Mercurial > libavcodec.hg
annotate dsputil.c @ 3823:41ad23b61a2c libavcodec
Original Commit: r17 | ods15 | 2006-09-22 12:28:28 +0300 (Fri, 22 Sep 2006) | 2 lines
bug fixes in codebook header syntax
author | ods15 |
---|---|
date | Mon, 02 Oct 2006 05:55:43 +0000 |
parents | 6a40092eb9e6 |
children | c8c591fe26f8 |
rev | line source |
---|---|
0 | 1 /* |
2 * DSP utils | |
429 | 3 * Copyright (c) 2000, 2001 Fabrice Bellard. |
1739
07a484280a82
copyright year update of the files i touched and remembered, things look annoyingly unmaintained otherwise
michael
parents:
1729
diff
changeset
|
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> |
0 | 5 * |
429 | 6 * This library is free software; you can redistribute it and/or |
7 * modify it under the terms of the GNU Lesser General Public | |
8 * License as published by the Free Software Foundation; either | |
9 * version 2 of the License, or (at your option) any later version. | |
0 | 10 * |
429 | 11 * This library is distributed in the hope that it will be useful, |
0 | 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
429 | 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
14 * Lesser General Public License for more details. | |
0 | 15 * |
429 | 16 * You should have received a copy of the GNU Lesser General Public |
17 * License along with this library; if not, write to the Free Software | |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3029
diff
changeset
|
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
256 | 19 * |
385 | 20 * gmc & q-pel & 32/64 bit based MC by Michael Niedermayer <michaelni@gmx.at> |
0 | 21 */ |
2967 | 22 |
1106 | 23 /** |
24 * @file dsputil.c | |
25 * DSP utils | |
26 */ | |
2967 | 27 |
0 | 28 #include "avcodec.h" |
29 #include "dsputil.h" | |
936 | 30 #include "mpegvideo.h" |
1092 | 31 #include "simple_idct.h" |
1557 | 32 #include "faandct.h" |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
33 #include "snow.h" |
676 | 34 |
2522
e25782262d7d
kill warnings patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2448
diff
changeset
|
35 /* snow.c */ |
e25782262d7d
kill warnings patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2448
diff
changeset
|
36 void ff_spatial_dwt(int *buffer, int width, int height, int stride, int type, int decomposition_count); |
e25782262d7d
kill warnings patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2448
diff
changeset
|
37 |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
38 /* vorbis.c */ |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
39 void vorbis_inverse_coupling(float *mag, float *ang, int blocksize); |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
40 |
2169
db8baace74d8
Minor Patch for shared libs on Mac OSX by (Bill May <wmay at cisco dot com>)
michael
parents:
2066
diff
changeset
|
41 uint8_t cropTbl[256 + 2 * MAX_NEG_CROP] = {0, }; |
db8baace74d8
Minor Patch for shared libs on Mac OSX by (Bill May <wmay at cisco dot com>)
michael
parents:
2066
diff
changeset
|
42 uint32_t squareTbl[512] = {0, }; |
0 | 43 |
1064 | 44 const uint8_t ff_zigzag_direct[64] = { |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
45 0, 1, 8, 16, 9, 2, 3, 10, |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
46 17, 24, 32, 25, 18, 11, 4, 5, |
34 | 47 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
|
48 27, 20, 13, 6, 7, 14, 21, 28, |
34 | 49 35, 42, 49, 56, 57, 50, 43, 36, |
50 29, 22, 15, 23, 30, 37, 44, 51, | |
51 58, 59, 52, 45, 38, 31, 39, 46, | |
52 53, 60, 61, 54, 47, 55, 62, 63 | |
53 }; | |
54 | |
1567 | 55 /* Specific zigzag scan for 248 idct. NOTE that unlike the |
56 specification, we interleave the fields */ | |
57 const uint8_t ff_zigzag248_direct[64] = { | |
58 0, 8, 1, 9, 16, 24, 2, 10, | |
59 17, 25, 32, 40, 48, 56, 33, 41, | |
60 18, 26, 3, 11, 4, 12, 19, 27, | |
61 34, 42, 49, 57, 50, 58, 35, 43, | |
62 20, 28, 5, 13, 6, 14, 21, 29, | |
63 36, 44, 51, 59, 52, 60, 37, 45, | |
64 22, 30, 7, 15, 23, 31, 38, 46, | |
65 53, 61, 54, 62, 39, 47, 55, 63, | |
66 }; | |
67 | |
220 | 68 /* not permutated inverse zigzag_direct + 1 for MMX quantizer */ |
3089 | 69 DECLARE_ALIGNED_8(uint16_t, inv_zigzag_direct16[64]) = {0, }; |
220 | 70 |
1064 | 71 const uint8_t ff_alternate_horizontal_scan[64] = { |
2967 | 72 0, 1, 2, 3, 8, 9, 16, 17, |
34 | 73 10, 11, 4, 5, 6, 7, 15, 14, |
2967 | 74 13, 12, 19, 18, 24, 25, 32, 33, |
34 | 75 26, 27, 20, 21, 22, 23, 28, 29, |
2967 | 76 30, 31, 34, 35, 40, 41, 48, 49, |
34 | 77 42, 43, 36, 37, 38, 39, 44, 45, |
2967 | 78 46, 47, 50, 51, 56, 57, 58, 59, |
34 | 79 52, 53, 54, 55, 60, 61, 62, 63, |
80 }; | |
81 | |
1064 | 82 const uint8_t ff_alternate_vertical_scan[64] = { |
2967 | 83 0, 8, 16, 24, 1, 9, 2, 10, |
34 | 84 17, 25, 32, 40, 48, 56, 57, 49, |
2967 | 85 41, 33, 26, 18, 3, 11, 4, 12, |
34 | 86 19, 27, 34, 42, 50, 58, 35, 43, |
2967 | 87 51, 59, 20, 28, 5, 13, 6, 14, |
34 | 88 21, 29, 36, 44, 52, 60, 37, 45, |
2967 | 89 53, 61, 22, 30, 7, 15, 23, 31, |
34 | 90 38, 46, 54, 62, 39, 47, 55, 63, |
91 }; | |
92 | |
220 | 93 /* a*inverse[b]>>32 == a/b for all 0<=a<=65536 && 2<=b<=255 */ |
1064 | 94 const uint32_t inverse[256]={ |
2967 | 95 0, 4294967295U,2147483648U,1431655766, 1073741824, 858993460, 715827883, 613566757, |
96 536870912, 477218589, 429496730, 390451573, 357913942, 330382100, 306783379, 286331154, | |
97 268435456, 252645136, 238609295, 226050911, 214748365, 204522253, 195225787, 186737709, | |
98 178956971, 171798692, 165191050, 159072863, 153391690, 148102321, 143165577, 138547333, | |
99 134217728, 130150525, 126322568, 122713352, 119304648, 116080198, 113025456, 110127367, | |
100 107374183, 104755300, 102261127, 99882961, 97612894, 95443718, 93368855, 91382283, | |
101 89478486, 87652394, 85899346, 84215046, 82595525, 81037119, 79536432, 78090315, | |
102 76695845, 75350304, 74051161, 72796056, 71582789, 70409300, 69273667, 68174085, | |
103 67108864, 66076420, 65075263, 64103990, 63161284, 62245903, 61356676, 60492498, | |
104 59652324, 58835169, 58040099, 57266231, 56512728, 55778797, 55063684, 54366675, | |
105 53687092, 53024288, 52377650, 51746594, 51130564, 50529028, 49941481, 49367441, | |
106 48806447, 48258060, 47721859, 47197443, 46684428, 46182445, 45691142, 45210183, | |
107 44739243, 44278014, 43826197, 43383509, 42949673, 42524429, 42107523, 41698712, | |
108 41297763, 40904451, 40518560, 40139882, 39768216, 39403370, 39045158, 38693400, | |
109 38347923, 38008561, 37675152, 37347542, 37025581, 36709123, 36398028, 36092163, | |
110 35791395, 35495598, 35204650, 34918434, 34636834, 34359739, 34087043, 33818641, | |
111 33554432, 33294321, 33038210, 32786010, 32537632, 32292988, 32051995, 31814573, | |
112 31580642, 31350127, 31122952, 30899046, 30678338, 30460761, 30246249, 30034737, | |
113 29826162, 29620465, 29417585, 29217465, 29020050, 28825284, 28633116, 28443493, | |
114 28256364, 28071682, 27889399, 27709467, 27531842, 27356480, 27183338, 27012373, | |
115 26843546, 26676816, 26512144, 26349493, 26188825, 26030105, 25873297, 25718368, | |
116 25565282, 25414008, 25264514, 25116768, 24970741, 24826401, 24683721, 24542671, | |
117 24403224, 24265352, 24129030, 23994231, 23860930, 23729102, 23598722, 23469767, | |
118 23342214, 23216040, 23091223, 22967740, 22845571, 22724695, 22605092, 22486740, | |
119 22369622, 22253717, 22139007, 22025474, 21913099, 21801865, 21691755, 21582751, | |
120 21474837, 21367997, 21262215, 21157475, 21053762, 20951060, 20849356, 20748635, | |
121 20648882, 20550083, 20452226, 20355296, 20259280, 20164166, 20069941, 19976593, | |
122 19884108, 19792477, 19701685, 19611723, 19522579, 19434242, 19346700, 19259944, | |
123 19173962, 19088744, 19004281, 18920561, 18837576, 18755316, 18673771, 18592933, | |
124 18512791, 18433337, 18354562, 18276457, 18199014, 18122225, 18046082, 17970575, | |
125 17895698, 17821442, 17747799, 17674763, 17602325, 17530479, 17459217, 17388532, | |
220 | 126 17318417, 17248865, 17179870, 17111424, 17043522, 16976156, 16909321, 16843010, |
127 }; | |
128 | |
1092 | 129 /* Input permutation for the simple_idct_mmx */ |
130 static const uint8_t simple_mmx_permutation[64]={ | |
2979 | 131 0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, |
132 0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, | |
133 0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, | |
134 0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, | |
135 0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, | |
136 0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, | |
137 0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, | |
138 0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F, | |
1092 | 139 }; |
140 | |
1064 | 141 static int pix_sum_c(uint8_t * pix, int line_size) |
612 | 142 { |
143 int s, i, j; | |
144 | |
145 s = 0; | |
146 for (i = 0; i < 16; i++) { | |
2979 | 147 for (j = 0; j < 16; j += 8) { |
148 s += pix[0]; | |
149 s += pix[1]; | |
150 s += pix[2]; | |
151 s += pix[3]; | |
152 s += pix[4]; | |
153 s += pix[5]; | |
154 s += pix[6]; | |
155 s += pix[7]; | |
156 pix += 8; | |
157 } | |
158 pix += line_size - 16; | |
612 | 159 } |
160 return s; | |
161 } | |
162 | |
1064 | 163 static int pix_norm1_c(uint8_t * pix, int line_size) |
612 | 164 { |
165 int s, i, j; | |
1064 | 166 uint32_t *sq = squareTbl + 256; |
612 | 167 |
168 s = 0; | |
169 for (i = 0; i < 16; i++) { | |
2979 | 170 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
|
171 #if 0 |
2979 | 172 s += sq[pix[0]]; |
173 s += sq[pix[1]]; | |
174 s += sq[pix[2]]; | |
175 s += sq[pix[3]]; | |
176 s += sq[pix[4]]; | |
177 s += sq[pix[5]]; | |
178 s += sq[pix[6]]; | |
179 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
|
180 #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
|
181 #if LONG_MAX > 2147483647 |
2979 | 182 register uint64_t x=*(uint64_t*)pix; |
183 s += sq[x&0xff]; | |
184 s += sq[(x>>8)&0xff]; | |
185 s += sq[(x>>16)&0xff]; | |
186 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
|
187 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
|
188 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
|
189 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
|
190 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
|
191 #else |
2979 | 192 register uint32_t x=*(uint32_t*)pix; |
193 s += sq[x&0xff]; | |
194 s += sq[(x>>8)&0xff]; | |
195 s += sq[(x>>16)&0xff]; | |
196 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
|
197 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
|
198 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
|
199 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
|
200 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
|
201 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
|
202 #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
|
203 #endif |
2979 | 204 pix += 8; |
205 } | |
206 pix += line_size - 16; | |
612 | 207 } |
208 return s; | |
209 } | |
210 | |
1273 | 211 static void bswap_buf(uint32_t *dst, uint32_t *src, int w){ |
212 int i; | |
2967 | 213 |
1273 | 214 for(i=0; i+8<=w; i+=8){ |
215 dst[i+0]= bswap_32(src[i+0]); | |
216 dst[i+1]= bswap_32(src[i+1]); | |
217 dst[i+2]= bswap_32(src[i+2]); | |
218 dst[i+3]= bswap_32(src[i+3]); | |
219 dst[i+4]= bswap_32(src[i+4]); | |
220 dst[i+5]= bswap_32(src[i+5]); | |
221 dst[i+6]= bswap_32(src[i+6]); | |
222 dst[i+7]= bswap_32(src[i+7]); | |
223 } | |
224 for(;i<w; i++){ | |
225 dst[i+0]= bswap_32(src[i+0]); | |
226 } | |
227 } | |
612 | 228 |
2184 | 229 static int sse4_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
230 { | |
231 int s, i; | |
232 uint32_t *sq = squareTbl + 256; | |
233 | |
234 s = 0; | |
235 for (i = 0; i < h; i++) { | |
236 s += sq[pix1[0] - pix2[0]]; | |
237 s += sq[pix1[1] - pix2[1]]; | |
238 s += sq[pix1[2] - pix2[2]]; | |
239 s += sq[pix1[3] - pix2[3]]; | |
240 pix1 += line_size; | |
241 pix2 += line_size; | |
242 } | |
243 return s; | |
244 } | |
245 | |
1708 | 246 static int sse8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) |
936 | 247 { |
248 int s, i; | |
1064 | 249 uint32_t *sq = squareTbl + 256; |
936 | 250 |
251 s = 0; | |
1708 | 252 for (i = 0; i < h; i++) { |
936 | 253 s += sq[pix1[0] - pix2[0]]; |
254 s += sq[pix1[1] - pix2[1]]; | |
255 s += sq[pix1[2] - pix2[2]]; | |
256 s += sq[pix1[3] - pix2[3]]; | |
257 s += sq[pix1[4] - pix2[4]]; | |
258 s += sq[pix1[5] - pix2[5]]; | |
259 s += sq[pix1[6] - pix2[6]]; | |
260 s += sq[pix1[7] - pix2[7]]; | |
261 pix1 += line_size; | |
262 pix2 += line_size; | |
263 } | |
264 return s; | |
265 } | |
266 | |
1708 | 267 static int sse16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
884 | 268 { |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
269 int s, i; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
270 uint32_t *sq = squareTbl + 256; |
884 | 271 |
272 s = 0; | |
1708 | 273 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
|
274 s += sq[pix1[ 0] - pix2[ 0]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
275 s += sq[pix1[ 1] - pix2[ 1]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
276 s += sq[pix1[ 2] - pix2[ 2]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
277 s += sq[pix1[ 3] - pix2[ 3]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
278 s += sq[pix1[ 4] - pix2[ 4]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
279 s += sq[pix1[ 5] - pix2[ 5]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
280 s += sq[pix1[ 6] - pix2[ 6]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
281 s += sq[pix1[ 7] - pix2[ 7]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
282 s += sq[pix1[ 8] - pix2[ 8]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
283 s += sq[pix1[ 9] - pix2[ 9]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
284 s += sq[pix1[10] - pix2[10]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
285 s += sq[pix1[11] - pix2[11]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
286 s += sq[pix1[12] - pix2[12]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
287 s += sq[pix1[13] - pix2[13]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
288 s += sq[pix1[14] - pix2[14]]; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
289 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
|
290 |
1012
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
291 pix1 += line_size; |
7a5038ec769b
sse16_c is totally fucked up (unaligned loads, LONG_MAX is undefined,
mellum
parents:
1011
diff
changeset
|
292 pix2 += line_size; |
884 | 293 } |
294 return s; | |
295 } | |
296 | |
2184 | 297 |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
298 #ifdef CONFIG_SNOW_ENCODER //dwt is in snow.c |
2184 | 299 static inline int w_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int w, int h, int type){ |
300 int s, i, j; | |
301 const int dec_count= w==8 ? 3 : 4; | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
302 int tmp[32*32]; |
2184 | 303 int level, ori; |
2967 | 304 static const int scale[2][2][4][4]={ |
2184 | 305 { |
306 { | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
307 // 9/7 8x8 dec=3 |
2184 | 308 {268, 239, 239, 213}, |
309 { 0, 224, 224, 152}, | |
310 { 0, 135, 135, 110}, | |
311 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
312 // 9/7 16x16 or 32x32 dec=4 |
2184 | 313 {344, 310, 310, 280}, |
314 { 0, 320, 320, 228}, | |
315 { 0, 175, 175, 136}, | |
316 { 0, 129, 129, 102}, | |
317 } | |
318 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
319 { |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
320 // 5/3 8x8 dec=3 |
2184 | 321 {275, 245, 245, 218}, |
322 { 0, 230, 230, 156}, | |
323 { 0, 138, 138, 113}, | |
324 },{ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
325 // 5/3 16x16 or 32x32 dec=4 |
2184 | 326 {352, 317, 317, 286}, |
327 { 0, 328, 328, 233}, | |
328 { 0, 180, 180, 140}, | |
329 { 0, 132, 132, 105}, | |
330 } | |
331 } | |
332 }; | |
333 | |
334 for (i = 0; i < h; i++) { | |
335 for (j = 0; j < w; j+=4) { | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
336 tmp[32*i+j+0] = (pix1[j+0] - pix2[j+0])<<4; |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
337 tmp[32*i+j+1] = (pix1[j+1] - pix2[j+1])<<4; |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
338 tmp[32*i+j+2] = (pix1[j+2] - pix2[j+2])<<4; |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
339 tmp[32*i+j+3] = (pix1[j+3] - pix2[j+3])<<4; |
2184 | 340 } |
341 pix1 += line_size; | |
342 pix2 += line_size; | |
343 } | |
2639 | 344 |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
345 ff_spatial_dwt(tmp, w, h, 32, type, dec_count); |
2184 | 346 |
347 s=0; | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
348 assert(w==h); |
2184 | 349 for(level=0; level<dec_count; level++){ |
350 for(ori= level ? 1 : 0; ori<4; ori++){ | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
351 int size= w>>(dec_count-level); |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
352 int sx= (ori&1) ? size : 0; |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
353 int stride= 32<<(dec_count-level); |
2184 | 354 int sy= (ori&2) ? stride>>1 : 0; |
2967 | 355 |
2184 | 356 for(i=0; i<size; i++){ |
357 for(j=0; j<size; j++){ | |
358 int v= tmp[sx + sy + i*stride + j] * scale[type][dec_count-3][level][ori]; | |
359 s += ABS(v); | |
360 } | |
361 } | |
362 } | |
363 } | |
2967 | 364 assert(s>=0); |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
365 return s>>9; |
2184 | 366 } |
367 | |
368 static int w53_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
369 return w_c(v, pix1, pix2, line_size, 8, h, 1); | |
370 } | |
371 | |
372 static int w97_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
373 return w_c(v, pix1, pix2, line_size, 8, h, 0); | |
374 } | |
375 | |
376 static int w53_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
377 return w_c(v, pix1, pix2, line_size, 16, h, 1); | |
378 } | |
379 | |
380 static int w97_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ | |
381 return w_c(v, pix1, pix2, line_size, 16, h, 0); | |
382 } | |
383 | |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
384 int w53_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
385 return w_c(v, pix1, pix2, line_size, 32, h, 1); |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
386 } |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
387 |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
388 int w97_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){ |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
389 return w_c(v, pix1, pix2, line_size, 32, h, 0); |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
390 } |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
391 #endif |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3248
diff
changeset
|
392 |
1064 | 393 static void get_pixels_c(DCTELEM *restrict block, const uint8_t *pixels, int line_size) |
0 | 394 { |
395 int i; | |
396 | |
397 /* read the pixels */ | |
398 for(i=0;i<8;i++) { | |
516 | 399 block[0] = pixels[0]; |
400 block[1] = pixels[1]; | |
401 block[2] = pixels[2]; | |
402 block[3] = pixels[3]; | |
403 block[4] = pixels[4]; | |
404 block[5] = pixels[5]; | |
405 block[6] = pixels[6]; | |
406 block[7] = pixels[7]; | |
407 pixels += line_size; | |
408 block += 8; | |
0 | 409 } |
410 } | |
411 | |
1064 | 412 static void diff_pixels_c(DCTELEM *restrict block, const uint8_t *s1, |
2979 | 413 const uint8_t *s2, int stride){ |
324 | 414 int i; |
415 | |
416 /* read the pixels */ | |
417 for(i=0;i<8;i++) { | |
516 | 418 block[0] = s1[0] - s2[0]; |
419 block[1] = s1[1] - s2[1]; | |
420 block[2] = s1[2] - s2[2]; | |
421 block[3] = s1[3] - s2[3]; | |
422 block[4] = s1[4] - s2[4]; | |
423 block[5] = s1[5] - s2[5]; | |
424 block[6] = s1[6] - s2[6]; | |
425 block[7] = s1[7] - s2[7]; | |
324 | 426 s1 += stride; |
427 s2 += stride; | |
516 | 428 block += 8; |
324 | 429 } |
430 } | |
431 | |
432 | |
1064 | 433 static void put_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 434 int line_size) |
0 | 435 { |
436 int i; | |
1064 | 437 uint8_t *cm = cropTbl + MAX_NEG_CROP; |
2967 | 438 |
0 | 439 /* read the pixels */ |
440 for(i=0;i<8;i++) { | |
516 | 441 pixels[0] = cm[block[0]]; |
442 pixels[1] = cm[block[1]]; | |
443 pixels[2] = cm[block[2]]; | |
444 pixels[3] = cm[block[3]]; | |
445 pixels[4] = cm[block[4]]; | |
446 pixels[5] = cm[block[5]]; | |
447 pixels[6] = cm[block[6]]; | |
448 pixels[7] = cm[block[7]]; | |
449 | |
450 pixels += line_size; | |
451 block += 8; | |
0 | 452 } |
453 } | |
454 | |
2256 | 455 static void put_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 456 int line_size) |
2256 | 457 { |
458 int i; | |
459 uint8_t *cm = cropTbl + MAX_NEG_CROP; | |
2967 | 460 |
2256 | 461 /* read the pixels */ |
462 for(i=0;i<4;i++) { | |
463 pixels[0] = cm[block[0]]; | |
464 pixels[1] = cm[block[1]]; | |
465 pixels[2] = cm[block[2]]; | |
466 pixels[3] = cm[block[3]]; | |
467 | |
468 pixels += line_size; | |
469 block += 8; | |
470 } | |
471 } | |
472 | |
2257 | 473 static void put_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, |
2979 | 474 int line_size) |
2257 | 475 { |
476 int i; | |
477 uint8_t *cm = cropTbl + MAX_NEG_CROP; | |
2967 | 478 |
2257 | 479 /* read the pixels */ |
480 for(i=0;i<2;i++) { | |
481 pixels[0] = cm[block[0]]; | |
482 pixels[1] = cm[block[1]]; | |
483 | |
484 pixels += line_size; | |
485 block += 8; | |
486 } | |
487 } | |
488 | |
2967 | 489 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
|
490 uint8_t *restrict pixels, |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
491 int line_size) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
492 { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
493 int i, j; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
494 |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
495 for (i = 0; i < 8; i++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
496 for (j = 0; j < 8; j++) { |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
497 if (*block < -128) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
498 *pixels = 0; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
499 else if (*block > 127) |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
500 *pixels = 255; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
501 else |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
502 *pixels = (uint8_t)(*block + 128); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
503 block++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
504 pixels++; |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
505 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
506 pixels += (line_size - 8); |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
507 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
508 } |
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
509 |
1064 | 510 static void add_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels, |
516 | 511 int line_size) |
0 | 512 { |
513 int i; | |
1064 | 514 uint8_t *cm = cropTbl + MAX_NEG_CROP; |
2967 | 515 |
0 | 516 /* read the pixels */ |
517 for(i=0;i<8;i++) { | |
516 | 518 pixels[0] = cm[pixels[0] + block[0]]; |
519 pixels[1] = cm[pixels[1] + block[1]]; | |
520 pixels[2] = cm[pixels[2] + block[2]]; | |
521 pixels[3] = cm[pixels[3] + block[3]]; | |
522 pixels[4] = cm[pixels[4] + block[4]]; | |
523 pixels[5] = cm[pixels[5] + block[5]]; | |
524 pixels[6] = cm[pixels[6] + block[6]]; | |
525 pixels[7] = cm[pixels[7] + block[7]]; | |
526 pixels += line_size; | |
527 block += 8; | |
0 | 528 } |
529 } | |
2256 | 530 |
531 static void add_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels, | |
532 int line_size) | |
533 { | |
534 int i; | |
