4)>>3]; | |
3943 } | |
3944 static void ff_jref_idct1_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3945 { | |
4176 | 3946 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2259 | 3947 |
3948 dest[0] = cm[dest[0] + ((block[0] + 4)>>3)]; | |
3949 } | |
3950 | |
5143 | 3951 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
|
3952 |
1201 | 3953 /* init static data */ |
4197 | 3954 void dsputil_static_init(void) |
0 | 3955 { |
751 | 3956 int i; |
0 | 3957 |
4176 | 3958 for(i=0;i<256;i++) ff_cropTbl[i + MAX_NEG_CROP] = i; |
1201 | 3959 for(i=0;i<MAX_NEG_CROP;i++) { |
4176 | 3960 ff_cropTbl[i] = 0; |
3961 ff_cropTbl[i + MAX_NEG_CROP + 256] = 255; | |
1201 | 3962 } |
2967 | 3963 |
1201 | 3964 for(i=0;i<512;i++) { |
4179 | 3965 ff_squareTbl[i] = (i - 256) * (i - 256); |
1201 | 3966 } |
2967 | 3967 |
4197 | 3968 for(i=0; i<64; i++) inv_zigzag_direct16[ff_zigzag_direct[i]]= i+1; |
1201 | 3969 } |
0 | 3970 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3971 int ff_check_alignment(void){ |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3972 static int did_fail=0; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3973 DECLARE_ALIGNED_16(int, aligned); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3974 |
5150 | 3975 if((long)&aligned & 15){ |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3976 if(!did_fail){ |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3977 #if defined(HAVE_MMX) || defined(HAVE_ALTIVEC) |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3978 av_log(NULL, AV_LOG_ERROR, |
4292 | 3979 "Compiler did not align stack variables. Libavcodec has been miscompiled\n" |
3980 "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
|
3981 "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
|
3982 "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
|
3983 #endif |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3984 did_fail=1; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3985 } |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3986 return -1; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3987 } |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3988 return 0; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3989 } |
861 | 3990 |
1201 | 3991 void dsputil_init(DSPContext* c, AVCodecContext *avctx) |
3992 { | |
3993 int i; | |
0 | 3994 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3995 ff_check_alignment(); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
3996 |
1092 | 3997 #ifdef CONFIG_ENCODERS |
1567 | 3998 if(avctx->dct_algo==FF_DCT_FASTINT) { |
1092 | 3999 c->fdct = fdct_ifast; |
2979 | 4000 c->fdct248 = fdct_ifast248; |
2967 | 4001 } |
1567 | 4002 else if(avctx->dct_algo==FF_DCT_FAAN) { |
1557 | 4003 c->fdct = ff_faandct; |
2979 | 4004 c->fdct248 = ff_faandct248; |
2967 | 4005 } |
1567 | 4006 else { |
1092 | 4007 c->fdct = ff_jpeg_fdct_islow; //slow/accurate/default |
2979 | 4008 c->fdct248 = ff_fdct248_islow; |
1567 | 4009 } |
1092 | 4010 #endif //CONFIG_ENCODERS |
4011 | |
2256 | 4012 if(avctx->lowres==1){ |
5064 | 4013 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
|
4014 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
|
4015 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
|
4016 }else{ |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4017 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
|
4018 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
|
4019 } |
2256 | 4020 c->idct = j_rev_dct4; |
1092 | 4021 c->idct_permutation_type= FF_NO_IDCT_PERM; |
2257 | 4022 }else if(avctx->lowres==2){ |
4023 c->idct_put= ff_jref_idct2_put; | |
4024 c->idct_add= ff_jref_idct2_add; | |
4025 c->idct = j_rev_dct2; | |
4026 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2259 | 4027 }else if(avctx->lowres==3){ |
4028 c->idct_put= ff_jref_idct1_put; | |
4029 c->idct_add= ff_jref_idct1_add; | |
4030 c->idct = j_rev_dct1; | |
4031 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2256 | 4032 }else{ |
4033 if(avctx->idct_algo==FF_IDCT_INT){ | |
4034 c->idct_put= ff_jref_idct_put; | |
