Mercurial > libavcodec.hg
annotate dsputil.c @ 12180:b24153464669 libavcodec
Attempt to fix x86-64 testsuite on fate.
author | rbultje |
---|---|
date | Fri, 16 Jul 2010 21:35:30 +0000 |
parents | 8b28e74de2c0 |
children | 5d73c4b4cd37 |
rev | line source |
---|---|
0 | 1 /* |
2 * DSP utils | |
8629
04423b2f6e0b
cosmetics: Remove pointless period after copyright statement non-sentences.
diego
parents:
8627
diff
changeset
|
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 /** |
11644
7dd2a45249a9
Remove explicit filename from Doxygen @file commands.
diego
parents:
11637
diff
changeset
|
26 * @file |
1106 | 27 * DSP utils |
28 */ | |
2967 | 29 |
0 | 30 #include "avcodec.h" |
31 #include "dsputil.h" | |
1092 | 32 #include "simple_idct.h" |
1557 | 33 #include "faandct.h" |
6407 | 34 #include "faanidct.h" |
8627
d6bab465b82c
moves mid_pred() into mathops.h (with arch specific code split by directory)
aurel
parents:
8596
diff
changeset
|
35 #include "mathops.h" |
10748
36611425fedb
Add required header #includes for mpegvideo.h and config.h.
diego
parents:
10644
diff
changeset
|
36 #include "mpegvideo.h" |
36611425fedb
Add required header #includes for mpegvideo.h and config.h.
diego
parents:
10644
diff
changeset
|
37 #include "config.h" |
11375
84963c795459
Move some prototypes from dsputil.c to reasonable header files
mru
parents:
11369
diff
changeset
|
38 #include "lpc.h" |
84963c795459
Move some prototypes from dsputil.c to reasonable header files
mru
parents:
11369
diff
changeset
|
39 #include "ac3dec.h" |
84963c795459
Move some prototypes from dsputil.c to reasonable header files
mru
parents:
11369
diff
changeset
|
40 #include "vorbis.h" |
84963c795459
Move some prototypes from dsputil.c to reasonable header files
mru
parents:
11369
diff
changeset
|
41 #include "png.h" |
11921 | 42 #include "vp8dsp.h" |
676 | 43 |
4176 | 44 uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP] = {0, }; |
4179 | 45 uint32_t ff_squareTbl[512] = {0, }; |
0 | 46 |
6387 | 47 // 0x7f7f7f7f or 0x7f7f7f7f7f7f7f7f or whatever, depending on the cpu's native arithmetic size |
48 #define pb_7f (~0UL/255 * 0x7f) | |
49 #define pb_80 (~0UL/255 * 0x80) | |
6385 | 50 |
1064 | 51 const uint8_t ff_zigzag_direct[64] = { |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
52 0, 1, 8, 16, 9, 2, 3, 10, |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
53 17, 24, 32, 25, 18, 11, 4, 5, |
34 | 54 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
|
55 27, 20, 13, 6, 7, 14, 21, 28, |
34 | 56 35, 42, 49, 56, 57, 50, 43, 36, |
57 29, 22, 15, 23, 30, 37, 44, 51, | |
58 58, 59, 52, 45, 38, 31, 39, 46, | |
59 53, 60, 61, 54, 47, 55, 62, 63 | |
60 }; | |
61 | |
1567 | 62 /* Specific zigzag scan for 248 idct. NOTE that unlike the |
63 specification, we interleave the fields */ | |
64 const uint8_t ff_zigzag248_direct[64] = { | |
65 0, 8, 1, 9, 16, 24, 2, 10, | |
66 17, 25, 32, 40, 48, 56, 33, 41, | |
67 18, 26, 3, 11, 4, 12, 19, 27, | |
68 34, 42, 49, 57, 50, 58, 35, 43, | |
69 20, 28, 5, 13, 6, 14, 21, 29, | |
70 36, 44, 51, 59, 52, 60, 37, 45, | |
71 22, 30, 7, 15, 23, 31, 38, 46, | |
72 53, 61, 54, 62, 39, 47, 55, 63, | |
73 }; | |
74 | |
220 | 75 /* not permutated inverse zigzag_direct + 1 for MMX quantizer */ |
11369 | 76 DECLARE_ALIGNED(16, uint16_t, inv_zigzag_direct16)[64]; |
220 | 77 |
1064 | 78 const uint8_t ff_alternate_horizontal_scan[64] = { |
2967 | 79 0, 1, 2, 3, 8, 9, 16, 17, |
34 | 80 10, 11, 4, 5, 6, 7, 15, 14, |
2967 | 81 13, 12, 19, 18, 24, 25, 32, 33, |
34 | 82 26, 27, 20, 21, 22, 23, 28, 29, |
2967 | 83 30, 31, 34, 35, 40, 41, 48, 49, |
34 | 84 42, 43, 36, 37, 38, 39, 44, 45, |
2967 | 85 46, 47, 50, 51, 56, 57, 58, 59, |
34 | 86 52, 53, 54, 55, 60, 61, 62, 63, |
87 }; | |
88 | |
1064 | 89 const uint8_t ff_alternate_vertical_scan[64] = { |
2967 | 90 0, 8, 16, 24, 1, 9, 2, 10, |
34 | 91 17, 25, 32, 40, 48, 56, 57, 49, |
2967 | 92 41, 33, 26, 18, 3, 11, 4, 12, |
34 | 93 19, 27, 34, 42, 50, 58, 35, 43, |
2967 | 94 51, 59, 20, 28, 5, 13, 6, 14, |
34 | 95 21, 29, 36, 44, 52, 60, 37, 45, |
2967 | 96 53, 61, 22, 30, 7, 15, 23, 31, |
34 | 97 38, 46, 54, 62, 39, 47, 55, 63, |
98 }; | |
99 | |
10207 | 100 /* a*inverse[b]>>32 == a/b for all 0<=a<=16909558 && 2<=b<=256 |
101 * for a>16909558, is an overestimate by less than 1 part in 1<<24 */ | |
102 const uint32_t ff_inverse[257]={ | |
2967 | 103 0, 4294967295U,2147483648U,1431655766, 1073741824, 858993460, 715827883, 613566757, |
104 536870912, 477218589, 429496730, 390451573, 357913942, 330382100, 306783379, 286331154, | |
105 268435456, 252645136, 238609295, 226050911, 214748365, 204522253, 195225787, 186737709, | |
106 178956971, 171798692, 165191050, 159072863, 153391690, 148102321, 143165577, 138547333, | |
107 134217728, 130150525, 126322568, 122713352, 119304648, 116080198, 113025456, 110127367, | |
108 107374183, 104755300, 102261127, 99882961, 97612894, 95443718, 93368855, 91382283, | |
109 89478486, 87652394, 85899346, 84215046, 82595525, 81037119, 79536432, 78090315, | |
110 76695845, 75350304, 74051161, 72796056, 71582789, 70409300, 69273667, 68174085, | |
111 67108864, 66076420, 65075263, 64103990, 63161284, 62245903, 61356676, 60492498, | |
112 59652324, 58835169, 58040099, 57266231, 56512728, 55778797, 55063684, 54366675, | |
113 53687092, 53024288, 52377650, 51746594, 51130564, 50529028, 49941481, 49367441, | |
114 48806447, 48258060, 47721859, 47197443, 46684428, 46182445, 45691142, 45210183, | |
115 44739243, 44278014, 43826197, 43383509, 42949673, 42524429, 42107523, 41698712, | |
116 41297763, 40904451, 40518560, 40139882, 39768216, 39403370, 39045158, 38693400, | |
117 38347923, 38008561, 37675152, 37347542, 37025581, 36709123, 36398028, 36092163, | |
118 35791395, 35495598, 35204650, 34918434, 34636834, 34359739, 34087043, 33818641, | |
119 33554432, 33294321, 33038210, 32786010, 32537632, 32292988, 32051995, 31814573, | |
120 31580642, 31350127, 31122952, 30899046, 30678338, 30460761, 30246249, 30034737, | |
121 29826162, 29620465, 29417585, 29217465, 29020050, 28825284, 28633116, 28443493, | |
122 28256364, 28071682, 27889399, 27709467, 27531842, 27356480, 27183338, 27012373, | |
123 26843546, 26676816, 26512144, 26349493, 26188825, 26030105, 25873297, 25718368, | |
124 25565282, 25414008, 25264514, 25116768, 24970741, 24826401, 24683721, 24542671, | |
125 24403224, 24265352, 24129030, 23994231, 23860930, 23729102, 23598722, 23469767, | |
126 23342214, 23216040, 23091223, 22967740, 22845571, 22724695, 22605092, 22486740, | |
127 22369622, 22253717, 22139007, 22025474, 21913099, 21801865, 21691755, 21582751, | |
128 21474837, 21367997, 21262215, 21157475, 21053762, 20951060, 20849356, 20748635, | |
129 20648882, 20550083, 20452226, 20355296, 20259280, 20164166, 20069941, 19976593, | |
130 19884108, 19792477, 19701685, 19611723, 19522579, 19434242, 19346700, 19259944, | |
131 19173962, 19088744, 19004281, 18920561, 18837576, 18755316, 18673771, 18592933, | |
132 18512791, 18433337, 18354562, 18276457, 18199014, 18122225, 18046082, 17970575, | |
133 17895698, 17821442, 17747799, 17674763, 17602325, 17530479, 17459217, 17388532, | |
220 | 134 17318417, 17248865, 17179870, 17111424, 17043522, 16976156, 16909321, 16843010, |
10207 | 135 16777216 |
220 | 136 }; |
137 | |
1092 | 138 /* Input permutation for the simple_idct_mmx */ |
139 static const uint8_t simple_mmx_permutation[64]={ | |
2979 | 140 0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, |
141 0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, | |
142 0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, | |
143 0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, | |
144 0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, | |
145 0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, | |
146 0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, | |
147 0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F, | |
1092 | 148 }; |
149 | |
6600
c3213c91124c
Add a new IDCT permutation, used in xvid_sse2 and possibly future similar IDCTs.
astrange
parents:
6450
diff
changeset
|
150 static const uint8_t idct_sse2_row_perm[8] = {0, 4, 1, 5, 2, 6, 3, 7}; |
c3213c91124c
Add a new IDCT permutation, used in xvid_sse2 and possibly future similar IDCTs.
astrange
parents:
6450
diff
changeset
|
151 |
6438 | 152 void ff_init_scantable(uint8_t *permutation, ScanTable *st, const uint8_t *src_scantable){ |
153 int i; | |
154 int end; | |
155 | |
156 st->scantable= src_scantable; | |
157 | |
158 for(i=0; i<64; i++){ | |
159 int j; | |
160 j = src_scantable[i]; | |
161 st->permutated[i] = permutation[j]; | |
8590 | 162 #if ARCH_PPC |
6438 | 163 st->inverse[j] = i; |
164 #endif | |
165 } | |
166 | |
167 end=-1; | |
168 for(i=0; i<64; i++){ | |
169 int j; | |
170 j = st->permutated[i]; | |
171 if(j>end) end=j; | |
172 st->raster_end[i]= end; | |
173 } | |
174 } | |
175 | |
1064 | 176 static int pix_sum_c(uint8_t * pix, int line_size) |
612 | 177 { |
178 int s, i, j; | |
179 | |
180 s = 0; | |
181 for (i = 0; i < 16; i++) { | |
2979 | 182 for (j = 0; j < 16; j += 8) { |
183 s += pix[0]; | |
184 s += pix[1]; | |
185 s += pix[2]; | |
186 s += pix[3]; | |
187 s += pix[4]; | |
188 s += pix[5]; | |
189 s += pix[6]; | |
190 s += pix[7]; | |
191 pix += 8; | |
192 } | |
193 pix += line_size - 16; | |
612 | 194 } |
195 return s; | |
196 } | |
197 | |
1064 | 198 static int pix_norm1_c(uint8_t * pix, int line_size) |
612 | 199 { |
200 int s, i, j; | |
4179 | 201 uint32_t *sq = ff_squareTbl + 256; |
612 | 202 |
203 s = 0; | |
204 for (i = 0; i < 16; i++) { | |
2979 | 205 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
|
206 #if 0 |
2979 | 207 s += sq[pix[0]]; |
208 s += sq[pix[1]]; | |
209 s += sq[pix[2]]; | |
210 s += sq[pix[3]]; | |
211 s += sq[pix[4]]; | |
212 s += sq[pix[5]]; | |
213 s += sq[pix[6]]; | |
214 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
|
215 #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
|
216 #if LONG_MAX > 2147483647 |
2979 | 217 register uint64_t x=*(uint64_t*)pix; |
218 s += sq[x&0xff]; | |
219 s += sq[(x>>8)&0xff]; | |
220 s += sq[(x>>16)&0xff]; | |
221 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
|
222 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
|
223 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
|
224 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
|
225 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
|
226 #else |
2979 | 227 register uint32_t x=*(uint32_t*)pix; |
228 s += sq[x&0xff]; | |
229 s += sq[(x>>8)&0xff]; | |
230 s += sq[(x>>16)&0xff]; | |
231 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
|
232 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
|
233 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
|
234 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
|
235 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
|
236 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
|
237 #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
|
238 #endif |
2979 | 239 pix += 8; |
240 } | |
241 pix += line_size - 16; | |
612 | 242 } |
243 return s; | |
244 } | |
245 | |
6241 | 246 static void bswap_buf(uint32_t *dst, const uint32_t *src, int w){ |
1273 | 247 int i; |
2967 | 248 |
1273 | 249 for(i=0; i+8<=w; i+=8){ |
12129 | 250 dst[i+0]= av_bswap32(src[i+0]); |
251 dst[i+1]= av_bswap32(src[i+1]); | |
252 dst[i+2]= av_bswap32(src[i+2]); | |
253 dst[i+3]= av_bswap32(src[i+3]); | |
254 dst[i+4]= av_bswap32(src[i+4]); | |
255 dst[i+5]= av_bswap32(src[i+5]); | |
256 dst[i+6]= av_bswap32(src[i+6]); | |
257 dst[i+7]= av_bswap32(src[i+7]); | |
1273 | 258 } |
259 for(;i<w; i++){ | |
12129 | 260 dst[i+0]= av_bswap32(src[i+0]); |
1273 | 261 } |
262 } | |
612 | 263 |
2184 | 264 static int sse4_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
265 { | |
266 int s, i; | |
4179 | 267 uint32_t *sq = ff_squareTbl + 256; |
2184 | 268 |
269 s = 0; | |
270 for (i = 0; i < h; i++) { | |
271 s += sq[pix1[0] - pix2[0]]; | |
272 s += sq[pix1[1] - pix2[1]]; | |
273 s += sq[pix1[2] - pix2[2]]; | |
274 s += sq[pix1[3] - pix2[3]]; | |
275 pix1 += line_size; | |
276 pix2 += line_size; | |
277 } | |
278 return s; | |
279 } | |
280 | |
1708 | 281 static int sse8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
936 | 282 { |
283 int s, i; | |
4179 | 284 uint32_t *sq = ff_squareTbl + 256; |
936 | 285 |
286 s = 0; | |
1708 | 287 for (i = 0; i < h; i++) { |
936 | 288 s += sq[pix1[0] - pix2[0]]; |
289 s += sq[pix1[1] - pix2[1]]; | |
290 s += sq[pix1[2] - pix2[2]]; | |
291 s += sq[pix1[3] - pix2[3]]; | |
292 s += sq[pix1[4] - pix2[4]]; | |
293 s += sq[pix1[5] - pix2[5]]; | |
294 s += sq[pix1[6] - pix2[6]]; | |
295 s += sq[pix1[7] - pix2[7]]; | |
296 pix1 += line_size; | |
297 pix2 += line_size; | |
298 } | |
299 return s; | |
300 } | |
301 | |
1708 | 302 static int sse16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
884 | 303 { |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
304 int s, i; |
4179 | 305 uint32_t *sq = ff_squareTbl + 256; |
884 | 306 |
307 s = 0; | |
1708 | 308 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
|
309 s += sq[pix1[ 0] - pix2[ 0]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
310 s += sq[pix1[ 1] - pix2[ 1]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
311 s += sq[pix1[ 2] - pix2[ 2]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
312 s += sq[pix1[ 3] - pix2[ 3]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
313 s += sq[pix1[ 4] - pix2[ 4]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
314 s += sq[pix1[ 5] - pix2[ 5]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
315 s += sq[pix1[ 6] - pix2[ 6]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
316 s += sq[pix1[ 7] - pix2[ 7]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
317 s += sq[pix1[ 8] - pix2[ 8]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
318 s += sq[pix1[ 9] - pix2[ 9]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
319 s += sq[pix1[10] - pix2[10]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
320 s += sq[pix1[11] - pix2[11]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
321 s += sq[pix1[12] - pix2[12]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
322 s += sq[pix1[13] - pix2[13]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
323 s += sq[pix1[14] - pix2[14]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
324 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
|
325 |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
326 pix1 += line_size; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
327 pix2 += line_size; |
884 | 328 } |
329 return s; | |
330 } | |
331 | |
6437 | 332 /* draw the edges of width 'w' of an image of size width, height */ |
333 //FIXME check that this is ok for mpeg4 interlaced | |
334 static void draw_edges_c(uint8_t *buf, int wrap, int width, int height, int w) | |
335 { | |
336 uint8_t *ptr, *last_line; | |
337 int i; | |
338 | |
339 last_line = buf + (height - 1) * wrap; | |
340 for(i=0;i<w;i++) { | |
341 /* top and bottom */ | |
342 memcpy(buf - (i + 1) * wrap, buf, width); | |
343 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
344 } | |
345 /* left and right */ | |
346 ptr = buf; | |
347 for(i=0;i<height;i++) { | |
348 memset(ptr - w, ptr[0], w); | |
349 memset(ptr + width, ptr[width-1], w); | |
350 ptr += wrap; | |
351 } | |
352 /* corners */ | |
353 for(i=0;i<w;i++) { | |
354 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
355 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
356 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
357 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
358 } | |
359 } | |
360 | |
6445 | 361 /** |
12024 | 362 * Copy a rectangular area of samples to a temporary buffer and replicate the border samples. |
6445 | 363 * @param buf destination buffer |
364 * @param src source buffer | |
365 * @param linesize number of bytes between 2 vertically adjacent samples in both the source and destination buffers | |
366 * @param block_w width of block | |
367 * @param block_h height of block | |
368 * @param src_x x coordinate of the top left sample of the block in the source buffer | |
369 * @param src_y y coordinate of the top left sample of the block in the source buffer | |
370 * @param w width of the source buffer | |
371 * @param h height of the source buffer | |
372 */ | |
11784 | 373 void ff_emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize, int block_w, int block_h, |
6445 | 374 int src_x, int src_y, int w, int h){ |
375 int x, y; | |
376 int start_y, start_x, end_y, end_x; | |
377 | |
378 if(src_y>= h){ | |
379 src+= (h-1-src_y)*linesize; | |
380 src_y=h-1; | |
381 }else if(src_y<=-block_h){ | |
382 src+= (1-block_h-src_y)*linesize; | |
383 src_y=1-block_h; | |
384 } | |
385 if(src_x>= w){ | |
386 src+= (w-1-src_x); | |
387 src_x=w-1; | |
388 }else if(src_x<=-block_w){ | |
389 src+= (1-block_w-src_x); | |
390 src_x=1-block_w; | |
391 } | |
392 | |
393 start_y= FFMAX(0, -src_y); | |
394 start_x= FFMAX(0, -src_x); | |
395 end_y= FFMIN(block_h, h-src_y); | |
396 end_x= FFMIN(block_w, w-src_x); | |
397 | |
398 // copy existing part | |
399 for(y=start_y; y<end_y; y++){ | |
400 for(x=start_x; x<end_x; x++){ | |
401 buf[x + y*linesize]= src[x + y*linesize]; | |
402 } | |
403 } | |
404 | |
405 //top | |
406 for(y=0; y<start_y; y++){ | |
407 for(x=start_x; x<end_x; x++){ | |
408 buf[x + y*linesize]= buf[x + start_y*linesize]; | |
409 } | |
410 } | |
411 | |
412 //bottom | |
413 for(y=end_y; y<block_h; y++){ | |
414 for(x=start_x; x<end_x; x++){ | |
415 buf[x + y*linesize]= buf[x + (end_y-1)*linesize]; | |
416 } | |
417 } | |
418 | |
419 for(y=0; y<block_h; y++){ | |
420 //left | |
421 for(x=0; x<start_x; x++){ | |
422 buf[x + y*linesize]= buf[start_x + y*linesize]; | |
423 } | |
424 | |
425 //right | |
426 for(x=end_x; x<block_w; x++){ | |
427 buf[x + y*linesize]= buf[end_x - 1 + y*linesize]; | |
428 } | |
429 } | |
430 } | |
431 | |
1064 | 432 static void get_pixels_c(DCTELEM *restrict block, const uint8_t *pixels, int line_size) |
0 | 433 { |
434 int i; | |
435 | |
436 /* read the pixels */ | |
437 for(i=0;i<8;i++) { | |
516 | 438 block[0] = pixels[0]; |
439 block[1] = pixels[1]; | |
440 block[2] = pixels[2]; | |
441 block[3] = pixels[3]; | |
442 block[4] = pixels[4]; | |
443 block[5] = pixels[5]; | |
444 block[6] = pixels[6]; | |
445 block[7] = pixels[7]; | |
446 pixels += line_size; | |
447 block += 8; | |
0 | 448 } |
449 } | |
450 | |
1064 | 451 static void diff_pixels_c(DCTELEM *restrict block, const uint8_t *s1, |
2979 | 452 const uint8_t *s2, int stride){ |
324 | 453 int i; |
454 | |
455 /* read the pixels */ | |
456 for(i=0;i<8;i++) { | |
516 | 457 block[0] = s1[0] - s2[0]; |
458 block[1] = s1[1] - s2[1]; | |
459 block[2] = s1[2] - s2[2]; | |
460 block[3] = s1[3] - s2[3]; | |
461 block[4] = s1[4] - s2[4]; | |
462 block[5] = s1[5] - s2[5]; | |
463 block[6] = s1[6] - s2[6]; | |
464 block[7] = s1[7] - s2[7]; | |
324 | 465 s1 += stride; |
466 s2 += stride; | |
516 | 467 block += 8; |
324 | 468 } |
469 } | |
470 | |
471 | |
1064 | 472 static void put_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 473 int line_size) |
0 | 474 { |
475 int i; | |
4176 | 476 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 477 |
0 | 478 /* read the pixels */ |
479 for(i=0;i<8;i++) { | |
516 | 480 pixels[0] = cm[block[0]]; |
481 pixels[1] = cm[block[1]]; | |
482 pixels[2] = cm[block[2]]; | |
483 pixels[3] = cm[block[3]]; | |
484 pixels[4] = cm[block[4]]; | |
485 pixels[5] = cm[block[5]]; | |
486 pixels[6] = cm[block[6]]; | |
487 pixels[7] = cm[block[7]]; | |
488 | |
489 pixels += line_size; | |
490 block += 8; | |
0 | 491 } |
492 } | |
493 | |
2256 | 494 static void put_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 495 int line_size) |
2256 | 496 { |
497 int i; | |
4176 | 498 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 499 |
2256 | 500 /* read the pixels */ |
501 for(i=0;i<4;i++) { | |
502 pixels[0] = cm[block[0]]; | |
503 pixels[1] = cm[block[1]]; | |
504 pixels[2] = cm[block[2]]; | |
505 pixels[3] = cm[block[3]]; | |
506 | |
507 pixels += line_size; | |
508 block += 8; | |
509 } | |
510 } | |
511 | |
2257 | 512 static void put_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 513 int line_size) |
2257 | 514 { |
515 int i; | |
4176 | 516 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 517 |
2257 | 518 /* read the pixels */ |
519 for(i=0;i<2;i++) { | |
520 pixels[0] = cm[block[0]]; | |
521 pixels[1] = cm[block[1]]; | |
522 | |
523 pixels += line_size; | |
524 block += 8; | |
525 } | |
526 } | |
527 | |
2967 | 528 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
|
529 uint8_t *restrict pixels, |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
530 int line_size) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
531 { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
532 int i, j; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
533 |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
534 for (i = 0; i < 8; i++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
535 for (j = 0; j < 8; j++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
536 if (*block < -128) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
537 *pixels = 0; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
538 else if (*block > 127) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
539 *pixels = 255; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
540 else |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
541 *pixels = (uint8_t)(*block + 128); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
542 block++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
543 pixels++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
544 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
545 pixels += (line_size - 8); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
546 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
547 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
548 |
11231 | 549 static void put_pixels_nonclamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
550 int line_size) | |
551 { | |
552 int i; | |
553 | |
554 /* read the pixels */ | |
555 for(i=0;i<8;i++) { | |
556 pixels[0] = block[0]; | |
557 pixels[1] = block[1]; | |
558 pixels[2] = block[2]; | |
559 pixels[3] = block[3]; | |
560 pixels[4] = block[4]; | |
561 pixels[5] = block[5]; | |
562 pixels[6] = block[6]; | |
563 pixels[7] = block[7]; | |
564 | |
565 pixels += line_size; | |
566 block += 8; | |
567 } | |
568 } | |
569 | |
1064 | 570 static void add_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
516 | 571 int line_size) |
0 | 572 { |
573 int i; | |
4176 | 574 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 575 |
0 | 576 /* read the pixels */ |
577 for(i=0;i<8;i++) { | |
516 | 578 pixels[0] = cm[pixels[0] + block[0]]; |
579 pixels[1] = cm[pixels[1] + block[1]]; | |
580 pixels[2] = cm[pixels[2] + block[2]]; | |
581 pixels[3] = cm[pixels[3] + block[3]]; | |
582 pixels[4] = cm[pixels[4] + block[4]]; | |
583 pixels[5] = cm[pixels[5] + block[5]]; | |
584 pixels[6] = cm[pixels[6] + block[6]]; | |
585 pixels[7] = cm[pixels[7] + block[7]]; | |
586 pixels += line_size; | |
587 block += 8; | |
0 | 588 } |
589 } | |
2256 | 590 |
591 static void add_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, | |
592 int line_size) | |
593 { | |
594 int i; | |
4176 | 595 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 596 |
2256 | 597 /* read the pixels */ |
598 for(i=0;i<4;i++) { | |
599 pixels[0] = cm[pixels[0] + block[0]]; | |
600 pixels[1] = cm[pixels[1] + block[1]]; | |
601 pixels[2] = cm[pixels[2] + block[2]]; | |
602 pixels[3] = cm[pixels[3] + block[3]]; | |
603 pixels += line_size; | |
604 block += 8; | |
605 } | |
606 } | |
2257 | 607 |
608 static void add_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, | |
609 int line_size) | |
610 { | |
611 int i; | |
4176 | 612 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2967 | 613 |
2257 | 614 /* read the pixels */ |
615 for(i=0;i<2;i++) { | |
616 pixels[0] = cm[pixels[0] + block[0]]; | |
617 pixels[1] = cm[pixels[1] + block[1]]; | |
618 pixels += line_size; | |
619 block += 8; | |
620 } | |
621 } | |
2763 | 622 |
623 static void add_pixels8_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
624 { | |
625 int i; | |
626 for(i=0;i<8;i++) { | |
627 pixels[0] += block[0]; | |
628 pixels[1] += block[1]; | |
629 pixels[2] += block[2]; | |
630 pixels[3] += block[3]; | |
631 pixels[4] += block[4]; | |
632 pixels[5] += block[5]; | |
633 pixels[6] += block[6]; | |
634 pixels[7] += block[7]; | |
635 pixels += line_size; | |
636 block += 8; | |
637 } | |
638 } | |
639 | |
640 static void add_pixels4_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
641 { | |
642 int i; | |
643 for(i=0;i<4;i++) { | |
644 pixels[0] += block[0]; | |
645 pixels[1] += block[1]; | |
646 pixels[2] += block[2]; | |
647 pixels[3] += block[3]; | |
648 pixels += line_size; | |
649 block += 4; | |
650 } | |
651 } | |
652 | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
653 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
|
654 { |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
655 int sum=0, i; |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
656 for(i=0; i<64; i++) |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
657 sum+= FFABS(block[i]); |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
658 return sum; |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
659 } |
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
660 |
11231 | 661 static void fill_block16_c(uint8_t *block, uint8_t value, int line_size, int h) |
662 { | |
663 int i; | |
664 | |
665 for (i = 0; i < h; i++) { | |
666 memset(block, value, 16); | |
667 block += line_size; | |
668 } | |
669 } | |
670 | |
671 static void fill_block8_c(uint8_t *block, uint8_t value, int line_size, int h) | |
672 { | |
673 int i; | |
674 | |
675 for (i = 0; i < h; i++) { | |
676 memset(block, value, 8); | |
677 block += line_size; | |
678 } | |
679 } | |
680 | |
681 static void scale_block_c(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize) | |
682 { | |
683 int i, j; | |
11459 | 684 uint16_t *dst1 = (uint16_t *) dst; |
685 uint16_t *dst2 = (uint16_t *)(dst + linesize); | |
11231 | 686 |
687 for (j = 0; j < 8; j++) { | |
688 for (i = 0; i < 8; i++) { | |
689 dst1[i] = dst2[i] = src[i] * 0x0101; | |
690 } | |
691 src += 8; | |
692 dst1 += linesize; | |