535 uint8_t *cm = cropTbl + MAX_NEG_CROP; | |
2967 | 536 |
2256 | 537 /* read the pixels */ |
538 for(i=0;i<4;i++) { | |
539 pixels[0] = cm[pixels[0] + block[0]]; | |
540 pixels[1] = cm[pixels[1] + block[1]]; | |
541 pixels[2] = cm[pixels[2] + block[2]]; | |
542 pixels[3] = cm[pixels[3] + block[3]]; | |
543 pixels += line_size; | |
544 block += 8; | |
545 } | |
546 } | |
2257 | 547 |
548 static void add_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels, | |
549 int line_size) | |
550 { | |
551 int i; | |
552 uint8_t *cm = cropTbl + MAX_NEG_CROP; | |
2967 | 553 |
2257 | 554 /* read the pixels */ |
555 for(i=0;i<2;i++) { | |
556 pixels[0] = cm[pixels[0] + block[0]]; | |
557 pixels[1] = cm[pixels[1] + block[1]]; | |
558 pixels += line_size; | |
559 block += 8; | |
560 } | |
561 } | |
2763 | 562 |
563 static void add_pixels8_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
564 { | |
565 int i; | |
566 for(i=0;i<8;i++) { | |
567 pixels[0] += block[0]; | |
568 pixels[1] += block[1]; | |
569 pixels[2] += block[2]; | |
570 pixels[3] += block[3]; | |
571 pixels[4] += block[4]; | |
572 pixels[5] += block[5]; | |
573 pixels[6] += block[6]; | |
574 pixels[7] += block[7]; | |
575 pixels += line_size; | |
576 block += 8; | |
577 } | |
578 } | |
579 | |
580 static void add_pixels4_c(uint8_t *restrict pixels, DCTELEM *block, int line_size) | |
581 { | |
582 int i; | |
583 for(i=0;i<4;i++) { | |
584 pixels[0] += block[0]; | |
585 pixels[1] += block[1]; | |
586 pixels[2] += block[2]; | |
587 pixels[3] += block[3]; | |
588 pixels += line_size; | |
589 block += 4; | |
590 } | |
591 } | |
592 | |
385 | 593 #if 0 |
594 | |
595 #define PIXOP2(OPNAME, OP) \ | |
651 | 596 static void OPNAME ## _pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 597 {\ |
598 int i;\ | |
599 for(i=0; i<h; i++){\ | |
600 OP(*((uint64_t*)block), LD64(pixels));\ | |
601 pixels+=line_size;\ | |
602 block +=line_size;\ | |
603 }\ | |
604 }\ | |
605 \ | |
859 | 606 static void OPNAME ## _no_rnd_pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 607 {\ |
608 int i;\ | |
609 for(i=0; i<h; i++){\ | |
610 const uint64_t a= LD64(pixels );\ | |
611 const uint64_t b= LD64(pixels+1);\ | |
612 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ | |
613 pixels+=line_size;\ | |
614 block +=line_size;\ | |
615 }\ | |
616 }\ | |
617 \ | |
859 | 618 static void OPNAME ## _pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 619 {\ |
620 int i;\ | |
621 for(i=0; i<h; i++){\ | |
622 const uint64_t a= LD64(pixels );\ | |
623 const uint64_t b= LD64(pixels+1);\ | |
624 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ | |
625 pixels+=line_size;\ | |
626 block +=line_size;\ | |
627 }\ | |
628 }\ | |
629 \ | |
859 | 630 static void OPNAME ## _no_rnd_pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 631 {\ |
632 int i;\ | |
633 for(i=0; i<h; i++){\ | |
634 const uint64_t a= LD64(pixels );\ | |
635 const uint64_t b= LD64(pixels+line_size);\ | |
636 OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ | |
637 pixels+=line_size;\ | |
638 block +=line_size;\ | |
639 }\ | |
640 }\ | |
641 \ | |
859 | 642 static void OPNAME ## _pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 643 {\ |
644 int i;\ | |
645 for(i=0; i<h; i++){\ | |
646 const uint64_t a= LD64(pixels );\ | |
647 const uint64_t b= LD64(pixels+line_size);\ | |
648 OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\ | |
649 pixels+=line_size;\ | |
650 block +=line_size;\ | |
651 }\ | |
652 }\ | |
653 \ | |
859 | 654 static void OPNAME ## _pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 655 {\ |
656 int i;\ | |
657 const uint64_t a= LD64(pixels );\ | |
658 const uint64_t b= LD64(pixels+1);\ | |
659 uint64_t l0= (a&0x0303030303030303ULL)\ | |
660 + (b&0x0303030303030303ULL)\ | |
661 + 0x0202020202020202ULL;\ | |
662 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
663 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
664 uint64_t l1,h1;\ | |
665 \ | |
666 pixels+=line_size;\ | |
667 for(i=0; i<h; i+=2){\ | |
668 uint64_t a= LD64(pixels );\ | |
669 uint64_t b= LD64(pixels+1);\ | |
670 l1= (a&0x0303030303030303ULL)\ | |
671 + (b&0x0303030303030303ULL);\ | |
672 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
673 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
674 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
675 pixels+=line_size;\ | |
676 block +=line_size;\ | |
677 a= LD64(pixels );\ | |
678 b= LD64(pixels+1);\ | |
679 l0= (a&0x0303030303030303ULL)\ | |
680 + (b&0x0303030303030303ULL)\ | |
681 + 0x0202020202020202ULL;\ | |
682 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
683 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
684 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
685 pixels+=line_size;\ | |
686 block +=line_size;\ | |
687 }\ | |
688 }\ | |
689 \ | |
859 | 690 static void OPNAME ## _no_rnd_pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 691 {\ |
692 int i;\ | |
693 const uint64_t a= LD64(pixels );\ | |
694 const uint64_t b= LD64(pixels+1);\ | |
695 uint64_t l0= (a&0x0303030303030303ULL)\ | |
696 + (b&0x0303030303030303ULL)\ | |
697 + 0x0101010101010101ULL;\ | |
698 uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
699 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
700 uint64_t l1,h1;\ | |
701 \ | |
702 pixels+=line_size;\ | |
703 for(i=0; i<h; i+=2){\ | |
704 uint64_t a= LD64(pixels );\ | |
705 uint64_t b= LD64(pixels+1);\ | |
706 l1= (a&0x0303030303030303ULL)\ | |
707 + (b&0x0303030303030303ULL);\ | |
708 h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
709 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
710 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
711 pixels+=line_size;\ | |
712 block +=line_size;\ | |
713 a= LD64(pixels );\ | |
714 b= LD64(pixels+1);\ | |
715 l0= (a&0x0303030303030303ULL)\ | |
716 + (b&0x0303030303030303ULL)\ | |
717 + 0x0101010101010101ULL;\ | |
718 h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\ | |
719 + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\ | |
720 OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\ | |
721 pixels+=line_size;\ | |
722 block +=line_size;\ | |
723 }\ | |
724 }\ | |
725 \ | |
859 | 726 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels_c , 8)\ |
727 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels_x2_c , 8)\ | |
728 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels_y2_c , 8)\ | |
729 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels_xy2_c, 8)\ | |
730 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels_x2_c , 8)\ | |
731 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels_y2_c , 8)\ | |
732 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels_xy2_c, 8) | |
385 | 733 |
734 #define op_avg(a, b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEFEFEFEFEULL)>>1) ) | |
735 #else // 64 bit variant | |
736 | |
737 #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
|
738 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
|
739 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
|
740 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
|
741 OP(*((uint16_t*)(block )), LD16(pixels ));\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
742 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
|
743 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
|
744 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
745 }\ |
1168 | 746 static void OPNAME ## _pixels4_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
747 int i;\ | |
748 for(i=0; i<h; i++){\ | |
749 OP(*((uint32_t*)(block )), LD32(pixels ));\ | |
750 pixels+=line_size;\ | |
751 block +=line_size;\ | |
752 }\ | |
753 }\ | |
859 | 754 static void OPNAME ## _pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
385 | 755 int i;\ |
756 for(i=0; i<h; i++){\ | |
757 OP(*((uint32_t*)(block )), LD32(pixels ));\ | |
758 OP(*((uint32_t*)(block+4)), LD32(pixels+4));\ | |
759 pixels+=line_size;\ | |
760 block +=line_size;\ | |
761 }\ | |
762 }\ | |
859 | 763 static inline void OPNAME ## _no_rnd_pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
764 OPNAME ## _pixels8_c(block, pixels, line_size, h);\ | |
651 | 765 }\ |
385 | 766 \ |
651 | 767 static inline void OPNAME ## _no_rnd_pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
768 int src_stride1, int src_stride2, int h){\ | |
385 | 769 int i;\ |
770 for(i=0; i<h; i++){\ | |
651 | 771 uint32_t a,b;\ |
772 a= LD32(&src1[i*src_stride1 ]);\ | |
773 b= LD32(&src2[i*src_stride2 ]);\ | |
1264 | 774 OP(*((uint32_t*)&dst[i*dst_stride ]), no_rnd_avg32(a, b));\ |
651 | 775 a= LD32(&src1[i*src_stride1+4]);\ |
776 b= LD32(&src2[i*src_stride2+4]);\ | |
1264 | 777 OP(*((uint32_t*)&dst[i*dst_stride+4]), no_rnd_avg32(a, b));\ |
385 | 778 }\ |
779 }\ | |
780 \ | |
651 | 781 static inline void OPNAME ## _pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
782 int src_stride1, int src_stride2, int h){\ | |
385 | 783 int i;\ |
784 for(i=0; i<h; i++){\ | |
651 | 785 uint32_t a,b;\ |
786 a= LD32(&src1[i*src_stride1 ]);\ | |
787 b= LD32(&src2[i*src_stride2 ]);\ | |
1264 | 788 OP(*((uint32_t*)&dst[i*dst_stride ]), rnd_avg32(a, b));\ |
651 | 789 a= LD32(&src1[i*src_stride1+4]);\ |
790 b= LD32(&src2[i*src_stride2+4]);\ | |
1264 | 791 OP(*((uint32_t*)&dst[i*dst_stride+4]), rnd_avg32(a, b));\ |
385 | 792 }\ |
793 }\ | |
794 \ | |
1168 | 795 static inline void OPNAME ## _pixels4_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
796 int src_stride1, int src_stride2, int h){\ | |
797 int i;\ | |
798 for(i=0; i<h; i++){\ | |
799 uint32_t a,b;\ | |
800 a= LD32(&src1[i*src_stride1 ]);\ | |
801 b= LD32(&src2[i*src_stride2 ]);\ | |
1264 | 802 OP(*((uint32_t*)&dst[i*dst_stride ]), rnd_avg32(a, b));\ |
1168 | 803 }\ |
804 }\ | |
805 \ | |
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
|
806 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
|
807 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
|
808 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
|
809 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
|
810 uint32_t 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
|
811 a= LD16(&src1[i*src_stride1 ]);\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
812 b= LD16(&src2[i*src_stride2 ]);\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
813 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
|
814 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
815 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
816 \ |
651 | 817 static inline void OPNAME ## _pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ |
818 int src_stride1, int src_stride2, int h){\ | |
819 OPNAME ## _pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
820 OPNAME ## _pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
821 }\ | |
822 \ | |
823 static inline void OPNAME ## _no_rnd_pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \ | |
824 int src_stride1, int src_stride2, int h){\ | |
825 OPNAME ## _no_rnd_pixels8_l2(dst , src1 , src2 , dst_stride, src_stride1, src_stride2, h);\ | |
826 OPNAME ## _no_rnd_pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\ | |
827 }\ | |
828 \ | |
859 | 829 static inline void OPNAME ## _no_rnd_pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 830 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
831 }\ | |
832 \ | |
859 | 833 static inline void OPNAME ## _pixels8_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 834 OPNAME ## _pixels8_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\ |
835 }\ | |
836 \ | |
859 | 837 static inline void OPNAME ## _no_rnd_pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 838 OPNAME ## _no_rnd_pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
839 }\ | |
840 \ | |
859 | 841 static inline void OPNAME ## _pixels8_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
651 | 842 OPNAME ## _pixels8_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\ |
385 | 843 }\ |
844 \ | |
651 | 845 static inline void OPNAME ## _pixels8_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
846 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
847 int i;\ | |
848 for(i=0; i<h; i++){\ | |
849 uint32_t a, b, c, d, l0, l1, h0, h1;\ | |
850 a= LD32(&src1[i*src_stride1]);\ | |
851 b= LD32(&src2[i*src_stride2]);\ | |
852 c= LD32(&src3[i*src_stride3]);\ | |
853 d= LD32(&src4[i*src_stride4]);\ | |
854 l0= (a&0x03030303UL)\ | |
855 + (b&0x03030303UL)\ | |
856 + 0x02020202UL;\ | |
857 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
858 + ((b&0xFCFCFCFCUL)>>2);\ | |
859 l1= (c&0x03030303UL)\ | |
860 + (d&0x03030303UL);\ | |
861 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
862 + ((d&0xFCFCFCFCUL)>>2);\ | |
863 OP(*((uint32_t*)&dst[i*dst_stride]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
864 a= LD32(&src1[i*src_stride1+4]);\ | |
865 b= LD32(&src2[i*src_stride2+4]);\ | |
866 c= LD32(&src3[i*src_stride3+4]);\ | |
867 d= LD32(&src4[i*src_stride4+4]);\ | |
868 l0= (a&0x03030303UL)\ | |
869 + (b&0x03030303UL)\ | |
870 + 0x02020202UL;\ | |
871 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
872 + ((b&0xFCFCFCFCUL)>>2);\ | |
873 l1= (c&0x03030303UL)\ | |
874 + (d&0x03030303UL);\ | |
875 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
876 + ((d&0xFCFCFCFCUL)>>2);\ | |
877 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
878 }\ | |
879 }\ | |
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
|
880 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
881 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
|
882 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
|
883 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
884 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
885 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
|
886 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
|
887 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
888 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
889 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
|
890 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
|
891 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
892 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
893 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
|
894 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
|
895 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
896 \ |
651 | 897 static inline void OPNAME ## _no_rnd_pixels8_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
898 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
385 | 899 int i;\ |
900 for(i=0; i<h; i++){\ | |
651 | 901 uint32_t a, b, c, d, l0, l1, h0, h1;\ |
902 a= LD32(&src1[i*src_stride1]);\ | |
903 b= LD32(&src2[i*src_stride2]);\ | |
904 c= LD32(&src3[i*src_stride3]);\ | |
905 d= LD32(&src4[i*src_stride4]);\ | |
906 l0= (a&0x03030303UL)\ | |
907 + (b&0x03030303UL)\ | |
908 + 0x01010101UL;\ | |
909 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
910 + ((b&0xFCFCFCFCUL)>>2);\ | |
911 l1= (c&0x03030303UL)\ | |
912 + (d&0x03030303UL);\ | |
913 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
914 + ((d&0xFCFCFCFCUL)>>2);\ | |
915 OP(*((uint32_t*)&dst[i*dst_stride]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
916 a= LD32(&src1[i*src_stride1+4]);\ | |
917 b= LD32(&src2[i*src_stride2+4]);\ | |
918 c= LD32(&src3[i*src_stride3+4]);\ | |
919 d= LD32(&src4[i*src_stride4+4]);\ | |
920 l0= (a&0x03030303UL)\ | |
921 + (b&0x03030303UL)\ | |
922 + 0x01010101UL;\ | |
923 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
924 + ((b&0xFCFCFCFCUL)>>2);\ | |
925 l1= (c&0x03030303UL)\ | |
926 + (d&0x03030303UL);\ | |
927 h1= ((c&0xFCFCFCFCUL)>>2)\ | |
928 + ((d&0xFCFCFCFCUL)>>2);\ | |
929 OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
385 | 930 }\ |
931 }\ | |
651 | 932 static inline void OPNAME ## _pixels16_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ |
933 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
934 OPNAME ## _pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
935 OPNAME ## _pixels8_l4(dst+8, src1+8, src2+8, src3+8, src4+8, dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
936 }\ | |
937 static inline void OPNAME ## _no_rnd_pixels16_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\ | |
938 int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\ | |
939 OPNAME ## _no_rnd_pixels8_l4(dst , src1 , src2 , src3 , src4 , dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\ | |
940 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);\ | |
941 }\ | |
385 | 942 \ |
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
|
943 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
|
944 {\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
945 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
|
946 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
|
947 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
|
948 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
|
949 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
|
950 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
951 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
|
952 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
|
953 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
|
954 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
|
955 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
|
956 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
|
957 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
958 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
|
959 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
|
960 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
961 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
|
962 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
|
963 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
964 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
|
965 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
|
966 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
|
967 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
|
968 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
969 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
|
970 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
|
971 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
|
972 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
|
973 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
974 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
975 \ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
976 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
|
977 {\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
978 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
|
979 const uint32_t a= LD32(pixels );\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
980 const uint32_t b= LD32(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
|
981 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
|
982 + (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
|
983 + 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
|
984 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
|
985 + ((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
|
986 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
|
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 pixels+=line_size;\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
989 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
|
990 uint32_t a= LD32(pixels );\ |
85b71f9f7450
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 uint32_t b= LD32(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
|
992 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
|
993 + (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
|
994 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
|
995 + ((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
|
996 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
|
997 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
|
998 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
|
999 a= LD32(pixels );\ |
85b71f9f7450
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 b= LD32(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
|
1001 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
|
1002 + (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
|
1003 + 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
|
1004 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
|
1005 + ((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
|
1006 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
|
1007 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
|
1008 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
|
1009 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1010 }\ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1011 \ |
859 | 1012 static inline void OPNAME ## _pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1013 {\ |
1014 int j;\ | |
1015 for(j=0; j<2; j++){\ | |
1016 int i;\ | |
1017 const uint32_t a= LD32(pixels );\ | |
1018 const uint32_t b= LD32(pixels+1);\ | |
1019 uint32_t l0= (a&0x03030303UL)\ | |
1020 + (b&0x03030303UL)\ | |
1021 + 0x02020202UL;\ | |
1022 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1023 + ((b&0xFCFCFCFCUL)>>2);\ | |
1024 uint32_t l1,h1;\ | |
1025 \ | |
1026 pixels+=line_size;\ | |
1027 for(i=0; i<h; i+=2){\ | |
1028 uint32_t a= LD32(pixels );\ | |
1029 uint32_t b= LD32(pixels+1);\ | |
1030 l1= (a&0x03030303UL)\ | |
1031 + (b&0x03030303UL);\ | |
1032 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1033 + ((b&0xFCFCFCFCUL)>>2);\ | |
1034 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1035 pixels+=line_size;\ | |
1036 block +=line_size;\ | |
1037 a= LD32(pixels );\ | |
1038 b= LD32(pixels+1);\ | |
1039 l0= (a&0x03030303UL)\ | |
1040 + (b&0x03030303UL)\ | |
1041 + 0x02020202UL;\ | |
1042 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1043 + ((b&0xFCFCFCFCUL)>>2);\ | |
1044 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1045 pixels+=line_size;\ | |
1046 block +=line_size;\ | |
1047 }\ | |
1048 pixels+=4-line_size*(h+1);\ | |
1049 block +=4-line_size*h;\ | |
1050 }\ | |
1051 }\ | |
1052 \ | |
859 | 1053 static inline void OPNAME ## _no_rnd_pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\ |
385 | 1054 {\ |
1055 int j;\ | |
1056 for(j=0; j<2; j++){\ | |
1057 int i;\ | |
1058 const uint32_t a= LD32(pixels );\ | |
1059 const uint32_t b= LD32(pixels+1);\ | |
1060 uint32_t l0= (a&0x03030303UL)\ | |
1061 + (b&0x03030303UL)\ | |
1062 + 0x01010101UL;\ | |
1063 uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1064 + ((b&0xFCFCFCFCUL)>>2);\ | |
1065 uint32_t l1,h1;\ | |
1066 \ | |
1067 pixels+=line_size;\ | |
1068 for(i=0; i<h; i+=2){\ | |
1069 uint32_t a= LD32(pixels );\ | |
1070 uint32_t b= LD32(pixels+1);\ | |
1071 l1= (a&0x03030303UL)\ | |
1072 + (b&0x03030303UL);\ | |
1073 h1= ((a&0xFCFCFCFCUL)>>2)\ | |
1074 + ((b&0xFCFCFCFCUL)>>2);\ | |
1075 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1076 pixels+=line_size;\ | |
1077 block +=line_size;\ | |
1078 a= LD32(pixels );\ | |
1079 b= LD32(pixels+1);\ | |
1080 l0= (a&0x03030303UL)\ | |
1081 + (b&0x03030303UL)\ | |
1082 + 0x01010101UL;\ | |
1083 h0= ((a&0xFCFCFCFCUL)>>2)\ | |
1084 + ((b&0xFCFCFCFCUL)>>2);\ | |
1085 OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\ | |
1086 pixels+=line_size;\ | |
1087 block +=line_size;\ | |
1088 }\ | |
1089 pixels+=4-line_size*(h+1);\ | |
1090 block +=4-line_size*h;\ | |
1091 }\ | |
1092 }\ | |
1093 \ | |
859 | 1094 CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels8_c , 8)\ |