4035 c->idct_add= ff_jref_idct_add; | |
4036 c->idct = j_rev_dct; | |
4037 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM; | |
5007 | 4038 }else if((ENABLE_VP3_DECODER || ENABLE_VP5_DECODER || ENABLE_VP6_DECODER || ENABLE_THEORA_DECODER ) && |
4039 avctx->idct_algo==FF_IDCT_VP3){ | |
2693 | 4040 c->idct_put= ff_vp3_idct_put_c; |
4041 c->idct_add= ff_vp3_idct_add_c; | |
4042 c->idct = ff_vp3_idct_c; | |
4043 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
5887 | 4044 }else if(avctx->idct_algo==FF_IDCT_WMV2){ |
4045 c->idct_put= ff_wmv2_idct_put_c; | |
4046 c->idct_add= ff_wmv2_idct_add_c; | |
4047 c->idct = ff_wmv2_idct_c; | |
4048 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2256 | 4049 }else{ //accurate/default |
6001 | 4050 c->idct_put= ff_simple_idct_put; |
4051 c->idct_add= ff_simple_idct_add; | |
4052 c->idct = ff_simple_idct; | |
2256 | 4053 c->idct_permutation_type= FF_NO_IDCT_PERM; |
4054 } | |
1092 | 4055 } |
4056 | |
5064 | 4057 if (ENABLE_H264_DECODER) { |
5065 | 4058 c->h264_idct_add= ff_h264_idct_add_c; |
4059 c->h264_idct8_add= ff_h264_idct8_add_c; | |
4060 c->h264_idct_dc_add= ff_h264_idct_dc_add_c; | |
4061 c->h264_idct8_dc_add= ff_h264_idct8_dc_add_c; | |
5064 | 4062 } |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4063 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4064 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
|
4065 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
|
4066 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
|
4067 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
|
4068 c->add_pixels_clamped = add_pixels_clamped_c; |
2763 | 4069 c->add_pixels8 = add_pixels8_c; |
4070 c->add_pixels4 = add_pixels4_c; | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
4071 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
|
4072 c->gmc1 = gmc1_c; |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
4073 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
|
4074 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
|
4075 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
|
4076 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
|
4077 |
859 | 4078 /* TODO [0] 16 [1] 8 */ |
1708 | 4079 c->pix_abs[0][0] = pix_abs16_c; |
4080 c->pix_abs[0][1] = pix_abs16_x2_c; | |
4081 c->pix_abs[0][2] = pix_abs16_y2_c; | |
4082 c->pix_abs[0][3] = pix_abs16_xy2_c; | |
4083 c->pix_abs[1][0] = pix_abs8_c; | |
4084 c->pix_abs[1][1] = pix_abs8_x2_c; | |
4085 c->pix_abs[1][2] = pix_abs8_y2_c; | |
4086 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
|
4087 |
859 | 4088 #define dspfunc(PFX, IDX, NUM) \ |
4089 c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## NUM ## _c; \ | |
4090 c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## NUM ## _x2_c; \ | |
4091 c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## NUM ## _y2_c; \ | |
4092 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
|
4093 |
859 | 4094 dspfunc(put, 0, 16); |
4095 dspfunc(put_no_rnd, 0, 16); | |
4096 dspfunc(put, 1, 8); | |
4097 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
|
4098 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
|
4099 dspfunc(put, 3, 2); |
0 | 4100 |
859 | 4101 dspfunc(avg, 0, 16); |
4102 dspfunc(avg_no_rnd, 0, 16); | |
4103 dspfunc(avg, 1, 8); | |
4104 dspfunc(avg_no_rnd, 1, 8); | |
1319 | 4105 dspfunc(avg, 2, 4); |
4106 dspfunc(avg, 3, 2); | |
859 | 4107 #undef dspfunc |
857 | 4108 |
1864 | 4109 c->put_no_rnd_pixels_l2[0]= put_no_rnd_pixels16_l2_c; |
4110 c->put_no_rnd_pixels_l2[1]= put_no_rnd_pixels8_l2_c; | |
4111 | |
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
|
4112 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
|
4113 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
|
4114 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
|
4115 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
|
4116 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
|
4117 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
|
4118 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
|
4119 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