693 dst2 += linesize; | |
694 } | |
695 } | |
696 | |
385 | 697 #if 0 |
698 | |
699 #define PIXOP2(OPNAME, OP) \ | |
651 | 700 static void OPNAME ## _pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 701 {\ |
702 int i;\ | |
703 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
|
704 OP(*((uint64_t*)block), AV_RN64(pixels));\ |
385 | 705 pixels+=line_size;\ |
706 block +=line_size;\ | |
707 }\ | |
708 }\ | |
709 \ | |
859 | 710 static void OPNAME ## _no_rnd_pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 711 {\ |
712 int i;\ | |
713 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
|
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 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
717 pixels+=line_size;\ | |
718 block +=line_size;\ | |
719 }\ | |
720 }\ | |
721 \ | |
859 | 722 static void OPNAME ## _pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 723 {\ |
724 int i;\ | |
725 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
|
726 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
|
727 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 728 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
729 pixels+=line_size;\ | |
730 block +=line_size;\ | |
731 }\ | |
732 }\ | |
733 \ | |
859 | 734 static void OPNAME ## _no_rnd_pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 735 {\ |
736 int i;\ | |
737 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
|
738 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
|
739 const uint64_t b= AV_RN64(pixels+line_size);\ |
385 | 740 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
741 pixels+=line_size;\ | |
742 block +=line_size;\ | |
743 }\ | |
744 }\ | |
745 \ | |
859 | 746 static void OPNAME ## _pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 747 {\ |
748 int i;\ | |
749 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
|
750 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
|
751 const uint64_t b= AV_RN64(pixels+line_size);\ |
385 | 752 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ |
753 pixels+=line_size;\ | |
754 block +=line_size;\ | |
755 }\ | |
756 }\ | |
757 \ | |
859 | 758 static void OPNAME ## _pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 759 {\ |
760 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
761 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
|
762 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 763 uint64_t l0= (a&0x0303030303030303ULL)\ |
764 + (b&0x0303030303030303ULL)\ | |
765 + 0x0202020202020202ULL;\ | |
766 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
767 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
768 uint64_t l1,h1;\ | |
769 \ | |
770 pixels+=line_size;\ | |
771 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
|
772 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
|
773 uint64_t b= AV_RN64(pixels+1);\ |
385 | 774 l1= (a&0x0303030303030303ULL)\ |
775 + (b&0x0303030303030303ULL);\ | |
776 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
777 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
778 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
779 pixels+=line_size;\ | |
780 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
781 a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
782 b= AV_RN64(pixels+1);\ |
385 | 783 l0= (a&0x0303030303030303ULL)\ |
784 + (b&0x0303030303030303ULL)\ | |
785 + 0x0202020202020202ULL;\ | |
786 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
787 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
788 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
789 pixels+=line_size;\ | |
790 block +=line_size;\ | |
791 }\ | |
792 }\ | |
793 \ | |
859 | 794 static void OPNAME ## _no_rnd_pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 795 {\ |
796 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
797 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
|
798 const uint64_t b= AV_RN64(pixels+1);\ |
385 | 799 uint64_t l0= (a&0x0303030303030303ULL)\ |
800 + (b&0x0303030303030303ULL)\ | |
801 + 0x0101010101010101ULL;\ | |
802 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
803 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
804 uint64_t l1,h1;\ | |
805 \ | |
806 pixels+=line_size;\ | |
807 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
|
808 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
|
809 uint64_t b= AV_RN64(pixels+1);\ |
385 | 810 l1= (a&0x0303030303030303ULL)\ |
811 + (b&0x0303030303030303ULL);\ | |
812 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
813 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
814 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
815 pixels+=line_size;\ | |
816 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
817 a= AV_RN64(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
818 b= AV_RN64(pixels+1);\ |
385 | 819 l0= (a&0x0303030303030303ULL)\ |
820 + (b&0x0303030303030303ULL)\ | |
821 + 0x0101010101010101ULL;\ | |
822 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
823 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
824 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
825 pixels+=line_size;\ | |
826 block +=line_size;\ | |
827 }\ | |
828 }\ | |
829 \ | |
859 | 830 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels_c , 8)\ |
831 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels_x2_c , 8)\ | |
832 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels_y2_c , 8)\ | |
833 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels_xy2_c, 8)\ | |
834 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels_x2_c , 8)\ | |
835 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels_y2_c , 8)\ | |
836 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels_xy2_c, 8) | |
385 | 837 |
838 #define op_avg(a, b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEFEFEFEFEULL)>>1) ) | |
839 #else // 64 bit variant | |
840 | |
841 #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
|
842 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
|
843 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
|
844 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
|
845 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
|
846 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
|
847 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
|
848 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
849 }\ |
1168 | 850 static void OPNAME ## _pixels4_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
851 int i;\ | |
852 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
|
853 OP(*((uint32_t*)(block )), AV_RN32(pixels ));\ |
1168 | 854 pixels+=line_size;\ |
855 block +=line_size;\ | |
856 }\ | |
857 }\ | |
859 | 858 static void OPNAME ## _pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
385 | 859 int i;\ |
860 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
|
861 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
|
862 OP(*((uint32_t*)(block+4)), AV_RN32(pixels+4));\ |
385 | 863 pixels+=line_size;\ |
864 block +=line_size;\ | |
865 }\ | |
866 }\ | |
859 | 867 static inline void OPNAME ## _no_rnd_pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
868 OPNAME ## _pixels8_c(block, pixels, line_size, h);\ | |
651 | 869 }\ |
385 | 870 \ |
651 | 871 static inline void OPNAME ## _no_rnd_pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
872 int src_stride1, int src_stride2, int h){\ | |
385 | 873 int i;\ |
874 for(i=0; i<h; i++){\ | |
651 | 875 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
876 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
|
877 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 878 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
|
879 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
|
880 b= AV_RN32(&src2[i*src_stride2+4]);\ |
1264 | 881 OP(*((uint32_t*)&dst[i*dst_stride+4]), no_rnd_avg32(a, b));\ |
385 | 882 }\ |
883 }\ | |
884 \ | |
651 | 885 static inline void OPNAME ## _pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
886 int src_stride1, int src_stride2, int h){\ | |
385 | 887 int i;\ |
888 for(i=0; i<h; i++){\ | |
651 | 889 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
890 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
|
891 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 892 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
|
893 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
|
894 b= AV_RN32(&src2[i*src_stride2+4]);\ |
1264 | 895 OP(*((uint32_t*)&dst[i*dst_stride+4]), rnd_avg32(a, b));\ |
385 | 896 }\ |
897 }\ | |
898 \ | |
1168 | 899 static inline void OPNAME ## _pixels4_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
900 int src_stride1, int src_stride2, int h){\ | |
901 int i;\ | |
902 for(i=0; i<h; i++){\ | |
903 uint32_t a,b;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
904 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
|
905 b= AV_RN32(&src2[i*src_stride2 ]);\ |
1264 | 906 OP(*((uint32_t*)&dst[i*dst_stride ]), rnd_avg32(a, b));\ |
1168 | 907 }\ |
908 }\ | |
909 \ | |
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
|
910 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
|
911 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
|
912 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
|
913 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
|
914 uint32_t a,b;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
915 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
|
916 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
|
917 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
|
918 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
919 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
920 \ |
651 | 921 static inline void OPNAME ## _pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
922 int src_stride1, int src_stride2, int h){\ | |
923 OPNAME ## _pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
924 OPNAME ## _pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
925 }\ | |
926 \ | |
927 static inline void OPNAME ## _no_rnd_pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ | |
928 int src_stride1, int src_stride2, int h){\ | |
929 OPNAME ## _no_rnd_pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
930 OPNAME ## _no_rnd_pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
931 }\ | |
932 \ | |
859 | 933 static inline void OPNAME ## _no_rnd_pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 934 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
935 }\ | |
936 \ | |
859 | 937 static inline void OPNAME ## _pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 938 OPNAME ## _pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
939 }\ | |
940 \ | |
859 | 941 static inline void OPNAME ## _no_rnd_pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 942 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
943 }\ | |
944 \ | |
859 | 945 static inline void OPNAME ## _pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 946 OPNAME ## _pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
385 | 947 }\ |
948 \ | |
11783 | 949 static inline void OPNAME ## _pixels8_l4(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, const uint8_t *src3, const uint8_t *src4,\ |
651 | 950 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ |
951 int i;\ | |
952 for(i=0; i<h; i++){\ | |
953 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
|
954 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
|
955 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
|
956 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
|
957 d= AV_RN32(&src4[i*src_stride4]);\ |
651 | 958 l0= (a&0x03030303UL)\ |
959 + (b&0x03030303UL)\ | |
960 + 0x02020202UL;\ | |
961 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
962 + ((b&0xFCFCFCFCUL)>>2);\ | |
963 l1= (c&0x03030303UL)\ | |
964 + (d&0x03030303UL);\ | |
965 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
966 + ((d&0xFCFCFCFCUL)>>2);\ | |
967 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
|
968 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
|
969 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
|
970 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
|
971 d= AV_RN32(&src4[i*src_stride4+4]);\ |
651 | 972 l0= (a&0x03030303UL)\ |
973 + (b&0x03030303UL)\ | |
974 + 0x02020202UL;\ | |
975 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
976 + ((b&0xFCFCFCFCUL)>>2);\ | |
977 l1= (c&0x03030303UL)\ | |
978 + (d&0x03030303UL);\ | |
979 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
980 + ((d&0xFCFCFCFCUL)>>2);\ | |
981 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
982 }\ | |
983 }\ | |
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
|
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 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
|
986 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
|
987 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
988 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
989 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
|
990 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
|
991 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
992 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
993 static inline void OPNAME ## _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
|
994 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
|
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 ## _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
|
998 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
|
999 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1000 \ |
11783 | 1001 static inline void OPNAME ## _no_rnd_pixels8_l4(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, const uint8_t *src3, const uint8_t *src4,\ |
651 | 1002 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ |
385 | 1003 int i;\ |
1004 for(i=0; i<h; i++){\ | |
651 | 1005 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
|
1006 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
|
1007 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
|
1008 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
|
1009 d= AV_RN32(&src4[i*src_stride4]);\ |
651 | 1010 l0= (a&0x03030303UL)\ |
1011 + (b&0x03030303UL)\ | |
1012 + 0x01010101UL;\ | |
1013 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1014 + ((b&0xFCFCFCFCUL)>>2);\ | |
1015 l1= (c&0x03030303UL)\ | |
1016 + (d&0x03030303UL);\ | |
1017 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
1018 + ((d&0xFCFCFCFCUL)>>2);\ | |
1019 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
|
1020 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
|
1021 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
|
1022 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
|
1023 d= AV_RN32(&src4[i*src_stride4+4]);\ |
651 | 1024 l0= (a&0x03030303UL)\ |
1025 + (b&0x03030303UL)\ | |
1026 + 0x01010101UL;\ | |
1027 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1028 + ((b&0xFCFCFCFCUL)>>2);\ | |
1029 l1= (c&0x03030303UL)\ | |
1030 + (d&0x03030303UL);\ | |
1031 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
1032 + ((d&0xFCFCFCFCUL)>>2);\ | |
1033 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
385 | 1034 }\ |
1035 }\ | |
11783 | 1036 static inline void OPNAME ## _pixels16_l4(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, const uint8_t *src3, const uint8_t *src4,\ |
651 | 1037 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ |
1038 OPNAME ## _pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
1039 OPNAME ## _pixels8_l4(dst+8, src1+8, src2+8, src3+8, src4+8, dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
1040 }\ | |
11783 | 1041 static inline void OPNAME ## _no_rnd_pixels16_l4(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, const uint8_t *src3, const uint8_t *src4,\ |
651 | 1042 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ |
1043 OPNAME ## _no_rnd_pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
1044 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);\ | |
1045 }\ | |
385 | 1046 \ |
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
|
1047 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
|
1048 {\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1049 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
|
1050 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
|
1051 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
|
1052 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
|
1053 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
|
1054 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1055 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
|
1056 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
|
1057 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
|
1058 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
|
1059 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
|
1060 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
|
1061 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1062 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
|
1063 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
|
1064 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1065 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
|
1066 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
|
1067 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1068 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
|
1069 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
|
1070 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
|
1071 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
|
1072 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1073 block[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
|
1074 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
|
1075 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
|
1076 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
|
1077 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1078 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1079 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1080 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
|
1081 {\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1082 int i;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1083 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
|
1084 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
|
1085 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
|
1086 + (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
|
1087 + 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
|
1088 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
|
1089 + ((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
|
1090 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
|
1091 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1092 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
|
1093 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
|
1094 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
|
1095 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
|
1096 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
|
1097 + (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
|
1098 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
|
1099 + ((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
|
1100 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
|
1101 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
|
1102 block +=line_size;\ |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1103 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1104 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
|
1105 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
|
1106 + (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
|
1107 + 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
|
1108 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
|
1109 + ((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
|
1110 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
|
1111 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
|
1112 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
|
1113 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1114 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1115 \ |
859 | 1116 static inline void OPNAME ## _pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1117 {\ |
1118 int j;\ | |
1119 for(j=0; j<2; j++){\ | |
1120 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1121 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
|
1122 const uint32_t b= AV_RN32(pixels+1);\ |
385 | 1123 uint32_t l0= (a&0x03030303UL)\ |
1124 + (b&0x03030303UL)\ | |
1125 + 0x02020202UL;\ | |
1126 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1127 + ((b&0xFCFCFCFCUL)>>2);\ | |
1128 uint32_t l1,h1;\ | |
1129 \ | |
1130 pixels+=line_size;\ | |
1131 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
|
1132 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
|
1133 uint32_t b= AV_RN32(pixels+1);\ |
385 | 1134 l1= (a&0x03030303UL)\ |
1135 + (b&0x03030303UL);\ | |
1136 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1137 + ((b&0xFCFCFCFCUL)>>2);\ | |
1138 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1139 pixels+=line_size;\ | |
1140 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1141 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1142 b= AV_RN32(pixels+1);\ |
385 | 1143 l0= (a&0x03030303UL)\ |
1144 + (b&0x03030303UL)\ | |
1145 + 0x02020202UL;\ | |
1146 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1147 + ((b&0xFCFCFCFCUL)>>2);\ | |
1148 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1149 pixels+=line_size;\ | |
1150 block +=line_size;\ | |
1151 }\ | |
1152 pixels+=4-line_size*(h+1);\ | |
1153 block +=4-line_size*h;\ | |
1154 }\ | |
1155 }\ | |
1156 \ | |
859 | 1157 static inline void OPNAME ## _no_rnd_pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1158 {\ |
1159 int j;\ | |
1160 for(j=0; j<2; j++){\ | |
1161 int i;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1162 const uint32_t a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1163 const uint32_t b= AV_RN32(pixels+1);\ |
385 | 1164 uint32_t l0= (a&0x03030303UL)\ |
1165 + (b&0x03030303UL)\ | |
1166 + 0x01010101UL;\ | |
1167 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1168 + ((b&0xFCFCFCFCUL)>>2);\ | |
1169 uint32_t l1,h1;\ | |
1170 \ | |
1171 pixels+=line_size;\ | |
1172 for(i=0; i<h; i+=2){\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1173 uint32_t a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1174 uint32_t b= AV_RN32(pixels+1);\ |
385 | 1175 l1= (a&0x03030303UL)\ |
1176 + (b&0x03030303UL);\ | |
1177 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1178 + ((b&0xFCFCFCFCUL)>>2);\ | |
1179 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1180 pixels+=line_size;\ | |
1181 block +=line_size;\ | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1182 a= AV_RN32(pixels );\ |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5411
diff
changeset
|
1183 b= AV_RN32(pixels+1);\ |
385 | 1184 l0= (a&0x03030303UL)\ |
1185 + (b&0x03030303UL)\ | |
1186 + 0x01010101UL;\ | |
1187 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1188 + ((b&0xFCFCFCFCUL)>>2);\ | |
1189 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1190 pixels+=line_size;\ | |
1191 block +=line_size;\ | |
1192 }\ | |
1193 pixels+=4-line_size*(h+1);\ | |
1194 block +=4-line_size*h;\ | |
1195 }\ | |
1196 }\ | |
1197 \ | |
859 | 1198 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels8_c , 8)\ |
1199 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels8_x2_c , 8)\ | |
1200 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels8_y2_c , 8)\ | |
1201 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels8_xy2_c, 8)\ | |
1202 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_c , OPNAME ## _pixels8_c , 8)\ | |
1203 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels8_x2_c , 8)\ | |
1204 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels8_y2_c , 8)\ | |
1205 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels8_xy2_c, 8)\ | |
651 | 1206 |
1264 | 1207 #define op_avg(a, b) a = rnd_avg32(a, b) |
385 | 1208 #endif |
1209 #define op_put(a, b) a = b | |
1210 | |
1211 PIXOP2(avg, op_avg) | |
1212 PIXOP2(put, op_put) | |
1213 #undef op_avg | |
1214 #undef op_put | |
1215 | |
0 | 1216 #define avg2(a,b) ((a+b+1)>>1) |
1217 #define avg4(a,b,c,d) ((a+b+c+d+2)>>2) | |
1218 | |
1864 | 1219 static void put_no_rnd_pixels16_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ |
1220 put_no_rnd_pixels16_l2(dst, a, b, stride, stride, stride, h); | |
1221 } | |
1222 | |
1223 static void put_no_rnd_pixels8_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ | |
1224 put_no_rnd_pixels8_l2(dst, a, b, stride, stride, stride, h); | |
1225 } | |
753 | 1226 |
1064 | 1227 static void gmc1_c(uint8_t *dst, uint8_t *src, int stride, int h, int x16, int y16, int rounder) |
255 | 1228 { |
1229 const int A=(16-x16)*(16-y16); | |
1230 const int B=( x16)*(16-y16); | |
1231 const int C=(16-x16)*( y16); | |
1232 const int D=( x16)*( y16); | |
1233 int i; | |
1234 | |
1235 for(i=0; i<h; i++) | |
1236 { | |
651 | 1237 dst[0]= (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + rounder)>>8; |
1238 dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8; | |
1239 dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8; | |
1240 dst[3]= (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + rounder)>>8; | |
1241 dst[4]= (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + rounder)>>8; | |
1242 dst[5]= (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + rounder)>>8; | |
1243 dst[6]= (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + rounder)>>8; | |
1244 dst[7]= (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + rounder)>>8; | |
1245 dst+= stride; | |
1246 src+= stride; | |
255 | 1247 } |
1248 } | |
1249 | |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
1250 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, |
753 | 1251 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height) |
1252 { | |
1253 int y, vx, vy; | |
1254 const int s= 1<<shift; | |
2967 | 1255 |
753 | 1256 width--; |
1257 height--; | |
1258 | |
1259 for(y=0; y<h; y++){ | |
1260 int x; | |
1261 | |
1262 vx= ox; | |
1263 vy= oy; | |
1264 for(x=0; x<8; x++){ //XXX FIXME optimize | |
1265 int src_x, src_y, frac_x, frac_y, index; | |
1266 | |
1267 src_x= vx>>16; | |