1095 CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels8_x2_c , 8)\ | |
1096 CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels8_y2_c , 8)\ | |
1097 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels8_xy2_c, 8)\ | |
1098 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_c , OPNAME ## _pixels8_c , 8)\ | |
1099 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels8_x2_c , 8)\ | |
1100 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels8_y2_c , 8)\ | |
1101 CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels8_xy2_c, 8)\ | |
651 | 1102 |
1264 | 1103 #define op_avg(a, b) a = rnd_avg32(a, b) |
385 | 1104 #endif |
1105 #define op_put(a, b) a = b | |
1106 | |
1107 PIXOP2(avg, op_avg) | |
1108 PIXOP2(put, op_put) | |
1109 #undef op_avg | |
1110 #undef op_put | |
1111 | |
0 | 1112 #define avg2(a,b) ((a+b+1)>>1) |
1113 #define avg4(a,b,c,d) ((a+b+c+d+2)>>2) | |
1114 | |
1864 | 1115 static void put_no_rnd_pixels16_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ |
1116 put_no_rnd_pixels16_l2(dst, a, b, stride, stride, stride, h); | |
1117 } | |
1118 | |
1119 static void put_no_rnd_pixels8_l2_c(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h){ | |
1120 put_no_rnd_pixels8_l2(dst, a, b, stride, stride, stride, h); | |
1121 } | |
753 | 1122 |
1064 | 1123 static void gmc1_c(uint8_t *dst, uint8_t *src, int stride, int h, int x16, int y16, int rounder) |
255 | 1124 { |
1125 const int A=(16-x16)*(16-y16); | |
1126 const int B=( x16)*(16-y16); | |
1127 const int C=(16-x16)*( y16); | |
1128 const int D=( x16)*( y16); | |
1129 int i; | |
1130 | |
1131 for(i=0; i<h; i++) | |
1132 { | |
651 | 1133 dst[0]= (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + rounder)>>8; |
1134 dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8; | |
1135 dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8; | |
1136 dst[3]= (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + rounder)>>8; | |
1137 dst[4]= (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + rounder)>>8; | |
1138 dst[5]= (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + rounder)>>8; | |
1139 dst[6]= (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + rounder)>>8; | |
1140 dst[7]= (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + rounder)>>8; | |
1141 dst+= stride; | |
1142 src+= stride; | |
255 | 1143 } |
1144 } | |
1145 | |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
1146 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, |
753 | 1147 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height) |
1148 { | |
1149 int y, vx, vy; | |
1150 const int s= 1<<shift; | |
2967 | 1151 |
753 | 1152 width--; |
1153 height--; | |
1154 | |
1155 for(y=0; y<h; y++){ | |
1156 int x; | |
1157 | |
1158 vx= ox; | |
1159 vy= oy; | |
1160 for(x=0; x<8; x++){ //XXX FIXME optimize | |
1161 int src_x, src_y, frac_x, frac_y, index; | |
1162 | |
1163 src_x= vx>>16; | |
1164 src_y= vy>>16; | |
1165 frac_x= src_x&(s-1); | |
1166 frac_y= src_y&(s-1); | |
1167 src_x>>=shift; | |
1168 src_y>>=shift; | |
2967 | 1169 |
753 | 1170 if((unsigned)src_x < width){ |
1171 if((unsigned)src_y < height){ | |
1172 index= src_x + src_y*stride; | |
1173 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) | |
1174 + src[index +1]* frac_x )*(s-frac_y) | |
1175 + ( src[index+stride ]*(s-frac_x) | |
1176 + src[index+stride+1]* frac_x )* frac_y | |
1177 + r)>>(shift*2); | |
1178 }else{ | |
2967 | 1179 index= src_x + clip(src_y, 0, height)*stride; |
1180 dst[y*stride + x]= ( ( src[index ]*(s-frac_x) | |
753 | 1181 + src[index +1]* frac_x )*s |
1182 + r)>>(shift*2); | |
1183 } | |
1184 }else{ | |
1185 if((unsigned)src_y < height){ | |
2967 | 1186 index= clip(src_x, 0, width) + src_y*stride; |
1187 dst[y*stride + x]= ( ( src[index ]*(s-frac_y) | |
753 | 1188 + src[index+stride ]* frac_y )*s |
1189 + r)>>(shift*2); | |
1190 }else{ | |
2967 | 1191 index= clip(src_x, 0, width) + clip(src_y, 0, height)*stride; |
753 | 1192 dst[y*stride + x]= src[index ]; |
1193 } | |
1194 } | |
2967 | 1195 |
753 | 1196 vx+= dxx; |
1197 vy+= dyx; | |
1198 } | |
1199 ox += dxy; | |
1200 oy += dyy; | |
1201 } | |
1202 } | |
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
|
1203 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1204 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
|
1205 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
|
1206 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
|
1207 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
|
1208 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
|
1209 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
|
1210 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1211 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1212 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1213 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
|
1214 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
|
1215 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
|
1216 for (j=0; j < width; j++) { |
2979 | 1217 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
|
1218 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1219 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
|
1220 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
|
1221 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1222 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1223 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1224 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
|
1225 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
|
1226 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
|
1227 for (j=0; j < width; j++) { |
2979 | 1228 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
|
1229 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1230 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
|
1231 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
|
1232 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1233 } |
2967 | 1234 |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1235 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
|
1236 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
|
1237 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
|
1238 for (j=0; j < width; j++) { |
2979 | 1239 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
|
1240 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1241 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
|
1242 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
|
1243 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1244 } |
2967 | 1245 |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1246 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
|
1247 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
|
1248 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
|
1249 for (j=0; j < width; j++) { |
2979 | 1250 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
|
1251 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1252 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
|
1253 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
|
1254 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1255 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1256 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1257 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
|
1258 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
|
1259 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
|
1260 for (j=0; j < width; j++) { |
2979 | 1261 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
|
1262 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1263 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
|
1264 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
|
1265 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1266 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1267 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1268 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
|
1269 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
|
1270 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
|
1271 for (j=0; j < width; j++) { |
2979 | 1272 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
|
1273 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1274 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
|
1275 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
|
1276 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1277 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1278 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1279 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
|
1280 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
|
1281 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
|
1282 for (j=0; j < width; j++) { |
2979 | 1283 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
|
1284 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1285 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
|
1286 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
|
1287 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1288 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1289 |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1290 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
|
1291 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
|
1292 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
|
1293 for (j=0; j < width; j++) { |
2979 | 1294 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
|
1295 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1296 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
|
1297 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
|
1298 } |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
1299 } |
1319 | 1300 |
1301 static inline void avg_tpel_pixels_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1302 switch(width){ | |
1303 case 2: avg_pixels2_c (dst, src, stride, height); break; | |
1304 case 4: avg_pixels4_c (dst, src, stride, height); break; | |
1305 case 8: avg_pixels8_c (dst, src, stride, height); break; | |
1306 case 16:avg_pixels16_c(dst, src, stride, height); break; | |
1307 } | |
1308 } | |
1309 | |
1310 static inline void avg_tpel_pixels_mc10_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1311 int i,j; | |
1312 for (i=0; i < height; i++) { | |
1313 for (j=0; j < width; j++) { | |
2979 | 1314 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1315 } |
1316 src += stride; | |
1317 dst += stride; | |
1318 } | |
1319 } | |
1320 | |
1321 static inline void avg_tpel_pixels_mc20_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1322 int i,j; | |
1323 for (i=0; i < height; i++) { | |
1324 for (j=0; j < width; j++) { | |
2979 | 1325 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+1] + 1)) >> 11) + 1) >> 1; |
1319 | 1326 } |
1327 src += stride; | |
1328 dst += stride; | |
1329 } | |
1330 } | |
2967 | 1331 |
1319 | 1332 static inline void avg_tpel_pixels_mc01_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1333 int i,j; | |
1334 for (i=0; i < height; i++) { | |
1335 for (j=0; j < width; j++) { | |
2979 | 1336 dst[j] = (dst[j] + ((683*(2*src[j] + src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1337 } |
1338 src += stride; | |
1339 dst += stride; | |
1340 } | |
1341 } | |
2967 | 1342 |
1319 | 1343 static inline void avg_tpel_pixels_mc11_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ |
1344 int i,j; | |
1345 for (i=0; i < height; i++) { | |
1346 for (j=0; j < width; j++) { | |
2979 | 1347 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 | 1348 } |
1349 src += stride; | |
1350 dst += stride; | |
1351 } | |
1352 } | |
1353 | |
1354 static inline void avg_tpel_pixels_mc12_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1355 int i,j; | |
1356 for (i=0; i < height; i++) { | |
1357 for (j=0; j < width; j++) { | |
2979 | 1358 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 | 1359 } |
1360 src += stride; | |
1361 dst += stride; | |
1362 } | |
1363 } | |
1364 | |
1365 static inline void avg_tpel_pixels_mc02_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1366 int i,j; | |
1367 for (i=0; i < height; i++) { | |
1368 for (j=0; j < width; j++) { | |
2979 | 1369 dst[j] = (dst[j] + ((683*(src[j] + 2*src[j+stride] + 1)) >> 11) + 1) >> 1; |
1319 | 1370 } |
1371 src += stride; | |
1372 dst += stride; | |
1373 } | |
1374 } | |
1375 | |
1376 static inline void avg_tpel_pixels_mc21_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1377 int i,j; | |
1378 for (i=0; i < height; i++) { | |
1379 for (j=0; j < width; j++) { | |
2979 | 1380 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 | 1381 } |
1382 src += stride; | |
1383 dst += stride; | |
1384 } | |
1385 } | |
1386 | |
1387 static inline void avg_tpel_pixels_mc22_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){ | |
1388 int i,j; | |
1389 for (i=0; i < height; i++) { | |
1390 for (j=0; j < width; j++) { | |
2979 | 1391 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 | 1392 } |
1393 src += stride; | |
1394 dst += stride; | |
1395 } | |
1396 } | |
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
|
1397 #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
|
1398 #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
|
1399 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
|
1400 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
|
1401 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
|
1402 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
|
1403 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
|
1404 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
|
1405 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
|
1406 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
|
1407 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
|
1408 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
|
1409 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
|
1410 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
|
1411 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
|
1412 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
|
1413 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
|
1414 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
|
1415 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
|
1416 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
|
1417 #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
|
1418 |
1168 | 1419 #define H264_CHROMA_MC(OPNAME, OP)\ |
1420 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){\ | |
1421 const int A=(8-x)*(8-y);\ | |
1422 const int B=( x)*(8-y);\ | |
1423 const int C=(8-x)*( y);\ | |
1424 const int D=( x)*( y);\ | |
1425 int i;\ | |
1426 \ | |
1427 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1428 \ | |
1429 for(i=0; i<h; i++)\ | |
1430 {\ | |
1431 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ | |
1432 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1433 dst+= stride;\ | |
1434 src+= stride;\ | |
1435 }\ | |
1436 }\ | |
1437 \ | |
1438 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){\ | |
1439 const int A=(8-x)*(8-y);\ | |
1440 const int B=( x)*(8-y);\ | |
1441 const int C=(8-x)*( y);\ | |
1442 const int D=( x)*( y);\ | |
1443 int i;\ | |
1444 \ | |
1445 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1446 \ | |
1447 for(i=0; i<h; i++)\ | |
1448 {\ | |
1449 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ | |
1450 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1451 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1452 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1453 dst+= stride;\ | |
1454 src+= stride;\ | |
1455 }\ | |
1456 }\ | |
1457 \ | |
1458 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){\ | |
1459 const int A=(8-x)*(8-y);\ | |
1460 const int B=( x)*(8-y);\ | |
1461 const int C=(8-x)*( y);\ | |
1462 const int D=( x)*( y);\ | |
1463 int i;\ | |
1464 \ | |
1465 assert(x<8 && y<8 && x>=0 && y>=0);\ | |
1466 \ | |
1467 for(i=0; i<h; i++)\ | |
1468 {\ | |
1469 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));\ | |
1470 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));\ | |
1471 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));\ | |
1472 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));\ | |
1473 OP(dst[4], (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5]));\ | |
1474 OP(dst[5], (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6]));\ | |
1475 OP(dst[6], (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7]));\ | |
1476 OP(dst[7], (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8]));\ | |
1477 dst+= stride;\ | |
1478 src+= stride;\ | |
1479 }\ | |
1480 } | |
1481 | |
1482 #define op_avg(a, b) a = (((a)+(((b) + 32)>>6)+1)>>1) | |
1483 #define op_put(a, b) a = (((b) + 32)>>6) | |
1484 | |
1485 H264_CHROMA_MC(put_ , op_put) | |
1486 H264_CHROMA_MC(avg_ , op_avg) | |
1487 #undef op_avg | |
1488 #undef op_put | |
1489 | |
3663 | 1490 static void put_no_rnd_h264_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){ |
1491 const int A=(8-x)*(8-y); | |
1492 const int B=( x)*(8-y); | |
1493 const int C=(8-x)*( y); | |
1494 const int D=( x)*( y); | |
1495 int i; | |
1496 | |
1497 assert(x<8 && y<8 && x>=0 && y>=0); | |
1498 | |
1499 for(i=0; i<h; i++) | |
1500 { | |
1501 dst[0] = (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6; | |
1502 dst[1] = (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6; | |
1503 dst[2] = (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6; | |
1504 dst[3] = (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6; | |
1505 dst[4] = (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6; | |
1506 dst[5] = (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6; | |
1507 dst[6] = (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6; | |
1508 dst[7] = (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6; | |
1509 dst+= stride; | |
1510 src+= stride; | |
1511 } | |
1512 } | |
1513 | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
1514 static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
1515 { |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
1516 int i; |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
1517 for(i=0; i<h; i++) |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
1518 { |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
1519 ST16(dst , LD16(src )); |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
1520 dst+=dstStride; |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
1521 src+=srcStride; |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
1522 } |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
1523 } |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
1524 |
1168 | 1525 static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) |
1526 { | |
1527 int i; | |
1528 for(i=0; i<h; i++) | |
1529 { | |
1530 ST32(dst , LD32(src )); | |
1531 dst+=dstStride; | |
1532 src+=srcStride; | |
1533 } | |
1534 } | |
1535 | |
1536 static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
1537 { | |
1538 int i; | |
1539 for(i=0; i<h; i++) | |
1540 { | |
1541 ST32(dst , LD32(src )); | |
1542 ST32(dst+4 , LD32(src+4 )); | |
1543 dst+=dstStride; | |
1544 src+=srcStride; | |
1545 } | |
1546 } | |
1547 | |
1548 static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
1549 { | |
1550 int i; | |
1551 for(i=0; i<h; i++) | |
1552 { | |
1553 ST32(dst , LD32(src )); | |
1554 ST32(dst+4 , LD32(src+4 )); | |
1555 ST32(dst+8 , LD32(src+8 )); | |
1556 ST32(dst+12, LD32(src+12)); | |
1557 dst+=dstStride; | |
1558 src+=srcStride; | |
1559 } | |
1560 } | |
753 | 1561 |
1064 | 1562 static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) |
255 | 1563 { |
1564 int i; | |
1565 for(i=0; i<h; i++) | |
1566 { | |
651 | 1567 ST32(dst , LD32(src )); |
1568 ST32(dst+4 , LD32(src+4 )); | |
1569 ST32(dst+8 , LD32(src+8 )); | |
1570 ST32(dst+12, LD32(src+12)); | |
1571 dst[16]= src[16]; | |
255 | 1572 dst+=dstStride; |
1573 src+=srcStride; | |
1574 } | |
1575 } | |
1576 | |
1064 | 1577 static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) |
255 | 1578 { |
1579 int i; | |
651 | 1580 for(i=0; i<h; i++) |
255 | 1581 { |
651 | 1582 ST32(dst , LD32(src )); |
1583 ST32(dst+4 , LD32(src+4 )); | |
1584 dst[8]= src[8]; | |
255 | 1585 dst+=dstStride; |
1586 src+=srcStride; | |
1587 } | |
1588 } | |
1589 | |
954 | 1590 |
651 | 1591 #define QPEL_MC(r, OPNAME, RND, OP) \ |
1064 | 1592 static void OPNAME ## mpeg4_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
1593 uint8_t *cm = cropTbl + MAX_NEG_CROP;\ | |
651 | 1594 int i;\ |
1595 for(i=0; i<h; i++)\ | |
1596 {\ | |
1597 OP(dst[0], (src[0]+src[1])*20 - (src[0]+src[2])*6 + (src[1]+src[3])*3 - (src[2]+src[4]));\ | |
1598 OP(dst[1], (src[1]+src[2])*20 - (src[0]+src[3])*6 + (src[0]+src[4])*3 - (src[1]+src[5]));\ | |
1599 OP(dst[2], (src[2]+src[3])*20 - (src[1]+src[4])*6 + (src[0]+src[5])*3 - (src[0]+src[6]));\ | |
1600 OP(dst[3], (src[3]+src[4])*20 - (src[2]+src[5])*6 + (src[1]+src[6])*3 - (src[0]+src[7]));\ | |
1601 OP(dst[4], (src[4]+src[5])*20 - (src[3]+src[6])*6 + (src[2]+src[7])*3 - (src[1]+src[8]));\ | |
1602 OP(dst[5], (src[5]+src[6])*20 - (src[4]+src[7])*6 + (src[3]+src[8])*3 - (src[2]+src[8]));\ | |
1603 OP(dst[6], (src[6]+src[7])*20 - (src[5]+src[8])*6 + (src[4]+src[8])*3 - (src[3]+src[7]));\ | |
1604 OP(dst[7], (src[7]+src[8])*20 - (src[6]+src[8])*6 + (src[5]+src[7])*3 - (src[4]+src[6]));\ | |
1605 dst+=dstStride;\ | |
1606 src+=srcStride;\ | |
1607 }\ | |
1608 }\ | |
1609 \ | |
1064 | 1610 static void OPNAME ## mpeg4_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
984 | 1611 const int w=8;\ |
1064 | 1612 uint8_t *cm = cropTbl + MAX_NEG_CROP;\ |
651 | 1613 int i;\ |
1614 for(i=0; i<w; i++)\ | |
1615 {\ | |
1616 const int src0= src[0*srcStride];\ | |
1617 const int src1= src[1*srcStride];\ | |
1618 const int src2= src[2*srcStride];\ | |
1619 const int src3= src[3*srcStride];\ | |
1620 const int src4= src[4*srcStride];\ | |
1621 const int src5= src[5*srcStride];\ | |
1622 const int src6= src[6*srcStride];\ | |
1623 const int src7= src[7*srcStride];\ | |
1624 const int src8= src[8*srcStride];\ | |
1625 OP(dst[0*dstStride], (src0+src1)*20 - (src0+src2)*6 + (src1+src3)*3 - (src2+src4));\ | |
1626 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*6 + (src0+src4)*3 - (src1+src5));\ | |
1627 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*6 + (src0+src5)*3 - (src0+src6));\ | |
1628 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*6 + (src1+src6)*3 - (src0+src7));\ | |
1629 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*6 + (src2+src7)*3 - (src1+src8));\ | |
1630 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*6 + (src3+src8)*3 - (src2+src8));\ | |
1631 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*6 + (src4+src8)*3 - (src3+src7));\ | |
1632 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src8)*6 + (src5+src7)*3 - (src4+src6));\ | |
1633 dst++;\ | |
1634 src++;\ | |
1635 }\ | |
1636 }\ | |
1637 \ | |
1064 | 1638 static void OPNAME ## mpeg4_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
1639 uint8_t *cm = cropTbl + MAX_NEG_CROP;\ | |
651 | 1640 int i;\ |
954 | 1641 \ |
651 | 1642 for(i=0; i<h; i++)\ |
1643 {\ | |
1644 OP(dst[ 0], (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]));\ | |
1645 OP(dst[ 1], (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]));\ | |
1646 OP(dst[ 2], (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]));\ | |
1647 OP(dst[ 3], (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]));\ | |
1648 OP(dst[ 4], (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]));\ | |