|
4120 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
|
4121 |
1319 | 4122 c->avg_tpel_pixels_tab[ 0] = avg_tpel_pixels_mc00_c; |
4123 c->avg_tpel_pixels_tab[ 1] = avg_tpel_pixels_mc10_c; | |
4124 c->avg_tpel_pixels_tab[ 2] = avg_tpel_pixels_mc20_c; | |
4125 c->avg_tpel_pixels_tab[ 4] = avg_tpel_pixels_mc01_c; | |
4126 c->avg_tpel_pixels_tab[ 5] = avg_tpel_pixels_mc11_c; | |
4127 c->avg_tpel_pixels_tab[ 6] = avg_tpel_pixels_mc21_c; | |
4128 c->avg_tpel_pixels_tab[ 8] = avg_tpel_pixels_mc02_c; | |
4129 c->avg_tpel_pixels_tab[ 9] = avg_tpel_pixels_mc12_c; | |
4130 c->avg_tpel_pixels_tab[10] = avg_tpel_pixels_mc22_c; | |
4131 | |
859 | 4132 #define dspfunc(PFX, IDX, NUM) \ |
4133 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_c; \ | |
4134 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_c; \ | |
4135 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_c; \ | |
4136 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_c; \ | |
4137 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_c; \ | |
4138 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_c; \ | |
4139 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_c; \ | |
4140 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_c; \ | |
4141 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_c; \ | |
4142 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_c; \ | |
4143 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_c; \ | |
4144 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_c; \ | |
4145 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_c; \ | |
4146 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_c; \ | |
4147 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_c; \ | |
4148 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_c | |
857 | 4149 |
859 | 4150 dspfunc(put_qpel, 0, 16); |
4151 dspfunc(put_no_rnd_qpel, 0, 16); | |
4152 | |
4153 dspfunc(avg_qpel, 0, 16); | |
4154 /* dspfunc(avg_no_rnd_qpel, 0, 16); */ | |
857 | 4155 |
859 | 4156 dspfunc(put_qpel, 1, 8); |
4157 dspfunc(put_no_rnd_qpel, 1, 8); | |
4158 | |
4159 dspfunc(avg_qpel, 1, 8); | |
4160 /* dspfunc(avg_no_rnd_qpel, 1, 8); */ | |
1168 | 4161 |
4162 dspfunc(put_h264_qpel, 0, 16); | |
4163 dspfunc(put_h264_qpel, 1, 8); | |
4164 dspfunc(put_h264_qpel, 2, 4); | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
4165 dspfunc(put_h264_qpel, 3, 2); |
1168 | 4166 dspfunc(avg_h264_qpel, 0, 16); |
4167 dspfunc(avg_h264_qpel, 1, 8); | |
4168 dspfunc(avg_h264_qpel, 2, 4); | |
4169 | |
859 | 4170 #undef dspfunc |
1168 | 4171 c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_c; |
4172 c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_c; | |
4173 c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_c; | |
4174 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_c; | |
4175 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_c; | |
4176 c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_c; | |
3663 | 4177 c->put_no_rnd_h264_chroma_pixels_tab[0]= put_no_rnd_h264_chroma_mc8_c; |
857 | 4178 |
2415 | 4179 c->weight_h264_pixels_tab[0]= weight_h264_pixels16x16_c; |
4180 c->weight_h264_pixels_tab[1]= weight_h264_pixels16x8_c; | |
4181 c->weight_h264_pixels_tab[2]= weight_h264_pixels8x16_c; | |
4182 c->weight_h264_pixels_tab[3]= weight_h264_pixels8x8_c; | |
4183 c->weight_h264_pixels_tab[4]= weight_h264_pixels8x4_c; | |
4184 c->weight_h264_pixels_tab[5]= weight_h264_pixels4x8_c; | |
4185 c->weight_h264_pixels_tab[6]= weight_h264_pixels4x4_c; | |
4186 c->weight_h264_pixels_tab[7]= weight_h264_pixels4x2_c; | |
4187 c->weight_h264_pixels_tab[8]= weight_h264_pixels2x4_c; | |
4188 c->weight_h264_pixels_tab[9]= weight_h264_pixels2x2_c; | |
4189 c->biweight_h264_pixels_tab[0]= biweight_h264_pixels16x16_c; | |
4190 c->biweight_h264_pixels_tab[1]= biweight_h264_pixels16x8_c; | |
4191 c->biweight_h264_pixels_tab[2]= biweight_h264_pixels8x16_c; | |
4192 c->biweight_h264_pixels_tab[3]= biweight_h264_pixels8x8_c; | |
4193 c->biweight_h264_pixels_tab[4]= biweight_h264_pixels8x4_c; | |
4194 c->biweight_h264_pixels_tab[5]= biweight_h264_pixels4x8_c; | |