1268 src_y= vy>>16; | |
1269 frac_x= src_x&(s-1); | |
1270 frac_y= src_y&(s-1); | |
1271 src_x>>=shift; | |
1272 src_y>>=shift; | |
2967 | 1273 |
753 | 1274 if((unsigned)src_x < width){ |
1275 if((unsigned)src_y < height){ | |
1276 index= src_x + src_y*stride; | |
1277 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) | |
1278 + src[index +1]* frac_x )*(s-frac_y) | |
1279 + ( src[index+stride ]*(s-frac_x) | |
1280 + src[index+stride+1]* frac_x )* frac_y | |
1281 + r)>>(shift*2); | |
1282 }else{ | |
4594 | 1283 index= src_x + av_clip(src_y, 0, height)*stride; |
2967 | 1284 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) |
753 | 1285 + src[index +1]* frac_x )*s |
1286 + r)>>(shift*2); | |
1287 } | |
1288 }else{ | |
1289 if((unsigned)src_y < height){ | |
4594 | 1290 index= av_clip(src_x, 0, width) + src_y*stride; |
2967 | 1291 dst[y*stride + x]= ( ( src[index ]*(s-frac_y) |
753 | 1292 + src[index+stride ]* frac_y )*s |
1293 + r)>>(shift*2); | |
1294 }else{ | |
4594 | 1295 index= av_clip(src_x, 0, width) + av_clip(src_y, 0, height)*stride; |
753 | 1296 dst[y*stride + x]= src[index ]; |
1297 } | |
1298 } | |
2967 | 1299 |
753 | 1300 vx+= dxx; |
1301 vy+= dyx; | |
1302 } | |
1303 ox += dxy; | |
1304 oy += dyy; | |
1305 } | |
1306 } | |
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
|
1307 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1308 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
|
1309 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
|
1310 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
|
1311 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
|
1312 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
|
1313 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
|
1314 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1315 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1316 |
85b71f9f7450
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 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
|
1318 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
|
1319 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
|
1320 for (j=0; j < width; j++) { |
2979 | 1321 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
|
1322 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1323 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
|
1324 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
|
1325 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1326 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1327 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1328 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
|
1329 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
|
1330 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
|
1331 for (j=0; j < width; j++) { |
2979 | 1332 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
|
1333 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1334 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
|
1335 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
|
1336 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1337 } |
2967 | 1338 |
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
|
1339 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
|
1340 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
|
1341 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
|
1342 for (j=0; j < width; j++) { |
2979 | 1343 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
|
1344 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1345 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
|
1346 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
|
1347 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1348 } |
2967 | 1349 |
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
|
1350 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
|
1351 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
|
1352 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
|
1353 for (j=0; j < width; j++) { |
2979 | 1354 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
|
1355 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1356 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
|
1357 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
|
1358 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1359 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1360 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1361 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
|
1362 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
|
1363 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
|
1364 for (j=0; j < width; j++) { |
2979 | 1365 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
|
1366 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1367 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
|
1368 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
|
1369 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1370 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1371 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1372 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
|
1373 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
|
1374 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
|
1375 for (j=0; j < width; j++) { |
2979 | 1376 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
|
1377 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1378 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
|
1379 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
|
1380 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1381 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1382 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1383 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
|
1384 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
|
1385 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
|
1386 for (j=0; j < width; j++) { |
2979 | 1387 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
|
1388 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1389 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
|
1390 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
|
1391 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1392 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1393 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1394 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
|
1395 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
|
1396 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
|
1397 for (j=0; j < width; j++) { |
2979 | 1398 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
|
1399 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1400 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
|
1401 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
|
1402 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1403 } |
1319 | 1404 |
1405 static inline void avg_tpel_pixels_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1406 switch(width){ | |
1407 case 2: avg_pixels2_c (dst, src, stride, height); break; | |
1408 case 4: avg_pixels4_c (dst, src, stride, height); break; | |
1409 case 8: avg_pixels8_c (dst, src, stride, height); break; | |
1410 case 16:avg_pixels16_c(dst, src, stride, height); break; | |
1411 } | |
1412 } | |
1413 | |
1414 static inline void avg_tpel_pixels_mc10_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1415 int i,j; | |
1416 for (i=0; i < height; i++) { | |
1417 for (j=0; j < width; j++) { | |
2979 | 1418 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1419 } |
1420 src += stride; | |
1421 dst += stride; | |
1422 } | |
1423 } | |
1424 | |
1425 static inline void avg_tpel_pixels_mc20_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1426 int i,j; | |
1427 for (i=0; i < height; i++) { | |
1428 for (j=0; j < width; j++) { | |
2979 | 1429 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1430 } |
1431 src += stride; | |
1432 dst += stride; | |
1433 } | |
1434 } | |
2967 | 1435 |
1319 | 1436 static inline void avg_tpel_pixels_mc01_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1437 int i,j; | |
1438 for (i=0; i < height; i++) { | |
1439 for (j=0; j < width; j++) { | |
2979 | 1440 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1441 } |
1442 src += stride; | |
1443 dst += stride; | |
1444 } | |
1445 } | |
2967 | 1446 |
1319 | 1447 static inline void avg_tpel_pixels_mc11_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1448 int i,j; | |
1449 for (i=0; i < height; i++) { | |
1450 for (j=0; j < width; j++) { | |
2979 | 1451 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 | 1452 } |
1453 src += stride; | |
1454 dst += stride; | |
1455 } | |
1456 } | |
1457 | |
1458 static inline void avg_tpel_pixels_mc12_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1459 int i,j; | |
1460 for (i=0; i < height; i++) { | |
1461 for (j=0; j < width; j++) { | |
2979 | 1462 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 | 1463 } |
1464 src += stride; | |
1465 dst += stride; | |
1466 } | |
1467 } | |
1468 | |
1469 static inline void avg_tpel_pixels_mc02_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1470 int i,j; | |
1471 for (i=0; i < height; i++) { | |
1472 for (j=0; j < width; j++) { | |
2979 | 1473 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1474 } |
1475 src += stride; | |
1476 dst += stride; | |
1477 } | |
1478 } | |
1479 | |
1480 static inline void avg_tpel_pixels_mc21_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1481 int i,j; | |
1482 for (i=0; i < height; i++) { | |
1483 for (j=0; j < width; j++) { | |
2979 | 1484 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 | 1485 } |
1486 src += stride; | |
1487 dst += stride; | |
1488 } | |
1489 } | |
1490 | |
1491 static inline void avg_tpel_pixels_mc22_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1492 int i,j; | |
1493 for (i=0; i < height; i++) { | |
1494 for (j=0; j < width; j++) { | |
2979 | 1495 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 | 1496 } |
1497 src += stride; | |
1498 dst += stride; | |
1499 } | |
1500 } | |
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
|
1501 #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
|
1502 #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
|
1503 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
|
1504 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
|
1505 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
|
1506 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
|
1507 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
|
1508 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
|
1509 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
|
1510 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
|
1511 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
|
1512 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
|
1513 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
|
1514 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
|
1515 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
|
1516 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
|
1517 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
|
1518 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
|
1519 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
|
1520 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
|
1521 #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
|
1522 |
1168 | 1523 #define H264_CHROMA_MC(OPNAME, OP)\ |
1524 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){\ | |
1525 const int A=(8-x)*(8-y);\ | |
1526 const int B=( x)*(8-y);\ | |
1527 const int C=(8-x)*( y);\ | |
1528 const int D=( x)*( y);\ | |
1529 int i;\ | |
1530 \ | |
1531 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1532 \ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1533 if(D){\ |
6054 | 1534 for(i=0; i<h; i++){\ |
6053 | 1535 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1536 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1537 dst+= stride;\ | |
1538 src+= stride;\ | |
1539 }\ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1540 }else{\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1541 const int E= B+C;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1542 const int step= C ? stride : 1;\ |
6054 | 1543 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
|
1544 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
|
1545 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
|
1546 dst+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1547 src+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1548 }\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1549 }\ |
1168 | 1550 }\ |
1551 \ | |
1552 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){\ | |
1553 const int A=(8-x)*(8-y);\ | |
1554 const int B=( x)*(8-y);\ | |
1555 const int C=(8-x)*( y);\ | |
1556 const int D=( x)*( y);\ | |
1557 int i;\ | |
1558 \ | |
1559 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1560 \ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1561 if(D){\ |
6054 | 1562 for(i=0; i<h; i++){\ |
6053 | 1563 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1564 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1565 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1566 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1567 dst+= stride;\ | |
1568 src+= stride;\ | |
1569 }\ | |
6052
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1570 }else{\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1571 const int E= B+C;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1572 const int step= C ? stride : 1;\ |
6054 | 1573 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
|
1574 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
|
1575 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
|
1576 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
|
1577 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
|
1578 dst+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1579 src+= stride;\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1580 }\ |
c90798ac28ee
~15% faster h264_chroma_mc2/4_c() these also prevent some possible out
michael
parents:
6051
diff
changeset
|
1581 }\ |
1168 | 1582 }\ |
1583 \ | |
1584 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){\ | |
1585 const int A=(8-x)*(8-y);\ | |
1586 const int B=( x)*(8-y);\ | |
1587 const int C=(8-x)*( y);\ | |
1588 const int D=( x)*( y);\ | |
1589 int i;\ | |
1590 \ | |
1591 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1592 \ | |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1593 if(D){\ |
6054 | 1594 for(i=0; i<h; i++){\ |
6053 | 1595 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ |
1596 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1597 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1598 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1599 OP(dst[4], (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5]));\ | |
1600 OP(dst[5], (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6]));\ | |
1601 OP(dst[6], (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7]));\ | |
1602 OP(dst[7], (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8]));\ | |
1603 dst+= stride;\ | |
1604 src+= stride;\ | |
1605 }\ | |
6051
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1606 }else{\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1607 const int E= B+C;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1608 const int step= C ? stride : 1;\ |
6054 | 1609 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
|
1610 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
|
1611 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
|
1612 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
|
1613 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
|
1614 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
|
1615 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
|
1616 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
|
1617 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
|
1618 dst+= stride;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1619 src+= stride;\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1620 }\ |
1e3b5597505a
30% faster h264_chroma_mc8_c(), this also prevents a possible out of
michael
parents:
6001
diff
changeset
|
1621 }\ |
1168 | 1622 } |
1623 | |
1624 #define op_avg(a, b) a = (((a)+(((b) + 32)>>6)+1)>>1) | |
1625 #define op_put(a, b) a = (((b) + 32)>>6) | |
1626 | |
1627 H264_CHROMA_MC(put_ , op_put) | |
1628 H264_CHROMA_MC(avg_ , op_avg) | |
1629 #undef op_avg | |
1630 #undef op_put | |
1631 | |
9439
ef3a7b711cc0
Rename put_no_rnd_h264_chroma* to reflect its usage in VC1 only
conrad
parents:
9437
diff
changeset
|
1632 static void put_no_rnd_vc1_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){ |
3663 | 1633 const int A=(8-x)*(8-y); |
1634 const int B=( x)*(8-y); | |
1635 const int C=(8-x)*( y); | |
1636 const int D=( x)*( y); | |
1637 int i; | |
1638 | |
1639 assert(x<8 && y<8 && x>=0 && y>=0); | |
1640 | |
1641 for(i=0; i<h; i++) | |
1642 { | |
1643 dst[0] = (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6; | |
1644 dst[1] = (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6; | |
1645 dst[2] = (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6; | |
1646 dst[3] = (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6; | |
1647 dst[4] = (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6; | |
1648 dst[5] = (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6; | |
1649 dst[6] = (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6; | |
1650 dst[7] = (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6; | |
1651 dst+= stride; | |
1652 src+= stride; | |
1653 } | |
1654 } | |
1655 | |
9440 | 1656 static void avg_no_rnd_vc1_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){ |
1657 const int A=(8-x)*(8-y); | |
1658 const int B=( x)*(8-y); | |
1659 const int C=(8-x)*( y); | |
1660 const int D=( x)*( y); | |
1661 int i; | |
1662 | |
1663 assert(x<8 && y<8 && x>=0 && y>=0); | |
1664 | |
1665 for(i=0; i<h; i++) | |
1666 { | |
1667 dst[0] = avg2(dst[0], ((A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6)); | |
1668 dst[1] = avg2(dst[1], ((A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6)); | |
1669 dst[2] = avg2(dst[2], ((A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6)); | |
1670 dst[3] = avg2(dst[3], ((A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6)); | |
1671 dst[4] = avg2(dst[4], ((A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6)); | |
1672 dst[5] = avg2(dst[5], ((A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6)); | |
1673 dst[6] = avg2(dst[6], ((A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6)); | |
1674 dst[7] = avg2(dst[7], ((A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6)); | |
1675 dst+= stride; | |
1676 src+= stride; | |
1677 } | |
1678 } | |
1679 | |
651 | 1680 #define QPEL_MC(r, OPNAME, RND, OP) \ |
1064 | 1681 static void OPNAME ## mpeg4_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
4176 | 1682 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1683 int i;\ |
1684 for(i=0; i<h; i++)\ | |
1685 {\ | |
1686 OP(dst[0], (src[0]+src[1])*20 - (src[0]+src[2])*6 + (src[1]+src[3])*3 - (src[2]+src[4]));\ | |
1687 OP(dst[1], (src[1]+src[2])*20 - (src[0]+src[3])*6 + (src[0]+src[4])*3 - (src[1]+src[5]));\ | |
1688 OP(dst[2], (src[2]+src[3])*20 - (src[1]+src[4])*6 + (src[0]+src[5])*3 - (src[0]+src[6]));\ | |
1689 OP(dst[3], (src[3]+src[4])*20 - (src[2]+src[5])*6 + (src[1]+src[6])*3 - (src[0]+src[7]));\ | |
1690 OP(dst[4], (src[4]+src[5])*20 - (src[3]+src[6])*6 + (src[2]+src[7])*3 - (src[1]+src[8]));\ | |
1691 OP(dst[5], (src[5]+src[6])*20 - (src[4]+src[7])*6 + (src[3]+src[8])*3 - (src[2]+src[8]));\ | |
1692 OP(dst[6], (src[6]+src[7])*20 - (src[5]+src[8])*6 + (src[4]+src[8])*3 - (src[3]+src[7]));\ | |
1693 OP(dst[7], (src[7]+src[8])*20 - (src[6]+src[8])*6 + (src[5]+src[7])*3 - (src[4]+src[6]));\ | |
1694 dst+=dstStride;\ | |
1695 src+=srcStride;\ | |
1696 }\ | |
1697 }\ | |
1698 \ | |
1064 | 1699 static void OPNAME ## mpeg4_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
984 | 1700 const int w=8;\ |
4176 | 1701 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1702 int i;\ |
1703 for(i=0; i<w; i++)\ | |
1704 {\ | |
1705 const int src0= src[0*srcStride];\ | |
1706 const int src1= src[1*srcStride];\ | |
1707 const int src2= src[2*srcStride];\ | |
1708 const int src3= src[3*srcStride];\ | |
1709 const int src4= src[4*srcStride];\ | |
1710 const int src5= src[5*srcStride];\ | |
1711 const int src6= src[6*srcStride];\ | |
1712 const int src7= src[7*srcStride];\ | |
1713 const int src8= src[8*srcStride];\ | |
1714 OP(dst[0*dstStride], (src0+src1)*20 - (src0+src2)*6 + (src1+src3)*3 - (src2+src4));\ | |
1715 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*6 + (src0+src4)*3 - (src1+src5));\ | |
1716 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*6 + (src0+src5)*3 - (src0+src6));\ | |
1717 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*6 + (src1+src6)*3 - (src0+src7));\ | |
1718 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*6 + (src2+src7)*3 - (src1+src8));\ | |
1719 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*6 + (src3+src8)*3 - (src2+src8));\ | |
1720 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*6 + (src4+src8)*3 - (src3+src7));\ | |
1721 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src8)*6 + (src5+src7)*3 - (src4+src6));\ | |
1722 dst++;\ | |
1723 src++;\ | |
1724 }\ | |
1725 }\ | |
1726 \ | |
1064 | 1727 static void OPNAME ## mpeg4_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
4176 | 1728 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1729 int i;\ |
954 | 1730 \ |
651 | 1731 for(i=0; i<h; i++)\ |
1732 {\ | |
1733 OP(dst[ 0], (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]));\ | |
1734 OP(dst[ 1], (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]));\ | |
1735 OP(dst[ 2], (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]));\ | |
1736 OP(dst[ 3], (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]));\ | |
1737 OP(dst[ 4], (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]));\ | |
1738 OP(dst[ 5], (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]));\ | |
1739 OP(dst[ 6], (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]));\ | |
1740 OP(dst[ 7], (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]));\ | |
1741 OP(dst[ 8], (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]));\ | |
1742 OP(dst[ 9], (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]));\ | |
1743 OP(dst[10], (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]));\ | |
1744 OP(dst[11], (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]));\ | |
1745 OP(dst[12], (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]));\ | |
1746 OP(dst[13], (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]));\ | |
1747 OP(dst[14], (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]));\ | |
1748 OP(dst[15], (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]));\ | |
1749 dst+=dstStride;\ | |
1750 src+=srcStride;\ | |
1751 }\ | |
255 | 1752 }\ |
1753 \ | |
1064 | 1754 static void OPNAME ## mpeg4_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
4176 | 1755 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
651 | 1756 int i;\ |
954 | 1757 const int w=16;\ |
651 | 1758 for(i=0; i<w; i++)\ |
1759 {\ | |
1760 const int src0= src[0*srcStride];\ | |
1761 const int src1= src[1*srcStride];\ | |
1762 const int src2= src[2*srcStride];\ | |
1763 const int src3= src[3*srcStride];\ | |
1764 const int src4= src[4*srcStride];\ | |
1765 const int src5= src[5*srcStride];\ | |
1766 const int src6= src[6*srcStride];\ | |
1767 const int src7= src[7*srcStride];\ | |
1768 const int src8= src[8*srcStride];\ | |
1769 const int src9= src[9*srcStride];\ | |
1770 const int src10= src[10*srcStride];\ | |
1771 const int src11= src[11*srcStride];\ | |
1772 const int src12= src[12*srcStride];\ | |
1773 const int src13= src[13*srcStride];\ | |
1774 const int src14= src[14*srcStride];\ | |
1775 const int src15= src[15*srcStride];\ | |
1776 const int src16= src[16*srcStride];\ | |
1777 OP(dst[ 0*dstStride], (src0 +src1 )*20 - (src0 +src2 )*6 + (src1 +src3 )*3 - (src2 +src4 ));\ | |
1778 OP(dst[ 1*dstStride], (src1 +src2 )*20 - (src0 +src3 )*6 + (src0 +src4 )*3 - (src1 +src5 ));\ | |
1779 OP(dst[ 2*dstStride], (src2 +src3 )*20 - (src1 +src4 )*6 + (src0 +src5 )*3 - (src0 +src6 ));\ | |
1780 OP(dst[ 3*dstStride], (src3 +src4 )*20 - (src2 +src5 )*6 + (src1 +src6 )*3 - (src0 +src7 ));\ | |
1781 OP(dst[ 4*dstStride], (src4 +src5 )*20 - (src3 +src6 )*6 + (src2 +src7 )*3 - (src1 +src8 ));\ | |
1782 OP(dst[ 5*dstStride], (src5 +src6 )*20 - (src4 +src7 )*6 + (src3 +src8 )*3 - (src2 +src9 ));\ | |
1783 OP(dst[ 6*dstStride], (src6 +src7 )*20 - (src5 +src8 )*6 + (src4 +src9 )*3 - (src3 +src10));\ | |
1784 OP(dst[ 7*dstStride], (src7 +src8 )*20 - (src6 +src9 )*6 + (src5 +src10)*3 - (src4 +src11));\ | |
1785 OP(dst[ 8*dstStride], (src8 +src9 )*20 - (src7 +src10)*6 + (src6 +src11)*3 - (src5 +src12));\ | |
1786 OP(dst[ 9*dstStride], (src9 +src10)*20 - (src8 +src11)*6 + (src7 +src12)*3 - (src6 +src13));\ | |
1787 OP(dst[10*dstStride], (src10+src11)*20 - (src9 +src12)*6 + (src8 +src13)*3 - (src7 +src14));\ | |
1788 OP(dst[11*dstStride], (src11+src12)*20 - (src10+src13)*6 + (src9 +src14)*3 - (src8 +src15));\ | |
1789 OP(dst[12*dstStride], (src12+src13)*20 - (src11+src14)*6 + (src10+src15)*3 - (src9 +src16));\ | |
1790 OP(dst[13*dstStride], (src13+src14)*20 - (src12+src15)*6 + (src11+src16)*3 - (src10+src16));\ | |
1791 OP(dst[14*dstStride], (src14+src15)*20 - (src13+src16)*6 + (src12+src16)*3 - (src11+src15));\ | |
1792 OP(dst[15*dstStride], (src15+src16)*20 - (src14+src16)*6 + (src13+src15)*3 - (src12+src14));\ | |
1793 dst++;\ | |
1794 src++;\ | |
1795 }\ | |
255 | 1796 }\ |
1797 \ | |
1064 | 1798 static void OPNAME ## qpel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1799 OPNAME ## pixels8_c(dst, src, stride, 8);\ |
255 | 1800 }\ |
1801 \ | |
1064 | 1802 static void OPNAME ## qpel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
1803 uint8_t half[64];\ | |
651 | 1804 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1805 OPNAME ## pixels8_l2(dst, src, half, stride, stride, 8, 8);\ | |
1806 }\ | |
1807 \ | |