1649 OP(dst[ 5], (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]));\ | |
1650 OP(dst[ 6], (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]));\ | |
1651 OP(dst[ 7], (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]));\ | |
1652 OP(dst[ 8], (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]));\ | |
1653 OP(dst[ 9], (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]));\ | |
1654 OP(dst[10], (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]));\ | |
1655 OP(dst[11], (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]));\ | |
1656 OP(dst[12], (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]));\ | |
1657 OP(dst[13], (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]));\ | |
1658 OP(dst[14], (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]));\ | |
1659 OP(dst[15], (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]));\ | |
1660 dst+=dstStride;\ | |
1661 src+=srcStride;\ | |
1662 }\ | |
255 | 1663 }\ |
1664 \ | |
1064 | 1665 static void OPNAME ## mpeg4_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
1666 uint8_t *cm = cropTbl + MAX_NEG_CROP;\ | |
651 | 1667 int i;\ |
954 | 1668 const int w=16;\ |
651 | 1669 for(i=0; i<w; i++)\ |
1670 {\ | |
1671 const int src0= src[0*srcStride];\ | |
1672 const int src1= src[1*srcStride];\ | |
1673 const int src2= src[2*srcStride];\ | |
1674 const int src3= src[3*srcStride];\ | |
1675 const int src4= src[4*srcStride];\ | |
1676 const int src5= src[5*srcStride];\ | |
1677 const int src6= src[6*srcStride];\ | |
1678 const int src7= src[7*srcStride];\ | |
1679 const int src8= src[8*srcStride];\ | |
1680 const int src9= src[9*srcStride];\ | |
1681 const int src10= src[10*srcStride];\ | |
1682 const int src11= src[11*srcStride];\ | |
1683 const int src12= src[12*srcStride];\ | |
1684 const int src13= src[13*srcStride];\ | |
1685 const int src14= src[14*srcStride];\ | |
1686 const int src15= src[15*srcStride];\ | |
1687 const int src16= src[16*srcStride];\ | |
1688 OP(dst[ 0*dstStride], (src0 +src1 )*20 - (src0 +src2 )*6 + (src1 +src3 )*3 - (src2 +src4 ));\ | |
1689 OP(dst[ 1*dstStride], (src1 +src2 )*20 - (src0 +src3 )*6 + (src0 +src4 )*3 - (src1 +src5 ));\ | |
1690 OP(dst[ 2*dstStride], (src2 +src3 )*20 - (src1 +src4 )*6 + (src0 +src5 )*3 - (src0 +src6 ));\ | |
1691 OP(dst[ 3*dstStride], (src3 +src4 )*20 - (src2 +src5 )*6 + (src1 +src6 )*3 - (src0 +src7 ));\ | |
1692 OP(dst[ 4*dstStride], (src4 +src5 )*20 - (src3 +src6 )*6 + (src2 +src7 )*3 - (src1 +src8 ));\ | |
1693 OP(dst[ 5*dstStride], (src5 +src6 )*20 - (src4 +src7 )*6 + (src3 +src8 )*3 - (src2 +src9 ));\ | |
1694 OP(dst[ 6*dstStride], (src6 +src7 )*20 - (src5 +src8 )*6 + (src4 +src9 )*3 - (src3 +src10));\ | |
1695 OP(dst[ 7*dstStride], (src7 +src8 )*20 - (src6 +src9 )*6 + (src5 +src10)*3 - (src4 +src11));\ | |
1696 OP(dst[ 8*dstStride], (src8 +src9 )*20 - (src7 +src10)*6 + (src6 +src11)*3 - (src5 +src12));\ | |
1697 OP(dst[ 9*dstStride], (src9 +src10)*20 - (src8 +src11)*6 + (src7 +src12)*3 - (src6 +src13));\ | |
1698 OP(dst[10*dstStride], (src10+src11)*20 - (src9 +src12)*6 + (src8 +src13)*3 - (src7 +src14));\ | |
1699 OP(dst[11*dstStride], (src11+src12)*20 - (src10+src13)*6 + (src9 +src14)*3 - (src8 +src15));\ | |
1700 OP(dst[12*dstStride], (src12+src13)*20 - (src11+src14)*6 + (src10+src15)*3 - (src9 +src16));\ | |
1701 OP(dst[13*dstStride], (src13+src14)*20 - (src12+src15)*6 + (src11+src16)*3 - (src10+src16));\ | |
1702 OP(dst[14*dstStride], (src14+src15)*20 - (src13+src16)*6 + (src12+src16)*3 - (src11+src15));\ | |
1703 OP(dst[15*dstStride], (src15+src16)*20 - (src14+src16)*6 + (src13+src15)*3 - (src12+src14));\ | |
1704 dst++;\ | |
1705 src++;\ | |
1706 }\ | |
255 | 1707 }\ |
1708 \ | |
1064 | 1709 static void OPNAME ## qpel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1710 OPNAME ## pixels8_c(dst, src, stride, 8);\ |
255 | 1711 }\ |
1712 \ | |
1064 | 1713 static void OPNAME ## qpel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
1714 uint8_t half[64];\ | |
651 | 1715 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1716 OPNAME ## pixels8_l2(dst, src, half, stride, stride, 8, 8);\ | |
1717 }\ | |
1718 \ | |
1064 | 1719 static void OPNAME ## qpel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 1720 OPNAME ## mpeg4_qpel8_h_lowpass(dst, src, stride, stride, 8);\ |
255 | 1721 }\ |
1722 \ | |
1064 | 1723 static void OPNAME ## qpel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
1724 uint8_t half[64];\ | |
651 | 1725 put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\ |
1726 OPNAME ## pixels8_l2(dst, src+1, half, stride, stride, 8, 8);\ | |
1727 }\ | |
1728 \ | |
1064 | 1729 static void OPNAME ## qpel8_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
1730 uint8_t full[16*9];\ | |
1731 uint8_t half[64];\ | |
651 | 1732 copy_block9(full, src, 16, stride, 9);\ |
984 | 1733 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1734 OPNAME ## pixels8_l2(dst, full, half, stride, 16, 8, 8);\ |
1735 }\ | |
1736 \ | |
1064 | 1737 static void OPNAME ## qpel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
1738 uint8_t full[16*9];\ | |
651 | 1739 copy_block9(full, src, 16, stride, 9);\ |
984 | 1740 OPNAME ## mpeg4_qpel8_v_lowpass(dst, full, stride, 16);\ |
255 | 1741 }\ |
1742 \ | |
1064 | 1743 static void OPNAME ## qpel8_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
1744 uint8_t full[16*9];\ | |
1745 uint8_t half[64];\ | |
651 | 1746 copy_block9(full, src, 16, stride, 9);\ |
984 | 1747 put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\ |
651 | 1748 OPNAME ## pixels8_l2(dst, full+16, half, stride, 16, 8, 8);\ |
1749 }\ | |
1064 | 1750 void ff_ ## OPNAME ## qpel8_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1751 uint8_t full[16*9];\ | |
1752 uint8_t halfH[72];\ | |
1753 uint8_t halfV[64];\ | |
1754 uint8_t halfHV[64];\ | |
651 | 1755 copy_block9(full, src, 16, stride, 9);\ |
1756 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1757 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1758 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1759 OPNAME ## pixels8_l4(dst, full, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1760 }\ |
1064 | 1761 static void OPNAME ## qpel8_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
1762 uint8_t full[16*9];\ | |
1763 uint8_t halfH[72];\ | |
1764 uint8_t halfHV[64];\ | |
984 | 1765 copy_block9(full, src, 16, stride, 9);\ |
1766 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1767 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1768 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1769 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1770 }\ | |
1064 | 1771 void ff_ ## OPNAME ## qpel8_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1772 uint8_t full[16*9];\ | |
1773 uint8_t halfH[72];\ | |
1774 uint8_t halfV[64];\ | |
1775 uint8_t halfHV[64];\ | |
651 | 1776 copy_block9(full, src, 16, stride, 9);\ |
1777 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1778 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1779 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1780 OPNAME ## pixels8_l4(dst, full+1, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1781 }\ |
1064 | 1782 static void OPNAME ## qpel8_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
1783 uint8_t full[16*9];\ | |
1784 uint8_t halfH[72];\ | |
1785 uint8_t halfHV[64];\ | |
984 | 1786 copy_block9(full, src, 16, stride, 9);\ |
1787 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1788 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1789 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1790 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ | |
1791 }\ | |
1064 | 1792 void ff_ ## OPNAME ## qpel8_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1793 uint8_t full[16*9];\ | |
1794 uint8_t halfH[72];\ | |
1795 uint8_t halfV[64];\ | |
1796 uint8_t halfHV[64];\ | |
651 | 1797 copy_block9(full, src, 16, stride, 9);\ |
1798 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1799 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1800 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1801 OPNAME ## pixels8_l4(dst, full+16, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
1802 }\ | |
1064 | 1803 static void OPNAME ## qpel8_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
1804 uint8_t full[16*9];\ | |
1805 uint8_t halfH[72];\ | |
1806 uint8_t halfHV[64];\ | |
984 | 1807 copy_block9(full, src, 16, stride, 9);\ |
1808 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1809 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1810 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1811 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1812 }\ | |
1064 | 1813 void ff_ ## OPNAME ## qpel8_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1814 uint8_t full[16*9];\ | |
1815 uint8_t halfH[72];\ | |
1816 uint8_t halfV[64];\ | |
1817 uint8_t halfHV[64];\ | |
651 | 1818 copy_block9(full, src, 16, stride, 9);\ |
1819 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full , 8, 16, 9);\ | |
984 | 1820 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1821 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1822 OPNAME ## pixels8_l4(dst, full+17, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\ |
255 | 1823 }\ |
1064 | 1824 static void OPNAME ## qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
1825 uint8_t full[16*9];\ | |
1826 uint8_t halfH[72];\ | |
1827 uint8_t halfHV[64];\ | |
984 | 1828 copy_block9(full, src, 16, stride, 9);\ |
1829 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1830 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1831 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
1832 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ | |
1833 }\ | |
1064 | 1834 static void OPNAME ## qpel8_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
1835 uint8_t halfH[72];\ | |
1836 uint8_t halfHV[64];\ | |
651 | 1837 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1838 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1839 OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\ |
1840 }\ | |
1064 | 1841 static void OPNAME ## qpel8_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
1842 uint8_t halfH[72];\ | |
1843 uint8_t halfHV[64];\ | |
651 | 1844 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1845 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ |
651 | 1846 OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\ |
1847 }\ | |
1064 | 1848 void ff_ ## OPNAME ## qpel8_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1849 uint8_t full[16*9];\ | |
1850 uint8_t halfH[72];\ | |
1851 uint8_t halfV[64];\ | |
1852 uint8_t halfHV[64];\ | |
651 | 1853 copy_block9(full, src, 16, stride, 9);\ |
1854 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1855 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\ |
1856 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1857 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
255 | 1858 }\ |
1064 | 1859 static void OPNAME ## qpel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
1860 uint8_t full[16*9];\ | |
1861 uint8_t halfH[72];\ | |
984 | 1862 copy_block9(full, src, 16, stride, 9);\ |
1863 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1864 put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\ | |
1865 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1866 }\ | |
1064 | 1867 void ff_ ## OPNAME ## qpel8_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1868 uint8_t full[16*9];\ | |
1869 uint8_t halfH[72];\ | |
1870 uint8_t halfV[64];\ | |
1871 uint8_t halfHV[64];\ | |
651 | 1872 copy_block9(full, src, 16, stride, 9);\ |
1873 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
984 | 1874 put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\ |
1875 put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\ | |
651 | 1876 OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\ |
1877 }\ | |
1064 | 1878 static void OPNAME ## qpel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
1879 uint8_t full[16*9];\ | |
1880 uint8_t halfH[72];\ | |
984 | 1881 copy_block9(full, src, 16, stride, 9);\ |
1882 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\ | |
1883 put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\ | |
1884 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ | |
1885 }\ | |
1064 | 1886 static void OPNAME ## qpel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
1887 uint8_t halfH[72];\ | |
651 | 1888 put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\ |
984 | 1889 OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\ |
651 | 1890 }\ |
1064 | 1891 static void OPNAME ## qpel16_mc00_c (uint8_t *dst, uint8_t *src, int stride){\ |
859 | 1892 OPNAME ## pixels16_c(dst, src, stride, 16);\ |
255 | 1893 }\ |
651 | 1894 \ |
1064 | 1895 static void OPNAME ## qpel16_mc10_c(uint8_t *dst, uint8_t *src, int stride){\ |
1896 uint8_t half[256];\ | |
651 | 1897 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
1898 OPNAME ## pixels16_l2(dst, src, half, stride, stride, 16, 16);\ | |
1899 }\ | |
1900 \ | |
1064 | 1901 static void OPNAME ## qpel16_mc20_c(uint8_t *dst, uint8_t *src, int stride){\ |
651 | 1902 OPNAME ## mpeg4_qpel16_h_lowpass(dst, src, stride, stride, 16);\ |
1903 }\ | |
1904 \ | |
1064 | 1905 static void OPNAME ## qpel16_mc30_c(uint8_t *dst, uint8_t *src, int stride){\ |
1906 uint8_t half[256];\ | |
651 | 1907 put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\ |
1908 OPNAME ## pixels16_l2(dst, src+1, half, stride, stride, 16, 16);\ | |
1909 }\ | |
1910 \ | |
1064 | 1911 static void OPNAME ## qpel16_mc01_c(uint8_t *dst, uint8_t *src, int stride){\ |
1912 uint8_t full[24*17];\ | |
1913 uint8_t half[256];\ | |
651 | 1914 copy_block17(full, src, 24, stride, 17);\ |
954 | 1915 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 1916 OPNAME ## pixels16_l2(dst, full, half, stride, 24, 16, 16);\ |
255 | 1917 }\ |
651 | 1918 \ |
1064 | 1919 static void OPNAME ## qpel16_mc02_c(uint8_t *dst, uint8_t *src, int stride){\ |
1920 uint8_t full[24*17];\ | |
651 | 1921 copy_block17(full, src, 24, stride, 17);\ |
954 | 1922 OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24);\ |
651 | 1923 }\ |
1924 \ | |
1064 | 1925 static void OPNAME ## qpel16_mc03_c(uint8_t *dst, uint8_t *src, int stride){\ |
1926 uint8_t full[24*17];\ | |
1927 uint8_t half[256];\ | |
651 | 1928 copy_block17(full, src, 24, stride, 17);\ |
954 | 1929 put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\ |
651 | 1930 OPNAME ## pixels16_l2(dst, full+24, half, stride, 24, 16, 16);\ |
255 | 1931 }\ |
1064 | 1932 void ff_ ## OPNAME ## qpel16_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1933 uint8_t full[24*17];\ | |
1934 uint8_t halfH[272];\ | |
1935 uint8_t halfV[256];\ | |
1936 uint8_t halfHV[256];\ | |
651 | 1937 copy_block17(full, src, 24, stride, 17);\ |
1938 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1939 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
1940 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1941 OPNAME ## pixels16_l4(dst, full, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
1942 }\ | |
1064 | 1943 static void OPNAME ## qpel16_mc11_c(uint8_t *dst, uint8_t *src, int stride){\ |
1944 uint8_t full[24*17];\ | |
1945 uint8_t halfH[272];\ | |
1946 uint8_t halfHV[256];\ | |
984 | 1947 copy_block17(full, src, 24, stride, 17);\ |
1948 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1949 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
1950 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1951 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
1952 }\ | |
1064 | 1953 void ff_ ## OPNAME ## qpel16_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1954 uint8_t full[24*17];\ | |
1955 uint8_t halfH[272];\ | |
1956 uint8_t halfV[256];\ | |
1957 uint8_t halfHV[256];\ | |
651 | 1958 copy_block17(full, src, 24, stride, 17);\ |
1959 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1960 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
1961 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1962 OPNAME ## pixels16_l4(dst, full+1, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
1963 }\ | |
1064 | 1964 static void OPNAME ## qpel16_mc31_c(uint8_t *dst, uint8_t *src, int stride){\ |
1965 uint8_t full[24*17];\ | |
1966 uint8_t halfH[272];\ | |
1967 uint8_t halfHV[256];\ | |
984 | 1968 copy_block17(full, src, 24, stride, 17);\ |
1969 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1970 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
1971 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1972 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ | |
1973 }\ | |
1064 | 1974 void ff_ ## OPNAME ## qpel16_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1975 uint8_t full[24*17];\ | |
1976 uint8_t halfH[272];\ | |
1977 uint8_t halfV[256];\ | |
1978 uint8_t halfHV[256];\ | |
651 | 1979 copy_block17(full, src, 24, stride, 17);\ |
1980 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 1981 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
1982 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 1983 OPNAME ## pixels16_l4(dst, full+24, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
255 | 1984 }\ |
1064 | 1985 static void OPNAME ## qpel16_mc13_c(uint8_t *dst, uint8_t *src, int stride){\ |
1986 uint8_t full[24*17];\ | |
1987 uint8_t halfH[272];\ | |
1988 uint8_t halfHV[256];\ | |
984 | 1989 copy_block17(full, src, 24, stride, 17);\ |
1990 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
1991 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
1992 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
1993 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
1994 }\ | |
1064 | 1995 void ff_ ## OPNAME ## qpel16_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
1996 uint8_t full[24*17];\ | |
1997 uint8_t halfH[272];\ | |
1998 uint8_t halfV[256];\ | |
1999 uint8_t halfHV[256];\ | |
651 | 2000 copy_block17(full, src, 24, stride, 17);\ |
2001 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full , 16, 24, 17);\ | |
954 | 2002 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
2003 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2004 OPNAME ## pixels16_l4(dst, full+25, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\ |
2005 }\ | |
1064 | 2006 static void OPNAME ## qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride){\ |
2007 uint8_t full[24*17];\ | |
2008 uint8_t halfH[272];\ | |
2009 uint8_t halfHV[256];\ | |
984 | 2010 copy_block17(full, src, 24, stride, 17);\ |
2011 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2012 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2013 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
2014 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ | |
2015 }\ | |
1064 | 2016 static void OPNAME ## qpel16_mc21_c(uint8_t *dst, uint8_t *src, int stride){\ |
2017 uint8_t halfH[272];\ | |
2018 uint8_t halfHV[256];\ | |
651 | 2019 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2020 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 2021 OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\ |
255 | 2022 }\ |
1064 | 2023 static void OPNAME ## qpel16_mc23_c(uint8_t *dst, uint8_t *src, int stride){\ |
2024 uint8_t halfH[272];\ | |
2025 uint8_t halfHV[256];\ | |
651 | 2026 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2027 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ |
651 | 2028 OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\ |
2029 }\ | |
1064 | 2030 void ff_ ## OPNAME ## qpel16_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2031 uint8_t full[24*17];\ | |
2032 uint8_t halfH[272];\ | |
2033 uint8_t halfV[256];\ | |
2034 uint8_t halfHV[256];\ | |
651 | 2035 copy_block17(full, src, 24, stride, 17);\ |
2036 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2037 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\ |
2038 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2039 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
255 | 2040 }\ |
1064 | 2041 static void OPNAME ## qpel16_mc12_c(uint8_t *dst, uint8_t *src, int stride){\ |
2042 uint8_t full[24*17];\ | |
2043 uint8_t halfH[272];\ | |
984 | 2044 copy_block17(full, src, 24, stride, 17);\ |
2045 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2046 put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\ | |
2047 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2048 }\ | |
1064 | 2049 void ff_ ## OPNAME ## qpel16_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\ |
2050 uint8_t full[24*17];\ | |
2051 uint8_t halfH[272];\ | |
2052 uint8_t halfV[256];\ | |
2053 uint8_t halfHV[256];\ | |
651 | 2054 copy_block17(full, src, 24, stride, 17);\ |
2055 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
954 | 2056 put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\ |
2057 put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\ | |
651 | 2058 OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\ |
2059 }\ | |
1064 | 2060 static void OPNAME ## qpel16_mc32_c(uint8_t *dst, uint8_t *src, int stride){\ |
2061 uint8_t full[24*17];\ | |
2062 uint8_t halfH[272];\ | |
984 | 2063 copy_block17(full, src, 24, stride, 17);\ |
2064 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\ | |
2065 put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\ | |
2066 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ | |
2067 }\ | |
1064 | 2068 static void OPNAME ## qpel16_mc22_c(uint8_t *dst, uint8_t *src, int stride){\ |
2069 uint8_t halfH[272];\ | |
651 | 2070 put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\ |
954 | 2071 OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\ |
859 | 2072 } |
255 | 2073 |
651 | 2074 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) |
2075 #define op_avg_no_rnd(a, b) a = (((a)+cm[((b) + 15)>>5])>>1) | |
2076 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2077 #define op_put_no_rnd(a, b) a = cm[((b) + 15)>>5] | |
2078 | |
2079 QPEL_MC(0, put_ , _ , op_put) | |
2080 QPEL_MC(1, put_no_rnd_, _no_rnd_, op_put_no_rnd) | |
2081 QPEL_MC(0, avg_ , _ , op_avg) | |
2082 //QPEL_MC(1, avg_no_rnd , _ , op_avg) | |
2083 #undef op_avg | |
2084 #undef op_avg_no_rnd | |
2085 #undef op_put | |
2086 #undef op_put_no_rnd | |
255 | 2087 |
1168 | 2088 #if 1 |
2089 #define H264_LOWPASS(OPNAME, OP, OP2) \ | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2090 static void OPNAME ## h264_qpel2_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2091 const int h=2;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2092 uint8_t *cm = cropTbl + MAX_NEG_CROP;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2093 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2094 for(i=0; i<h; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2095 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2096 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
|
2097 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
|
2098 dst+=dstStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2099 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2100 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2101 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2102 \ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2103 static void OPNAME ## h264_qpel2_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2104 const int w=2;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2105 uint8_t *cm = cropTbl + MAX_NEG_CROP;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2106 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2107 for(i=0; i<w; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2108 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2109 const int srcB= src[-2*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2110 const int srcA= src[-1*srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2111 const int src0= src[0 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2112 const int src1= src[1 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2113 const int src2= src[2 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2114 const int src3= src[3 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2115 const int src4= src[4 *srcStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2116 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