4195 c->biweight_h264_pixels_tab[6]= biweight_h264_pixels4x4_c; | |
4196 c->biweight_h264_pixels_tab[7]= biweight_h264_pixels4x2_c; | |
4197 c->biweight_h264_pixels_tab[8]= biweight_h264_pixels2x4_c; | |
4198 c->biweight_h264_pixels_tab[9]= biweight_h264_pixels2x2_c; | |
4199 | |
3432 | 4200 #ifdef CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4201 ff_cavsdsp_init(c,avctx); |
3432 | 4202 #endif |
3526 | 4203 #if defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
4204 ff_vc1dsp_init(c,avctx); | |
4205 #endif | |
5887 | 4206 #if defined(CONFIG_WMV2_DECODER) || defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
4207 ff_intrax8dsp_init(c,avctx); | |
4208 #endif | |
4296 | 4209 #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
|
4210 ff_h264dspenc_init(c,avctx); |
4296 | 4211 #endif |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4212 |
936 | 4213 c->put_mspel_pixels_tab[0]= put_mspel8_mc00_c; |
4214 c->put_mspel_pixels_tab[1]= put_mspel8_mc10_c; | |
4215 c->put_mspel_pixels_tab[2]= put_mspel8_mc20_c; | |
4216 c->put_mspel_pixels_tab[3]= put_mspel8_mc30_c; | |
4217 c->put_mspel_pixels_tab[4]= put_mspel8_mc02_c; | |
4218 c->put_mspel_pixels_tab[5]= put_mspel8_mc12_c; | |
4219 c->put_mspel_pixels_tab[6]= put_mspel8_mc22_c; | |
4220 c->put_mspel_pixels_tab[7]= put_mspel8_mc32_c; | |
2967 | 4221 |
1708 | 4222 #define SET_CMP_FUNC(name) \ |
4223 c->name[0]= name ## 16_c;\ | |
4224 c->name[1]= name ## 8x8_c; | |
2967 | 4225 |
1708 | 4226 SET_CMP_FUNC(hadamard8_diff) |
1729 | 4227 c->hadamard8_diff[4]= hadamard8_intra16_c; |
1708 | 4228 SET_CMP_FUNC(dct_sad) |
2382 | 4229 SET_CMP_FUNC(dct_max) |
3013 | 4230 #ifdef CONFIG_GPL |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
4231 SET_CMP_FUNC(dct264_sad) |
3013 | 4232 #endif |
1708 | 4233 c->sad[0]= pix_abs16_c; |
4234 c->sad[1]= pix_abs8_c; | |
4235 c->sse[0]= sse16_c; | |
4236 c->sse[1]= sse8_c; | |
2184 | 4237 c->sse[2]= sse4_c; |
1708 | 4238 SET_CMP_FUNC(quant_psnr) |
4239 SET_CMP_FUNC(rd) | |
4240 SET_CMP_FUNC(bit) | |
1729 | 4241 c->vsad[0]= vsad16_c; |
4242 c->vsad[4]= vsad_intra16_c; | |
4243 c->vsse[0]= vsse16_c; | |
4244 c->vsse[4]= vsse_intra16_c; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4245 c->nsse[0]= nsse16_c; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4246 c->nsse[1]= nsse8_c; |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
4247 #ifdef CONFIG_SNOW_ENCODER |
2184 | 4248 c->w53[0]= w53_16_c; |
4249 c->w53[1]= w53_8_c; | |
4250 c->w97[0]= w97_16_c; | |
4251 c->w97[1]= w97_8_c; | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
4252 #endif |
2184 | 4253 |
4749 | 4254 c->ssd_int8_vs_int16 = ssd_int8_vs_int16_c; |
4255 | |
866 | 4256 c->add_bytes= add_bytes_c; |
6384 | 4257 c->add_bytes_l2= add_bytes_l2_c; |
866 | 4258 c->diff_bytes= diff_bytes_c; |
1527 | 4259 c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_c; |
1273 | 4260 c->bswap_buf= bswap_buf; |
6384 | 4261 #ifdef CONFIG_PNG_DECODER |
4262 c->add_png_paeth_prediction= ff_add_png_paeth_prediction; | |
4263 #endif | |
2633 | 4264 |
4265 c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_c; | |
4266 c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_c; | |
4267 c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_c; | |
4268 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
|
4269 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
|
4270 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
|
4271 c->h264_loop_filter_strength= NULL; |
2967 | 4272 |
5277
7b3fcb7c61ce
Avoid linking with h263.c functions when the relevant codecs
aurel
parents:
5256
diff
changeset
|
4273 if (ENABLE_ANY_H263) { |
5278 | 4274 c->h263_h_loop_filter= h263_h_loop_filter_c; |
4275 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
|
4276 } |
2967 | 4277 |
2045 | 4278 c->h261_loop_filter= h261_loop_filter_c; |
2967 | 4279 |
1784 | 4280 c->try_8x8basis= try_8x8basis_c; |
4281 c->add_8x8basis= add_8x8basis_c; | |
866 | 4282 |
4598
5111fceeb971
The Snow DEcoding routines should be under CONFIG_SNOW_DECODER.