1064 | 1808 static void OPNAME ## qpel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 1809 OPNAME ## mpeg4_qpel8_h_lowpass(dst, src, stride, stride, 8);\ |
255 | 1810 }\ |
1811 \ | |
1064 | 1812 static void OPNAME ## qpel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
1813 uint8_t half[64];\ | |
651 | 1814 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1815 OPNAME ## pixels8_l2(dst, src+1, half, stride, stride, 8, 8);\ | |
1816 }\ | |
1817 \ | |
1064 | 1818 static void OPNAME ## qpel8_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
1819 uint8_t full[16*9];\ | |
1820 uint8_t half[64];\ | |
651 | 1821 copy_block9(full, src, 16, stride, 9);\ |
984 | 1822 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1823 OPNAME ## pixels8_l2(dst, full, half, stride, 16, 8, 8);\ |
1824 }\ | |
1825 \ | |
1064 | 1826 static void OPNAME ## qpel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
1827 uint8_t full[16*9];\ | |
651 | 1828 copy_block9(full, src, 16, stride, 9);\ |
984 | 1829 OPNAME ## mpeg4_qpel8_v_lowpass(dst, full, stride, 16);\ |
255 | 1830 }\ |
1831 \ | |
1064 | 1832 static void OPNAME ## qpel8_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
1833 uint8_t full[16*9];\ | |
1834 uint8_t half[64];\ | |
651 | 1835 copy_block9(full, src, 16, stride, 9);\ |
984 | 1836 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1837 OPNAME ## pixels8_l2(dst, full+16, half, stride, 16, 8, 8);\ |
1838 }\ | |
1064 | 1839 void ff_ ## OPNAME ## qpel8_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1840 uint8_t full[16*9];\ | |
1841 uint8_t halfH[72];\ | |
1842 uint8_t halfV[64];\ | |
1843 uint8_t halfHV[64];\ | |
651 | 1844 copy_block9(full, src, 16, stride, 9);\ |
1845 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1846 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1847 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1848 OPNAME ## pixels8_l4(dst, full, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1849 }\ |
1064 | 1850 static void OPNAME ## qpel8_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
1851 uint8_t full[16*9];\ | |
1852 uint8_t halfH[72];\ | |
1853 uint8_t halfHV[64];\ | |
984 | 1854 copy_block9(full, src, 16, stride, 9);\ |
1855 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1856 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1857 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1858 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1859 }\ | |
1064 | 1860 void ff_ ## OPNAME ## qpel8_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1861 uint8_t full[16*9];\ | |
1862 uint8_t halfH[72];\ | |
1863 uint8_t halfV[64];\ | |
1864 uint8_t halfHV[64];\ | |
651 | 1865 copy_block9(full, src, 16, stride, 9);\ |
1866 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1867 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1868 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1869 OPNAME ## pixels8_l4(dst, full+1, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1870 }\ |
1064 | 1871 static void OPNAME ## qpel8_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
1872 uint8_t full[16*9];\ | |
1873 uint8_t halfH[72];\ | |
1874 uint8_t halfHV[64];\ | |
984 | 1875 copy_block9(full, src, 16, stride, 9);\ |
1876 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1877 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1878 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1879 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1880 }\ | |
1064 | 1881 void ff_ ## OPNAME ## qpel8_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1882 uint8_t full[16*9];\ | |
1883 uint8_t halfH[72];\ | |
1884 uint8_t halfV[64];\ | |
1885 uint8_t halfHV[64];\ | |
651 | 1886 copy_block9(full, src, 16, stride, 9);\ |
1887 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1888 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1889 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1890 OPNAME ## pixels8_l4(dst, full+16, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
1891 }\ | |
1064 | 1892 static void OPNAME ## qpel8_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
1893 uint8_t full[16*9];\ | |
1894 uint8_t halfH[72];\ | |
1895 uint8_t halfHV[64];\ | |
984 | 1896 copy_block9(full, src, 16, stride, 9);\ |
1897 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1898 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1899 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1900 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1901 }\ | |
1064 | 1902 void ff_ ## OPNAME ## qpel8_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1903 uint8_t full[16*9];\ | |
1904 uint8_t halfH[72];\ | |
1905 uint8_t halfV[64];\ | |
1906 uint8_t halfHV[64];\ | |
651 | 1907 copy_block9(full, src, 16, stride, 9);\ |
1908 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full , 8, 16, 9);\ | |
984 | 1909 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1910 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1911 OPNAME ## pixels8_l4(dst, full+17, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1912 }\ |
1064 | 1913 static void OPNAME ## qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
1914 uint8_t full[16*9];\ | |
1915 uint8_t halfH[72];\ | |
1916 uint8_t halfHV[64];\ | |
984 | 1917 copy_block9(full, src, 16, stride, 9);\ |
1918 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1919 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1920 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1921 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1922 }\ | |
1064 | 1923 static void OPNAME ## qpel8_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
1924 uint8_t halfH[72];\ | |
1925 uint8_t halfHV[64];\ | |
651 | 1926 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1927 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1928 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ |
1929 }\ | |
1064 | 1930 static void OPNAME ## qpel8_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
1931 uint8_t halfH[72];\ | |
1932 uint8_t halfHV[64];\ | |
651 | 1933 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1934 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1935 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ |
1936 }\ | |
1064 | 1937 void ff_ ## OPNAME ## qpel8_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1938 uint8_t full[16*9];\ | |
1939 uint8_t halfH[72];\ | |
1940 uint8_t halfV[64];\ | |
1941 uint8_t halfHV[64];\ | |
651 | 1942 copy_block9(full, src, 16, stride, 9);\ |
1943 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1944 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1945 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1946 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
255 | 1947 }\ |
1064 | 1948 static void OPNAME ## qpel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
1949 uint8_t full[16*9];\ | |
1950 uint8_t halfH[72];\ | |
984 | 1951 copy_block9(full, src, 16, stride, 9);\ |
1952 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1953 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1954 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1955 }\ | |
1064 | 1956 void ff_ ## OPNAME ## qpel8_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1957 uint8_t full[16*9];\ | |
1958 uint8_t halfH[72];\ | |
1959 uint8_t halfV[64];\ | |
1960 uint8_t halfHV[64];\ | |
651 | 1961 copy_block9(full, src, 16, stride, 9);\ |
1962 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1963 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1964 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1965 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
1966 }\ | |
1064 | 1967 static void OPNAME ## qpel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
1968 uint8_t full[16*9];\ | |
1969 uint8_t halfH[72];\ | |
984 | 1970 copy_block9(full, src, 16, stride, 9);\ |
1971 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1972 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1973 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1974 }\ | |
1064 | 1975 static void OPNAME ## qpel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
1976 uint8_t halfH[72];\ | |
651 | 1977 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1978 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ |
651 | 1979 }\ |
1064 | 1980 static void OPNAME ## qpel16_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1981 OPNAME ## pixels16_c(dst, src, stride, 16);\ |
255 | 1982 }\ |
651 | 1983 \ |
1064 | 1984 static void OPNAME ## qpel16_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
1985 uint8_t half[256];\ | |
651 | 1986 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
1987 OPNAME ## pixels16_l2(dst, src, half, stride, stride, 16, 16);\ | |
1988 }\ | |
1989 \ | |
1064 | 1990 static void OPNAME ## qpel16_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 1991 OPNAME ## mpeg4_qpel16_h_lowpass(dst, src, stride, stride, 16);\ |
1992 }\ | |
1993 \ | |
1064 | 1994 static void OPNAME ## qpel16_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
1995 uint8_t half[256];\ | |
651 | 1996 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
1997 OPNAME ## pixels16_l2(dst, src+1, half, stride, stride, 16, 16);\ | |
1998 }\ | |
1999 \ | |
1064 | 2000 static void OPNAME ## qpel16_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
2001 uint8_t full[24*17];\ | |
2002 uint8_t half[256];\ | |
651 | 2003 copy_block17(full, src, 24, stride, 17);\ |
954 | 2004 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 2005 OPNAME ## pixels16_l2(dst, full, half, stride, 24, 16, 16);\ |
255 | 2006 }\ |
651 | 2007 \ |
1064 | 2008 static void OPNAME ## qpel16_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
2009 uint8_t full[24*17];\ | |
651 | 2010 copy_block17(full, src, 24, stride, 17);\ |
954 | 2011 OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24);\ |
651 | 2012 }\ |
2013 \ | |
1064 | 2014 static void OPNAME ## qpel16_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
2015 uint8_t full[24*17];\ | |
2016 uint8_t half[256];\ | |
651 | 2017 copy_block17(full, src, 24, stride, 17);\ |
954 | 2018 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 2019 OPNAME ## pixels16_l2(dst, full+24, half, stride, 24, 16, 16);\ |
255 | 2020 }\ |
1064 | 2021 void ff_ ## OPNAME ## qpel16_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2022 uint8_t full[24*17];\ | |
2023 uint8_t halfH[272];\ | |
2024 uint8_t halfV[256];\ | |
2025 uint8_t halfHV[256];\ | |
651 | 2026 copy_block17(full, src, 24, stride, 17);\ |
2027 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2028 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
2029 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2030 OPNAME ## pixels16_l4(dst, full, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
2031 }\ | |
1064 | 2032 static void OPNAME ## qpel16_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
2033 uint8_t full[24*17];\ | |
2034 uint8_t halfH[272];\ | |
2035 uint8_t halfHV[256];\ | |
984 | 2036 copy_block17(full, src, 24, stride, 17);\ |
2037 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2038 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
2039 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
2040 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
2041 }\ | |
1064 | 2042 void ff_ ## OPNAME ## qpel16_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2043 uint8_t full[24*17];\ | |
2044 uint8_t halfH[272];\ | |
2045 uint8_t halfV[256];\ | |
2046 uint8_t halfHV[256];\ | |
651 | 2047 copy_block17(full, src, 24, stride, 17);\ |
2048 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2049 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
2050 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2051 OPNAME ## pixels16_l4(dst, full+1, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
2052 }\ | |
1064 | 2053 static void OPNAME ## qpel16_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
2054 uint8_t full[24*17];\ | |
2055 uint8_t halfH[272];\ | |
2056 uint8_t halfHV[256];\ | |
984 | 2057 copy_block17(full, src, 24, stride, 17);\ |
2058 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2059 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2060 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
2061 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
2062 }\ | |
1064 | 2063 void ff_ ## OPNAME ## qpel16_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2064 uint8_t full[24*17];\ | |
2065 uint8_t halfH[272];\ | |
2066 uint8_t halfV[256];\ | |
2067 uint8_t halfHV[256];\ | |
651 | 2068 copy_block17(full, src, 24, stride, 17);\ |
2069 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2070 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
2071 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2072 OPNAME ## pixels16_l4(dst, full+24, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
255 | 2073 }\ |
1064 | 2074 static void OPNAME ## qpel16_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
2075 uint8_t full[24*17];\ | |
2076 uint8_t halfH[272];\ | |
2077 uint8_t halfHV[256];\ | |
984 | 2078 copy_block17(full, src, 24, stride, 17);\ |
2079 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2080 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
2081 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
2082 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
2083 }\ | |
1064 | 2084 void ff_ ## OPNAME ## qpel16_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2085 uint8_t full[24*17];\ | |
2086 uint8_t halfH[272];\ | |
2087 uint8_t halfV[256];\ | |
2088 uint8_t halfHV[256];\ | |
651 | 2089 copy_block17(full, src, 24, stride, 17);\ |
2090 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full , 16, 24, 17);\ | |
954 | 2091 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
2092 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2093 OPNAME ## pixels16_l4(dst, full+25, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
2094 }\ | |
1064 | 2095 static void OPNAME ## qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
2096 uint8_t full[24*17];\ | |
2097 uint8_t halfH[272];\ | |
2098 uint8_t halfHV[256];\ | |
984 | 2099 copy_block17(full, src, 24, stride, 17);\ |
2100 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2101 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2102 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
2103 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
2104 }\ | |
1064 | 2105 static void OPNAME ## qpel16_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
2106 uint8_t halfH[272];\ | |
2107 uint8_t halfHV[256];\ | |
651 | 2108 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2109 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 2110 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ |
255 | 2111 }\ |
1064 | 2112 static void OPNAME ## qpel16_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
2113 uint8_t halfH[272];\ | |
2114 uint8_t halfHV[256];\ | |
651 | 2115 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2116 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 2117 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ |
2118 }\ | |
1064 | 2119 void ff_ ## OPNAME ## qpel16_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2120 uint8_t full[24*17];\ | |
2121 uint8_t halfH[272];\ | |
2122 uint8_t halfV[256];\ | |
2123 uint8_t halfHV[256];\ | |
651 | 2124 copy_block17(full, src, 24, stride, 17);\ |
2125 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2126 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
2127 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2128 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
255 | 2129 }\ |
1064 | 2130 static void OPNAME ## qpel16_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
2131 uint8_t full[24*17];\ | |
2132 uint8_t halfH[272];\ | |
984 | 2133 copy_block17(full, src, 24, stride, 17);\ |
2134 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2135 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
2136 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2137 }\ | |
1064 | 2138 void ff_ ## OPNAME ## qpel16_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2139 uint8_t full[24*17];\ | |
2140 uint8_t halfH[272];\ | |
2141 uint8_t halfV[256];\ | |
2142 uint8_t halfHV[256];\ | |
651 | 2143 copy_block17(full, src, 24, stride, 17);\ |
2144 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2145 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
2146 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2147 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
2148 }\ | |
1064 | 2149 static void OPNAME ## qpel16_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
2150 uint8_t full[24*17];\ | |
2151 uint8_t halfH[272];\ | |
984 | 2152 copy_block17(full, src, 24, stride, 17);\ |
2153 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2154 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2155 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2156 }\ | |
1064 | 2157 static void OPNAME ## qpel16_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
2158 uint8_t halfH[272];\ | |
651 | 2159 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2160 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ |
859 | 2161 } |
255 | 2162 |
651 | 2163 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) |
2164 #define op_avg_no_rnd(a, b) a = (((a)+cm[((b) + 15)>>5])>>1) | |
2165 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2166 #define op_put_no_rnd(a, b) a = cm[((b) + 15)>>5] | |
2167 | |
2168 QPEL_MC(0, put_ , _ , op_put) | |
2169 QPEL_MC(1, put_no_rnd_, _no_rnd_, op_put_no_rnd) | |
2170 QPEL_MC(0, avg_ , _ , op_avg) | |
2171 //QPEL_MC(1, avg_no_rnd , _ , op_avg) | |
2172 #undef op_avg | |
2173 #undef op_avg_no_rnd | |
2174 #undef op_put | |
2175 #undef op_put_no_rnd | |
255 | 2176 |
1168 | 2177 #if 1 |
2178 #define H264_LOWPASS(OPNAME, OP, OP2) \ | |
5151 | 2179 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
|
2180 const int h=2;\ |
4176 | 2181 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
|
2182 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2183 for(i=0; i<h; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2184 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2185 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
|
2186 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
|
2187 dst+=dstStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2188 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2189 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2190 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2191 \ |
5151 | 2192 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
|
2193 const int w=2;\ |
4176 | 2194 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
|
2195 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2196 for(i=0; i<w; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2197 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2198 const int srcB= src[-2*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2199 const int srcA= src[-1*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2200 const int src0= src[0 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2201 const int src1= src[1 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2202 const int src2= src[2 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2203 const int src3= src[3 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2204 const int src4= src[4 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2205 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
|
2206 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
|
2207 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2208 src++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2209 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2210 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2211 \ |
5151 | 2212 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
|
2213 const int h=2;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2214 const int w=2;\ |
4176 | 2215 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
|
2216 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2217 src -= 2*srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2218 for(i=0; i<h+5; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2219 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2220 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
|
2221 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
|
2222 tmp+=tmpStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2223 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2224 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2225 tmp -= tmpStride*(h+5-2);\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2226 for(i=0; i<w; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2227 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2228 const int tmpB= tmp[-2*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2229 const int tmpA= tmp[-1*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2230 const int tmp0= tmp[0 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2231 const int tmp1= tmp[1 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2232 const int tmp2= tmp[2 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2233 const int tmp3= tmp[3 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2234 const int tmp4= tmp[4 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2235 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
|
2236 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
|
2237 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2238 tmp++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2239 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2240 }\ |
1168 | 2241 static void OPNAME ## h264_qpel4_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
2242 const int h=4;\ | |
4176 | 2243 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2244 int i;\ |
2245 for(i=0; i<h; i++)\ | |
2246 {\ | |
2247 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]));\ | |
2248 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]));\ | |
2249 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]));\ | |
2250 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]));\ | |
2251 dst+=dstStride;\ | |
2252 src+=srcStride;\ | |
2253 }\ | |
2254 }\ | |
2255 \ | |
2256 static void OPNAME ## h264_qpel4_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2257 const int w=4;\ | |
4176 | 2258 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2259 int i;\ |
2260 for(i=0; i<w; i++)\ | |
2261 {\ | |
2262 const int srcB= src[-2*srcStride];\ | |
2263 const int srcA= src[-1*srcStride];\ | |
2264 const int src0= src[0 *srcStride];\ | |
2265 const int src1= src[1 *srcStride];\ | |
2266 const int src2= src[2 *srcStride];\ | |
2267 const int src3= src[3 *srcStride];\ | |
2268 const int src4= src[4 *srcStride];\ | |
2269 const int src5= src[5 *srcStride];\ | |
2270 const int src6= src[6 *srcStride];\ | |
2271 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2272 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2273 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2274 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2275 dst++;\ | |
2276 src++;\ | |
2277 }\ | |
2278 }\ | |
2279 \ | |
2280 static void OPNAME ## h264_qpel4_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2281 const int h=4;\ | |
2282 const int w=4;\ | |
4176 | 2283 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2284 int i;\ |
2285 src -= 2*srcStride;\ | |
2286 for(i=0; i<h+5; i++)\ | |
2287 {\ | |
2288 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]);\ | |
2289 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]);\ | |
2290 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]);\ | |
2291 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]);\ | |
2292 tmp+=tmpStride;\ | |
2293 src+=srcStride;\ | |
2294 }\ | |
2295 tmp -= tmpStride*(h+5-2);\ | |
2296 for(i=0; i<w; i++)\ | |
2297 {\ | |
2298 const int tmpB= tmp[-2*tmpStride];\ | |
2299 const int tmpA= tmp[-1*tmpStride];\ | |
2300 const int tmp0= tmp[0 *tmpStride];\ | |
2301 const int tmp1= tmp[1 *tmpStride];\ | |
2302 const int tmp2= tmp[2 *tmpStride];\ | |
2303 const int tmp3= tmp[3 *tmpStride];\ | |
2304 const int tmp4= tmp[4 *tmpStride];\ | |
2305 const int tmp5= tmp[5 *tmpStride];\ | |
2306 const int tmp6= tmp[6 *tmpStride];\ | |
2307 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2308 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2309 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2310 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2311 dst++;\ | |
2312 tmp++;\ | |
2313 }\ | |
2314 }\ | |
2315 \ | |
2316 static void OPNAME ## h264_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2317 const int h=8;\ | |
4176 | 2318 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2319 int i;\ |
2320 for(i=0; i<h; i++)\ | |
2321 {\ | |
2322 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]));\ | |
2323 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]));\ | |
2324 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]));\ | |
2325 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]));\ | |
2326 OP(dst[4], (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]));\ | |
2327 OP(dst[5], (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]));\ | |
2328 OP(dst[6], (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]));\ | |
2329 OP(dst[7], (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]));\ | |
2330 dst+=dstStride;\ | |
2331 src+=srcStride;\ | |
2332 }\ | |
2333 }\ | |
2334 \ | |
2335 static void OPNAME ## h264_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2336 const int w=8;\ | |
4176 | 2337 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2338 int i;\ |
2339 for(i=0; i<w; i++)\ | |
2340 {\ | |
2341 const int srcB= src[-2*srcStride];\ | |
2342 const int srcA= src[-1*srcStride];\ | |
2343 const int src0= src[0 *srcStride];\ | |
2344 const int src1= src[1 *srcStride];\ | |
2345 const int src2= src[2 *srcStride];\ | |
2346 const int src3= src[3 *srcStride];\ | |
2347 const int src4= src[4 *srcStride];\ | |
2348 const int src5= src[5 *srcStride];\ | |
2349 const int src6= src[6 *srcStride];\ | |
2350 const int src7= src[7 *srcStride];\ | |
2351 const int src8= src[8 *srcStride];\ | |
2352 const int src9= src[9 *srcStride];\ | |
2353 const int src10=src[10*srcStride];\ | |