|
2117 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
|
2118 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2119 src++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2120 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2121 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2122 \ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2123 static void OPNAME ## h264_qpel2_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2124 const int h=2;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2125 const int w=2;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2126 uint8_t *cm = cropTbl + MAX_NEG_CROP;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2127 int i;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2128 src -= 2*srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2129 for(i=0; i<h+5; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2130 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2131 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
|
2132 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
|
2133 tmp+=tmpStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2134 src+=srcStride;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2135 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2136 tmp -= tmpStride*(h+5-2);\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2137 for(i=0; i<w; i++)\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2138 {\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2139 const int tmpB= tmp[-2*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2140 const int tmpA= tmp[-1*tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2141 const int tmp0= tmp[0 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2142 const int tmp1= tmp[1 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2143 const int tmp2= tmp[2 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2144 const int tmp3= tmp[3 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2145 const int tmp4= tmp[4 *tmpStride];\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2146 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
|
2147 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
|
2148 dst++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2149 tmp++;\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2150 }\ |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
2151 }\ |
1168 | 2152 static void OPNAME ## h264_qpel4_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
2153 const int h=4;\ | |
2154 uint8_t *cm = cropTbl + MAX_NEG_CROP;\ | |
2155 int i;\ | |
2156 for(i=0; i<h; i++)\ | |
2157 {\ | |
2158 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]));\ | |
2159 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]));\ | |
2160 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]));\ | |
2161 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]));\ | |
2162 dst+=dstStride;\ | |
2163 src+=srcStride;\ | |
2164 }\ | |
2165 }\ | |
2166 \ | |
2167 static void OPNAME ## h264_qpel4_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2168 const int w=4;\ | |
2169 uint8_t *cm = cropTbl + MAX_NEG_CROP;\ | |
2170 int i;\ | |
2171 for(i=0; i<w; i++)\ | |
2172 {\ | |
2173 const int srcB= src[-2*srcStride];\ | |
2174 const int srcA= src[-1*srcStride];\ | |
2175 const int src0= src[0 *srcStride];\ | |
2176 const int src1= src[1 *srcStride];\ | |
2177 const int src2= src[2 *srcStride];\ | |
2178 const int src3= src[3 *srcStride];\ | |
2179 const int src4= src[4 *srcStride];\ | |
2180 const int src5= src[5 *srcStride];\ | |
2181 const int src6= src[6 *srcStride];\ | |
2182 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2183 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2184 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2185 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2186 dst++;\ | |
2187 src++;\ | |
2188 }\ | |
2189 }\ | |
2190 \ | |
2191 static void OPNAME ## h264_qpel4_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2192 const int h=4;\ | |
2193 const int w=4;\ | |
2194 uint8_t *cm = cropTbl + MAX_NEG_CROP;\ | |
2195 int i;\ | |
2196 src -= 2*srcStride;\ | |
2197 for(i=0; i<h+5; i++)\ | |
2198 {\ | |
2199 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]);\ | |
2200 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]);\ | |
2201 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]);\ | |
2202 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]);\ | |
2203 tmp+=tmpStride;\ | |
2204 src+=srcStride;\ | |
2205 }\ | |
2206 tmp -= tmpStride*(h+5-2);\ | |
2207 for(i=0; i<w; i++)\ | |
2208 {\ | |
2209 const int tmpB= tmp[-2*tmpStride];\ | |
2210 const int tmpA= tmp[-1*tmpStride];\ | |
2211 const int tmp0= tmp[0 *tmpStride];\ | |
2212 const int tmp1= tmp[1 *tmpStride];\ | |
2213 const int tmp2= tmp[2 *tmpStride];\ | |
2214 const int tmp3= tmp[3 *tmpStride];\ | |
2215 const int tmp4= tmp[4 *tmpStride];\ | |
2216 const int tmp5= tmp[5 *tmpStride];\ | |
2217 const int tmp6= tmp[6 *tmpStride];\ | |
2218 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2219 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2220 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2221 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2222 dst++;\ | |
2223 tmp++;\ | |
2224 }\ | |
2225 }\ | |
2226 \ | |
2227 static void OPNAME ## h264_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2228 const int h=8;\ | |
2229 uint8_t *cm = cropTbl + MAX_NEG_CROP;\ | |
2230 int i;\ | |
2231 for(i=0; i<h; i++)\ | |
2232 {\ | |
2233 OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]));\ | |
2234 OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]));\ | |
2235 OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]));\ | |
2236 OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]));\ | |
2237 OP(dst[4], (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]));\ | |
2238 OP(dst[5], (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]));\ | |
2239 OP(dst[6], (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]));\ | |
2240 OP(dst[7], (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]));\ | |
2241 dst+=dstStride;\ | |
2242 src+=srcStride;\ | |
2243 }\ | |
2244 }\ | |
2245 \ | |
2246 static void OPNAME ## h264_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2247 const int w=8;\ | |
2248 uint8_t *cm = cropTbl + MAX_NEG_CROP;\ | |
2249 int i;\ | |
2250 for(i=0; i<w; i++)\ | |
2251 {\ | |
2252 const int srcB= src[-2*srcStride];\ | |
2253 const int srcA= src[-1*srcStride];\ | |
2254 const int src0= src[0 *srcStride];\ | |
2255 const int src1= src[1 *srcStride];\ | |
2256 const int src2= src[2 *srcStride];\ | |
2257 const int src3= src[3 *srcStride];\ | |
2258 const int src4= src[4 *srcStride];\ | |
2259 const int src5= src[5 *srcStride];\ | |
2260 const int src6= src[6 *srcStride];\ | |
2261 const int src7= src[7 *srcStride];\ | |
2262 const int src8= src[8 *srcStride];\ | |
2263 const int src9= src[9 *srcStride];\ | |
2264 const int src10=src[10*srcStride];\ | |
2265 OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\ | |
2266 OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\ | |
2267 OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\ | |
2268 OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\ | |
2269 OP(dst[4*dstStride], (src4+src5)*20 - (src3+src6)*5 + (src2+src7));\ | |
2270 OP(dst[5*dstStride], (src5+src6)*20 - (src4+src7)*5 + (src3+src8));\ | |
2271 OP(dst[6*dstStride], (src6+src7)*20 - (src5+src8)*5 + (src4+src9));\ | |
2272 OP(dst[7*dstStride], (src7+src8)*20 - (src6+src9)*5 + (src5+src10));\ | |
2273 dst++;\ | |
2274 src++;\ | |
2275 }\ | |
2276 }\ | |
2277 \ | |
2278 static void OPNAME ## h264_qpel8_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2279 const int h=8;\ | |
2280 const int w=8;\ | |
2281 uint8_t *cm = cropTbl + MAX_NEG_CROP;\ | |
2282 int i;\ | |
2283 src -= 2*srcStride;\ | |
2284 for(i=0; i<h+5; i++)\ | |
2285 {\ | |
2286 tmp[0]= (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3 ]);\ | |
2287 tmp[1]= (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4 ]);\ | |
2288 tmp[2]= (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5 ]);\ | |
2289 tmp[3]= (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6 ]);\ | |
2290 tmp[4]= (src[4]+src[5])*20 - (src[3 ]+src[6])*5 + (src[2 ]+src[7 ]);\ | |
2291 tmp[5]= (src[5]+src[6])*20 - (src[4 ]+src[7])*5 + (src[3 ]+src[8 ]);\ | |
2292 tmp[6]= (src[6]+src[7])*20 - (src[5 ]+src[8])*5 + (src[4 ]+src[9 ]);\ | |
2293 tmp[7]= (src[7]+src[8])*20 - (src[6 ]+src[9])*5 + (src[5 ]+src[10]);\ | |
2294 tmp+=tmpStride;\ | |
2295 src+=srcStride;\ | |
2296 }\ | |
2297 tmp -= tmpStride*(h+5-2);\ | |
2298 for(i=0; i<w; i++)\ | |
2299 {\ | |
2300 const int tmpB= tmp[-2*tmpStride];\ | |
2301 const int tmpA= tmp[-1*tmpStride];\ | |
2302 const int tmp0= tmp[0 *tmpStride];\ | |
2303 const int tmp1= tmp[1 *tmpStride];\ | |
2304 const int tmp2= tmp[2 *tmpStride];\ | |
2305 const int tmp3= tmp[3 *tmpStride];\ | |
2306 const int tmp4= tmp[4 *tmpStride];\ | |
2307 const int tmp5= tmp[5 *tmpStride];\ | |
2308 const int tmp6= tmp[6 *tmpStride];\ | |
2309 const int tmp7= tmp[7 *tmpStride];\ | |
2310 const int tmp8= tmp[8 *tmpStride];\ | |
2311 const int tmp9= tmp[9 *tmpStride];\ | |
2312 const int tmp10=tmp[10*tmpStride];\ | |
2313 OP2(dst[0*dstStride], (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));\ | |
2314 OP2(dst[1*dstStride], (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));\ | |
2315 OP2(dst[2*dstStride], (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));\ | |
2316 OP2(dst[3*dstStride], (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));\ | |
2317 OP2(dst[4*dstStride], (tmp4+tmp5)*20 - (tmp3+tmp6)*5 + (tmp2+tmp7));\ | |
2318 OP2(dst[5*dstStride], (tmp5+tmp6)*20 - (tmp4+tmp7)*5 + (tmp3+tmp8));\ | |
2319 OP2(dst[6*dstStride], (tmp6+tmp7)*20 - (tmp5+tmp8)*5 + (tmp4+tmp9));\ | |
2320 OP2(dst[7*dstStride], (tmp7+tmp8)*20 - (tmp6+tmp9)*5 + (tmp5+tmp10));\ | |
2321 dst++;\ | |
2322 tmp++;\ | |
2323 }\ | |
2324 }\ | |
2325 \ | |
2326 static void OPNAME ## h264_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2327 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2328 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2329 src += 8*srcStride;\ | |
2330 dst += 8*dstStride;\ | |
2331 OPNAME ## h264_qpel8_v_lowpass(dst , src , dstStride, srcStride);\ | |
2332 OPNAME ## h264_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2333 }\ | |
2334 \ | |
2335 static void OPNAME ## h264_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ | |
2336 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2337 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2338 src += 8*srcStride;\ | |
2339 dst += 8*dstStride;\ | |
2340 OPNAME ## h264_qpel8_h_lowpass(dst , src , dstStride, srcStride);\ | |
2341 OPNAME ## h264_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride);\ | |
2342 }\ | |
2343 \ | |
2344 static void OPNAME ## h264_qpel16_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\ | |
2345 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2346 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2347 src += 8*srcStride;\ | |
2348 dst += 8*dstStride;\ | |
2349 OPNAME ## h264_qpel8_hv_lowpass(dst , tmp , src , dstStride, tmpStride, srcStride);\ | |
2350 OPNAME ## h264_qpel8_hv_lowpass(dst+8, tmp+8, src+8, dstStride, tmpStride, srcStride);\ | |
2351 }\ | |
2352 | |
2353 #define H264_MC(OPNAME, SIZE) \ | |
2354 static void OPNAME ## h264_qpel ## SIZE ## _mc00_c (uint8_t *dst, uint8_t *src, int stride){\ | |
2355 OPNAME ## pixels ## SIZE ## _c(dst, src, stride, SIZE);\ | |
2356 }\ | |
2357 \ | |
2358 static void OPNAME ## h264_qpel ## SIZE ## _mc10_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2359 uint8_t half[SIZE*SIZE];\ | |
2360 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2361 OPNAME ## pixels ## SIZE ## _l2(dst, src, half, stride, stride, SIZE, SIZE);\ | |
2362 }\ | |
2363 \ | |
2364 static void OPNAME ## h264_qpel ## SIZE ## _mc20_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2365 OPNAME ## h264_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride);\ | |
2366 }\ | |
2367 \ | |
2368 static void OPNAME ## h264_qpel ## SIZE ## _mc30_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2369 uint8_t half[SIZE*SIZE];\ | |
2370 put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\ | |
2371 OPNAME ## pixels ## SIZE ## _l2(dst, src+1, half, stride, stride, SIZE, SIZE);\ | |
2372 }\ | |
2373 \ | |
2374 static void OPNAME ## h264_qpel ## SIZE ## _mc01_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2375 uint8_t full[SIZE*(SIZE+5)];\ | |
2376 uint8_t * const full_mid= full + SIZE*2;\ | |
2377 uint8_t half[SIZE*SIZE];\ | |
2378 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2379 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2380 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid, half, stride, SIZE, SIZE, SIZE);\ | |
2381 }\ | |
2382 \ | |
2383 static void OPNAME ## h264_qpel ## SIZE ## _mc02_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2384 uint8_t full[SIZE*(SIZE+5)];\ | |
2385 uint8_t * const full_mid= full + SIZE*2;\ | |
2386 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2387 OPNAME ## h264_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE);\ | |
2388 }\ | |
2389 \ | |
2390 static void OPNAME ## h264_qpel ## SIZE ## _mc03_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2391 uint8_t full[SIZE*(SIZE+5)];\ | |
2392 uint8_t * const full_mid= full + SIZE*2;\ | |
2393 uint8_t half[SIZE*SIZE];\ | |
2394 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2395 put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\ | |
2396 OPNAME ## pixels ## SIZE ## _l2(dst, full_mid+SIZE, half, stride, SIZE, SIZE, SIZE);\ | |
2397 }\ | |
2398 \ | |
2399 static void OPNAME ## h264_qpel ## SIZE ## _mc11_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2400 uint8_t full[SIZE*(SIZE+5)];\ | |
2401 uint8_t * const full_mid= full + SIZE*2;\ | |
2402 uint8_t halfH[SIZE*SIZE];\ | |
2403 uint8_t halfV[SIZE*SIZE];\ | |
2404 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2405 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2406 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2407 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2408 }\ | |
2409 \ | |
2410 static void OPNAME ## h264_qpel ## SIZE ## _mc31_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2411 uint8_t full[SIZE*(SIZE+5)];\ | |
2412 uint8_t * const full_mid= full + SIZE*2;\ | |
2413 uint8_t halfH[SIZE*SIZE];\ | |
2414 uint8_t halfV[SIZE*SIZE];\ | |
2415 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2416 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2417 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2418 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2419 }\ | |
2420 \ | |
2421 static void OPNAME ## h264_qpel ## SIZE ## _mc13_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2422 uint8_t full[SIZE*(SIZE+5)];\ | |
2423 uint8_t * const full_mid= full + SIZE*2;\ | |
2424 uint8_t halfH[SIZE*SIZE];\ | |
2425 uint8_t halfV[SIZE*SIZE];\ | |
2426 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2427 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2428 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2429 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2430 }\ | |
2431 \ | |
2432 static void OPNAME ## h264_qpel ## SIZE ## _mc33_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2433 uint8_t full[SIZE*(SIZE+5)];\ | |
2434 uint8_t * const full_mid= full + SIZE*2;\ | |
2435 uint8_t halfH[SIZE*SIZE];\ | |
2436 uint8_t halfV[SIZE*SIZE];\ | |
2437 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2438 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2439 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2440 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\ | |
2441 }\ | |
2442 \ | |
2443 static void OPNAME ## h264_qpel ## SIZE ## _mc22_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2444 int16_t tmp[SIZE*(SIZE+5)];\ | |
2445 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass(dst, tmp, src, stride, SIZE, stride);\ | |
2446 }\ | |
2447 \ | |
2448 static void OPNAME ## h264_qpel ## SIZE ## _mc21_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2449 int16_t tmp[SIZE*(SIZE+5)];\ | |
2450 uint8_t halfH[SIZE*SIZE];\ | |
2451 uint8_t halfHV[SIZE*SIZE];\ | |
2452 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\ | |
2453 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2454 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2455 }\ | |
2456 \ | |
2457 static void OPNAME ## h264_qpel ## SIZE ## _mc23_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2458 int16_t tmp[SIZE*(SIZE+5)];\ | |
2459 uint8_t halfH[SIZE*SIZE];\ | |
2460 uint8_t halfHV[SIZE*SIZE];\ | |
2461 put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\ | |
2462 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2463 OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2464 }\ | |
2465 \ | |
2466 static void OPNAME ## h264_qpel ## SIZE ## _mc12_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2467 uint8_t full[SIZE*(SIZE+5)];\ | |
2468 uint8_t * const full_mid= full + SIZE*2;\ | |
2469 int16_t tmp[SIZE*(SIZE+5)];\ | |
2470 uint8_t halfV[SIZE*SIZE];\ | |
2471 uint8_t halfHV[SIZE*SIZE];\ | |
2472 copy_block ## SIZE (full, src - stride*2, SIZE, stride, SIZE + 5);\ | |
2473 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2474 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2475 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2476 }\ | |
2477 \ | |
2478 static void OPNAME ## h264_qpel ## SIZE ## _mc32_c(uint8_t *dst, uint8_t *src, int stride){\ | |
2479 uint8_t full[SIZE*(SIZE+5)];\ | |
2480 uint8_t * const full_mid= full + SIZE*2;\ | |
2481 int16_t tmp[SIZE*(SIZE+5)];\ | |
2482 uint8_t halfV[SIZE*SIZE];\ | |
2483 uint8_t halfHV[SIZE*SIZE];\ | |
2484 copy_block ## SIZE (full, src - stride*2 + 1, SIZE, stride, SIZE + 5);\ | |
2485 put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\ | |
2486 put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\ | |
2487 OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\ | |
2488 }\ | |
2489 | |
2490 #define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1) | |
2491 //#define op_avg2(a, b) a = (((a)*w1+cm[((b) + 16)>>5]*w2 + o + 64)>>7) | |
2492 #define op_put(a, b) a = cm[((b) + 16)>>5] | |
2493 #define op2_avg(a, b) a = (((a)+cm[((b) + 512)>>10]+1)>>1) | |
2494 #define op2_put(a, b) a = cm[((b) + 512)>>10] | |
2495 | |
2496 H264_LOWPASS(put_ , op_put, op2_put) | |
2497 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
|
2498 H264_MC(put_, 2) |
1168 | 2499 H264_MC(put_, 4) |
2500 H264_MC(put_, 8) | |
2501 H264_MC(put_, 16) | |
2502 H264_MC(avg_, 4) | |
2503 H264_MC(avg_, 8) | |
2504 H264_MC(avg_, 16) | |
2505 | |
2506 #undef op_avg | |
2507 #undef op_put | |
2508 #undef op2_avg | |
2509 #undef op2_put | |
2510 #endif | |
2511 | |
2448 | 2512 #define op_scale1(x) block[x] = clip_uint8( (block[x]*weight + offset) >> log2_denom ) |
2513 #define op_scale2(x) dst[x] = clip_uint8( (src[x]*weights + dst[x]*weightd + offset) >> (log2_denom+1)) | |
2415 | 2514 #define H264_WEIGHT(W,H) \ |
2515 static void weight_h264_pixels ## W ## x ## H ## _c(uint8_t *block, int stride, int log2_denom, int weight, int offset){ \ | |
3029 | 2516 int y; \ |
2415 | 2517 offset <<= log2_denom; \ |
2518 if(log2_denom) offset += 1<<(log2_denom-1); \ | |
2519 for(y=0; y<H; y++, block += stride){ \ | |
2520 op_scale1(0); \ | |
2521 op_scale1(1); \ | |
2522 if(W==2) continue; \ | |
2523 op_scale1(2); \ | |
2524 op_scale1(3); \ | |
2525 if(W==4) continue; \ | |
2526 op_scale1(4); \ | |
2527 op_scale1(5); \ | |
2528 op_scale1(6); \ | |
2529 op_scale1(7); \ | |
2530 if(W==8) continue; \ | |
2531 op_scale1(8); \ | |
2532 op_scale1(9); \ | |
2533 op_scale1(10); \ | |
2534 op_scale1(11); \ | |
2535 op_scale1(12); \ | |
2536 op_scale1(13); \ | |
2537 op_scale1(14); \ | |
2538 op_scale1(15); \ | |
2539 } \ | |
2540 } \ | |
3029 | 2541 static void biweight_h264_pixels ## W ## x ## H ## _c(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset){ \ |
2542 int y; \ | |
2543 offset = ((offset + 1) | 1) << log2_denom; \ | |
2415 | 2544 for(y=0; y<H; y++, dst += stride, src += stride){ \ |
2545 op_scale2(0); \ | |
2546 op_scale2(1); \ | |
2547 if(W==2) continue; \ | |
2548 op_scale2(2); \ | |
2549 op_scale2(3); \ | |
2550 if(W==4) continue; \ | |
2551 op_scale2(4); \ | |
2552 op_scale2(5); \ | |
2553 op_scale2(6); \ | |
2554 op_scale2(7); \ | |
2555 if(W==8) continue; \ | |
2556 op_scale2(8); \ | |
2557 op_scale2(9); \ | |
2558 op_scale2(10); \ | |
2559 op_scale2(11); \ | |
2560 op_scale2(12); \ | |
2561 op_scale2(13); \ | |
2562 op_scale2(14); \ | |
2563 op_scale2(15); \ | |
2564 } \ | |
2565 } | |
2566 | |
2567 H264_WEIGHT(16,16) | |
2568 H264_WEIGHT(16,8) | |
2569 H264_WEIGHT(8,16) | |
2570 H264_WEIGHT(8,8) | |
2571 H264_WEIGHT(8,4) | |
2572 H264_WEIGHT(4,8) | |
2573 H264_WEIGHT(4,4) | |
2574 H264_WEIGHT(4,2) | |
2575 H264_WEIGHT(2,4) | |
2576 H264_WEIGHT(2,2) | |
2577 | |
2578 #undef op_scale1 | |
2579 #undef op_scale2 | |
2580 #undef H264_WEIGHT | |
2581 | |
936 | 2582 static void wmv2_mspel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){ |
2583 uint8_t *cm = cropTbl + MAX_NEG_CROP; | |
2584 int i; | |
2585 | |
2586 for(i=0; i<h; i++){ | |
2587 dst[0]= cm[(9*(src[0] + src[1]) - (src[-1] + src[2]) + 8)>>4]; | |
2588 dst[1]= cm[(9*(src[1] + src[2]) - (src[ 0] + src[3]) + 8)>>4]; | |
2589 dst[2]= cm[(9*(src[2] + src[3]) - (src[ 1] + src[4]) + 8)>>4]; | |
2590 dst[3]= cm[(9*(src[3] + src[4]) - (src[ 2] + src[5]) + 8)>>4]; | |
2591 dst[4]= cm[(9*(src[4] + src[5]) - (src[ 3] + src[6]) + 8)>>4]; | |
2592 dst[5]= cm[(9*(src[5] + src[6]) - (src[ 4] + src[7]) + 8)>>4]; | |
2593 dst[6]= cm[(9*(src[6] + src[7]) - (src[ 5] + src[8]) + 8)>>4]; | |
2594 dst[7]= cm[(9*(src[7] + src[8]) - (src[ 6] + src[9]) + 8)>>4]; | |
2595 dst+=dstStride; | |
2967 | 2596 src+=srcStride; |
936 | 2597 } |
2598 } | |
2599 | |
3432 | 2600 #ifdef CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2601 /* AVS specific */ |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2602 void ff_cavsdsp_init(DSPContext* c, AVCodecContext *avctx); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2603 |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2604 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
|
2605 put_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2606 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2607 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
|
2608 avg_pixels8_c(dst, src, stride, 8); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2609 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2610 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
|