diego
parents:
4594
diff
changeset
|
4283 #ifdef CONFIG_SNOW_DECODER |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4284 c->vertical_compose97i = ff_snow_vertical_compose97i; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4285 c->horizontal_compose97i = ff_snow_horizontal_compose97i; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4286 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
|
4287 #endif |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4288 |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4289 #ifdef CONFIG_VORBIS_DECODER |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4290 c->vorbis_inverse_coupling = vorbis_inverse_coupling; |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4291 #endif |
5737 | 4292 #ifdef CONFIG_FLAC_ENCODER |
4293 c->flac_compute_autocorr = ff_flac_compute_autocorr; | |
4294 #endif | |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4295 c->vector_fmul = vector_fmul_c; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4296 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
|
4297 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
|
4298 c->float_to_int16 = ff_float_to_int16_c; |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4299 |
3245 | 4300 c->shrink[0]= ff_img_copy_plane; |
4301 c->shrink[1]= ff_shrink22; | |
4302 c->shrink[2]= ff_shrink44; | |
4303 c->shrink[3]= ff_shrink88; | |
4304 | |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4305 c->prefetch= just_return; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4306 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4307 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
|
4308 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
|
4309 |
5146
5db8d038eb1f
use if() instead of #ifdef and ensure all possible optimisations are enabled
aurel
parents:
5143
diff
changeset
|
4310 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
|
4311 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
|
4312 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
|
4313 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
|
4314 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
|
4315 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
|
4316 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
|
4317 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
|
4318 if (ENABLE_BFIN) dsputil_init_bfin (c, avctx); |
1092 | 4319 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4320 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
|
4321 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
|
4322 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
|
4323 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
|
4324 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
|
4325 } |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4326 |
1092 | 4327 switch(c->idct_permutation_type){ |
4328 case FF_NO_IDCT_PERM: | |
4329 for(i=0; i<64; i++) | |
4330 c->idct_permutation[i]= i; | |
4331 break; | |
4332 case FF_LIBMPEG2_IDCT_PERM: | |
4333 for(i=0; i<64; i++) | |
4334 c->idct_permutation[i]= (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2); | |
4335 break; | |
4336 case FF_SIMPLE_IDCT_PERM: | |
4337 for(i=0; i<64; i++) | |
4338 c->idct_permutation[i]= simple_mmx_permutation[i]; | |
4339 break; | |
4340 case FF_TRANSPOSE_IDCT_PERM: | |
4341 for(i=0; i<64; i++) | |
4342 c->idct_permutation[i]= ((i&7)<<3) | (i>>3); | |
4343 break; | |
2696
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4344 case FF_PARTTRANS_IDCT_PERM: |
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4345 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
|
4346 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
|
4347 break; |
1092 | 4348 default: |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1571
diff
changeset
|
4349 av_log(avctx, AV_LOG_ERROR, "Internal error, IDCT permutation not set\n"); |
1092 | 4350 } |
0 | 4351 } |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
220
diff
changeset
|
4352 |