2354 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2355 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2356 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2357 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2358 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*5 + (src2+src7));\ | |
2359 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*5 + (src3+src8));\ | |
2360 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*5 + (src4+src9));\ | |
2361 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src9)*5 + (src5+src10));\ | |
2362 dst++;\ | |
2363 src++;\ | |
2364 }\ | |
2365 }\ | |
2366 \ | |
2367 static void OPNAME ## h264_qpel8_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2368 const int h=8;\ | |
2369 const int w=8;\ | |
4176 | 2370 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;\ |
1168 | 2371 int i;\ |
2372 src -= 2*srcStride;\ | |
2373 for(i=0; i<h+5; i++)\ | |
2374 {\ | |
2375 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]);\ | |
2376 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]);\ | |
2377 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]);\ | |
2378 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]);\ | |
2379 tmp[4]= (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]);\ | |
2380 tmp[5]= (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]);\ | |
2381 tmp[6]= (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]);\ | |
2382 tmp[7]= (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]);\ | |
2383 tmp+=tmpStride;\ | |
2384 src+=srcStride;\ | |
2385 }\ | |
2386 tmp -= tmpStride*(h+5-2);\ | |
2387 for(i=0; i<w; i++)\ | |
2388 {\ | |
2389 const int tmpB= tmp[-2*tmpStride];\ | |
2390 const int tmpA= tmp[-1*tmpStride];\ | |
2391 const int tmp0= tmp[0 *tmpStride];\ | |
2392 const int tmp1= tmp[1 *tmpStride];\ | |
2393 const int tmp2= tmp[2 *tmpStride];\ | |
2394 const int tmp3= tmp[3 *tmpStride];\ | |
2395 const int tmp4= tmp[4 *tmpStride];\ | |
2396 const int tmp5= tmp[5 *tmpStride];\ | |
2397 const int tmp6= tmp[6 *tmpStride];\ | |
2398 const int tmp7= tmp[7 *tmpStride];\ | |
2399 const int tmp8= tmp[8 *tmpStride];\ | |
2400 const int tmp9= tmp[9 *tmpStride];\ | |
2401 const int tmp10=tmp[10*tmpStride];\ | |
2402 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2403 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2404 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2405 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2406 OP2(dst[4*dstStride], (tmp4+tmp5)*20 - (tmp3+tmp6)*5 + (tmp2+tmp7));\ | |
2407 OP2(dst[5*dstStride], (tmp5+tmp6)*20 - (tmp4+tmp7)*5 + (tmp3+tmp8));\ | |
2408 OP2(dst[6*dstStride], (tmp6+tmp7)*20 - (tmp5+tmp8)*5 + (tmp4+tmp9));\ | |
2409 OP2(dst[7*dstStride], (tmp7+tmp8)*20 - (tmp6+tmp9)*5 + (tmp5+tmp10));\ | |
2410 dst++;\ | |
2411 tmp++;\ | |
2412 }\ | |
2413 }\ | |
2414 \ | |
2415 static void OPNAME ## h264_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2416 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2417 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2418 src += 8*srcStride;\ | |
2419 dst += 8*dstStride;\ | |
2420 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2421 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2422 }\ | |
2423 \ | |
2424 static void OPNAME ## h264_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2425 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2426 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2427 src += 8*srcStride;\ | |
2428 dst += 8*dstStride;\ | |
2429 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2430 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2431 }\ | |
2432 \ | |
2433 static void OPNAME ## h264_qpel16_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2434 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2435 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2436 src += 8*srcStride;\ | |
2437 dst += 8*dstStride;\ | |
2438 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2439 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2440 }\ | |
2441 | |
2442 #define H264_MC(OPNAME, SIZE) \ | |
2443 static void OPNAME ## h264_qpel ## SIZE ## _mc00_c (uint8_t *dst, uint8_t *src, int stride){\ | |
2444 OPNAME ## pixels ## SIZE ## _c(dst, src, stride, SIZE);\ | |
2445 }\ | |
2446 \ | |
2447 static void OPNAME ## h264_qpel ## SIZE ## _mc10_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2448 uint8_t half[SIZE*SIZE];\ | |
2449 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2450 OPNAME ## pixels ## SIZE ## _l2(dst, src, half, stride, stride, SIZE, SIZE);\ | |
2451 }\ | |
2452 \ | |
2453 static void OPNAME ## h264_qpel ## SIZE ## _mc20_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2454 OPNAME ## h264_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride);\ | |
2455 }\ | |
2456 \ | |
2457 static void OPNAME ## h264_qpel ## SIZE ## _mc30_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2458 uint8_t half[SIZE*SIZE];\ | |
2459 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2460 OPNAME ## pixels ## SIZE ## _l2(dst, src+1, half, stride, stride, SIZE, SIZE);\ | |
2461 }\ | |
2462 \ | |
2463 static void OPNAME ## h264_qpel ## SIZE ## _mc01_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2464 uint8_t full[SIZE*(SIZE+5)];\ | |
2465 uint8_t * const full_mid= full + SIZE*2;\ | |
2466 uint8_t half[SIZE*SIZE];\ | |
2467 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2468 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2469 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid, half, stride, SIZE, SIZE, SIZE);\ | |
2470 }\ | |
2471 \ | |
2472 static void OPNAME ## h264_qpel ## SIZE ## _mc02_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2473 uint8_t full[SIZE*(SIZE+5)];\ | |
2474 uint8_t * const full_mid= full + SIZE*2;\ | |
2475 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2476 OPNAME ## h264_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE);\ | |
2477 }\ | |
2478 \ | |
2479 static void OPNAME ## h264_qpel ## SIZE ## _mc03_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2480 uint8_t full[SIZE*(SIZE+5)];\ | |
2481 uint8_t * const full_mid= full + SIZE*2;\ | |
2482 uint8_t half[SIZE*SIZE];\ | |
2483 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2484 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2485 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid+SIZE, half, stride, SIZE, SIZE, SIZE);\ | |
2486 }\ | |
2487 \ | |
2488 static void OPNAME ## h264_qpel ## SIZE ## _mc11_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2489 uint8_t full[SIZE*(SIZE+5)];\ | |
2490 uint8_t * const full_mid= full + SIZE*2;\ | |
2491 uint8_t halfH[SIZE*SIZE];\ | |
2492 uint8_t halfV[SIZE*SIZE];\ | |
2493 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2494 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2495 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2496 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2497 }\ | |
2498 \ | |
2499 static void OPNAME ## h264_qpel ## SIZE ## _mc31_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2500 uint8_t full[SIZE*(SIZE+5)];\ | |
2501 uint8_t * const full_mid= full + SIZE*2;\ | |
2502 uint8_t halfH[SIZE*SIZE];\ | |
2503 uint8_t halfV[SIZE*SIZE];\ | |
2504 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2505 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2506 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2507 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2508 }\ | |
2509 \ | |
2510 static void OPNAME ## h264_qpel ## SIZE ## _mc13_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2511 uint8_t full[SIZE*(SIZE+5)];\ | |
2512 uint8_t * const full_mid= full + SIZE*2;\ | |
2513 uint8_t halfH[SIZE*SIZE];\ | |
2514 uint8_t halfV[SIZE*SIZE];\ | |
2515 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2516 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2517 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2518 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2519 }\ | |
2520 \ | |
2521 static void OPNAME ## h264_qpel ## SIZE ## _mc33_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2522 uint8_t full[SIZE*(SIZE+5)];\ | |
2523 uint8_t * const full_mid= full + SIZE*2;\ | |
2524 uint8_t halfH[SIZE*SIZE];\ | |
2525 uint8_t halfV[SIZE*SIZE];\ | |
2526 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2527 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2528 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2529 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2530 }\ | |
2531 \ | |
2532 static void OPNAME ## h264_qpel ## SIZE ## _mc22_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2533 int16_t tmp[SIZE*(SIZE+5)];\ | |
2534 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass(dst, tmp, src, stride, SIZE, stride);\ | |
2535 }\ | |
2536 \ | |
2537 static void OPNAME ## h264_qpel ## SIZE ## _mc21_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2538 int16_t tmp[SIZE*(SIZE+5)];\ | |
2539 uint8_t halfH[SIZE*SIZE];\ | |
2540 uint8_t halfHV[SIZE*SIZE];\ | |
2541 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2542 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2543 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2544 }\ | |
2545 \ | |
2546 static void OPNAME ## h264_qpel ## SIZE ## _mc23_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2547 int16_t tmp[SIZE*(SIZE+5)];\ | |
2548 uint8_t halfH[SIZE*SIZE];\ | |
2549 uint8_t halfHV[SIZE*SIZE];\ | |
2550 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2551 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2552 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2553 }\ | |
2554 \ | |
2555 static void OPNAME ## h264_qpel ## SIZE ## _mc12_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2556 uint8_t full[SIZE*(SIZE+5)];\ | |
2557 uint8_t * const full_mid= full + SIZE*2;\ | |
2558 int16_t tmp[SIZE*(SIZE+5)];\ | |
2559 uint8_t halfV[SIZE*SIZE];\ | |
2560 uint8_t halfHV[SIZE*SIZE];\ | |
2561 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2562 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2563 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2564 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2565 }\ | |
2566 \ | |
2567 static void OPNAME ## h264_qpel ## SIZE ## _mc32_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2568 uint8_t full[SIZE*(SIZE+5)];\ | |
2569 uint8_t * const full_mid= full + SIZE*2;\ | |
2570 int16_t tmp[SIZE*(SIZE+5)];\ | |
2571 uint8_t halfV[SIZE*SIZE];\ | |
2572 uint8_t halfHV[SIZE*SIZE];\ | |
2573 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2574 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2575 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2576 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2577 }\ | |
2578 | |
2579 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) | |
2580 //#define op_avg2(a, b) a = (((a)*w1+cm[((b) + 16)>>5]*w2 + o + 64)>>7) | |
2581 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2582 #define op2_avg(a, b) a = (((a)+cm[((b) + 512)>>10]+1)>>1) | |
2583 #define op2_put(a, b) a = cm[((b) + 512)>>10] | |
2584 | |
2585 H264_LOWPASS(put_ , op_put, op2_put) | |
2586 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
|
2587 H264_MC(put_, 2) |
1168 | 2588 H264_MC(put_, 4) |
2589 H264_MC(put_, 8) | |
2590 H264_MC(put_, 16) | |
2591 H264_MC(avg_, 4) | |
2592 H264_MC(avg_, 8) | |
2593 H264_MC(avg_, 16) | |
2594 | |
2595 #undef op_avg | |
2596 #undef op_put | |
2597 #undef op2_avg | |
2598 #undef op2_put | |
2599 #endif | |
2600 | |
936 | 2601 static void wmv2_mspel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){ |
4176 | 2602 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
936 | 2603 int i; |
2604 | |
2605 for(i=0; i<h; i++){ | |
2606 dst[0]= cm[(9*(src[0] + src[1]) - (src[-1] + src[2]) + 8)>>4]; | |
2607 dst[1]= cm[(9*(src[1] + src[2]) - (src[ 0] + src[3]) + 8)>>4]; | |
2608 dst[2]= cm[(9*(src[2] + src[3]) - (src[ 1] + src[4]) + 8)>>4]; | |
2609 dst[3]= cm[(9*(src[3] + src[4]) - (src[ 2] + src[5]) + 8)>>4]; | |
2610 dst[4]= cm[(9*(src[4] + src[5]) - (src[ 3] + src[6]) + 8)>>4]; | |
2611 dst[5]= cm[(9*(src[5] + src[6]) - (src[ 4] + src[7]) + 8)>>4]; | |
2612 dst[6]= cm[(9*(src[6] + src[7]) - (src[ 5] + src[8]) + 8)>>4]; | |
2613 dst[7]= cm[(9*(src[7] + src[8]) - (src[ 6] + src[9]) + 8)>>4]; | |
2614 dst+=dstStride; | |
2967 | 2615 src+=srcStride; |
936 | 2616 } |
2617 } | |
2618 | |
8590 | 2619 #if CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2620 /* AVS specific */ |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2621 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
|
2622 put_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2623 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2624 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
|
2625 avg_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2626 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2627 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
|
2628 put_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2629 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2630 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
|
2631 avg_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2632 } |
3432 | 2633 #endif /* CONFIG_CAVS_DECODER */ |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2634 |
9995
3141f69e3905
Do not check for both CONFIG_VC1_DECODER and CONFIG_WMV3_DECODER,
diego
parents:
9975
diff
changeset
|
2635 #if CONFIG_VC1_DECODER |
3526 | 2636 /* VC-1 specific */ |
11378
f46b68960464
Move some VC1 dsp prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11376
diff
changeset
|
2637 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { |
3526 | 2638 put_pixels8_c(dst, src, stride, 8); |
2639 } | |
11378
f46b68960464
Move some VC1 dsp prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11376
diff
changeset
|
2640 void ff_avg_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { |
9437 | 2641 avg_pixels8_c(dst, src, stride, 8); |
2642 } | |
9995
3141f69e3905
Do not check for both CONFIG_VC1_DECODER and CONFIG_WMV3_DECODER,
diego
parents:
9975
diff
changeset
|
2643 #endif /* CONFIG_VC1_DECODER */ |
3526 | 2644 |
8590 | 2645 #if CONFIG_RV40_DECODER |
8232 | 2646 static void put_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride){ |
2647 put_pixels16_xy2_c(dst, src, stride, 16); | |
2648 } | |
2649 static void avg_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride){ | |
2650 avg_pixels16_xy2_c(dst, src, stride, 16); | |
2651 } | |
2652 static void put_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride){ | |
2653 put_pixels8_xy2_c(dst, src, stride, 8); | |
2654 } | |
2655 static void avg_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride){ | |
2656 avg_pixels8_xy2_c(dst, src, stride, 8); | |
2657 } | |
2658 #endif /* CONFIG_RV40_DECODER */ | |
2659 | |
936 | 2660 static void wmv2_mspel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int w){ |
4176 | 2661 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
936 | 2662 int i; |
2663 | |
2664 for(i=0; i<w; i++){ | |
2665 const int src_1= src[ -srcStride]; | |
2666 const int src0 = src[0 ]; | |
2667 const int src1 = src[ srcStride]; | |
2668 const int src2 = src[2*srcStride]; | |
2669 const int src3 = src[3*srcStride]; | |
2670 const int src4 = src[4*srcStride]; | |
2671 const int src5 = src[5*srcStride]; | |
2672 const int src6 = src[6*srcStride]; | |
2673 const int src7 = src[7*srcStride]; | |
2674 const int src8 = src[8*srcStride]; | |
2675 const int src9 = src[9*srcStride]; | |
2676 dst[0*dstStride]= cm[(9*(src0 + src1) - (src_1 + src2) + 8)>>4]; | |
2677 dst[1*dstStride]= cm[(9*(src1 + src2) - (src0 + src3) + 8)>>4]; | |
2678 dst[2*dstStride]= cm[(9*(src2 + src3) - (src1 + src4) + 8)>>4]; | |
2679 dst[3*dstStride]= cm[(9*(src3 + src4) - (src2 + src5) + 8)>>4]; | |
2680 dst[4*dstStride]= cm[(9*(src4 + src5) - (src3 + src6) + 8)>>4]; | |
2681 dst[5*dstStride]= cm[(9*(src5 + src6) - (src4 + src7) + 8)>>4]; | |
2682 dst[6*dstStride]= cm[(9*(src6 + src7) - (src5 + src8) + 8)>>4]; | |
2683 dst[7*dstStride]= cm[(9*(src7 + src8) - (src6 + src9) + 8)>>4]; | |
2684 src++; | |
2685 dst++; | |
2686 } | |
2687 } | |
2688 | |
2689 static void put_mspel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){ | |
2690 put_pixels8_c(dst, src, stride, 8); | |
2691 } | |
2692 | |
2693 static void put_mspel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){ | |
2694 uint8_t half[64]; | |
2695 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2696 put_pixels8_l2(dst, src, half, stride, stride, 8, 8); | |
2697 } | |
2698 | |
2699 static void put_mspel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){ | |
2700 wmv2_mspel8_h_lowpass(dst, src, stride, stride, 8); | |
2701 } | |
2702 | |
2703 static void put_mspel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){ | |
2704 uint8_t half[64]; | |
2705 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2706 put_pixels8_l2(dst, src+1, half, stride, stride, 8, 8); | |
2707 } | |
2708 | |
2709 static void put_mspel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){ | |
2710 wmv2_mspel8_v_lowpass(dst, src, stride, stride, 8); | |
2711 } | |
2712 | |
2713 static void put_mspel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){ | |
2714 uint8_t halfH[88]; | |
2715 uint8_t halfV[64]; | |
2716 uint8_t halfHV[64]; | |
2717 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2718 wmv2_mspel8_v_lowpass(halfV, src, 8, stride, 8); | |
2719 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2720 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2721 } | |
2722 static void put_mspel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){ | |
2723 uint8_t halfH[88]; | |
2724 uint8_t halfV[64]; | |
2725 uint8_t halfHV[64]; | |
2726 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2727 wmv2_mspel8_v_lowpass(halfV, src+1, 8, stride, 8); | |
2728 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2729 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2730 } | |
2731 static void put_mspel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){ | |
2732 uint8_t halfH[88]; | |
2733 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2734 wmv2_mspel8_v_lowpass(dst, halfH+8, stride, 8, 8); | |
2735 } | |
2736 | |
1644 | 2737 static void h263_v_loop_filter_c(uint8_t *src, int stride, int qscale){ |
10749
5cca4b6c459d
Get rid of pointless CONFIG_ANY_H263 preprocessor definition.
diego
parents:
10748
diff
changeset
|
2738 if(CONFIG_H263_DECODER || CONFIG_H263_ENCODER) { |
1644 | 2739 int x; |
2740 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2741 |
1644 | 2742 for(x=0; x<8; x++){ |
2743 int d1, d2, ad1; | |
2744 int p0= src[x-2*stride]; | |
2745 int p1= src[x-1*stride]; | |
2746 int p2= src[x+0*stride]; | |
2747 int p3= src[x+1*stride]; | |
2748 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2749 | |
2750 if (d<-2*strength) d1= 0; | |
2751 else if(d<- strength) d1=-2*strength - d; | |
2752 else if(d< strength) d1= d; | |
2753 else if(d< 2*strength) d1= 2*strength - d; | |
2754 else d1= 0; | |
2967 | 2755 |
1644 | 2756 p1 += d1; |
2757 p2 -= d1; | |
2758 if(p1&256) p1= ~(p1>>31); | |
2759 if(p2&256) p2= ~(p2>>31); | |
2967 | 2760 |
1644 | 2761 src[x-1*stride] = p1; |
2762 src[x+0*stride] = p2; | |
2763 | |
4001 | 2764 ad1= FFABS(d1)>>1; |
2967 | 2765 |
4594 | 2766 d2= av_clip((p0-p3)/4, -ad1, ad1); |
2967 | 2767 |
1644 | 2768 src[x-2*stride] = p0 - d2; |
2769 src[x+ stride] = p3 + d2; | |
2770 } | |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2771 } |
1644 | 2772 } |
2773 | |
2774 static void h263_h_loop_filter_c(uint8_t *src, int stride, int qscale){ | |
10749
5cca4b6c459d
Get rid of pointless CONFIG_ANY_H263 preprocessor definition.
diego
parents:
10748
diff
changeset
|
2775 if(CONFIG_H263_DECODER || CONFIG_H263_ENCODER) { |
1644 | 2776 int y; |
2777 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2778 |
1644 | 2779 for(y=0; y<8; y++){ |
2780 int d1, d2, ad1; | |
2781 int p0= src[y*stride-2]; | |
2782 int p1= src[y*stride-1]; | |
2783 int p2= src[y*stride+0]; | |
2784 int p3= src[y*stride+1]; | |
2785 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2786 | |
2787 if (d<-2*strength) d1= 0; | |
2788 else if(d<- strength) d1=-2*strength - d; | |
2789 else if(d< strength) d1= d; | |
2790 else if(d< 2*strength) d1= 2*strength - d; | |
2791 else d1= 0; | |
2967 | 2792 |
1644 | 2793 p1 += d1; |
2794 p2 -= d1; | |
2795 if(p1&256) p1= ~(p1>>31); | |
2796 if(p2&256) p2= ~(p2>>31); | |
2967 | 2797 |
1644 | 2798 src[y*stride-1] = p1; |
2799 src[y*stride+0] = p2; | |
2800 | |
4001 | 2801 ad1= FFABS(d1)>>1; |
2967 | 2802 |
4594 | 2803 d2= av_clip((p0-p3)/4, -ad1, ad1); |
2967 | 2804 |
1644 | 2805 src[y*stride-2] = p0 - d2; |
2806 src[y*stride+1] = p3 + d2; | |
2807 } | |
5394
e9a6215f4e3a
help some gcc version to optimize out those functions
aurel
parents:
5291
diff
changeset
|
2808 } |
1644 | 2809 } |
936 | 2810 |
2045 | 2811 static void h261_loop_filter_c(uint8_t *src, int stride){ |
2812 int x,y,xy,yz; | |
2813 int temp[64]; | |
2814 | |
2815 for(x=0; x<8; x++){ | |
2816 temp[x ] = 4*src[x ]; | |
2817 temp[x + 7*8] = 4*src[x + 7*stride]; | |
2818 } | |
2819 for(y=1; y<7; y++){ | |
2820 for(x=0; x<8; x++){ | |
2821 xy = y * stride + x; | |
2822 yz = y * 8 + x; | |
2823 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
|
2824 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2825 } |
2967 | 2826 |
2045 | 2827 for(y=0; y<8; y++){ |
2828 src[ y*stride] = (temp[ y*8] + 2)>>2; | |
2829 src[7+y*stride] = (temp[7+y*8] + 2)>>2; | |
2830 for(x=1; x<7; x++){ | |
2831 xy = y * stride + x; | |
2832 yz = y * 8 + x; | |
2833 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
|
2834 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2835 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2836 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2837 |
1708 | 2838 static inline int pix_abs16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2839 { |
2840 int s, i; | |
2841 | |
2842 s = 0; | |
1708 | 2843 for(i=0;i<h;i++) { |
0 | 2844 s += abs(pix1[0] - pix2[0]); |
2845 s += abs(pix1[1] - pix2[1]); | |
2846 s += abs(pix1[2] - pix2[2]); | |
2847 s += abs(pix1[3] - pix2[3]); | |
2848 s += abs(pix1[4] - pix2[4]); | |
2849 s += abs(pix1[5] - pix2[5]); | |
2850 s += abs(pix1[6] - pix2[6]); | |
2851 s += abs(pix1[7] - pix2[7]); | |
2852 s += abs(pix1[8] - pix2[8]); | |
2853 s += abs(pix1[9] - pix2[9]); | |
2854 s += abs(pix1[10] - pix2[10]); | |
2855 s += abs(pix1[11] - pix2[11]); | |
2856 s += abs(pix1[12] - pix2[12]); | |
2857 s += abs(pix1[13] - pix2[13]); | |
2858 s += abs(pix1[14] - pix2[14]); | |
2859 s += abs(pix1[15] - pix2[15]); | |
2860 pix1 += line_size; | |
2861 pix2 += line_size; | |
2862 } | |
2863 return s; | |
2864 } | |
2865 | |
1708 | 2866 static int pix_abs16_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2867 { |
2868 int s, i; | |
2869 | |
2870 s = 0; | |
1708 | 2871 for(i=0;i<h;i++) { |
0 | 2872 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
2873 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
2874 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
2875 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
2876 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
2877 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
2878 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
2879 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
2880 s += abs(pix1[8] - avg2(pix2[8], pix2[9])); | |
2881 s += abs(pix1[9] - avg2(pix2[9], pix2[10])); | |
2882 s += abs(pix1[10] - avg2(pix2[10], pix2[11])); | |
2883 s += abs(pix1[11] - avg2(pix2[11], pix2[12])); | |
2884 s += abs(pix1[12] - avg2(pix2[12], pix2[13])); | |
2885 s += abs(pix1[13] - avg2(pix2[13], pix2[14])); | |
2886 s += abs(pix1[14] - avg2(pix2[14], pix2[15])); | |
2887 s += abs(pix1[15] - avg2(pix2[15], pix2[16])); | |
2888 pix1 += line_size; | |
2889 pix2 += line_size; | |
2890 } | |
2891 return s; | |
2892 } | |
2893 | |
1708 | 2894 static int pix_abs16_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2895 { |
2896 int s, i; | |
1064 | 2897 uint8_t *pix3 = pix2 + line_size; |
0 | 2898 |
2899 s = 0; | |
1708 | 2900 for(i=0;i<h;i++) { |
0 | 2901 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
2902 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
2903 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
2904 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
2905 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
2906 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
2907 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
2908 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
2909 s += abs(pix1[8] - avg2(pix2[8], pix3[8])); | |
2910 s += abs(pix1[9] - avg2(pix2[9], pix3[9])); | |
2911 s += abs(pix1[10] - avg2(pix2[10], pix3[10])); | |
2912 s += abs(pix1[11] - avg2(pix2[11], pix3[11])); | |
2913 s += abs(pix1[12] - avg2(pix2[12], pix3[12])); | |
2914 s += abs(pix1[13] - avg2(pix2[13], pix3[13])); | |
2915 s += abs(pix1[14] - avg2(pix2[14], pix3[14])); | |
2916 s += abs(pix1[15] - avg2(pix2[15], pix3[15])); | |
2917 pix1 += line_size; | |
2918 pix2 += line_size; | |
2919 pix3 += line_size; | |
2920 } | |
2921 return s; | |
2922 } | |
2923 | |
1708 | 2924 static int pix_abs16_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2925 { |
2926 int s, i; | |
1064 | 2927 uint8_t *pix3 = pix2 + line_size; |
0 | 2928 |
2929 s = 0; | |
1708 | 2930 for(i=0;i<h;i++) { |
0 | 2931 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
2932 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
2933 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
2934 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
2935 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
2936 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
2937 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
2938 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
2939 s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9])); | |
2940 s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10])); | |
2941 s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11])); | |
2942 s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12])); | |
2943 s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13])); | |
2944 s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14])); | |
2945 s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15])); | |