2611 put_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2612 } |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2613 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
|
2614 avg_pixels16_c(dst, src, stride, 16); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2615 } |
3432 | 2616 #endif /* CONFIG_CAVS_DECODER */ |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
2617 |
3526 | 2618 #if defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
2619 /* VC-1 specific */ | |
2620 void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx); | |
2621 | |
2622 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, uint8_t *src, int stride, int rnd) { | |
2623 put_pixels8_c(dst, src, stride, 8); | |
2624 } | |
2625 #endif /* CONFIG_VC1_DECODER||CONFIG_WMV3_DECODER */ | |
2626 | |
936 | 2627 static void wmv2_mspel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int w){ |
2628 uint8_t *cm = cropTbl + MAX_NEG_CROP; | |
2629 int i; | |
2630 | |
2631 for(i=0; i<w; i++){ | |
2632 const int src_1= src[ -srcStride]; | |
2633 const int src0 = src[0 ]; | |
2634 const int src1 = src[ srcStride]; | |
2635 const int src2 = src[2*srcStride]; | |
2636 const int src3 = src[3*srcStride]; | |
2637 const int src4 = src[4*srcStride]; | |
2638 const int src5 = src[5*srcStride]; | |
2639 const int src6 = src[6*srcStride]; | |
2640 const int src7 = src[7*srcStride]; | |
2641 const int src8 = src[8*srcStride]; | |
2642 const int src9 = src[9*srcStride]; | |
2643 dst[0*dstStride]= cm[(9*(src0 + src1) - (src_1 + src2) + 8)>>4]; | |
2644 dst[1*dstStride]= cm[(9*(src1 + src2) - (src0 + src3) + 8)>>4]; | |
2645 dst[2*dstStride]= cm[(9*(src2 + src3) - (src1 + src4) + 8)>>4]; | |
2646 dst[3*dstStride]= cm[(9*(src3 + src4) - (src2 + src5) + 8)>>4]; | |
2647 dst[4*dstStride]= cm[(9*(src4 + src5) - (src3 + src6) + 8)>>4]; | |
2648 dst[5*dstStride]= cm[(9*(src5 + src6) - (src4 + src7) + 8)>>4]; | |
2649 dst[6*dstStride]= cm[(9*(src6 + src7) - (src5 + src8) + 8)>>4]; | |
2650 dst[7*dstStride]= cm[(9*(src7 + src8) - (src6 + src9) + 8)>>4]; | |
2651 src++; | |
2652 dst++; | |
2653 } | |
2654 } | |
2655 | |
2656 static void put_mspel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){ | |
2657 put_pixels8_c(dst, src, stride, 8); | |
2658 } | |
2659 | |
2660 static void put_mspel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){ | |
2661 uint8_t half[64]; | |
2662 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2663 put_pixels8_l2(dst, src, half, stride, stride, 8, 8); | |
2664 } | |
2665 | |
2666 static void put_mspel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){ | |
2667 wmv2_mspel8_h_lowpass(dst, src, stride, stride, 8); | |
2668 } | |
2669 | |
2670 static void put_mspel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){ | |
2671 uint8_t half[64]; | |
2672 wmv2_mspel8_h_lowpass(half, src, 8, stride, 8); | |
2673 put_pixels8_l2(dst, src+1, half, stride, stride, 8, 8); | |
2674 } | |
2675 | |
2676 static void put_mspel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){ | |
2677 wmv2_mspel8_v_lowpass(dst, src, stride, stride, 8); | |
2678 } | |
2679 | |
2680 static void put_mspel8_mc12_c(uint8_t *dst, uint8_t *src, int stride){ | |
2681 uint8_t halfH[88]; | |
2682 uint8_t halfV[64]; | |
2683 uint8_t halfHV[64]; | |
2684 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2685 wmv2_mspel8_v_lowpass(halfV, src, 8, stride, 8); | |
2686 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2687 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2688 } | |
2689 static void put_mspel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){ | |
2690 uint8_t halfH[88]; | |
2691 uint8_t halfV[64]; | |
2692 uint8_t halfHV[64]; | |
2693 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2694 wmv2_mspel8_v_lowpass(halfV, src+1, 8, stride, 8); | |
2695 wmv2_mspel8_v_lowpass(halfHV, halfH+8, 8, 8, 8); | |
2696 put_pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8); | |
2697 } | |
2698 static void put_mspel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){ | |
2699 uint8_t halfH[88]; | |
2700 wmv2_mspel8_h_lowpass(halfH, src-stride, 8, stride, 11); | |
2701 wmv2_mspel8_v_lowpass(dst, halfH+8, stride, 8, 8); | |
2702 } | |
2703 | |
1644 | 2704 static void h263_v_loop_filter_c(uint8_t *src, int stride, int qscale){ |
2705 int x; | |
2706 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2707 |
1644 | 2708 for(x=0; x<8; x++){ |
2709 int d1, d2, ad1; | |
2710 int p0= src[x-2*stride]; | |
2711 int p1= src[x-1*stride]; | |
2712 int p2= src[x+0*stride]; | |
2713 int p3= src[x+1*stride]; | |
2714 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2715 | |
2716 if (d<-2*strength) d1= 0; | |
2717 else if(d<- strength) d1=-2*strength - d; | |
2718 else if(d< strength) d1= d; | |
2719 else if(d< 2*strength) d1= 2*strength - d; | |
2720 else d1= 0; | |
2967 | 2721 |
1644 | 2722 p1 += d1; |
2723 p2 -= d1; | |
2724 if(p1&256) p1= ~(p1>>31); | |
2725 if(p2&256) p2= ~(p2>>31); | |
2967 | 2726 |
1644 | 2727 src[x-1*stride] = p1; |
2728 src[x+0*stride] = p2; | |
2729 | |
1645 | 2730 ad1= ABS(d1)>>1; |
2967 | 2731 |
1644 | 2732 d2= clip((p0-p3)/4, -ad1, ad1); |
2967 | 2733 |
1644 | 2734 src[x-2*stride] = p0 - d2; |
2735 src[x+ stride] = p3 + d2; | |
2736 } | |
2737 } | |
2738 | |
2739 static void h263_h_loop_filter_c(uint8_t *src, int stride, int qscale){ | |
2740 int y; | |
2741 const int strength= ff_h263_loop_filter_strength[qscale]; | |
2967 | 2742 |
1644 | 2743 for(y=0; y<8; y++){ |
2744 int d1, d2, ad1; | |
2745 int p0= src[y*stride-2]; | |
2746 int p1= src[y*stride-1]; | |
2747 int p2= src[y*stride+0]; | |
2748 int p3= src[y*stride+1]; | |
2749 int d = (p0 - p3 + 4*(p2 - p1)) / 8; | |
2750 | |
2751 if (d<-2*strength) d1= 0; | |
2752 else if(d<- strength) d1=-2*strength - d; | |
2753 else if(d< strength) d1= d; | |
2754 else if(d< 2*strength) d1= 2*strength - d; | |
2755 else d1= 0; | |
2967 | 2756 |
1644 | 2757 p1 += d1; |
2758 p2 -= d1; | |
2759 if(p1&256) p1= ~(p1>>31); | |
2760 if(p2&256) p2= ~(p2>>31); | |
2967 | 2761 |
1644 | 2762 src[y*stride-1] = p1; |
2763 src[y*stride+0] = p2; | |
2764 | |
2765 ad1= ABS(d1)>>1; | |
2967 | 2766 |
1644 | 2767 d2= clip((p0-p3)/4, -ad1, ad1); |
2967 | 2768 |
1644 | 2769 src[y*stride-2] = p0 - d2; |
2770 src[y*stride+1] = p3 + d2; | |
2771 } | |
2772 } | |
936 | 2773 |
2045 | 2774 static void h261_loop_filter_c(uint8_t *src, int stride){ |
2775 int x,y,xy,yz; | |
2776 int temp[64]; | |
2777 | |
2778 for(x=0; x<8; x++){ | |
2779 temp[x ] = 4*src[x ]; | |
2780 temp[x + 7*8] = 4*src[x + 7*stride]; | |
2781 } | |
2782 for(y=1; y<7; y++){ | |
2783 for(x=0; x<8; x++){ | |
2784 xy = y * stride + x; | |
2785 yz = y * 8 + x; | |
2786 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
|
2787 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2788 } |
2967 | 2789 |
2045 | 2790 for(y=0; y<8; y++){ |
2791 src[ y*stride] = (temp[ y*8] + 2)>>2; | |
2792 src[7+y*stride] = (temp[7+y*8] + 2)>>2; | |
2793 for(x=1; x<7; x++){ | |
2794 xy = y * stride + x; | |
2795 yz = y * 8 + x; | |
2796 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
|
2797 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2798 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2799 } |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
1984
diff
changeset
|
2800 |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2801 static inline void h264_loop_filter_luma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0) |
2633 | 2802 { |
2803 int i, d; | |
2804 for( i = 0; i < 4; i++ ) { | |
2805 if( tc0[i] < 0 ) { | |
2806 pix += 4*ystride; | |
2807 continue; | |
2808 } | |
2809 for( d = 0; d < 4; d++ ) { | |
2810 const int p0 = pix[-1*xstride]; | |
2811 const int p1 = pix[-2*xstride]; | |
2812 const int p2 = pix[-3*xstride]; | |
2813 const int q0 = pix[0]; | |
2814 const int q1 = pix[1*xstride]; | |
2815 const int q2 = pix[2*xstride]; | |
2967 | 2816 |
2633 | 2817 if( ABS( p0 - q0 ) < alpha && |
2818 ABS( p1 - p0 ) < beta && | |
2819 ABS( q1 - q0 ) < beta ) { | |
2967 | 2820 |
2633 | 2821 int tc = tc0[i]; |
2822 int i_delta; | |
2967 | 2823 |
2633 | 2824 if( ABS( p2 - p0 ) < beta ) { |
2651 | 2825 pix[-2*xstride] = p1 + clip( (( p2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - p1, -tc0[i], tc0[i] ); |
2633 | 2826 tc++; |
2827 } | |
2828 if( ABS( q2 - q0 ) < beta ) { | |
2651 | 2829 pix[ xstride] = q1 + clip( (( q2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - q1, -tc0[i], tc0[i] ); |
2633 | 2830 tc++; |
2831 } | |
2967 | 2832 |
2633 | 2833 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
2834 pix[-xstride] = clip_uint8( p0 + i_delta ); /* p0' */ | |
2835 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */ | |
2836 } | |
2837 pix += ystride; | |
2838 } | |
2839 } | |
2840 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2841 static void h264_v_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2842 { |
2843 h264_loop_filter_luma_c(pix, stride, 1, alpha, beta, tc0); | |
2844 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2845 static void h264_h_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2846 { |
2847 h264_loop_filter_luma_c(pix, 1, stride, alpha, beta, tc0); | |
2848 } | |
2849 | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2850 static inline void h264_loop_filter_chroma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0) |
2633 | 2851 { |
2852 int i, d; | |
2853 for( i = 0; i < 4; i++ ) { | |
2854 const int tc = tc0[i]; | |
2855 if( tc <= 0 ) { | |
2856 pix += 2*ystride; | |
2857 continue; | |
2858 } | |
2859 for( d = 0; d < 2; d++ ) { | |
2860 const int p0 = pix[-1*xstride]; | |
2861 const int p1 = pix[-2*xstride]; | |
2862 const int q0 = pix[0]; | |
2863 const int q1 = pix[1*xstride]; | |
2864 | |
2865 if( ABS( p0 - q0 ) < alpha && | |
2866 ABS( p1 - p0 ) < beta && | |
2867 ABS( q1 - q0 ) < beta ) { | |
2868 | |
2869 int delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); | |
2870 | |
2871 pix[-xstride] = clip_uint8( p0 + delta ); /* p0' */ | |
2872 pix[0] = clip_uint8( q0 - delta ); /* q0' */ | |
2873 } | |
2874 pix += ystride; | |
2875 } | |
2876 } | |
2877 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2878 static void h264_v_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2879 { |
2880 h264_loop_filter_chroma_c(pix, stride, 1, alpha, beta, tc0); | |
2881 } | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2882 static void h264_h_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) |
2633 | 2883 { |
2884 h264_loop_filter_chroma_c(pix, 1, stride, alpha, beta, tc0); | |
2885 } | |
2886 | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2887 static inline void h264_loop_filter_chroma_intra_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta) |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2888 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2889 int d; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2890 for( d = 0; d < 8; d++ ) { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2891 const int p0 = pix[-1*xstride]; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2892 const int p1 = pix[-2*xstride]; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2893 const int q0 = pix[0]; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2894 const int q1 = pix[1*xstride]; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2895 |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2896 if( ABS( p0 - q0 ) < alpha && |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2897 ABS( p1 - p0 ) < beta && |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2898 ABS( q1 - q0 ) < beta ) { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2899 |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2900 pix[-xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */ |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2901 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */ |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2902 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2903 pix += ystride; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2904 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2905 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2906 static void h264_v_loop_filter_chroma_intra_c(uint8_t *pix, int stride, int alpha, int beta) |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2907 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2908 h264_loop_filter_chroma_intra_c(pix, stride, 1, alpha, beta); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2909 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2910 static void h264_h_loop_filter_chroma_intra_c(uint8_t *pix, int stride, int alpha, int beta) |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2911 { |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2912 h264_loop_filter_chroma_intra_c(pix, 1, stride, alpha, beta); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2913 } |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
2914 |
1708 | 2915 static inline int pix_abs16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2916 { |
2917 int s, i; | |
2918 | |
2919 s = 0; | |
1708 | 2920 for(i=0;i<h;i++) { |
0 | 2921 s += abs(pix1[0] - pix2[0]); |
2922 s += abs(pix1[1] - pix2[1]); | |
2923 s += abs(pix1[2] - pix2[2]); | |
2924 s += abs(pix1[3] - pix2[3]); | |
2925 s += abs(pix1[4] - pix2[4]); | |
2926 s += abs(pix1[5] - pix2[5]); | |
2927 s += abs(pix1[6] - pix2[6]); | |
2928 s += abs(pix1[7] - pix2[7]); | |
2929 s += abs(pix1[8] - pix2[8]); | |
2930 s += abs(pix1[9] - pix2[9]); | |
2931 s += abs(pix1[10] - pix2[10]); | |
2932 s += abs(pix1[11] - pix2[11]); | |
2933 s += abs(pix1[12] - pix2[12]); | |
2934 s += abs(pix1[13] - pix2[13]); | |
2935 s += abs(pix1[14] - pix2[14]); | |
2936 s += abs(pix1[15] - pix2[15]); | |
2937 pix1 += line_size; | |
2938 pix2 += line_size; | |
2939 } | |
2940 return s; | |
2941 } | |
2942 | |
1708 | 2943 static int pix_abs16_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2944 { |
2945 int s, i; | |
2946 | |
2947 s = 0; | |
1708 | 2948 for(i=0;i<h;i++) { |
0 | 2949 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
2950 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
2951 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
2952 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
2953 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
2954 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
2955 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
2956 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
2957 s += abs(pix1[8] - avg2(pix2[8], pix2[9])); | |
2958 s += abs(pix1[9] - avg2(pix2[9], pix2[10])); | |
2959 s += abs(pix1[10] - avg2(pix2[10], pix2[11])); | |
2960 s += abs(pix1[11] - avg2(pix2[11], pix2[12])); | |
2961 s += abs(pix1[12] - avg2(pix2[12], pix2[13])); | |
2962 s += abs(pix1[13] - avg2(pix2[13], pix2[14])); | |
2963 s += abs(pix1[14] - avg2(pix2[14], pix2[15])); | |
2964 s += abs(pix1[15] - avg2(pix2[15], pix2[16])); | |
2965 pix1 += line_size; | |
2966 pix2 += line_size; | |
2967 } | |
2968 return s; | |
2969 } | |
2970 | |
1708 | 2971 static int pix_abs16_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 2972 { |
2973 int s, i; | |
1064 | 2974 uint8_t *pix3 = pix2 + line_size; |
0 | 2975 |
2976 s = 0; | |
1708 | 2977 for(i=0;i<h;i++) { |
0 | 2978 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
2979 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
2980 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
2981 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
2982 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
2983 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
2984 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
2985 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
2986 s += abs(pix1[8] - avg2(pix2[8], pix3[8])); | |
2987 s += abs(pix1[9] - avg2(pix2[9], pix3[9])); | |
2988 s += abs(pix1[10] - avg2(pix2[10], pix3[10])); | |
2989 s += abs(pix1[11] - avg2(pix2[11], pix3[11])); | |
2990 s += abs(pix1[12] - avg2(pix2[12], pix3[12])); | |
2991 s += abs(pix1[13] - avg2(pix2[13], pix3[13])); | |
2992 s += abs(pix1[14] - avg2(pix2[14], pix3[14])); | |
2993 s += abs(pix1[15] - avg2(pix2[15], pix3[15])); | |
2994 pix1 += line_size; | |
2995 pix2 += line_size; | |
2996 pix3 += line_size; | |
2997 } | |
2998 return s; | |
2999 } | |
3000 | |
1708 | 3001 static int pix_abs16_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
0 | 3002 { |
3003 int s, i; | |
1064 | 3004 uint8_t *pix3 = pix2 + line_size; |
0 | 3005 |
3006 s = 0; | |
1708 | 3007 for(i=0;i<h;i++) { |
0 | 3008 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
3009 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
3010 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
3011 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
3012 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
3013 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
3014 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
3015 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
3016 s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9])); | |
3017 s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10])); | |
3018 s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11])); | |
3019 s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12])); | |
3020 s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13])); | |
3021 s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14])); | |
3022 s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15])); | |
3023 s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16])); | |
3024 pix1 += line_size; | |
3025 pix2 += line_size; | |
3026 pix3 += line_size; | |
3027 } | |
3028 return s; | |
3029 } | |
3030 | |
1708 | 3031 static inline int pix_abs8_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3032 { |
3033 int s, i; | |
3034 | |
3035 s = 0; | |
1708 | 3036 for(i=0;i<h;i++) { |
294 | 3037 s += abs(pix1[0] - pix2[0]); |
3038 s += abs(pix1[1] - pix2[1]); | |
3039 s += abs(pix1[2] - pix2[2]); | |
3040 s += abs(pix1[3] - pix2[3]); | |
3041 s += abs(pix1[4] - pix2[4]); | |
3042 s += abs(pix1[5] - pix2[5]); | |
3043 s += abs(pix1[6] - pix2[6]); | |
3044 s += abs(pix1[7] - pix2[7]); | |
3045 pix1 += line_size; | |
3046 pix2 += line_size; | |
3047 } | |
3048 return s; | |
3049 } | |
3050 | |
1708 | 3051 static int pix_abs8_x2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3052 { |
3053 int s, i; | |
3054 | |
3055 s = 0; | |
1708 | 3056 for(i=0;i<h;i++) { |
294 | 3057 s += abs(pix1[0] - avg2(pix2[0], pix2[1])); |
3058 s += abs(pix1[1] - avg2(pix2[1], pix2[2])); | |
3059 s += abs(pix1[2] - avg2(pix2[2], pix2[3])); | |
3060 s += abs(pix1[3] - avg2(pix2[3], pix2[4])); | |
3061 s += abs(pix1[4] - avg2(pix2[4], pix2[5])); | |
3062 s += abs(pix1[5] - avg2(pix2[5], pix2[6])); | |
3063 s += abs(pix1[6] - avg2(pix2[6], pix2[7])); | |
3064 s += abs(pix1[7] - avg2(pix2[7], pix2[8])); | |
3065 pix1 += line_size; | |
3066 pix2 += line_size; | |
3067 } | |
3068 return s; | |
3069 } | |
3070 | |
1708 | 3071 static int pix_abs8_y2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3072 { |
3073 int s, i; | |
1064 | 3074 uint8_t *pix3 = pix2 + line_size; |
294 | 3075 |
3076 s = 0; | |
1708 | 3077 for(i=0;i<h;i++) { |
294 | 3078 s += abs(pix1[0] - avg2(pix2[0], pix3[0])); |
3079 s += abs(pix1[1] - avg2(pix2[1], pix3[1])); | |
3080 s += abs(pix1[2] - avg2(pix2[2], pix3[2])); | |
3081 s += abs(pix1[3] - avg2(pix2[3], pix3[3])); | |
3082 s += abs(pix1[4] - avg2(pix2[4], pix3[4])); | |
3083 s += abs(pix1[5] - avg2(pix2[5], pix3[5])); | |
3084 s += abs(pix1[6] - avg2(pix2[6], pix3[6])); | |
3085 s += abs(pix1[7] - avg2(pix2[7], pix3[7])); | |
3086 pix1 += line_size; | |
3087 pix2 += line_size; | |
3088 pix3 += line_size; | |
3089 } | |
3090 return s; | |
3091 } | |
3092 | |
1708 | 3093 static int pix_abs8_xy2_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) |
294 | 3094 { |
3095 int s, i; | |
1064 | 3096 uint8_t *pix3 = pix2 + line_size; |
294 | 3097 |
3098 s = 0; | |
1708 | 3099 for(i=0;i<h;i++) { |
294 | 3100 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1])); |
3101 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2])); | |
3102 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3])); | |
3103 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4])); | |
3104 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5])); | |
3105 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6])); | |
3106 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7])); | |
3107 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8])); | |
3108 pix1 += line_size; | |
3109 pix2 += line_size; | |
3110 pix3 += line_size; | |
3111 } | |
3112 return s; | |
3113 } | |
3114 | |
2834 | 3115 static int nsse16_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3116 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3117 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3118 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3119 int x,y; |
2066 | 3120 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3121 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3122 for(x=0; x<16; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3123 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3124 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3125 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3126 for(x=0; x<15; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3127 score2+= ABS( s1[x ] - s1[x +stride] |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3128 - s1[x+1] + s1[x+1+stride]) |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3129 -ABS( s2[x ] - s2[x +stride] |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3130 - s2[x+1] + s2[x+1+stride]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3131 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3132 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3133 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3134 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3135 } |
2066 | 3136 |
3137 if(c) return score1 + ABS(score2)*c->avctx->nsse_weight; | |
3138 else return score1 + ABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3139 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3140 |
2834 | 3141 static int nsse8_c(void *v, uint8_t *s1, uint8_t *s2, int stride, int h){ |
3142 MpegEncContext *c = v; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3143 int score1=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3144 int score2=0; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3145 int x,y; |
2967 | 3146 |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3147 for(y=0; y<h; y++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3148 for(x=0; x<8; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3149 score1+= (s1[x ] - s2[x ])*(s1[x ] - s2[x ]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3150 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3151 if(y+1<h){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3152 for(x=0; x<7; x++){ |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3153 score2+= ABS( s1[x ] - s1[x +stride] |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3154 - s1[x+1] + s1[x+1+stride]) |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3155 -ABS( s2[x ] - s2[x +stride] |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3156 - s2[x+1] + s2[x+1+stride]); |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3157 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3158 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3159 s1+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3160 s2+= stride; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3161 } |