2946 s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16])); | |
2947 pix1 += line_size; | |
2948 pix2 += line_size; | |
2949 pix3 += line_size; | |
2950 } | |
2951 return s; | |
2952 } | |
2953 | |
1708 | 2954 static inline int pix_abs8_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 2955 { |
2956 int s, i; | |
2957 | |
2958 s = 0; | |
1708 | 2959 for(i=0;i<h;i++) { |
294 | 2960 s += abs(pix1[0] - pix2[0]); |
2961 s += abs(pix1[1] - pix2[1]); | |
2962 s += abs(pix1[2] - pix2[2]); | |
2963 s += abs(pix1[3] - pix2[3]); | |
2964 s += abs(pix1[4] - pix2[4]); | |
2965 s += abs(pix1[5] - pix2[5]); | |
2966 s += abs(pix1[6] - pix2[6]); | |
2967 s += abs(pix1[7] - pix2[7]); | |
2968 pix1 += line_size; | |
2969 pix2 += line_size; | |
2970 } | |
2971 return s; | |
2972 } | |
2973 | |
1708 | 2974 static int pix_abs8_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 2975 { |
2976 int s, i; | |
2977 | |
2978 s = 0; | |
1708 | 2979 for(i=0;i<h;i++) { |
294 | 2980 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
2981 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
2982 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
2983 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
2984 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
2985 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
2986 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
2987 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
2988 pix1 += line_size; | |
2989 pix2 += line_size; | |
2990 } | |
2991 return s; | |
2992 } | |
2993 | |
1708 | 2994 static int pix_abs8_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 2995 { |
2996 int s, i; | |
1064 | 2997 uint8_t *pix3 = pix2 + line_size; |
294 | 2998 |
2999 s = 0; | |
1708 | 3000 for(i=0;i<h;i++) { |
294 | 3001 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
3002 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
3003 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
3004 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
3005 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
3006 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
3007 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
3008 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
3009 pix1 += line_size; | |
3010 pix2 += line_size; | |
3011 pix3 += line_size; | |
3012 } | |
3013 return s; | |
3014 } | |
3015 | |
1708 | 3016 static int pix_abs8_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3017 { |
3018 int s, i; | |
1064 | 3019 uint8_t *pix3 = pix2 + line_size; |
294 | 3020 |
3021 s = 0; | |
1708 | 3022 for(i=0;i<h;i++) { |
294 | 3023 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
3024 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
3025 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
3026 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
3027 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
3028 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
3029 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
3030 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
3031 pix1 += line_size; | |
3032 pix2 += line_size; | |
3033 pix3 += line_size; | |
3034 } | |
3035 return s; | |
3036 } | |
3037 | |
2834 | 3038 static int nsse16_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3039 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3040 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3041 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3042 int x,y; |
2066 | 3043 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3044 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3045 for(x=0; x<16; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3046 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3047 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3048 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3049 for(x=0; x<15; x++){ |
4001 | 3050 score2+= FFABS( s1[x ] - s1[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3051 - s1[x+1] + s1[x+1+stride]) |
4001 | 3052 -FFABS( s2[x ] - s2[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3053 - s2[x+1] + s2[x+1+stride]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3054 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3055 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3056 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3057 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3058 } |
2066 | 3059 |
4001 | 3060 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight; |
3061 else return score1 + FFABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3062 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3063 |
2834 | 3064 static int nsse8_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3065 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3066 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3067 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3068 int x,y; |
2967 | 3069 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3070 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3071 for(x=0; x<8; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3072 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3073 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3074 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3075 for(x=0; x<7; x++){ |
4001 | 3076 score2+= FFABS( s1[x ] - s1[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3077 - s1[x+1] + s1[x+1+stride]) |
4001 | 3078 -FFABS( s2[x ] - s2[x +stride] |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3079 - s2[x+1] + s2[x+1+stride]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3080 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3081 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3082 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3083 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3084 } |
2967 | 3085 |
4001 | 3086 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight; |
3087 else return score1 + FFABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3088 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3089 |
1784 | 3090 static int try_8x8basis_c(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale){ |
3091 int i; | |
3092 unsigned int sum=0; | |
3093 | |
3094 for(i=0; i<8*8; i++){ | |
3095 int b= rem[i] + ((basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT)); | |
3096 int w= weight[i]; | |
3097 b>>= RECON_SHIFT; | |
3098 assert(-512<b && b<512); | |
3099 | |
3100 sum += (w*b)*(w*b)>>4; | |
3101 } | |
3102 return sum>>2; | |
3103 } | |
3104 | |
3105 static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale){ | |
3106 int i; | |
3107 | |
3108 for(i=0; i<8*8; i++){ | |
3109 rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT); | |
2967 | 3110 } |
1784 | 3111 } |
3112 | |
1100 | 3113 /** |
3114 * permutes an 8x8 block. | |
1101 | 3115 * @param block the block which will be permuted according to the given permutation vector |
1100 | 3116 * @param permutation the permutation vector |
3117 * @param last the last non zero coefficient in scantable order, used to speed the permutation up | |
2967 | 3118 * @param scantable the used scantable, this is only used to speed the permutation up, the block is not |
1101 | 3119 * (inverse) permutated to scantable order! |
1100 | 3120 */ |
1064 | 3121 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
|
3122 { |
764 | 3123 int i; |
945 | 3124 DCTELEM temp[64]; |
2967 | 3125 |
764 | 3126 if(last<=0) return; |
5129 | 3127 //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
|
3128 |
764 | 3129 for(i=0; i<=last; i++){ |
3130 const int j= scantable[i]; | |
3131 temp[j]= block[j]; | |
3132 block[j]=0; | |
3133 } | |
2967 | 3134 |
764 | 3135 for(i=0; i<=last; i++){ |
3136 const int j= scantable[i]; | |
3137 const int perm_j= permutation[j]; | |
3138 block[perm_j]= temp[j]; | |
3139 } | |
174
ac5075a55488
new IDCT code by Michael Niedermayer (michaelni@gmx.at) - #define SIMPLE_IDCT to enable
arpi_esp
parents:
88
diff
changeset
|
3140 } |
34 | 3141 |
1729 | 3142 static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride, int h){ |
3143 return 0; | |
3144 } | |
3145 | |
3146 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type){ | |
3147 int i; | |
2967 | 3148 |
8976
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
3149 memset(cmp, 0, sizeof(void*)*6); |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
3150 |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
3151 for(i=0; i<6; i++){ |
1729 | 3152 switch(type&0xFF){ |
3153 case FF_CMP_SAD: | |
3154 cmp[i]= c->sad[i]; | |
3155 break; | |
3156 case FF_CMP_SATD: | |
3157 cmp[i]= c->hadamard8_diff[i]; | |
3158 break; | |
3159 case FF_CMP_SSE: | |
3160 cmp[i]= c->sse[i]; | |
3161 break; | |
3162 case FF_CMP_DCT: | |
3163 cmp[i]= c->dct_sad[i]; | |
3164 break; | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3165 case FF_CMP_DCT264: |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3166 cmp[i]= c->dct264_sad[i]; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3167 break; |
2382 | 3168 case FF_CMP_DCTMAX: |
3169 cmp[i]= c->dct_max[i]; | |
3170 break; | |
1729 | 3171 case FF_CMP_PSNR: |
3172 cmp[i]= c->quant_psnr[i]; | |
3173 break; | |
3174 case FF_CMP_BIT: | |
3175 cmp[i]= c->bit[i]; | |
3176 break; | |
3177 case FF_CMP_RD: | |
3178 cmp[i]= c->rd[i]; | |
3179 break; | |
3180 case FF_CMP_VSAD: | |
3181 cmp[i]= c->vsad[i]; | |
3182 break; | |
3183 case FF_CMP_VSSE: | |
3184 cmp[i]= c->vsse[i]; | |
3185 break; | |
3186 case FF_CMP_ZERO: | |
3187 cmp[i]= zero_cmp; | |
3188 break; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3189 case FF_CMP_NSSE: |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3190 cmp[i]= c->nsse[i]; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3191 break; |
11485 | 3192 #if CONFIG_DWT |
2184 | 3193 case FF_CMP_W53: |
3194 cmp[i]= c->w53[i]; | |
3195 break; | |
3196 case FF_CMP_W97: | |
3197 cmp[i]= c->w97[i]; | |
3198 break; | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
3199 #endif |
1729 | 3200 default: |
3201 av_log(NULL, AV_LOG_ERROR,"internal error in cmp function selection\n"); | |
3202 } | |
3203 } | |
3204 } | |
3205 | |
8288 | 3206 static void clear_block_c(DCTELEM *block) |
3207 { | |
3208 memset(block, 0, sizeof(DCTELEM)*64); | |
3209 } | |
3210 | |
1101 | 3211 /** |
3212 * memset(blocks, 0, sizeof(DCTELEM)*6*64) | |
3213 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
3214 static void clear_blocks_c(DCTELEM *blocks) |
296 | 3215 { |
3216 memset(blocks, 0, sizeof(DCTELEM)*6*64); | |
3217 } | |
3218 | |
866 | 3219 static void add_bytes_c(uint8_t *dst, uint8_t *src, int w){ |
6385 | 3220 long i; |
3221 for(i=0; i<=w-sizeof(long); i+=sizeof(long)){ | |
3222 long a = *(long*)(src+i); | |
3223 long b = *(long*)(dst+i); | |
3224 *(long*)(dst+i) = ((a&pb_7f) + (b&pb_7f)) ^ ((a^b)&pb_80); | |
866 | 3225 } |
3226 for(; i<w; i++) | |
3227 dst[i+0] += src[i+0]; | |
3228 } | |
3229 | |
6384 | 3230 static void add_bytes_l2_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ |
6385 | 3231 long i; |
6384 | 3232 for(i=0; i<=w-sizeof(long); i+=sizeof(long)){ |
3233 long a = *(long*)(src1+i); | |
3234 long b = *(long*)(src2+i); | |
6385 | 3235 *(long*)(dst+i) = ((a&pb_7f) + (b&pb_7f)) ^ ((a^b)&pb_80); |
6384 | 3236 } |
3237 for(; i<w; i++) | |
3238 dst[i] = src1[i]+src2[i]; | |
3239 } | |
3240 | |
866 | 3241 static void diff_bytes_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ |
6385 | 3242 long i; |
8590 | 3243 #if !HAVE_FAST_UNALIGNED |
6385 | 3244 if((long)src2 & (sizeof(long)-1)){ |
6386 | 3245 for(i=0; i+7<w; i+=8){ |
3246 dst[i+0] = src1[i+0]-src2[i+0]; | |
3247 dst[i+1] = src1[i+1]-src2[i+1]; | |
3248 dst[i+2] = src1[i+2]-src2[i+2]; | |
3249 dst[i+3] = src1[i+3]-src2[i+3]; | |
3250 dst[i+4] = src1[i+4]-src2[i+4]; | |
3251 dst[i+5] = src1[i+5]-src2[i+5]; | |
3252 dst[i+6] = src1[i+6]-src2[i+6]; | |
3253 dst[i+7] = src1[i+7]-src2[i+7]; | |
3254 } | |
6385 | 3255 }else |
3256 #endif | |
3257 for(i=0; i<=w-sizeof(long); i+=sizeof(long)){ | |
3258 long a = *(long*)(src1+i); | |
3259 long b = *(long*)(src2+i); | |
3260 *(long*)(dst+i) = ((a|pb_80) - (b&pb_7f)) ^ ((a^b^pb_80)&pb_80); | |
3261 } | |
866 | 3262 for(; i<w; i++) |
3263 dst[i+0] = src1[i+0]-src2[i+0]; | |
3264 } | |
3265 | |
10431 | 3266 static void add_hfyu_median_prediction_c(uint8_t *dst, const uint8_t *src1, const uint8_t *diff, int w, int *left, int *left_top){ |
8760 | 3267 int i; |
3268 uint8_t l, lt; | |
3269 | |
3270 l= *left; | |
3271 lt= *left_top; | |
3272 | |
3273 for(i=0; i<w; i++){ | |
3274 l= mid_pred(l, src1[i], (l + src1[i] - lt)&0xFF) + diff[i]; | |
3275 lt= src1[i]; | |
3276 dst[i]= l; | |
3277 } | |
3278 | |
3279 *left= l; | |
3280 *left_top= lt; | |
3281 } | |
3282 | |
10431 | 3283 static void sub_hfyu_median_prediction_c(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top){ |
1527 | 3284 int i; |
3285 uint8_t l, lt; | |
3286 | |
3287 l= *left; | |
3288 lt= *left_top; | |
3289 | |
3290 for(i=0; i<w; i++){ | |
3291 const int pred= mid_pred(l, src1[i], (l + src1[i] - lt)&0xFF); | |
3292 lt= src1[i]; | |
3293 l= src2[i]; | |
3294 dst[i]= l - pred; | |
2967 | 3295 } |
1527 | 3296 |
3297 *left= l; | |
3298 *left_top= lt; | |
3299 } | |
3300 | |
10420
442ab0c41eae
Huffyuv: Add missing const to src pointers in dsputil functions.
astrange
parents:
10370
diff
changeset
|
3301 static int add_hfyu_left_prediction_c(uint8_t *dst, const uint8_t *src, int w, int acc){ |
10370 | 3302 int i; |
3303 | |
3304 for(i=0; i<w-1; i++){ | |
3305 acc+= src[i]; | |
3306 dst[i]= acc; | |
3307 i++; | |
3308 acc+= src[i]; | |
3309 dst[i]= acc; | |
3310 } | |
3311 | |
3312 for(; i<w; i++){ | |
3313 acc+= src[i]; | |
3314 dst[i]= acc; | |
3315 } | |
3316 | |
3317 return acc; | |
3318 } | |
3319 | |
3320 #if HAVE_BIGENDIAN | |
3321 #define B 3 | |
3322 #define G 2 | |
3323 #define R 1 | |
10878
a8620b001ed3
Implement alpha channel decoding for BGR HuffYUV.
astrange
parents:
10867
diff
changeset
|
3324 #define A 0 |
10370 | 3325 #else |
3326 #define B 0 | |
3327 #define G 1 | |
3328 #define R 2 | |
10878
a8620b001ed3
Implement alpha channel decoding for BGR HuffYUV.
astrange
parents:
10867
diff
changeset
|
3329 #define A 3 |
10370 | 3330 #endif |
10878
a8620b001ed3
Implement alpha channel decoding for BGR HuffYUV.
astrange
parents:
10867
diff
changeset
|
3331 static void add_hfyu_left_prediction_bgr32_c(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha){ |
10370 | 3332 int i; |
10878
a8620b001ed3
Implement alpha channel decoding for BGR HuffYUV.
astrange
parents:
10867
diff
changeset
|
3333 int r,g,b,a; |
10370 | 3334 r= *red; |
3335 g= *green; | |
3336 b= *blue; | |
10878
a8620b001ed3
Implement alpha channel decoding for BGR HuffYUV.
astrange
parents:
10867
diff
changeset
|
3337 a= *alpha; |
10370 | 3338 |
3339 for(i=0; i<w; i++){ | |
3340 b+= src[4*i+B]; | |
3341 g+= src[4*i+G]; | |
3342 r+= src[4*i+R]; | |
10878
a8620b001ed3
Implement alpha channel decoding for BGR HuffYUV.
astrange
parents:
10867
diff
changeset
|
3343 a+= src[4*i+A]; |
10370 | 3344 |
3345 dst[4*i+B]= b; | |
3346 dst[4*i+G]= g; | |
3347 dst[4*i+R]= r; | |
10878
a8620b001ed3
Implement alpha channel decoding for BGR HuffYUV.
astrange
parents:
10867
diff
changeset
|
3348 dst[4*i+A]= a; |
10370 | 3349 } |
3350 | |
3351 *red= r; | |
3352 *green= g; | |
3353 *blue= b; | |
10878
a8620b001ed3
Implement alpha channel decoding for BGR HuffYUV.
astrange
parents:
10867
diff
changeset
|
3354 *alpha= a; |
10370 | 3355 } |
3356 #undef B | |
3357 #undef G | |
3358 #undef R | |
10878
a8620b001ed3
Implement alpha channel decoding for BGR HuffYUV.
astrange
parents:
10867
diff
changeset
|
3359 #undef A |
10370 | 3360 |
936 | 3361 #define BUTTERFLY2(o1,o2,i1,i2) \ |
3362 o1= (i1)+(i2);\ | |
3363 o2= (i1)-(i2); | |
3364 | |
3365 #define BUTTERFLY1(x,y) \ | |
3366 {\ | |
3367 int a,b;\ | |
3368 a= x;\ | |
3369 b= y;\ | |
3370 x= a+b;\ | |
3371 y= a-b;\ | |
3372 } | |
3373 | |
4001 | 3374 #define BUTTERFLYA(x,y) (FFABS((x)+(y)) + FFABS((x)-(y))) |
936 | 3375 |
1708 | 3376 static int hadamard8_diff8x8_c(/*MpegEncContext*/ void *s, uint8_t *dst, uint8_t *src, int stride, int h){ |
936 | 3377 int i; |
3378 int temp[64]; | |
3379 int sum=0; | |
2967 | 3380 |
1708 | 3381 assert(h==8); |
936 | 3382 |
3383 for(i=0; i<8; i++){ | |
3384 //FIXME try pointer walks | |
3385 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]); | |
3386 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]); | |
3387 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]); | |
3388 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 | 3389 |
936 | 3390 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3391 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3392 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3393 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3394 |
936 | 3395 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3396 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3397 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3398 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3399 } | |
3400 | |
3401 for(i=0; i<8; i++){ | |
3402 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3403 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3404 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3405 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3406 |
936 | 3407 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3408 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3409 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3410 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
3411 | |
2967 | 3412 sum += |
936 | 3413 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3414 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3415 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3416 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3417 } | |
3418 #if 0 | |
3419 static int maxi=0; | |
3420 if(sum>maxi){ | |
3421 maxi=sum; | |
3422 printf("MAX:%d\n", maxi); | |
3423 } | |
3424 #endif | |
3425 return sum; | |
3426 } | |
3427 | |
1729 | 3428 static int hadamard8_intra8x8_c(/*MpegEncContext*/ void *s, uint8_t *src, uint8_t *dummy, int stride, int h){ |
936 | 3429 int i; |
3430 int temp[64]; | |
3431 int sum=0; | |
2967 | 3432 |
1729 | 3433 assert(h==8); |
2967 | 3434 |
936 | 3435 for(i=0; i<8; i++){ |
3436 //FIXME try pointer walks | |
1729 | 3437 BUTTERFLY2(temp[8*i+0], temp[8*i+1], src[stride*i+0],src[stride*i+1]); |
3438 BUTTERFLY2(temp[8*i+2], temp[8*i+3], src[stride*i+2],src[stride*i+3]); | |
3439 BUTTERFLY2(temp[8*i+4], temp[8*i+5], src[stride*i+4],src[stride*i+5]); | |
3440 BUTTERFLY2(temp[8*i+6], temp[8*i+7], src[stride*i+6],src[stride*i+7]); | |
2967 | 3441 |
936 | 3442 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3443 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3444 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3445 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3446 |
936 | 3447 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3448 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3449 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3450 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3451 } | |
3452 | |
3453 for(i=0; i<8; i++){ | |
3454 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3455 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3456 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3457 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3458 |
936 | 3459 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3460 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3461 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3462 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
2967 | 3463 |
3464 sum += | |
936 | 3465 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3466 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3467 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3468 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3469 } | |
2967 | 3470 |
4001 | 3471 sum -= FFABS(temp[8*0] + temp[8*4]); // -mean |
2967 | 3472 |
936 | 3473 return sum; |
3474 } | |
3475 | |
1708 | 3476 static int dct_sad8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3477 MpegEncContext * const s= (MpegEncContext *)c; |
11195 | 3478 LOCAL_ALIGNED_16(DCTELEM, temp, [64]); |
2967 | 3479 |
1708 | 3480 assert(h==8); |
936 | 3481 |
3482 s->dsp.diff_pixels(temp, src1, src2, stride); | |
1092 | 3483 s->dsp.fdct(temp); |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
3484 return s->dsp.sum_abs_dctelem(temp); |
936 | 3485 } |
3486 | |
8590 | 3487 #if CONFIG_GPL |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3488 #define DCT8_1D {\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3489 const int s07 = SRC(0) + SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3490 const int s16 = SRC(1) + SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3491 const int s25 = SRC(2) + SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3492 const int s34 = SRC(3) + SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3493 const int a0 = s07 + s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3494 const int a1 = s16 + s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3495 const int a2 = s07 - s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3496 const int a3 = s16 - s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3497 const int d07 = SRC(0) - SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3498 const int d16 = SRC(1) - SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3499 const int d25 = SRC(2) - SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3500 const int d34 = SRC(3) - SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3501 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
|
3502 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
|
3503 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
|
3504 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
|
3505 DST(0, a0 + a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3506 DST(1, a4 + (a7>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3507 DST(2, a2 + (a3>>1)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3508 DST(3, a5 + (a6>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3509 DST(4, a0 - a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3510 DST(5, a6 - (a5>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3511 DST(6, (a2>>1) - a3 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3512 DST(7, (a4>>2) - a7 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3513 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3514 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3515 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
|
3516 MpegEncContext * const s= (MpegEncContext *)c; |
5256 | 3517 DCTELEM dct[8][8]; |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3518 int i; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3519 int sum=0; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3520 |
5256 | 3521 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
|
3522 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3523 #define SRC(x) dct[i][x] |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3524 #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
|
3525 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3526 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3527 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3528 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3529 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3530 #define SRC(x) dct[x][i] |
4001 | 3531 #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
|
3532 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3533 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3534 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3535 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3536 return sum; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3537 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3538 #endif |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3539 |
2382 | 3540 static int dct_max8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
3541 MpegEncContext * const s= (MpegEncContext *)c; | |
11195 | 3542 LOCAL_ALIGNED_16(DCTELEM, temp, [64]); |
2382 | 3543 int sum=0, i; |
2967 | 3544 |
2382 | 3545 assert(h==8); |
3546 | |
3547 s->dsp.diff_pixels(temp, src1, src2, stride); | |
3548 s->dsp.fdct(temp); | |
3549 | |
3550 for(i=0; i<64; i++) | |
4001 | 3551 sum= FFMAX(sum, FFABS(temp[i])); |
2967 | 3552 |
2382 | 3553 return sum; |
3554 } | |
3555 | |
1708 | 3556 static int quant_psnr8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3557 MpegEncContext * const s= (MpegEncContext *)c; |
11195 | 3558 LOCAL_ALIGNED_16(DCTELEM, temp, [64*2]); |
11193 | 3559 DCTELEM * const bak = temp+64; |
936 | 3560 int sum=0, i; |
3561 | |
1708 | 3562 assert(h==8); |
936 | 3563 s->mb_intra=0; |
2967 | 3564 |
936 | 3565 s->dsp.diff_pixels(temp, src1, src2, stride); |
2967 | 3566 |
936 | 3567 memcpy(bak, temp, 64*sizeof(DCTELEM)); |
2967 | 3568 |
1013 | 3569 s->block_last_index[0/*FIXME*/]= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
1689 | 3570 s->dct_unquantize_inter(s, temp, 0, s->qscale); |
6001 | 3571 ff_simple_idct(temp); //FIXME |
2967 | 3572 |
936 | 3573 for(i=0; i<64; i++) |
3574 sum+= (temp[i]-bak[i])*(temp[i]-bak[i]); | |
2967 | 3575 |
936 | 3576 return sum; |
3577 } | |
3578 | |
1708 | 3579 static int rd8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3580 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3581 const uint8_t *scantable= s->intra_scantable.permutated; |
11195 | 3582 LOCAL_ALIGNED_16(DCTELEM, temp, [64]); |
3583 LOCAL_ALIGNED_16(uint8_t, lsrc1, [64]); | |
3584 LOCAL_ALIGNED_16(uint8_t, lsrc2, [64]); | |
6719 | 3585 int i, last, run, bits, level, distortion, start_i; |
1007 | 3586 const int esc_length= s->ac_esc_length; |
3587 uint8_t * length; | |
3588 uint8_t * last_length; | |
2967 | 3589 |
1708 | 3590 assert(h==8); |
3591 | |
10068 | 3592 copy_block8(lsrc1, src1, 8, stride, 8); |
3593 copy_block8(lsrc2, src2, 8, stride, 8); | |
3594 | |
3595 s->dsp.diff_pixels(temp, lsrc1, lsrc2, 8); | |
1007 | 3596 |
1013 | 3597 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3598 | |
3599 bits=0; | |
2967 | 3600 |
1013 | 3601 if (s->mb_intra) { |
2967 | 3602 start_i = 1; |
1013 | 3603 length = s->intra_ac_vlc_length; |
3604 last_length= s->intra_ac_vlc_last_length; | |
3605 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma | |
3606 } else { | |
3607 start_i = 0; | |
3608 length = s->inter_ac_vlc_length; | |
3609 last_length= s->inter_ac_vlc_last_length; | |
3610 } | |
2967 | 3611 |
1013 | 3612 if(last>=start_i){ |
1007 | 3613 run=0; |
3614 for(i=start_i; i<last; i++){ | |
3615 int j= scantable[i]; | |
3616 level= temp[j]; | |
2967 | 3617 |
1007 | 3618 if(level){ |
3619 level+=64; | |
3620 if((level&(~127)) == 0){ | |
3621 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3622 }else | |
3623 bits+= esc_length; | |
3624 run=0; | |
3625 }else | |
3626 run++; | |
3627 } | |
3628 i= scantable[last]; | |
2967 | 3629 |
1011 | 3630 level= temp[i] + 64; |
3631 | |
3632 assert(level - 64); | |
2967 | 3633 |
1007 | 3634 if((level&(~127)) == 0){ |
3635 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3636 }else | |
3637 bits+= esc_length; | |
2967 | 3638 |
1013 | 3639 } |
3640 | |
3641 if(last>=0){ | |
1689 | 3642 if(s->mb_intra) |
3643 s->dct_unquantize_intra(s, temp, 0, s->qscale); | |
3644 else | |
3645 s->dct_unquantize_inter(s, temp, 0, s->qscale); | |
1007 | 3646 } |
2967 | 3647 |
10068 | 3648 s->dsp.idct_add(lsrc2, 8, temp); |
3649 | |
3650 distortion= s->dsp.sse[1](NULL, lsrc2, lsrc1, 8, 8); | |
6719 | 3651 |
3652 return distortion + ((bits*s->qscale*s->qscale*109 + 64)>>7); | |
1007 | 3653 } |
3654 | |
1708 | 3655 static int bit8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3656 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3657 const uint8_t *scantable= s->intra_scantable.permutated; |
11195 | 3658 LOCAL_ALIGNED_16(DCTELEM, temp, [64]); |
1007 | 3659 int i, last, run, bits, level, start_i; |
3660 const int esc_length= s->ac_esc_length; | |
3661 uint8_t * length; | |
3662 uint8_t * last_length; | |
1708 | 3663 |
3664 assert(h==8); | |
2967 | 3665 |
1013 | 3666 s->dsp.diff_pixels(temp, src1, src2, stride); |
1007 | 3667 |
1013 | 3668 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3669 | |
3670 bits=0; | |
2967 | 3671 |
1007 | 3672 if (s->mb_intra) { |
2967 | 3673 start_i = 1; |
1007 | 3674 length = s->intra_ac_vlc_length; |
3675 last_length= s->intra_ac_vlc_last_length; | |
1013 | 3676 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma |
1007 | 3677 } else { |
3678 start_i = 0; | |
3679 length = s->inter_ac_vlc_length; | |
3680 last_length= s->inter_ac_vlc_last_length; | |
3681 } | |
2967 | 3682 |
1013 | 3683 if(last>=start_i){ |
1007 | 3684 run=0; |
3685 for(i=start_i; i<last; i++){ | |
3686 int j= scantable[i]; | |
3687 level= temp[j]; | |
2967 | 3688 |
1007 | 3689 if(level){ |
3690 level+=64; | |
3691 if((level&(~127)) == 0){ | |
3692 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3693 }else | |
3694 bits+= esc_length; | |
3695 run=0; | |
3696 }else | |
3697 run++; | |
3698 } | |
3699 i= scantable[last]; | |
2967 | 3700 |
1013 | 3701 level= temp[i] + 64; |
2967 | 3702 |
1013 | 3703 assert(level - 64); |
2967 | 3704 |
1007 | 3705 if((level&(~127)) == 0){ |
3706 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3707 }else | |
3708 bits+= esc_length; | |
3709 } | |
3710 | |
3711 return bits; | |
3712 } | |
3713 | |
8978 | 3714 #define VSAD_INTRA(size) \ |
3715 static int vsad_intra##size##_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ \ | |
3716 int score=0; \ | |
3717 int x,y; \ | |
3718 \ | |
3719 for(y=1; y<h; y++){ \ | |
3720 for(x=0; x<size; x+=4){ \ | |
3721 score+= FFABS(s[x ] - s[x +stride]) + FFABS(s[x+1] - s[x+1+stride]) \ | |
3722 +FFABS(s[x+2] - s[x+2+stride]) + FFABS(s[x+3] - s[x+3+stride]); \ | |
3723 } \ | |
3724 s+= stride; \ | |
3725 } \ | |
3726 \ | |
3727 return score; \ | |
1729 | 3728 } |
8978 | 3729 VSAD_INTRA(8) |
3730 VSAD_INTRA(16) | |
1729 | 3731 |
3732 static int vsad16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3733 int score=0; | |
3734 int x,y; | |
2967 | 3735 |
1729 | 3736 for(y=1; y<h; y++){ |
3737 for(x=0; x<16; x++){ | |
4001 | 3738 score+= FFABS(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); |
1729 | 3739 } |
3740 s1+= stride; | |
3741 s2+= stride; | |
3742 } | |
2967 | 3743 |
1729 | 3744 return score; |
3745 } | |
3746 | |
3747 #define SQ(a) ((a)*(a)) | |
8978 | 3748 #define VSSE_INTRA(size) \ |
3749 static int vsse_intra##size##_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ \ | |
3750 int score=0; \ | |
3751 int x,y; \ | |
3752 \ | |
3753 for(y=1; y<h; y++){ \ | |
3754 for(x=0; x<size; x+=4){ \ | |
3755 score+= SQ(s[x ] - s[x +stride]) + SQ(s[x+1] - s[x+1+stride]) \ | |
3756 +SQ(s[x+2] - s[x+2+stride]) + SQ(s[x+3] - s[x+3+stride]); \ | |
3757 } \ | |
3758 s+= stride; \ | |
3759 } \ | |
3760 \ | |
3761 return score; \ | |
1729 | 3762 } |
8978 | 3763 VSSE_INTRA(8) |
3764 VSSE_INTRA(16) | |
1729 | 3765 |
3766 static int vsse16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3767 int score=0; | |
3768 int x,y; | |
2967 | 3769 |
1729 | 3770 for(y=1; y<h; y++){ |
3771 for(x=0; x<16; x++){ | |
3772 score+= SQ(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); | |
3773 } | |
3774 s1+= stride; | |
3775 s2+= stride; | |
3776 } | |
2967 | 3777 |
1729 | 3778 return score; |
3779 } | |
3780 | |
5255 | 3781 static int ssd_int8_vs_int16_c(const int8_t *pix1, const int16_t *pix2, |
3782 int size){ | |
4749 | 3783 int score=0; |
3784 int i; | |
3785 for(i=0; i<size; i++) | |
3786 score += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]); | |
3787 return score; | |
3788 } | |
3789 | |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3790 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
|
3791 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
|
3792 WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c) |
8590 | 3793 #if CONFIG_GPL |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3794 WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c) |
3013 | 3795 #endif |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3796 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
|
3797 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
|
3798 WRAPPER8_16_SQ(rd8x8_c, rd16_c) |
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
6054
diff
changeset
|
3799 WRAPPER8_16_SQ(bit8x8_c, bit16_c) |
936 | 3800 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3801 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
|
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] *= src[i]; |
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 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
|
3808 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3809 src1 += len-1; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3810 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3811 dst[i] = src0[i] * src1[-i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3812 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3813 |
10300
4d1b9ca628fc
Drop unused args from vector_fmul_add_add, simpify code, and rename
mru
parents:
10219
diff
changeset
|
3814 static void vector_fmul_add_c(float *dst, const float *src0, const float *src1, const float *src2, int len){ |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3815 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3816 for(i=0; i<len; i++) |
10300
4d1b9ca628fc
Drop unused args from vector_fmul_add_add, simpify code, and rename
mru
parents:
10219
diff
changeset
|
3817 dst[i] = src0[i] * src1[i] + src2[i]; |
3568
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 |
7261 | 3820 void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len){ |
7263 | 3821 int i,j; |
3822 dst += len; | |
3823 win += len; | |
3824 src0+= len; | |
3825 for(i=-len, j=len-1; i<0; i++, j--) { | |
3826 float s0 = src0[i]; | |
3827 float s1 = src1[j]; | |
3828 float wi = win[i]; | |
3829 float wj = win[j]; | |
3830 dst[i] = s0*wj - s1*wi + add_bias; | |
3831 dst[j] = s0*wi + s1*wj + add_bias; | |
3832 } | |
7261 | 3833 } |
3834 | |
10219 | 3835 static void vector_fmul_scalar_c(float *dst, const float *src, float mul, |
3836 int len) | |
3837 { | |
3838 int i; | |
3839 for (i = 0; i < len; i++) | |
3840 dst[i] = src[i] * mul; | |
3841 } | |
3842 | |
3843 static void vector_fmul_sv_scalar_2_c(float *dst, const float *src, | |
3844 const float **sv, float mul, int len) | |
3845 { | |
3846 int i; | |
3847 for (i = 0; i < len; i += 2, sv++) { | |
3848 dst[i ] = src[i ] * sv[0][0] * mul; | |
3849 dst[i+1] = src[i+1] * sv[0][1] * mul; | |
3850 } | |
3851 } | |
3852 | |
3853 static void vector_fmul_sv_scalar_4_c(float *dst, const float *src, | |
3854 const float **sv, float mul, int len) | |
3855 { | |
3856 int i; | |
3857 for (i = 0; i < len; i += 4, sv++) { | |
3858 dst[i ] = src[i ] * sv[0][0] * mul; | |
3859 dst[i+1] = src[i+1] * sv[0][1] * mul; | |
3860 dst[i+2] = src[i+2] * sv[0][2] * mul; | |
3861 dst[i+3] = src[i+3] * sv[0][3] * mul; | |
3862 } | |
3863 } | |
3864 | |
3865 static void sv_fmul_scalar_2_c(float *dst, const float **sv, float mul, | |
3866 int len) | |
3867 { | |
3868 int i; | |
3869 for (i = 0; i < len; i += 2, sv++) { | |
3870 dst[i ] = sv[0][0] * mul; | |
3871 dst[i+1] = sv[0][1] * mul; | |
3872 } | |
3873 } | |
3874 | |
3875 static void sv_fmul_scalar_4_c(float *dst, const float **sv, float mul, | |
3876 int len) | |
3877 { | |
3878 int i; | |
3879 for (i = 0; i < len; i += 4, sv++) { | |
3880 dst[i ] = sv[0][0] * mul; | |
3881 dst[i+1] = sv[0][1] * mul; | |
3882 dst[i+2] = sv[0][2] * mul; | |
3883 dst[i+3] = sv[0][3] * mul; | |
3884 } | |
3885 } | |
3886 | |
3887 static void butterflies_float_c(float *restrict v1, float *restrict v2, | |
3888 int len) | |
3889 { | |
3890 int i; | |
3891 for (i = 0; i < len; i++) { | |
3892 float t = v1[i] - v2[i]; | |
3893 v1[i] += v2[i]; | |
3894 v2[i] = t; | |
3895 } | |
3896 } | |
3897 | |
3898 static float scalarproduct_float_c(const float *v1, const float *v2, int len) | |
3899 { | |
3900 float p = 0.0; | |
3901 int i; | |
3902 | |
3903 for (i = 0; i < len; i++) | |
3904 p += v1[i] * v2[i]; | |
3905 | |
3906 return p; | |
3907 } | |
3908 | |
7564 | 3909 static void int32_to_float_fmul_scalar_c(float *dst, const int *src, float mul, int len){ |
3910 int i; | |
3911 for(i=0; i<len; i++) | |
3912 dst[i] = src[i] * mul; | |
3913 } | |
3914 | |
10104
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3915 static inline uint32_t clipf_c_one(uint32_t a, uint32_t mini, |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3916 uint32_t maxi, uint32_t maxisign) |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3917 { |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3918 |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3919 if(a > mini) return mini; |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3920 else if((a^(1<<31)) > maxisign) return maxi; |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3921 else return a; |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3922 } |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3923 |
10105 | 3924 static void vector_clipf_c_opposite_sign(float *dst, const float *src, float *min, float *max, int len){ |
10104
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3925 int i; |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3926 uint32_t mini = *(uint32_t*)min; |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3927 uint32_t maxi = *(uint32_t*)max; |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3928 uint32_t maxisign = maxi ^ (1<<31); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3929 uint32_t *dsti = (uint32_t*)dst; |
10105 | 3930 const uint32_t *srci = (const uint32_t*)src; |
10104
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3931 for(i=0; i<len; i+=8) { |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3932 dsti[i + 0] = clipf_c_one(srci[i + 0], mini, maxi, maxisign); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3933 dsti[i + 1] = clipf_c_one(srci[i + 1], mini, maxi, maxisign); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3934 dsti[i + 2] = clipf_c_one(srci[i + 2], mini, maxi, maxisign); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3935 dsti[i + 3] = clipf_c_one(srci[i + 3], mini, maxi, maxisign); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3936 dsti[i + 4] = clipf_c_one(srci[i + 4], mini, maxi, maxisign); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3937 dsti[i + 5] = clipf_c_one(srci[i + 5], mini, maxi, maxisign); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3938 dsti[i + 6] = clipf_c_one(srci[i + 6], mini, maxi, maxisign); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3939 dsti[i + 7] = clipf_c_one(srci[i + 7], mini, maxi, maxisign); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3940 } |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3941 } |
10105 | 3942 static void vector_clipf_c(float *dst, const float *src, float min, float max, int len){ |
10104
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3943 int i; |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3944 if(min < 0 && max > 0) { |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3945 vector_clipf_c_opposite_sign(dst, src, &min, &max, len); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3946 } else { |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3947 for(i=0; i < len; i+=8) { |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3948 dst[i ] = av_clipf(src[i ], min, max); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3949 dst[i + 1] = av_clipf(src[i + 1], min, max); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3950 dst[i + 2] = av_clipf(src[i + 2], min, max); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3951 dst[i + 3] = av_clipf(src[i + 3], min, max); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3952 dst[i + 4] = av_clipf(src[i + 4], min, max); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3953 dst[i + 5] = av_clipf(src[i + 5], min, max); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3954 dst[i + 6] = av_clipf(src[i + 6], min, max); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3955 dst[i + 7] = av_clipf(src[i + 7], min, max); |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3956 } |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3957 } |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3958 } |
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
3959 |
7261 | 3960 static av_always_inline int float_to_int16_one(const float *src){ |
3961 int_fast32_t tmp = *(const int32_t*)src; | |
3962 if(tmp & 0xf0000){ | |
3963 tmp = (0x43c0ffff - tmp)>>31; | |
3964 // is this faster on some gcc/cpu combinations? | |
3965 // if(tmp > 0x43c0ffff) tmp = 0xFFFF; | |
3966 // else tmp = 0; | |
3967 } | |
3968 return tmp - 0x8000; | |
3969 } | |
3970 | |
7218 | 3971 void ff_float_to_int16_c(int16_t *dst, const float *src, long len){ |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3972 int i; |
7261 | 3973 for(i=0; i<len; i++) |
3974 dst[i] = float_to_int16_one(src+i); | |
3975 } | |
3976 | |
7286
e267f2519248
float_to_int16_interleave: change src to an array of pointers instead of assuming it's contiguous.
lorenm
parents:
7263
diff
changeset
|
3977 void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels){ |
7261 | 3978 int i,j,c; |
3979 if(channels==2){ | |
3980 for(i=0; i<len; i++){ | |
7286
e267f2519248
float_to_int16_interleave: change src to an array of pointers instead of assuming it's contiguous.
lorenm
parents:
7263
diff
changeset
|
3981 dst[2*i] = float_to_int16_one(src[0]+i); |
e267f2519248
float_to_int16_interleave: change src to an array of pointers instead of assuming it's contiguous.
lorenm
parents:
7263
diff
changeset
|
3982 dst[2*i+1] = float_to_int16_one(src[1]+i); |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3983 } |
7261 | 3984 }else{ |
7286
e267f2519248
float_to_int16_interleave: change src to an array of pointers instead of assuming it's contiguous.
lorenm
parents:
7263
diff
changeset
|
3985 for(c=0; c<channels; c++) |
7261 | 3986 for(i=0, j=c; i<len; i++, j+=channels) |
7286
e267f2519248
float_to_int16_interleave: change src to an array of pointers instead of assuming it's contiguous.
lorenm
parents:
7263
diff
changeset
|
3987 dst[j] = float_to_int16_one(src[c]+i); |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3988 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3989 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3990 |
11981 | 3991 static int32_t scalarproduct_int16_c(const int16_t * v1, const int16_t * v2, int order, int shift) |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
6719
diff
changeset
|
3992 { |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
6719
diff
changeset
|
3993 int res = 0; |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
6719
diff
changeset
|
3994 |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
6719
diff
changeset
|
3995 while (order--) |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
6719
diff
changeset
|
3996 res += (*v1++ * *v2++) >> shift; |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
6719
diff
changeset
|
3997 |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
6719
diff
changeset
|
3998 return res; |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
6719
diff
changeset
|
3999 } |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
6719
diff
changeset
|
4000 |
11981 | 4001 static int32_t scalarproduct_and_madd_int16_c(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul) |
10644 | 4002 { |
4003 int res = 0; | |
4004 while (order--) { | |
4005 res += *v1 * *v2++; | |
4006 *v1++ += mul * *v3++; | |
4007 } | |
4008 return res; | |
4009 } | |
4010 | |
5887 | 4011 #define W0 2048 |
4012 #define W1 2841 /* 2048*sqrt (2)*cos (1*pi/16) */ | |
4013 #define W2 2676 /* 2048*sqrt (2)*cos (2*pi/16) */ | |
4014 #define W3 2408 /* 2048*sqrt (2)*cos (3*pi/16) */ | |
4015 #define W4 2048 /* 2048*sqrt (2)*cos (4*pi/16) */ | |
4016 #define W5 1609 /* 2048*sqrt (2)*cos (5*pi/16) */ | |
4017 #define W6 1108 /* 2048*sqrt (2)*cos (6*pi/16) */ | |
4018 #define W7 565 /* 2048*sqrt (2)*cos (7*pi/16) */ | |
4019 | |
4020 static void wmv2_idct_row(short * b) | |
4021 { | |
4022 int s1,s2; | |
4023 int a0,a1,a2,a3,a4,a5,a6,a7; | |
4024 /*step 1*/ | |
4025 a1 = W1*b[1]+W7*b[7]; | |
4026 a7 = W7*b[1]-W1*b[7]; | |
4027 a5 = W5*b[5]+W3*b[3]; | |
4028 a3 = W3*b[5]-W5*b[3]; | |
4029 a2 = W2*b[2]+W6*b[6]; | |
4030 a6 = W6*b[2]-W2*b[6]; | |
4031 a0 = W0*b[0]+W0*b[4]; | |
4032 a4 = W0*b[0]-W0*b[4]; | |
4033 /*step 2*/ | |
4034 s1 = (181*(a1-a5+a7-a3)+128)>>8;//1,3,5,7, | |
4035 s2 = (181*(a1-a5-a7+a3)+128)>>8; | |
4036 /*step 3*/ | |
4037 b[0] = (a0+a2+a1+a5 + (1<<7))>>8; | |
4038 b[1] = (a4+a6 +s1 + (1<<7))>>8; | |
4039 b[2] = (a4-a6 +s2 + (1<<7))>>8; | |
4040 b[3] = (a0-a2+a7+a3 + (1<<7))>>8; | |
4041 b[4] = (a0-a2-a7-a3 + (1<<7))>>8; | |
4042 b[5] = (a4-a6 -s2 + (1<<7))>>8; | |
4043 b[6] = (a4+a6 -s1 + (1<<7))>>8; | |
4044 b[7] = (a0+a2-a1-a5 + (1<<7))>>8; | |
4045 } | |
4046 static void wmv2_idct_col(short * b) | |
4047 { | |
4048 int s1,s2; | |
4049 int a0,a1,a2,a3,a4,a5,a6,a7; | |
4050 /*step 1, with extended precision*/ | |
4051 a1 = (W1*b[8*1]+W7*b[8*7] + 4)>>3; | |
4052 a7 = (W7*b[8*1]-W1*b[8*7] + 4)>>3; | |
4053 a5 = (W5*b[8*5]+W3*b[8*3] + 4)>>3; | |
4054 a3 = (W3*b[8*5]-W5*b[8*3] + 4)>>3; | |
4055 a2 = (W2*b[8*2]+W6*b[8*6] + 4)>>3; | |
4056 a6 = (W6*b[8*2]-W2*b[8*6] + 4)>>3; | |
4057 a0 = (W0*b[8*0]+W0*b[8*4] )>>3; | |
4058 a4 = (W0*b[8*0]-W0*b[8*4] )>>3; | |
4059 /*step 2*/ | |
4060 s1 = (181*(a1-a5+a7-a3)+128)>>8; | |
4061 s2 = (181*(a1-a5-a7+a3)+128)>>8; | |
4062 /*step 3*/ | |
4063 b[8*0] = (a0+a2+a1+a5 + (1<<13))>>14; | |
4064 b[8*1] = (a4+a6 +s1 + (1<<13))>>14; | |
4065 b[8*2] = (a4-a6 +s2 + (1<<13))>>14; | |
4066 b[8*3] = (a0-a2+a7+a3 + (1<<13))>>14; | |
4067 | |
4068 b[8*4] = (a0-a2-a7-a3 + (1<<13))>>14; | |
4069 b[8*5] = (a4-a6 -s2 + (1<<13))>>14; | |
4070 b[8*6] = (a4+a6 -s1 + (1<<13))>>14; | |
4071 b[8*7] = (a0+a2-a1-a5 + (1<<13))>>14; | |
4072 } | |
4073 void ff_wmv2_idct_c(short * block){ | |
4074 int i; | |
4075 | |
4076 for(i=0;i<64;i+=8){ | |
4077 wmv2_idct_row(block+i); | |
4078 } | |
4079 for(i=0;i<8;i++){ | |
4080 wmv2_idct_col(block+i); | |
4081 } | |
4082 } | |
1092 | 4083 /* XXX: those functions should be suppressed ASAP when all IDCTs are |
4084 converted */ | |
5887 | 4085 static void ff_wmv2_idct_put_c(uint8_t *dest, int line_size, DCTELEM *block) |
4086 { | |
4087 ff_wmv2_idct_c(block); | |
4088 put_pixels_clamped_c(block, dest, line_size); | |
4089 } | |
4090 static void ff_wmv2_idct_add_c(uint8_t *dest, int line_size, DCTELEM *block) | |
4091 { | |
4092 ff_wmv2_idct_c(block); | |
4093 add_pixels_clamped_c(block, dest, line_size); | |
4094 } | |
1092 | 4095 static void ff_jref_idct_put(uint8_t *dest, int line_size, DCTELEM *block) |
4096 { | |
4097 j_rev_dct (block); | |
4098 put_pixels_clamped_c(block, dest, line_size); | |
4099 } | |
4100 static void ff_jref_idct_add(uint8_t *dest, int line_size, DCTELEM *block) | |
4101 { | |
4102 j_rev_dct (block); | |
4103 add_pixels_clamped_c(block, dest, line_size); | |
4104 } | |
4105 | |
2256 | 4106 static void ff_jref_idct4_put(uint8_t *dest, int line_size, DCTELEM *block) |
4107 { | |
4108 j_rev_dct4 (block); | |
4109 put_pixels_clamped4_c(block, dest, line_size); | |
4110 } | |
4111 static void ff_jref_idct4_add(uint8_t *dest, int line_size, DCTELEM *block) | |
4112 { | |
4113 j_rev_dct4 (block); | |
4114 add_pixels_clamped4_c(block, dest, line_size); | |
4115 } | |
4116 | |
2257 | 4117 static void ff_jref_idct2_put(uint8_t *dest, int line_size, DCTELEM *block) |
4118 { | |
4119 j_rev_dct2 (block); | |
4120 put_pixels_clamped2_c(block, dest, line_size); | |
4121 } | |
4122 static void ff_jref_idct2_add(uint8_t *dest, int line_size, DCTELEM *block) | |
4123 { | |
4124 j_rev_dct2 (block); | |
4125 add_pixels_clamped2_c(block, dest, line_size); | |
4126 } | |
4127 | |
2259 | 4128 static void ff_jref_idct1_put(uint8_t *dest, int line_size, DCTELEM *block) |
4129 { | |
4176 | 4130 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2259 | 4131 |
4132 dest[0] = cm[(block[0] + 4)>>3]; | |
4133 } | |
4134 static void ff_jref_idct1_add(uint8_t *dest, int line_size, DCTELEM *block) | |
4135 { | |
4176 | 4136 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
2259 | 4137 |
4138 dest[0] = cm[dest[0] + ((block[0] + 4)>>3)]; | |
4139 } | |
4140 | |
5143 | 4141 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
|
4142 |
1201 | 4143 /* init static data */ |
10867 | 4144 av_cold void dsputil_static_init(void) |
0 | 4145 { |
751 | 4146 int i; |
0 | 4147 |
4176 | 4148 for(i=0;i<256;i++) ff_cropTbl[i + MAX_NEG_CROP] = i; |
1201 | 4149 for(i=0;i<MAX_NEG_CROP;i++) { |
4176 | 4150 ff_cropTbl[i] = 0; |
4151 ff_cropTbl[i + MAX_NEG_CROP + 256] = 255; | |
1201 | 4152 } |
2967 | 4153 |
1201 | 4154 for(i=0;i<512;i++) { |
4179 | 4155 ff_squareTbl[i] = (i - 256) * (i - 256); |
1201 | 4156 } |
2967 | 4157 |
4197 | 4158 for(i=0; i<64; i++) inv_zigzag_direct16[ff_zigzag_direct[i]]= i+1; |
1201 | 4159 } |
0 | 4160 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4161 int ff_check_alignment(void){ |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4162 static int did_fail=0; |
11369 | 4163 DECLARE_ALIGNED(16, int, aligned); |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4164 |
9259 | 4165 if((intptr_t)&aligned & 15){ |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4166 if(!did_fail){ |
8590 | 4167 #if HAVE_MMX || HAVE_ALTIVEC |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4168 av_log(NULL, AV_LOG_ERROR, |
4292 | 4169 "Compiler did not align stack variables. Libavcodec has been miscompiled\n" |
4170 "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
|
4171 "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
|
4172 "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
|
4173 #endif |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4174 did_fail=1; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4175 } |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4176 return -1; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4177 } |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4178 return 0; |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4179 } |
861 | 4180 |
10867 | 4181 av_cold void dsputil_init(DSPContext* c, AVCodecContext *avctx) |
1201 | 4182 { |
4183 int i; | |
0 | 4184 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4185 ff_check_alignment(); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4240
diff
changeset
|
4186 |
8590 | 4187 #if CONFIG_ENCODERS |
1567 | 4188 if(avctx->dct_algo==FF_DCT_FASTINT) { |
1092 | 4189 c->fdct = fdct_ifast; |
2979 | 4190 c->fdct248 = fdct_ifast248; |
2967 | 4191 } |
1567 | 4192 else if(avctx->dct_algo==FF_DCT_FAAN) { |
1557 | 4193 c->fdct = ff_faandct; |
2979 | 4194 c->fdct248 = ff_faandct248; |
2967 | 4195 } |
1567 | 4196 else { |
1092 | 4197 c->fdct = ff_jpeg_fdct_islow; //slow/accurate/default |
2979 | 4198 c->fdct248 = ff_fdct248_islow; |
1567 | 4199 } |
1092 | 4200 #endif //CONFIG_ENCODERS |
4201 | |
2256 | 4202 if(avctx->lowres==1){ |
8596
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
4203 if(avctx->idct_algo==FF_IDCT_INT || avctx->idct_algo==FF_IDCT_AUTO || !CONFIG_H264_DECODER){ |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4204 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
|
4205 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
|
4206 }else{ |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
4207 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
|
4208 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
|
4209 } |
2256 | 4210 c->idct = j_rev_dct4; |
1092 | 4211 c->idct_permutation_type= FF_NO_IDCT_PERM; |
2257 | 4212 }else if(avctx->lowres==2){ |
4213 c->idct_put= ff_jref_idct2_put; | |
4214 c->idct_add= ff_jref_idct2_add; | |
4215 c->idct = j_rev_dct2; | |
4216 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2259 | 4217 }else if(avctx->lowres==3){ |
4218 c->idct_put= ff_jref_idct1_put; | |
4219 c->idct_add= ff_jref_idct1_add; | |
4220 c->idct = j_rev_dct1; | |
4221 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2256 | 4222 }else{ |
4223 if(avctx->idct_algo==FF_IDCT_INT){ | |
4224 c->idct_put= ff_jref_idct_put; | |
4225 c->idct_add= ff_jref_idct_add; | |
4226 c->idct = j_rev_dct; | |
4227 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM; | |
9975
d6d7e8d4a04d
Do not redundantly check for both CONFIG_THEORA_DECODER and CONFIG_VP3_DECODER.