2967 | 3162 |
2066 | 3163 if(c) return score1 + ABS(score2)*c->avctx->nsse_weight; |
3164 else return score1 + ABS(score2)*8; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3165 } |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3166 |
1784 | 3167 static int try_8x8basis_c(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale){ |
3168 int i; | |
3169 unsigned int sum=0; | |
3170 | |
3171 for(i=0; i<8*8; i++){ | |
3172 int b= rem[i] + ((basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT)); | |
3173 int w= weight[i]; | |
3174 b>>= RECON_SHIFT; | |
3175 assert(-512<b && b<512); | |
3176 | |
3177 sum += (w*b)*(w*b)>>4; | |
3178 } | |
3179 return sum>>2; | |
3180 } | |
3181 | |
3182 static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale){ | |
3183 int i; | |
3184 | |
3185 for(i=0; i<8*8; i++){ | |
3186 rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT); | |
2967 | 3187 } |
1784 | 3188 } |
3189 | |
1100 | 3190 /** |
3191 * permutes an 8x8 block. | |
1101 | 3192 * @param block the block which will be permuted according to the given permutation vector |
1100 | 3193 * @param permutation the permutation vector |
3194 * @param last the last non zero coefficient in scantable order, used to speed the permutation up | |
2967 | 3195 * @param scantable the used scantable, this is only used to speed the permutation up, the block is not |
1101 | 3196 * (inverse) permutated to scantable order! |
1100 | 3197 */ |
1064 | 3198 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
|
3199 { |
764 | 3200 int i; |
945 | 3201 DCTELEM temp[64]; |
2967 | 3202 |
764 | 3203 if(last<=0) return; |
882 | 3204 //if(permutation[1]==1) return; //FIXME its ok but not clean and might fail for some perms |
174
ac5075a55488
new IDCT code by Michael Niedermayer (michaelni@gmx.at) - #define SIMPLE_IDCT to enable
arpi_esp
parents:
88
diff
changeset
|
3205 |
764 | 3206 for(i=0; i<=last; i++){ |
3207 const int j= scantable[i]; | |
3208 temp[j]= block[j]; | |
3209 block[j]=0; | |
3210 } | |
2967 | 3211 |
764 | 3212 for(i=0; i<=last; i++){ |
3213 const int j= scantable[i]; | |
3214 const int perm_j= permutation[j]; | |
3215 block[perm_j]= temp[j]; | |
3216 } | |
174
ac5075a55488
new IDCT code by Michael Niedermayer (michaelni@gmx.at) - #define SIMPLE_IDCT to enable
arpi_esp
parents:
88
diff
changeset
|
3217 } |
34 | 3218 |
1729 | 3219 static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride, int h){ |
3220 return 0; | |
3221 } | |
3222 | |
3223 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type){ | |
3224 int i; | |
2967 | 3225 |
1729 | 3226 memset(cmp, 0, sizeof(void*)*5); |
2967 | 3227 |
1729 | 3228 for(i=0; i<5; i++){ |
3229 switch(type&0xFF){ | |
3230 case FF_CMP_SAD: | |
3231 cmp[i]= c->sad[i]; | |
3232 break; | |
3233 case FF_CMP_SATD: | |
3234 cmp[i]= c->hadamard8_diff[i]; | |
3235 break; | |
3236 case FF_CMP_SSE: | |
3237 cmp[i]= c->sse[i]; | |
3238 break; | |
3239 case FF_CMP_DCT: | |
3240 cmp[i]= c->dct_sad[i]; | |
3241 break; | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3242 case FF_CMP_DCT264: |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3243 cmp[i]= c->dct264_sad[i]; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3244 break; |
2382 | 3245 case FF_CMP_DCTMAX: |
3246 cmp[i]= c->dct_max[i]; | |
3247 break; | |
1729 | 3248 case FF_CMP_PSNR: |
3249 cmp[i]= c->quant_psnr[i]; | |
3250 break; | |
3251 case FF_CMP_BIT: | |
3252 cmp[i]= c->bit[i]; | |
3253 break; | |
3254 case FF_CMP_RD: | |
3255 cmp[i]= c->rd[i]; | |
3256 break; | |
3257 case FF_CMP_VSAD: | |
3258 cmp[i]= c->vsad[i]; | |
3259 break; | |
3260 case FF_CMP_VSSE: | |
3261 cmp[i]= c->vsse[i]; | |
3262 break; | |
3263 case FF_CMP_ZERO: | |
3264 cmp[i]= zero_cmp; | |
3265 break; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3266 case FF_CMP_NSSE: |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3267 cmp[i]= c->nsse[i]; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
3268 break; |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
3269 #ifdef CONFIG_SNOW_ENCODER |
2184 | 3270 case FF_CMP_W53: |
3271 cmp[i]= c->w53[i]; | |
3272 break; | |
3273 case FF_CMP_W97: | |
3274 cmp[i]= c->w97[i]; | |
3275 break; | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
3276 #endif |
1729 | 3277 default: |
3278 av_log(NULL, AV_LOG_ERROR,"internal error in cmp function selection\n"); | |
3279 } | |
3280 } | |
3281 } | |
3282 | |
1101 | 3283 /** |
3284 * memset(blocks, 0, sizeof(DCTELEM)*6*64) | |
3285 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
3286 static void clear_blocks_c(DCTELEM *blocks) |
296 | 3287 { |
3288 memset(blocks, 0, sizeof(DCTELEM)*6*64); | |
3289 } | |
3290 | |
866 | 3291 static void add_bytes_c(uint8_t *dst, uint8_t *src, int w){ |
3292 int i; | |
996
ad44196ea483
add/diff_bytes bugfix patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>)
michaelni
parents:
984
diff
changeset
|
3293 for(i=0; i+7<w; i+=8){ |
866 | 3294 dst[i+0] += src[i+0]; |
3295 dst[i+1] += src[i+1]; | |
3296 dst[i+2] += src[i+2]; | |
3297 dst[i+3] += src[i+3]; | |
3298 dst[i+4] += src[i+4]; | |
3299 dst[i+5] += src[i+5]; | |
3300 dst[i+6] += src[i+6]; | |
3301 dst[i+7] += src[i+7]; | |
3302 } | |
3303 for(; i<w; i++) | |
3304 dst[i+0] += src[i+0]; | |
3305 } | |
3306 | |
3307 static void diff_bytes_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ | |
3308 int i; | |
996
ad44196ea483
add/diff_bytes bugfix patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>)
michaelni
parents:
984
diff
changeset
|
3309 for(i=0; i+7<w; i+=8){ |
866 | 3310 dst[i+0] = src1[i+0]-src2[i+0]; |
3311 dst[i+1] = src1[i+1]-src2[i+1]; | |
3312 dst[i+2] = src1[i+2]-src2[i+2]; | |
3313 dst[i+3] = src1[i+3]-src2[i+3]; | |
3314 dst[i+4] = src1[i+4]-src2[i+4]; | |
3315 dst[i+5] = src1[i+5]-src2[i+5]; | |
3316 dst[i+6] = src1[i+6]-src2[i+6]; | |
3317 dst[i+7] = src1[i+7]-src2[i+7]; | |
3318 } | |
3319 for(; i<w; i++) | |
3320 dst[i+0] = src1[i+0]-src2[i+0]; | |
3321 } | |
3322 | |
1527 | 3323 static void sub_hfyu_median_prediction_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){ |
3324 int i; | |
3325 uint8_t l, lt; | |
3326 | |
3327 l= *left; | |
3328 lt= *left_top; | |
3329 | |
3330 for(i=0; i<w; i++){ | |
3331 const int pred= mid_pred(l, src1[i], (l + src1[i] - lt)&0xFF); | |
3332 lt= src1[i]; | |
3333 l= src2[i]; | |
3334 dst[i]= l - pred; | |
2967 | 3335 } |
1527 | 3336 |
3337 *left= l; | |
3338 *left_top= lt; | |
3339 } | |
3340 | |
936 | 3341 #define BUTTERFLY2(o1,o2,i1,i2) \ |
3342 o1= (i1)+(i2);\ | |
3343 o2= (i1)-(i2); | |
3344 | |
3345 #define BUTTERFLY1(x,y) \ | |
3346 {\ | |
3347 int a,b;\ | |
3348 a= x;\ | |
3349 b= y;\ | |
3350 x= a+b;\ | |
3351 y= a-b;\ | |
3352 } | |
3353 | |
3354 #define BUTTERFLYA(x,y) (ABS((x)+(y)) + ABS((x)-(y))) | |
3355 | |
1708 | 3356 static int hadamard8_diff8x8_c(/*MpegEncContext*/ void *s, uint8_t *dst, uint8_t *src, int stride, int h){ |
936 | 3357 int i; |
3358 int temp[64]; | |
3359 int sum=0; | |
2967 | 3360 |
1708 | 3361 assert(h==8); |
936 | 3362 |
3363 for(i=0; i<8; i++){ | |
3364 //FIXME try pointer walks | |
3365 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]); | |
3366 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]); | |
3367 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]); | |
3368 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 | 3369 |
936 | 3370 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3371 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3372 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3373 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3374 |
936 | 3375 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3376 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3377 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3378 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3379 } | |
3380 | |
3381 for(i=0; i<8; i++){ | |
3382 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3383 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3384 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3385 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3386 |
936 | 3387 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3388 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3389 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3390 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
3391 | |
2967 | 3392 sum += |
936 | 3393 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3394 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3395 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3396 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3397 } | |
3398 #if 0 | |
3399 static int maxi=0; | |
3400 if(sum>maxi){ | |
3401 maxi=sum; | |
3402 printf("MAX:%d\n", maxi); | |
3403 } | |
3404 #endif | |
3405 return sum; | |
3406 } | |
3407 | |
1729 | 3408 static int hadamard8_intra8x8_c(/*MpegEncContext*/ void *s, uint8_t *src, uint8_t *dummy, int stride, int h){ |
936 | 3409 int i; |
3410 int temp[64]; | |
3411 int sum=0; | |
2967 | 3412 |
1729 | 3413 assert(h==8); |
2967 | 3414 |
936 | 3415 for(i=0; i<8; i++){ |
3416 //FIXME try pointer walks | |
1729 | 3417 BUTTERFLY2(temp[8*i+0], temp[8*i+1], src[stride*i+0],src[stride*i+1]); |
3418 BUTTERFLY2(temp[8*i+2], temp[8*i+3], src[stride*i+2],src[stride*i+3]); | |
3419 BUTTERFLY2(temp[8*i+4], temp[8*i+5], src[stride*i+4],src[stride*i+5]); | |
3420 BUTTERFLY2(temp[8*i+6], temp[8*i+7], src[stride*i+6],src[stride*i+7]); | |
2967 | 3421 |
936 | 3422 BUTTERFLY1(temp[8*i+0], temp[8*i+2]); |
3423 BUTTERFLY1(temp[8*i+1], temp[8*i+3]); | |
3424 BUTTERFLY1(temp[8*i+4], temp[8*i+6]); | |
3425 BUTTERFLY1(temp[8*i+5], temp[8*i+7]); | |
2967 | 3426 |
936 | 3427 BUTTERFLY1(temp[8*i+0], temp[8*i+4]); |
3428 BUTTERFLY1(temp[8*i+1], temp[8*i+5]); | |
3429 BUTTERFLY1(temp[8*i+2], temp[8*i+6]); | |
3430 BUTTERFLY1(temp[8*i+3], temp[8*i+7]); | |
3431 } | |
3432 | |
3433 for(i=0; i<8; i++){ | |
3434 BUTTERFLY1(temp[8*0+i], temp[8*1+i]); | |
3435 BUTTERFLY1(temp[8*2+i], temp[8*3+i]); | |
3436 BUTTERFLY1(temp[8*4+i], temp[8*5+i]); | |
3437 BUTTERFLY1(temp[8*6+i], temp[8*7+i]); | |
2967 | 3438 |
936 | 3439 BUTTERFLY1(temp[8*0+i], temp[8*2+i]); |
3440 BUTTERFLY1(temp[8*1+i], temp[8*3+i]); | |
3441 BUTTERFLY1(temp[8*4+i], temp[8*6+i]); | |
3442 BUTTERFLY1(temp[8*5+i], temp[8*7+i]); | |
2967 | 3443 |
3444 sum += | |
936 | 3445 BUTTERFLYA(temp[8*0+i], temp[8*4+i]) |
3446 +BUTTERFLYA(temp[8*1+i], temp[8*5+i]) | |
3447 +BUTTERFLYA(temp[8*2+i], temp[8*6+i]) | |
3448 +BUTTERFLYA(temp[8*3+i], temp[8*7+i]); | |
3449 } | |
2967 | 3450 |
1729 | 3451 sum -= ABS(temp[8*0] + temp[8*4]); // -mean |
2967 | 3452 |
936 | 3453 return sum; |
3454 } | |
3455 | |
1708 | 3456 static int dct_sad8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3457 MpegEncContext * const s= (MpegEncContext *)c; |
3089 | 3458 DECLARE_ALIGNED_8(uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
1016 | 3459 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
936 | 3460 int sum=0, i; |
2967 | 3461 |
1708 | 3462 assert(h==8); |
936 | 3463 |
3464 s->dsp.diff_pixels(temp, src1, src2, stride); | |
1092 | 3465 s->dsp.fdct(temp); |
936 | 3466 |
3467 for(i=0; i<64; i++) | |
3468 sum+= ABS(temp[i]); | |
2967 | 3469 |
936 | 3470 return sum; |
3471 } | |
3472 | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3473 #ifdef CONFIG_GPL |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3474 #define DCT8_1D {\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3475 const int s07 = SRC(0) + SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3476 const int s16 = SRC(1) + SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3477 const int s25 = SRC(2) + SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3478 const int s34 = SRC(3) + SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3479 const int a0 = s07 + s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3480 const int a1 = s16 + s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3481 const int a2 = s07 - s34;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3482 const int a3 = s16 - s25;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3483 const int d07 = SRC(0) - SRC(7);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3484 const int d16 = SRC(1) - SRC(6);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3485 const int d25 = SRC(2) - SRC(5);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3486 const int d34 = SRC(3) - SRC(4);\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3487 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
|
3488 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
|
3489 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
|
3490 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
|
3491 DST(0, a0 + a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3492 DST(1, a4 + (a7>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3493 DST(2, a2 + (a3>>1)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3494 DST(3, a5 + (a6>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3495 DST(4, a0 - a1 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3496 DST(5, a6 - (a5>>2)) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3497 DST(6, (a2>>1) - a3 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3498 DST(7, (a4>>2) - a7 ) ;\ |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3499 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3500 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3501 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
|
3502 MpegEncContext * const s= (MpegEncContext *)c; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3503 int16_t dct[8][8]; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3504 int i; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3505 int sum=0; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3506 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3507 s->dsp.diff_pixels(dct, src1, src2, stride); |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3508 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3509 #define SRC(x) dct[i][x] |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3510 #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
|
3511 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3512 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3513 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3514 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3515 |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3516 #define SRC(x) dct[x][i] |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3517 #define DST(x,v) sum += ABS(v) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3518 for( i = 0; i < 8; i++ ) |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3519 DCT8_1D |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3520 #undef SRC |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3521 #undef DST |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3522 return sum; |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3523 } |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3524 #endif |
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3525 |
2382 | 3526 static int dct_max8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
3527 MpegEncContext * const s= (MpegEncContext *)c; | |
3089 | 3528 DECLARE_ALIGNED_8(uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
2382 | 3529 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3530 int sum=0, i; | |
2967 | 3531 |
2382 | 3532 assert(h==8); |
3533 | |
3534 s->dsp.diff_pixels(temp, src1, src2, stride); | |
3535 s->dsp.fdct(temp); | |
3536 | |
3537 for(i=0; i<64; i++) | |
3538 sum= FFMAX(sum, ABS(temp[i])); | |
2967 | 3539 |
2382 | 3540 return sum; |
3541 } | |
3542 | |
1708 | 3543 static int quant_psnr8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
936 | 3544 MpegEncContext * const s= (MpegEncContext *)c; |
3089 | 3545 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64*2/8]); |
1016 | 3546 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3547 DCTELEM * const bak = ((DCTELEM*)aligned_temp)+64; | |
936 | 3548 int sum=0, i; |
3549 | |
1708 | 3550 assert(h==8); |
936 | 3551 s->mb_intra=0; |
2967 | 3552 |
936 | 3553 s->dsp.diff_pixels(temp, src1, src2, stride); |
2967 | 3554 |
936 | 3555 memcpy(bak, temp, 64*sizeof(DCTELEM)); |
2967 | 3556 |
1013 | 3557 s->block_last_index[0/*FIXME*/]= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
1689 | 3558 s->dct_unquantize_inter(s, temp, 0, s->qscale); |
2967 | 3559 simple_idct(temp); //FIXME |
3560 | |
936 | 3561 for(i=0; i<64; i++) |
3562 sum+= (temp[i]-bak[i])*(temp[i]-bak[i]); | |
2967 | 3563 |
936 | 3564 return sum; |
3565 } | |
3566 | |
1708 | 3567 static int rd8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3568 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3569 const uint8_t *scantable= s->intra_scantable.permutated; |
3089 | 3570 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
3571 DECLARE_ALIGNED_8 (uint64_t, aligned_bak[stride]); | |
1016 | 3572 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
3573 uint8_t * const bak= (uint8_t*)aligned_bak; | |
1007 | 3574 int i, last, run, bits, level, distoration, start_i; |
3575 const int esc_length= s->ac_esc_length; | |
3576 uint8_t * length; | |
3577 uint8_t * last_length; | |
2967 | 3578 |
1708 | 3579 assert(h==8); |
3580 | |
1007 | 3581 for(i=0; i<8; i++){ |
3582 ((uint32_t*)(bak + i*stride))[0]= ((uint32_t*)(src2 + i*stride))[0]; | |
3583 ((uint32_t*)(bak + i*stride))[1]= ((uint32_t*)(src2 + i*stride))[1]; | |
3584 } | |
3585 | |
3586 s->dsp.diff_pixels(temp, src1, src2, stride); | |
3587 | |
1013 | 3588 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3589 | |
3590 bits=0; | |
2967 | 3591 |
1013 | 3592 if (s->mb_intra) { |
2967 | 3593 start_i = 1; |
1013 | 3594 length = s->intra_ac_vlc_length; |
3595 last_length= s->intra_ac_vlc_last_length; | |
3596 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma | |
3597 } else { | |
3598 start_i = 0; | |
3599 length = s->inter_ac_vlc_length; | |
3600 last_length= s->inter_ac_vlc_last_length; | |
3601 } | |
2967 | 3602 |
1013 | 3603 if(last>=start_i){ |
1007 | 3604 run=0; |
3605 for(i=start_i; i<last; i++){ | |
3606 int j= scantable[i]; | |
3607 level= temp[j]; | |
2967 | 3608 |
1007 | 3609 if(level){ |
3610 level+=64; | |
3611 if((level&(~127)) == 0){ | |
3612 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3613 }else | |
3614 bits+= esc_length; | |
3615 run=0; | |
3616 }else | |
3617 run++; | |
3618 } | |
3619 i= scantable[last]; | |
2967 | 3620 |
1011 | 3621 level= temp[i] + 64; |
3622 | |
3623 assert(level - 64); | |
2967 | 3624 |
1007 | 3625 if((level&(~127)) == 0){ |
3626 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3627 }else | |
3628 bits+= esc_length; | |
2967 | 3629 |
1013 | 3630 } |
3631 | |
3632 if(last>=0){ | |
1689 | 3633 if(s->mb_intra) |
3634 s->dct_unquantize_intra(s, temp, 0, s->qscale); | |
3635 else | |
3636 s->dct_unquantize_inter(s, temp, 0, s->qscale); | |
1007 | 3637 } |
2967 | 3638 |
1092 | 3639 s->dsp.idct_add(bak, stride, temp); |
2967 | 3640 |
1708 | 3641 distoration= s->dsp.sse[1](NULL, bak, src1, stride, 8); |
1007 | 3642 |
1013 | 3643 return distoration + ((bits*s->qscale*s->qscale*109 + 64)>>7); |
1007 | 3644 } |
3645 | |
1708 | 3646 static int bit8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){ |
1007 | 3647 MpegEncContext * const s= (MpegEncContext *)c; |
1064 | 3648 const uint8_t *scantable= s->intra_scantable.permutated; |
3089 | 3649 DECLARE_ALIGNED_8 (uint64_t, aligned_temp[sizeof(DCTELEM)*64/8]); |
1016 | 3650 DCTELEM * const temp= (DCTELEM*)aligned_temp; |
1007 | 3651 int i, last, run, bits, level, start_i; |
3652 const int esc_length= s->ac_esc_length; | |
3653 uint8_t * length; | |
3654 uint8_t * last_length; | |
1708 | 3655 |
3656 assert(h==8); | |
2967 | 3657 |
1013 | 3658 s->dsp.diff_pixels(temp, src1, src2, stride); |
1007 | 3659 |
1013 | 3660 s->block_last_index[0/*FIXME*/]= last= s->fast_dct_quantize(s, temp, 0/*FIXME*/, s->qscale, &i); |
3661 | |
3662 bits=0; | |
2967 | 3663 |
1007 | 3664 if (s->mb_intra) { |
2967 | 3665 start_i = 1; |
1007 | 3666 length = s->intra_ac_vlc_length; |
3667 last_length= s->intra_ac_vlc_last_length; | |
1013 | 3668 bits+= s->luma_dc_vlc_length[temp[0] + 256]; //FIXME chroma |
1007 | 3669 } else { |
3670 start_i = 0; | |
3671 length = s->inter_ac_vlc_length; | |
3672 last_length= s->inter_ac_vlc_last_length; | |
3673 } | |
2967 | 3674 |
1013 | 3675 if(last>=start_i){ |
1007 | 3676 run=0; |
3677 for(i=start_i; i<last; i++){ | |
3678 int j= scantable[i]; | |
3679 level= temp[j]; | |
2967 | 3680 |
1007 | 3681 if(level){ |
3682 level+=64; | |
3683 if((level&(~127)) == 0){ | |
3684 bits+= length[UNI_AC_ENC_INDEX(run, level)]; | |
3685 }else | |
3686 bits+= esc_length; | |
3687 run=0; | |
3688 }else | |
3689 run++; | |
3690 } | |
3691 i= scantable[last]; | |
2967 | 3692 |
1013 | 3693 level= temp[i] + 64; |
2967 | 3694 |
1013 | 3695 assert(level - 64); |
2967 | 3696 |
1007 | 3697 if((level&(~127)) == 0){ |
3698 bits+= last_length[UNI_AC_ENC_INDEX(run, level)]; | |
3699 }else | |
3700 bits+= esc_length; | |
3701 } | |
3702 | |
3703 return bits; | |
3704 } | |
3705 | |
1729 | 3706 static int vsad_intra16_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ |
3707 int score=0; | |
3708 int x,y; | |
2967 | 3709 |
1729 | 3710 for(y=1; y<h; y++){ |
3711 for(x=0; x<16; x+=4){ | |
2967 | 3712 score+= ABS(s[x ] - s[x +stride]) + ABS(s[x+1] - s[x+1+stride]) |
1729 | 3713 +ABS(s[x+2] - s[x+2+stride]) + ABS(s[x+3] - s[x+3+stride]); |
3714 } | |
3715 s+= stride; | |
3716 } | |
2967 | 3717 |
1729 | 3718 return score; |
3719 } | |
3720 | |
3721 static int vsad16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3722 int score=0; | |
3723 int x,y; | |
2967 | 3724 |
1729 | 3725 for(y=1; y<h; y++){ |
3726 for(x=0; x<16; x++){ | |
3727 score+= ABS(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); | |
3728 } | |
3729 s1+= stride; | |
3730 s2+= stride; | |
3731 } | |
2967 | 3732 |
1729 | 3733 return score; |
3734 } | |
3735 | |
3736 #define SQ(a) ((a)*(a)) | |
3737 static int vsse_intra16_c(/*MpegEncContext*/ void *c, uint8_t *s, uint8_t *dummy, int stride, int h){ | |
3738 int score=0; | |
3739 int x,y; | |
2967 | 3740 |
1729 | 3741 for(y=1; y<h; y++){ |
3742 for(x=0; x<16; x+=4){ | |
2967 | 3743 score+= SQ(s[x ] - s[x +stride]) + SQ(s[x+1] - s[x+1+stride]) |
1729 | 3744 +SQ(s[x+2] - s[x+2+stride]) + SQ(s[x+3] - s[x+3+stride]); |
3745 } | |
3746 s+= stride; | |
3747 } | |
2967 | 3748 |
1729 | 3749 return score; |
3750 } | |
3751 | |
3752 static int vsse16_c(/*MpegEncContext*/ void *c, uint8_t *s1, uint8_t *s2, int stride, int h){ | |
3753 int score=0; | |
3754 int x,y; | |
2967 | 3755 |
1729 | 3756 for(y=1; y<h; y++){ |
3757 for(x=0; x<16; x++){ | |
3758 score+= SQ(s1[x ] - s2[x ] - s1[x +stride] + s2[x +stride]); | |
3759 } | |
3760 s1+= stride; | |
3761 s2+= stride; | |
3762 } | |
2967 | 3763 |
1729 | 3764 return score; |
3765 } | |
3766 | |
1708 | 3767 WARPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c) |
1729 | 3768 WARPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c) |
1708 | 3769 WARPER8_16_SQ(dct_sad8x8_c, dct_sad16_c) |
3013 | 3770 #ifdef CONFIG_GPL |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
3771 WARPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c) |
3013 | 3772 #endif |
2382 | 3773 WARPER8_16_SQ(dct_max8x8_c, dct_max16_c) |
1708 | 3774 WARPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c) |