diego
parents:
9586
diff
changeset
|
4228 }else if((CONFIG_VP3_DECODER || CONFIG_VP5_DECODER || CONFIG_VP6_DECODER ) && |
5007 | 4229 avctx->idct_algo==FF_IDCT_VP3){ |
2693 | 4230 c->idct_put= ff_vp3_idct_put_c; |
4231 c->idct_add= ff_vp3_idct_add_c; | |
4232 c->idct = ff_vp3_idct_c; | |
4233 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
5887 | 4234 }else if(avctx->idct_algo==FF_IDCT_WMV2){ |
4235 c->idct_put= ff_wmv2_idct_put_c; | |
4236 c->idct_add= ff_wmv2_idct_add_c; | |
4237 c->idct = ff_wmv2_idct_c; | |
4238 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
6407 | 4239 }else if(avctx->idct_algo==FF_IDCT_FAAN){ |
4240 c->idct_put= ff_faanidct_put; | |
4241 c->idct_add= ff_faanidct_add; | |
4242 c->idct = ff_faanidct; | |
4243 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
8596
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
4244 }else if(CONFIG_EATGQ_DECODER && avctx->idct_algo==FF_IDCT_EA) { |
8120 | 4245 c->idct_put= ff_ea_idct_put_c; |
4246 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
11231 | 4247 }else if(CONFIG_BINK_DECODER && avctx->idct_algo==FF_IDCT_BINK) { |
4248 c->idct = ff_bink_idct_c; | |
4249 c->idct_add = ff_bink_idct_add_c; | |
4250 c->idct_put = ff_bink_idct_put_c; | |
4251 c->idct_permutation_type = FF_NO_IDCT_PERM; | |
2256 | 4252 }else{ //accurate/default |
6001 | 4253 c->idct_put= ff_simple_idct_put; |
4254 c->idct_add= ff_simple_idct_add; | |
4255 c->idct = ff_simple_idct; | |
2256 | 4256 c->idct_permutation_type= FF_NO_IDCT_PERM; |
4257 } | |
1092 | 4258 } |
4259 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4260 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
|
4261 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
|
4262 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
|
4263 c->put_signed_pixels_clamped = put_signed_pixels_clamped_c; |
11231 | 4264 c->put_pixels_nonclamped = put_pixels_nonclamped_c; |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4265 c->add_pixels_clamped = add_pixels_clamped_c; |
2763 | 4266 c->add_pixels8 = add_pixels8_c; |
4267 c->add_pixels4 = add_pixels4_c; | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4749
diff
changeset
|
4268 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
|
4269 c->gmc1 = gmc1_c; |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
4270 c->gmc = ff_gmc_c; |
8288 | 4271 c->clear_block = clear_block_c; |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
4272 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
|
4273 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
|
4274 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
|
4275 |
11231 | 4276 c->fill_block_tab[0] = fill_block16_c; |
4277 c->fill_block_tab[1] = fill_block8_c; | |
4278 c->scale_block = scale_block_c; | |
4279 | |
859 | 4280 /* TODO [0] 16 [1] 8 */ |
1708 | 4281 c->pix_abs[0][0] = pix_abs16_c; |
4282 c->pix_abs[0][1] = pix_abs16_x2_c; | |
4283 c->pix_abs[0][2] = pix_abs16_y2_c; | |
4284 c->pix_abs[0][3] = pix_abs16_xy2_c; | |
4285 c->pix_abs[1][0] = pix_abs8_c; | |
4286 c->pix_abs[1][1] = pix_abs8_x2_c; | |
4287 c->pix_abs[1][2] = pix_abs8_y2_c; | |
4288 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
|
4289 |
859 | 4290 #define dspfunc(PFX, IDX, NUM) \ |
4291 c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## NUM ## _c; \ | |
4292 c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## NUM ## _x2_c; \ | |
4293 c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## NUM ## _y2_c; \ | |
4294 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
|
4295 |
859 | 4296 dspfunc(put, 0, 16); |
4297 dspfunc(put_no_rnd, 0, 16); | |
4298 dspfunc(put, 1, 8); | |
4299 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
|
4300 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
|
4301 dspfunc(put, 3, 2); |
0 | 4302 |
859 | 4303 dspfunc(avg, 0, 16); |
4304 dspfunc(avg_no_rnd, 0, 16); | |
4305 dspfunc(avg, 1, 8); | |
4306 dspfunc(avg_no_rnd, 1, 8); | |
1319 | 4307 dspfunc(avg, 2, 4); |
4308 dspfunc(avg, 3, 2); | |
859 | 4309 #undef dspfunc |
857 | 4310 |
1864 | 4311 c->put_no_rnd_pixels_l2[0]= put_no_rnd_pixels16_l2_c; |
4312 c->put_no_rnd_pixels_l2[1]= put_no_rnd_pixels8_l2_c; | |
4313 | |
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
|
4314 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
|
4315 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
|
4316 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
|
4317 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
|
4318 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
|
4319 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
|
4320 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
|
4321 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
|
4322 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
|
4323 |
1319 | 4324 c->avg_tpel_pixels_tab[ 0] = avg_tpel_pixels_mc00_c; |
4325 c->avg_tpel_pixels_tab[ 1] = avg_tpel_pixels_mc10_c; | |
4326 c->avg_tpel_pixels_tab[ 2] = avg_tpel_pixels_mc20_c; | |
4327 c->avg_tpel_pixels_tab[ 4] = avg_tpel_pixels_mc01_c; | |
4328 c->avg_tpel_pixels_tab[ 5] = avg_tpel_pixels_mc11_c; | |
4329 c->avg_tpel_pixels_tab[ 6] = avg_tpel_pixels_mc21_c; | |
4330 c->avg_tpel_pixels_tab[ 8] = avg_tpel_pixels_mc02_c; | |
4331 c->avg_tpel_pixels_tab[ 9] = avg_tpel_pixels_mc12_c; | |
4332 c->avg_tpel_pixels_tab[10] = avg_tpel_pixels_mc22_c; | |
4333 | |
859 | 4334 #define dspfunc(PFX, IDX, NUM) \ |
4335 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_c; \ | |
4336 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_c; \ | |
4337 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_c; \ | |
4338 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_c; \ | |
4339 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_c; \ | |
4340 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_c; \ | |
4341 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_c; \ | |
4342 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_c; \ | |
4343 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_c; \ | |
4344 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_c; \ | |
4345 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_c; \ | |
4346 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_c; \ | |
4347 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_c; \ | |
4348 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_c; \ | |
4349 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_c; \ | |
4350 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_c | |
857 | 4351 |
859 | 4352 dspfunc(put_qpel, 0, 16); |
4353 dspfunc(put_no_rnd_qpel, 0, 16); | |
4354 | |
4355 dspfunc(avg_qpel, 0, 16); | |
4356 /* dspfunc(avg_no_rnd_qpel, 0, 16); */ | |
857 | 4357 |
859 | 4358 dspfunc(put_qpel, 1, 8); |
4359 dspfunc(put_no_rnd_qpel, 1, 8); | |
4360 | |
4361 dspfunc(avg_qpel, 1, 8); | |
4362 /* dspfunc(avg_no_rnd_qpel, 1, 8); */ | |
1168 | 4363 |
4364 dspfunc(put_h264_qpel, 0, 16); | |
4365 dspfunc(put_h264_qpel, 1, 8); | |
4366 dspfunc(put_h264_qpel, 2, 4); | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
4367 dspfunc(put_h264_qpel, 3, 2); |
1168 | 4368 dspfunc(avg_h264_qpel, 0, 16); |
4369 dspfunc(avg_h264_qpel, 1, 8); | |
4370 dspfunc(avg_h264_qpel, 2, 4); | |
4371 | |
859 | 4372 #undef dspfunc |
1168 | 4373 c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_c; |
4374 c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_c; | |
4375 c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_c; | |
4376 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_c; | |
4377 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_c; | |
4378 c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_c; | |
9439
ef3a7b711cc0
Rename put_no_rnd_h264_chroma* to reflect its usage in VC1 only
conrad
parents:
9437
diff
changeset
|
4379 c->put_no_rnd_vc1_chroma_pixels_tab[0]= put_no_rnd_vc1_chroma_mc8_c; |
9440 | 4380 c->avg_no_rnd_vc1_chroma_pixels_tab[0]= avg_no_rnd_vc1_chroma_mc8_c; |
857 | 4381 |
6437 | 4382 c->draw_edges = draw_edges_c; |
4383 | |
8590 | 4384 #if CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4385 ff_cavsdsp_init(c,avctx); |
3432 | 4386 #endif |
9585 | 4387 |
4388 #if CONFIG_MLP_DECODER || CONFIG_TRUEHD_DECODER | |
4389 ff_mlp_init(c, avctx); | |
4390 #endif | |
9995
3141f69e3905
Do not check for both CONFIG_VC1_DECODER and CONFIG_WMV3_DECODER,
diego
parents:
9975
diff
changeset
|
4391 #if CONFIG_VC1_DECODER |
3526 | 4392 ff_vc1dsp_init(c,avctx); |
4393 #endif | |
9995
3141f69e3905
Do not check for both CONFIG_VC1_DECODER and CONFIG_WMV3_DECODER,
diego
parents:
9975
diff
changeset
|
4394 #if CONFIG_WMV2_DECODER || CONFIG_VC1_DECODER |
5887 | 4395 ff_intrax8dsp_init(c,avctx); |
4396 #endif | |
8590 | 4397 #if CONFIG_RV30_DECODER |
8410 | 4398 ff_rv30dsp_init(c,avctx); |
4399 #endif | |
8590 | 4400 #if CONFIG_RV40_DECODER |
8232 | 4401 ff_rv40dsp_init(c,avctx); |
4402 c->put_rv40_qpel_pixels_tab[0][15] = put_rv40_qpel16_mc33_c; | |
4403 c->avg_rv40_qpel_pixels_tab[0][15] = avg_rv40_qpel16_mc33_c; | |
4404 c->put_rv40_qpel_pixels_tab[1][15] = put_rv40_qpel8_mc33_c; | |
4405 c->avg_rv40_qpel_pixels_tab[1][15] = avg_rv40_qpel8_mc33_c; | |
4406 #endif | |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4407 |
936 | 4408 c->put_mspel_pixels_tab[0]= put_mspel8_mc00_c; |
4409 c->put_mspel_pixels_tab[1]= put_mspel8_mc10_c; | |
4410 c->put_mspel_pixels_tab[2]= put_mspel8_mc20_c; | |
4411 c->put_mspel_pixels_tab[3]= put_mspel8_mc30_c; | |
4412 c->put_mspel_pixels_tab[4]= put_mspel8_mc02_c; | |
4413 c->put_mspel_pixels_tab[5]= put_mspel8_mc12_c; | |
4414 c->put_mspel_pixels_tab[6]= put_mspel8_mc22_c; | |
4415 c->put_mspel_pixels_tab[7]= put_mspel8_mc32_c; | |
2967 | 4416 |
1708 | 4417 #define SET_CMP_FUNC(name) \ |
4418 c->name[0]= name ## 16_c;\ | |
4419 c->name[1]= name ## 8x8_c; | |
2967 | 4420 |
1708 | 4421 SET_CMP_FUNC(hadamard8_diff) |
1729 | 4422 c->hadamard8_diff[4]= hadamard8_intra16_c; |
8978 | 4423 c->hadamard8_diff[5]= hadamard8_intra8x8_c; |
1708 | 4424 SET_CMP_FUNC(dct_sad) |
2382 | 4425 SET_CMP_FUNC(dct_max) |
8590 | 4426 #if CONFIG_GPL |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
4427 SET_CMP_FUNC(dct264_sad) |
3013 | 4428 #endif |
1708 | 4429 c->sad[0]= pix_abs16_c; |
4430 c->sad[1]= pix_abs8_c; | |
4431 c->sse[0]= sse16_c; | |
4432 c->sse[1]= sse8_c; | |
2184 | 4433 c->sse[2]= sse4_c; |
1708 | 4434 SET_CMP_FUNC(quant_psnr) |
4435 SET_CMP_FUNC(rd) | |
4436 SET_CMP_FUNC(bit) | |
1729 | 4437 c->vsad[0]= vsad16_c; |
4438 c->vsad[4]= vsad_intra16_c; | |
8978 | 4439 c->vsad[5]= vsad_intra8_c; |
1729 | 4440 c->vsse[0]= vsse16_c; |
4441 c->vsse[4]= vsse_intra16_c; | |
8978 | 4442 c->vsse[5]= vsse_intra8_c; |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4443 c->nsse[0]= nsse16_c; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4444 c->nsse[1]= nsse8_c; |
11485 | 4445 #if CONFIG_DWT |
4446 ff_dsputil_init_dwt(c); | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
4447 #endif |
2184 | 4448 |
4749 | 4449 c->ssd_int8_vs_int16 = ssd_int8_vs_int16_c; |
4450 | |
866 | 4451 c->add_bytes= add_bytes_c; |
6384 | 4452 c->add_bytes_l2= add_bytes_l2_c; |
866 | 4453 c->diff_bytes= diff_bytes_c; |
8760 | 4454 c->add_hfyu_median_prediction= add_hfyu_median_prediction_c; |
1527 | 4455 c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_c; |
10370 | 4456 c->add_hfyu_left_prediction = add_hfyu_left_prediction_c; |
4457 c->add_hfyu_left_prediction_bgr32 = add_hfyu_left_prediction_bgr32_c; | |
1273 | 4458 c->bswap_buf= bswap_buf; |
8590 | 4459 #if CONFIG_PNG_DECODER |
6384 | 4460 c->add_png_paeth_prediction= ff_add_png_paeth_prediction; |
4461 #endif | |
2633 | 4462 |
10749
5cca4b6c459d
Get rid of pointless CONFIG_ANY_H263 preprocessor definition.
diego
parents:
10748
diff
changeset
|
4463 if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) { |
5278 | 4464 c->h263_h_loop_filter= h263_h_loop_filter_c; |
4465 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
|
4466 } |
2967 | 4467 |
9975
d6d7e8d4a04d
Do not redundantly check for both CONFIG_THEORA_DECODER and CONFIG_VP3_DECODER.
diego
parents:
9586
diff
changeset
|
4468 if (CONFIG_VP3_DECODER) { |
7995 | 4469 c->vp3_h_loop_filter= ff_vp3_h_loop_filter_c; |
4470 c->vp3_v_loop_filter= ff_vp3_v_loop_filter_c; | |
11637 | 4471 c->vp3_idct_dc_add= ff_vp3_idct_dc_add_c; |
7995 | 4472 } |
8785
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
4473 if (CONFIG_VP6_DECODER) { |
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
4474 c->vp6_filter_diag4= ff_vp6_filter_diag4_c; |
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
4475 } |
7995 | 4476 |
2045 | 4477 c->h261_loop_filter= h261_loop_filter_c; |
2967 | 4478 |
1784 | 4479 c->try_8x8basis= try_8x8basis_c; |
4480 c->add_8x8basis= add_8x8basis_c; | |
866 | 4481 |
8590 | 4482 #if CONFIG_VORBIS_DECODER |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4483 c->vorbis_inverse_coupling = vorbis_inverse_coupling; |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4484 #endif |
8590 | 4485 #if CONFIG_AC3_DECODER |
7563 | 4486 c->ac3_downmix = ff_ac3_downmix_c; |
4487 #endif | |
10429
289dd8daf4ee
add CONFIG_LPC to the build system for lpc dsputil functions. fixes build
jbr
parents:
10424
diff
changeset
|
4488 #if CONFIG_LPC |
10424
94595d0e617c
Move autocorrelation function from flacenc.c to lpc.c. Also rename the
jbr
parents:
10421
diff
changeset
|
4489 c->lpc_compute_autocorr = ff_lpc_compute_autocorr; |
10429
289dd8daf4ee
add CONFIG_LPC to the build system for lpc dsputil functions. fixes build
jbr
parents:
10424
diff
changeset
|
4490 #endif |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4491 c->vector_fmul = vector_fmul_c; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4492 c->vector_fmul_reverse = vector_fmul_reverse_c; |
10300
4d1b9ca628fc
Drop unused args from vector_fmul_add_add, simpify code, and rename
mru
parents:
10219
diff
changeset
|
4493 c->vector_fmul_add = vector_fmul_add_c; |
7261 | 4494 c->vector_fmul_window = ff_vector_fmul_window_c; |
7564 | 4495 c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_c; |
10104
0fa3d21b317e
SSE optimized vector_clipf(). 10% faster TwinVQ decoding.
vitor
parents:
10094
diff
changeset
|
4496 c->vector_clipf = vector_clipf_c; |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4497 c->float_to_int16 = ff_float_to_int16_c; |
7261 | 4498 c->float_to_int16_interleave = ff_float_to_int16_interleave_c; |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
6719
diff
changeset
|
4499 c->scalarproduct_int16 = scalarproduct_int16_c; |
10644 | 4500 c->scalarproduct_and_madd_int16 = scalarproduct_and_madd_int16_c; |
10219 | 4501 c->scalarproduct_float = scalarproduct_float_c; |
4502 c->butterflies_float = butterflies_float_c; | |
4503 c->vector_fmul_scalar = vector_fmul_scalar_c; | |
4504 | |
4505 c->vector_fmul_sv_scalar[0] = vector_fmul_sv_scalar_2_c; | |
4506 c->vector_fmul_sv_scalar[1] = vector_fmul_sv_scalar_4_c; | |
4507 | |
4508 c->sv_fmul_scalar[0] = sv_fmul_scalar_2_c; | |
4509 c->sv_fmul_scalar[1] = sv_fmul_scalar_4_c; | |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4510 |
3245 | 4511 c->shrink[0]= ff_img_copy_plane; |
4512 c->shrink[1]= ff_shrink22; | |
4513 c->shrink[2]= ff_shrink44; | |
4514 c->shrink[3]= ff_shrink88; | |
4515 | |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4516 c->prefetch= just_return; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4517 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4518 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
|
4519 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
|
4520 |
8596
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
4521 if (HAVE_MMX) dsputil_init_mmx (c, avctx); |
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
4522 if (ARCH_ARM) dsputil_init_arm (c, avctx); |
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
4523 if (CONFIG_MLIB) dsputil_init_mlib (c, avctx); |
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
4524 if (HAVE_VIS) dsputil_init_vis (c, avctx); |
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
4525 if (ARCH_ALPHA) dsputil_init_alpha (c, avctx); |
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
4526 if (ARCH_PPC) dsputil_init_ppc (c, avctx); |
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
4527 if (HAVE_MMI) dsputil_init_mmi (c, avctx); |
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
4528 if (ARCH_SH4) dsputil_init_sh4 (c, avctx); |
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
4529 if (ARCH_BFIN) dsputil_init_bfin (c, avctx); |
1092 | 4530 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4531 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
|
4532 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
|
4533 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
|
4534 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
|
4535 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
|
4536 } |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4537 |
11988
e382860b855f
Set rv34 (0,0) subpel mc functions to the optimised h264 ones
mru
parents:
11981
diff
changeset
|
4538 c->put_rv30_tpel_pixels_tab[0][0] = c->put_h264_qpel_pixels_tab[0][0]; |
e382860b855f
Set rv34 (0,0) subpel mc functions to the optimised h264 ones
mru
parents:
11981
diff
changeset
|
4539 c->put_rv30_tpel_pixels_tab[1][0] = c->put_h264_qpel_pixels_tab[1][0]; |
e382860b855f
Set rv34 (0,0) subpel mc functions to the optimised h264 ones
mru
parents:
11981
diff
changeset
|
4540 c->avg_rv30_tpel_pixels_tab[0][0] = c->avg_h264_qpel_pixels_tab[0][0]; |
e382860b855f
Set rv34 (0,0) subpel mc functions to the optimised h264 ones
mru
parents:
11981
diff
changeset
|
4541 c->avg_rv30_tpel_pixels_tab[1][0] = c->avg_h264_qpel_pixels_tab[1][0]; |
e382860b855f
Set rv34 (0,0) subpel mc functions to the optimised h264 ones
mru
parents:
11981
diff
changeset
|
4542 |
e382860b855f
Set rv34 (0,0) subpel mc functions to the optimised h264 ones
mru
parents:
11981
diff
changeset
|
4543 c->put_rv40_qpel_pixels_tab[0][0] = c->put_h264_qpel_pixels_tab[0][0]; |
e382860b855f
Set rv34 (0,0) subpel mc functions to the optimised h264 ones
mru
parents:
11981
diff
changeset
|
4544 c->put_rv40_qpel_pixels_tab[1][0] = c->put_h264_qpel_pixels_tab[1][0]; |
e382860b855f
Set rv34 (0,0) subpel mc functions to the optimised h264 ones
mru
parents:
11981
diff
changeset
|
4545 c->avg_rv40_qpel_pixels_tab[0][0] = c->avg_h264_qpel_pixels_tab[0][0]; |
e382860b855f
Set rv34 (0,0) subpel mc functions to the optimised h264 ones
mru
parents:
11981
diff
changeset
|
4546 c->avg_rv40_qpel_pixels_tab[1][0] = c->avg_h264_qpel_pixels_tab[1][0]; |
e382860b855f
Set rv34 (0,0) subpel mc functions to the optimised h264 ones
mru
parents:
11981
diff
changeset
|
4547 |
1092 | 4548 switch(c->idct_permutation_type){ |
4549 case FF_NO_IDCT_PERM: | |
4550 for(i=0; i<64; i++) | |
4551 c->idct_permutation[i]= i; | |
4552 break; | |
4553 case FF_LIBMPEG2_IDCT_PERM: | |
4554 for(i=0; i<64; i++) | |
4555 c->idct_permutation[i]= (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2); | |
4556 break; | |
4557 case FF_SIMPLE_IDCT_PERM: | |
4558 for(i=0; i<64; i++) | |
4559 c->idct_permutation[i]= simple_mmx_permutation[i]; | |
4560 break; | |
4561 case FF_TRANSPOSE_IDCT_PERM: | |
4562 for(i=0; i<64; i++) | |
4563 c->idct_permutation[i]= ((i&7)<<3) | (i>>3); | |
4564 break; | |
2696
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4565 case FF_PARTTRANS_IDCT_PERM: |
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4566 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
|
4567 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
|
4568 break; |
6600
c3213c91124c
Add a new IDCT permutation, used in xvid_sse2 and possibly future similar IDCTs.
astrange
parents:
6450
diff
changeset
|
4569 case FF_SSE2_IDCT_PERM: |
c3213c91124c
Add a new IDCT permutation, used in xvid_sse2 and possibly future similar IDCTs.
astrange
parents:
6450
diff
changeset
|
4570 for(i=0; i<64; i++) |
c3213c91124c
Add a new IDCT permutation, used in xvid_sse2 and possibly future similar IDCTs.
astrange
parents:
6450
diff
changeset
|
4571 c->idct_permutation[i]= (i&0x38) | idct_sse2_row_perm[i&7]; |
c3213c91124c
Add a new IDCT permutation, used in xvid_sse2 and possibly future similar IDCTs.
astrange
parents:
6450
diff
changeset
|
4572 break; |
1092 | 4573 default: |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1571
diff
changeset
|
4574 av_log(avctx, AV_LOG_ERROR, "Internal error, IDCT permutation not set\n"); |
1092 | 4575 } |
0 | 4576 } |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
220
diff
changeset
|
4577 |