3775 WARPER8_16_SQ(rd8x8_c, rd16_c) | |
3776 WARPER8_16_SQ(bit8x8_c, bit16_c) | |
936 | 3777 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3778 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
|
3779 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3780 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3781 dst[i] *= src[i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3782 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3783 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3784 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
|
3785 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3786 src1 += len-1; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3787 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3788 dst[i] = src0[i] * src1[-i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3789 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3790 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3791 void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step){ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3792 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3793 for(i=0; i<len; i++) |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3794 dst[i*step] = src0[i] * src1[i] + src2[i] + src3; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3795 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3796 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3797 void ff_float_to_int16_c(int16_t *dst, const float *src, int len){ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3798 int i; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3799 for(i=0; i<len; i++) { |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3800 int_fast32_t tmp = ((int32_t*)src)[i]; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3801 if(tmp & 0xf0000){ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3802 tmp = (0x43c0ffff - tmp)>>31; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3803 // is this faster on some gcc/cpu combinations? |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3804 // if(tmp > 0x43c0ffff) tmp = 0xFFFF; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3805 // else tmp = 0; |
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 dst[i] = tmp - 0x8000; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3808 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3809 } |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
3810 |
1092 | 3811 /* XXX: those functions should be suppressed ASAP when all IDCTs are |
3812 converted */ | |
3813 static void ff_jref_idct_put(uint8_t *dest, int line_size, DCTELEM *block) | |
3814 { | |
3815 j_rev_dct (block); | |
3816 put_pixels_clamped_c(block, dest, line_size); | |
3817 } | |
3818 static void ff_jref_idct_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3819 { | |
3820 j_rev_dct (block); | |
3821 add_pixels_clamped_c(block, dest, line_size); | |
3822 } | |
3823 | |
2256 | 3824 static void ff_jref_idct4_put(uint8_t *dest, int line_size, DCTELEM *block) |
3825 { | |
3826 j_rev_dct4 (block); | |
3827 put_pixels_clamped4_c(block, dest, line_size); | |
3828 } | |
3829 static void ff_jref_idct4_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3830 { | |
3831 j_rev_dct4 (block); | |
3832 add_pixels_clamped4_c(block, dest, line_size); | |
3833 } | |
3834 | |
2257 | 3835 static void ff_jref_idct2_put(uint8_t *dest, int line_size, DCTELEM *block) |
3836 { | |
3837 j_rev_dct2 (block); | |
3838 put_pixels_clamped2_c(block, dest, line_size); | |
3839 } | |
3840 static void ff_jref_idct2_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3841 { | |
3842 j_rev_dct2 (block); | |
3843 add_pixels_clamped2_c(block, dest, line_size); | |
3844 } | |
3845 | |
2259 | 3846 static void ff_jref_idct1_put(uint8_t *dest, int line_size, DCTELEM *block) |
3847 { | |
3848 uint8_t *cm = cropTbl + MAX_NEG_CROP; | |
3849 | |
3850 dest[0] = cm[(block[0] + 4)>>3]; | |
3851 } | |
3852 static void ff_jref_idct1_add(uint8_t *dest, int line_size, DCTELEM *block) | |
3853 { | |
3854 uint8_t *cm = cropTbl + MAX_NEG_CROP; | |
3855 | |
3856 dest[0] = cm[dest[0] + ((block[0] + 4)>>3)]; | |
3857 } | |
3858 | |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
3859 static void just_return() { return; } |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
3860 |
1201 | 3861 /* init static data */ |
3862 void dsputil_static_init(void) | |
0 | 3863 { |
751 | 3864 int i; |
0 | 3865 |
1201 | 3866 for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i; |
3867 for(i=0;i<MAX_NEG_CROP;i++) { | |
3868 cropTbl[i] = 0; | |
3869 cropTbl[i + MAX_NEG_CROP + 256] = 255; | |
3870 } | |
2967 | 3871 |
1201 | 3872 for(i=0;i<512;i++) { |
3873 squareTbl[i] = (i - 256) * (i - 256); | |
3874 } | |
2967 | 3875 |
1201 | 3876 for(i=0; i<64; i++) inv_zigzag_direct16[ff_zigzag_direct[i]]= i+1; |
3877 } | |
0 | 3878 |
861 | 3879 |
1201 | 3880 void dsputil_init(DSPContext* c, AVCodecContext *avctx) |
3881 { | |
3882 int i; | |
0 | 3883 |
1092 | 3884 #ifdef CONFIG_ENCODERS |
1567 | 3885 if(avctx->dct_algo==FF_DCT_FASTINT) { |
1092 | 3886 c->fdct = fdct_ifast; |
2979 | 3887 c->fdct248 = fdct_ifast248; |
2967 | 3888 } |
1567 | 3889 else if(avctx->dct_algo==FF_DCT_FAAN) { |
1557 | 3890 c->fdct = ff_faandct; |
2979 | 3891 c->fdct248 = ff_faandct248; |
2967 | 3892 } |
1567 | 3893 else { |
1092 | 3894 c->fdct = ff_jpeg_fdct_islow; //slow/accurate/default |
2979 | 3895 c->fdct248 = ff_fdct248_islow; |
1567 | 3896 } |
1092 | 3897 #endif //CONFIG_ENCODERS |
3898 | |
2256 | 3899 if(avctx->lowres==1){ |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
3900 if(avctx->idct_algo==FF_IDCT_INT || avctx->idct_algo==FF_IDCT_AUTO){ |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
3901 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
|
3902 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
|
3903 }else{ |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
3904 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
|
3905 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
|
3906 } |
2256 | 3907 c->idct = j_rev_dct4; |
1092 | 3908 c->idct_permutation_type= FF_NO_IDCT_PERM; |
2257 | 3909 }else if(avctx->lowres==2){ |
3910 c->idct_put= ff_jref_idct2_put; | |
3911 c->idct_add= ff_jref_idct2_add; | |
3912 c->idct = j_rev_dct2; | |
3913 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2259 | 3914 }else if(avctx->lowres==3){ |
3915 c->idct_put= ff_jref_idct1_put; | |
3916 c->idct_add= ff_jref_idct1_add; | |
3917 c->idct = j_rev_dct1; | |
3918 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2256 | 3919 }else{ |
3920 if(avctx->idct_algo==FF_IDCT_INT){ | |
3921 c->idct_put= ff_jref_idct_put; | |
3922 c->idct_add= ff_jref_idct_add; | |
3923 c->idct = j_rev_dct; | |
3924 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM; | |
2693 | 3925 }else if(avctx->idct_algo==FF_IDCT_VP3){ |
3926 c->idct_put= ff_vp3_idct_put_c; | |
3927 c->idct_add= ff_vp3_idct_add_c; | |
3928 c->idct = ff_vp3_idct_c; | |
3929 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
2256 | 3930 }else{ //accurate/default |
3931 c->idct_put= simple_idct_put; | |
3932 c->idct_add= simple_idct_add; | |
3933 c->idct = simple_idct; | |
3934 c->idct_permutation_type= FF_NO_IDCT_PERM; | |
3935 } | |
1092 | 3936 } |
3937 | |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
3938 c->h264_idct_add= ff_h264_idct_add_c; |
2755 | 3939 c->h264_idct8_add= ff_h264_idct8_add_c; |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
3940 c->h264_idct_dc_add= ff_h264_idct_dc_add_c; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
3941 c->h264_idct8_dc_add= ff_h264_idct8_dc_add_c; |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
3942 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
3943 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
|
3944 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
|
3945 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
|
3946 c->put_signed_pixels_clamped = put_signed_pixels_clamped_c; |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
3947 c->add_pixels_clamped = add_pixels_clamped_c; |
2763 | 3948 c->add_pixels8 = add_pixels8_c; |
3949 c->add_pixels4 = add_pixels4_c; | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
3950 c->gmc1 = gmc1_c; |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
3951 c->gmc = ff_gmc_c; |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
764
diff
changeset
|
3952 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
|
3953 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
|
3954 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
|
3955 |
859 | 3956 /* TODO [0] 16 [1] 8 */ |
1708 | 3957 c->pix_abs[0][0] = pix_abs16_c; |
3958 c->pix_abs[0][1] = pix_abs16_x2_c; | |
3959 c->pix_abs[0][2] = pix_abs16_y2_c; | |
3960 c->pix_abs[0][3] = pix_abs16_xy2_c; | |
3961 c->pix_abs[1][0] = pix_abs8_c; | |
3962 c->pix_abs[1][1] = pix_abs8_x2_c; | |
3963 c->pix_abs[1][2] = pix_abs8_y2_c; | |
3964 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
|
3965 |
859 | 3966 #define dspfunc(PFX, IDX, NUM) \ |
3967 c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## NUM ## _c; \ | |
3968 c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## NUM ## _x2_c; \ | |
3969 c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## NUM ## _y2_c; \ | |
3970 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
|
3971 |
859 | 3972 dspfunc(put, 0, 16); |
3973 dspfunc(put_no_rnd, 0, 16); | |
3974 dspfunc(put, 1, 8); | |
3975 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
|
3976 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
|
3977 dspfunc(put, 3, 2); |
0 | 3978 |
859 | 3979 dspfunc(avg, 0, 16); |
3980 dspfunc(avg_no_rnd, 0, 16); | |
3981 dspfunc(avg, 1, 8); | |
3982 dspfunc(avg_no_rnd, 1, 8); | |
1319 | 3983 dspfunc(avg, 2, 4); |
3984 dspfunc(avg, 3, 2); | |
859 | 3985 #undef dspfunc |
857 | 3986 |
1864 | 3987 c->put_no_rnd_pixels_l2[0]= put_no_rnd_pixels16_l2_c; |
3988 c->put_no_rnd_pixels_l2[1]= put_no_rnd_pixels8_l2_c; | |
3989 | |
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
|
3990 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
|
3991 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
|
3992 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
|
3993 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
|
3994 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
|
3995 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
|
3996 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
|
3997 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
|
3998 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
|
3999 |
1319 | 4000 c->avg_tpel_pixels_tab[ 0] = avg_tpel_pixels_mc00_c; |
4001 c->avg_tpel_pixels_tab[ 1] = avg_tpel_pixels_mc10_c; | |
4002 c->avg_tpel_pixels_tab[ 2] = avg_tpel_pixels_mc20_c; | |
4003 c->avg_tpel_pixels_tab[ 4] = avg_tpel_pixels_mc01_c; | |
4004 c->avg_tpel_pixels_tab[ 5] = avg_tpel_pixels_mc11_c; | |
4005 c->avg_tpel_pixels_tab[ 6] = avg_tpel_pixels_mc21_c; | |
4006 c->avg_tpel_pixels_tab[ 8] = avg_tpel_pixels_mc02_c; | |
4007 c->avg_tpel_pixels_tab[ 9] = avg_tpel_pixels_mc12_c; | |
4008 c->avg_tpel_pixels_tab[10] = avg_tpel_pixels_mc22_c; | |
4009 | |
859 | 4010 #define dspfunc(PFX, IDX, NUM) \ |
4011 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_c; \ | |
4012 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_c; \ | |
4013 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_c; \ | |
4014 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_c; \ | |
4015 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_c; \ | |
4016 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_c; \ | |
4017 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_c; \ | |
4018 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_c; \ | |
4019 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_c; \ | |
4020 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_c; \ | |
4021 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_c; \ | |
4022 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_c; \ | |
4023 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_c; \ | |
4024 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_c; \ | |
4025 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_c; \ | |
4026 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_c | |
857 | 4027 |
859 | 4028 dspfunc(put_qpel, 0, 16); |
4029 dspfunc(put_no_rnd_qpel, 0, 16); | |
4030 | |
4031 dspfunc(avg_qpel, 0, 16); | |
4032 /* dspfunc(avg_no_rnd_qpel, 0, 16); */ | |
857 | 4033 |
859 | 4034 dspfunc(put_qpel, 1, 8); |
4035 dspfunc(put_no_rnd_qpel, 1, 8); | |
4036 | |
4037 dspfunc(avg_qpel, 1, 8); | |
4038 /* dspfunc(avg_no_rnd_qpel, 1, 8); */ | |
1168 | 4039 |
4040 dspfunc(put_h264_qpel, 0, 16); | |
4041 dspfunc(put_h264_qpel, 1, 8); | |
4042 dspfunc(put_h264_qpel, 2, 4); | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3013
diff
changeset
|
4043 dspfunc(put_h264_qpel, 3, 2); |
1168 | 4044 dspfunc(avg_h264_qpel, 0, 16); |
4045 dspfunc(avg_h264_qpel, 1, 8); | |
4046 dspfunc(avg_h264_qpel, 2, 4); | |
4047 | |
859 | 4048 #undef dspfunc |
1168 | 4049 c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_c; |
4050 c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_c; | |
4051 c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_c; | |
4052 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_c; | |
4053 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_c; | |
4054 c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_c; | |
3663 | 4055 c->put_no_rnd_h264_chroma_pixels_tab[0]= put_no_rnd_h264_chroma_mc8_c; |
857 | 4056 |
2415 | 4057 c->weight_h264_pixels_tab[0]= weight_h264_pixels16x16_c; |
4058 c->weight_h264_pixels_tab[1]= weight_h264_pixels16x8_c; | |
4059 c->weight_h264_pixels_tab[2]= weight_h264_pixels8x16_c; | |
4060 c->weight_h264_pixels_tab[3]= weight_h264_pixels8x8_c; | |
4061 c->weight_h264_pixels_tab[4]= weight_h264_pixels8x4_c; | |
4062 c->weight_h264_pixels_tab[5]= weight_h264_pixels4x8_c; | |
4063 c->weight_h264_pixels_tab[6]= weight_h264_pixels4x4_c; | |
4064 c->weight_h264_pixels_tab[7]= weight_h264_pixels4x2_c; | |
4065 c->weight_h264_pixels_tab[8]= weight_h264_pixels2x4_c; | |
4066 c->weight_h264_pixels_tab[9]= weight_h264_pixels2x2_c; | |
4067 c->biweight_h264_pixels_tab[0]= biweight_h264_pixels16x16_c; | |
4068 c->biweight_h264_pixels_tab[1]= biweight_h264_pixels16x8_c; | |
4069 c->biweight_h264_pixels_tab[2]= biweight_h264_pixels8x16_c; | |
4070 c->biweight_h264_pixels_tab[3]= biweight_h264_pixels8x8_c; | |
4071 c->biweight_h264_pixels_tab[4]= biweight_h264_pixels8x4_c; | |
4072 c->biweight_h264_pixels_tab[5]= biweight_h264_pixels4x8_c; | |
4073 c->biweight_h264_pixels_tab[6]= biweight_h264_pixels4x4_c; | |
4074 c->biweight_h264_pixels_tab[7]= biweight_h264_pixels4x2_c; | |
4075 c->biweight_h264_pixels_tab[8]= biweight_h264_pixels2x4_c; | |
4076 c->biweight_h264_pixels_tab[9]= biweight_h264_pixels2x2_c; | |
4077 | |
3432 | 4078 #ifdef CONFIG_CAVS_DECODER |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4079 ff_cavsdsp_init(c,avctx); |
3432 | 4080 #endif |
3526 | 4081 #if defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER) |
4082 ff_vc1dsp_init(c,avctx); | |
4083 #endif | |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3373
diff
changeset
|
4084 |
936 | 4085 c->put_mspel_pixels_tab[0]= put_mspel8_mc00_c; |
4086 c->put_mspel_pixels_tab[1]= put_mspel8_mc10_c; | |
4087 c->put_mspel_pixels_tab[2]= put_mspel8_mc20_c; | |
4088 c->put_mspel_pixels_tab[3]= put_mspel8_mc30_c; | |
4089 c->put_mspel_pixels_tab[4]= put_mspel8_mc02_c; | |
4090 c->put_mspel_pixels_tab[5]= put_mspel8_mc12_c; | |
4091 c->put_mspel_pixels_tab[6]= put_mspel8_mc22_c; | |
4092 c->put_mspel_pixels_tab[7]= put_mspel8_mc32_c; | |
2967 | 4093 |
1708 | 4094 #define SET_CMP_FUNC(name) \ |
4095 c->name[0]= name ## 16_c;\ | |
4096 c->name[1]= name ## 8x8_c; | |
2967 | 4097 |
1708 | 4098 SET_CMP_FUNC(hadamard8_diff) |
1729 | 4099 c->hadamard8_diff[4]= hadamard8_intra16_c; |
1708 | 4100 SET_CMP_FUNC(dct_sad) |
2382 | 4101 SET_CMP_FUNC(dct_max) |
3013 | 4102 #ifdef CONFIG_GPL |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
4103 SET_CMP_FUNC(dct264_sad) |
3013 | 4104 #endif |
1708 | 4105 c->sad[0]= pix_abs16_c; |
4106 c->sad[1]= pix_abs8_c; | |
4107 c->sse[0]= sse16_c; | |
4108 c->sse[1]= sse8_c; | |
2184 | 4109 c->sse[2]= sse4_c; |
1708 | 4110 SET_CMP_FUNC(quant_psnr) |
4111 SET_CMP_FUNC(rd) | |
4112 SET_CMP_FUNC(bit) | |
1729 | 4113 c->vsad[0]= vsad16_c; |
4114 c->vsad[4]= vsad_intra16_c; | |
4115 c->vsse[0]= vsse16_c; | |
4116 c->vsse[4]= vsse_intra16_c; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4117 c->nsse[0]= nsse16_c; |
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
4118 c->nsse[1]= nsse8_c; |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
4119 #ifdef CONFIG_SNOW_ENCODER |
2184 | 4120 c->w53[0]= w53_16_c; |
4121 c->w53[1]= w53_8_c; | |
4122 c->w97[0]= w97_16_c; | |
4123 c->w97[1]= w97_8_c; | |
3373
b8996cc5ccae
Disable w53 and w97 cmp methods when snow encoder is disabled
gpoirier
parents:
3323
diff
changeset
|
4124 #endif |
2184 | 4125 |
866 | 4126 c->add_bytes= add_bytes_c; |
4127 c->diff_bytes= diff_bytes_c; | |
1527 | 4128 c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_c; |
1273 | 4129 c->bswap_buf= bswap_buf; |
2633 | 4130 |
4131 c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_c; | |
4132 c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_c; | |
4133 c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_c; | |
4134 c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_c; | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
4135 c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_c; |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
4136 c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_c; |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3568
diff
changeset
|
4137 c->h264_loop_filter_strength= NULL; |
2967 | 4138 |
1644 | 4139 c->h263_h_loop_filter= h263_h_loop_filter_c; |
4140 c->h263_v_loop_filter= h263_v_loop_filter_c; | |
2967 | 4141 |
2045 | 4142 c->h261_loop_filter= h261_loop_filter_c; |
2967 | 4143 |
1784 | 4144 c->try_8x8basis= try_8x8basis_c; |
4145 c->add_8x8basis= add_8x8basis_c; | |
866 | 4146 |
3199
1651e69b9f7a
10l: Only set *compose97i *add_yblock to dsputils context if we are building with Snow enabled
gpoirier
parents:
3198
diff
changeset
|
4147 #ifdef CONFIG_SNOW_ENCODER |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4148 c->vertical_compose97i = ff_snow_vertical_compose97i; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4149 c->horizontal_compose97i = ff_snow_horizontal_compose97i; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4150 c->inner_add_yblock = ff_snow_inner_add_yblock; |
3199
1651e69b9f7a
10l: Only set *compose97i *add_yblock to dsputils context if we are building with Snow enabled
gpoirier
parents:
3198
diff
changeset
|
4151 #endif |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3105
diff
changeset
|
4152 |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4153 #ifdef CONFIG_VORBIS_DECODER |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4154 c->vorbis_inverse_coupling = vorbis_inverse_coupling; |
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4155 #endif |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4156 c->vector_fmul = vector_fmul_c; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4157 c->vector_fmul_reverse = vector_fmul_reverse_c; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4158 c->vector_fmul_add_add = ff_vector_fmul_add_add_c; |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3536
diff
changeset
|
4159 c->float_to_int16 = ff_float_to_int16_c; |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
4160 |
3245 | 4161 c->shrink[0]= ff_img_copy_plane; |
4162 c->shrink[1]= ff_shrink22; | |
4163 c->shrink[2]= ff_shrink44; | |
4164 c->shrink[3]= ff_shrink88; | |
4165 | |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4166 c->prefetch= just_return; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3199
diff
changeset
|
4167 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4168 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
|
4169 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
|
4170 |
2 | 4171 #ifdef HAVE_MMX |
1092 | 4172 dsputil_init_mmx(c, avctx); |
0 | 4173 #endif |
62 | 4174 #ifdef ARCH_ARMV4L |
1092 | 4175 dsputil_init_armv4l(c, avctx); |
62 | 4176 #endif |
88 | 4177 #ifdef HAVE_MLIB |
1092 | 4178 dsputil_init_mlib(c, avctx); |
88 | 4179 #endif |
1959
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1866
diff
changeset
|
4180 #ifdef ARCH_SPARC |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1866
diff
changeset
|
4181 dsputil_init_vis(c,avctx); |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1866
diff
changeset
|
4182 #endif |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
209
diff
changeset
|
4183 #ifdef ARCH_ALPHA |
1092 | 4184 dsputil_init_alpha(c, avctx); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
209
diff
changeset
|
4185 #endif |
623
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
4186 #ifdef ARCH_POWERPC |
1092 | 4187 dsputil_init_ppc(c, avctx); |
626
23a093d6e450
patch by Heliodoro Tammaro <helio at interactives dot org>
michaelni
parents:
625
diff
changeset
|
4188 #endif |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
676
diff
changeset
|
4189 #ifdef HAVE_MMI |
1092 | 4190 dsputil_init_mmi(c, avctx); |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
676
diff
changeset
|
4191 #endif |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1201
diff
changeset
|
4192 #ifdef ARCH_SH4 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1201
diff
changeset
|
4193 dsputil_init_sh4(c,avctx); |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1201
diff
changeset
|
4194 #endif |
3728 | 4195 #ifdef ARCH_BFIN |
4196 dsputil_init_bfin(c,avctx); | |
4197 #endif | |
1092 | 4198 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4199 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
|
4200 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
|
4201 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
|
4202 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
|
4203 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
|
4204 } |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3728
diff
changeset
|
4205 |
1092 | 4206 switch(c->idct_permutation_type){ |
4207 case FF_NO_IDCT_PERM: | |
4208 for(i=0; i<64; i++) | |
4209 c->idct_permutation[i]= i; | |
4210 break; | |
4211 case FF_LIBMPEG2_IDCT_PERM: | |
4212 for(i=0; i<64; i++) | |
4213 c->idct_permutation[i]= (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2); | |
4214 break; | |
4215 case FF_SIMPLE_IDCT_PERM: | |
4216 for(i=0; i<64; i++) | |
4217 c->idct_permutation[i]= simple_mmx_permutation[i]; | |
4218 break; | |
4219 case FF_TRANSPOSE_IDCT_PERM: | |
4220 for(i=0; i<64; i++) | |
4221 c->idct_permutation[i]= ((i&7)<<3) | (i>>3); | |
4222 break; | |
2696
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4223 case FF_PARTTRANS_IDCT_PERM: |
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
4224 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
|
4225 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
|
4226 break; |
1092 | 4227 default: |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1571
diff
changeset
|
4228 av_log(avctx, AV_LOG_ERROR, "Internal error, IDCT permutation not set\n"); |
1092 | 4229 } |
0 | 4230 } |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
220
diff
changeset
|
4231 |