Mercurial > libavcodec.hg
annotate i386/motion_est_mmx.c @ 1560:07dac732074a libavcodec
10l
author | alex |
---|---|
date | Wed, 22 Oct 2003 22:12:25 +0000 |
parents | c4539ef4d8cb |
children | 1f8d1e1173d8 |
rev | line source |
---|---|
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
1 /* |
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
2 * MMX optimized motion estimation |
429 | 3 * Copyright (c) 2001 Fabrice Bellard. |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
4 * |
429 | 5 * This library is free software; you can redistribute it and/or |
6 * modify it under the terms of the GNU Lesser General Public | |
7 * License as published by the Free Software Foundation; either | |
8 * version 2 of the License, or (at your option) any later version. | |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
9 * |
429 | 10 * This library is distributed in the hope that it will be useful, |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
429 | 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
13 * Lesser General Public License for more details. | |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
14 * |
429 | 15 * You should have received a copy of the GNU Lesser General Public |
16 * License along with this library; if not, write to the Free Software | |
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
18 * |
294 | 19 * mostly by Michael Niedermayer <michaelni@gmx.at> |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
20 */ |
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
21 #include "../dsputil.h" |
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
22 |
1064 | 23 static const __attribute__ ((aligned(8))) uint64_t round_tab[3]={ |
294 | 24 0x0000000000000000, |
25 0x0001000100010001, | |
26 0x0002000200020002, | |
27 }; | |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
28 |
1455 | 29 static __attribute__ ((aligned(8), unused)) uint64_t bone= 0x0101010101010101LL; |
330 | 30 |
1064 | 31 static inline void sad8_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h) |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
32 { |
294 | 33 int len= -(stride<<h); |
34 asm volatile( | |
35 ".balign 16 \n\t" | |
36 "1: \n\t" | |
37 "movq (%1, %%eax), %%mm0 \n\t" | |
38 "movq (%2, %%eax), %%mm2 \n\t" | |
39 "movq (%2, %%eax), %%mm4 \n\t" | |
40 "addl %3, %%eax \n\t" | |
41 "psubusb %%mm0, %%mm2 \n\t" | |
42 "psubusb %%mm4, %%mm0 \n\t" | |
43 "movq (%1, %%eax), %%mm1 \n\t" | |
44 "movq (%2, %%eax), %%mm3 \n\t" | |
45 "movq (%2, %%eax), %%mm5 \n\t" | |
46 "psubusb %%mm1, %%mm3 \n\t" | |
47 "psubusb %%mm5, %%mm1 \n\t" | |
48 "por %%mm2, %%mm0 \n\t" | |
49 "por %%mm1, %%mm3 \n\t" | |
50 "movq %%mm0, %%mm1 \n\t" | |
51 "movq %%mm3, %%mm2 \n\t" | |
52 "punpcklbw %%mm7, %%mm0 \n\t" | |
53 "punpckhbw %%mm7, %%mm1 \n\t" | |
54 "punpcklbw %%mm7, %%mm3 \n\t" | |
55 "punpckhbw %%mm7, %%mm2 \n\t" | |
56 "paddw %%mm1, %%mm0 \n\t" | |
57 "paddw %%mm3, %%mm2 \n\t" | |
58 "paddw %%mm2, %%mm0 \n\t" | |
59 "paddw %%mm0, %%mm6 \n\t" | |
60 "addl %3, %%eax \n\t" | |
61 " js 1b \n\t" | |
62 : "+a" (len) | |
63 : "r" (blk1 - len), "r" (blk2 - len), "r" (stride) | |
64 ); | |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
65 } |
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
66 |
1064 | 67 static inline void sad8_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h) |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
68 { |
294 | 69 int len= -(stride<<h); |
70 asm volatile( | |
71 ".balign 16 \n\t" | |
72 "1: \n\t" | |
73 "movq (%1, %%eax), %%mm0 \n\t" | |
74 "movq (%2, %%eax), %%mm2 \n\t" | |
75 "psadbw %%mm2, %%mm0 \n\t" | |
76 "addl %3, %%eax \n\t" | |
77 "movq (%1, %%eax), %%mm1 \n\t" | |
78 "movq (%2, %%eax), %%mm3 \n\t" | |
79 "psadbw %%mm1, %%mm3 \n\t" | |
80 "paddw %%mm3, %%mm0 \n\t" | |
81 "paddw %%mm0, %%mm6 \n\t" | |
82 "addl %3, %%eax \n\t" | |
83 " js 1b \n\t" | |
84 : "+a" (len) | |
85 : "r" (blk1 - len), "r" (blk2 - len), "r" (stride) | |
86 ); | |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
87 } |
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
88 |
1064 | 89 static inline void sad8_2_mmx2(uint8_t *blk1a, uint8_t *blk1b, uint8_t *blk2, int stride, int h) |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
90 { |
294 | 91 int len= -(stride<<h); |
92 asm volatile( | |
93 ".balign 16 \n\t" | |
94 "1: \n\t" | |
95 "movq (%1, %%eax), %%mm0 \n\t" | |
96 "movq (%2, %%eax), %%mm2 \n\t" | |
97 "pavgb %%mm2, %%mm0 \n\t" | |
98 "movq (%3, %%eax), %%mm2 \n\t" | |
99 "psadbw %%mm2, %%mm0 \n\t" | |
100 "addl %4, %%eax \n\t" | |
101 "movq (%1, %%eax), %%mm1 \n\t" | |
102 "movq (%2, %%eax), %%mm3 \n\t" | |
103 "pavgb %%mm1, %%mm3 \n\t" | |
104 "movq (%3, %%eax), %%mm1 \n\t" | |
105 "psadbw %%mm1, %%mm3 \n\t" | |
106 "paddw %%mm3, %%mm0 \n\t" | |
107 "paddw %%mm0, %%mm6 \n\t" | |
108 "addl %4, %%eax \n\t" | |
109 " js 1b \n\t" | |
110 : "+a" (len) | |
111 : "r" (blk1a - len), "r" (blk1b -len), "r" (blk2 - len), "r" (stride) | |
112 ); | |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
113 } |
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
114 |
1064 | 115 static inline void sad8_4_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h) |
294 | 116 { //FIXME reuse src |
117 int len= -(stride<<h); | |
118 asm volatile( | |
119 ".balign 16 \n\t" | |
330 | 120 "movq "MANGLE(bone)", %%mm5 \n\t" |
1057 | 121 "1: \n\t" |
294 | 122 "movq (%1, %%eax), %%mm0 \n\t" |
123 "movq (%2, %%eax), %%mm2 \n\t" | |
124 "movq 1(%1, %%eax), %%mm1 \n\t" | |
125 "movq 1(%2, %%eax), %%mm3 \n\t" | |
126 "pavgb %%mm2, %%mm0 \n\t" | |
127 "pavgb %%mm1, %%mm3 \n\t" | |
330 | 128 "psubusb %%mm5, %%mm3 \n\t" |
294 | 129 "pavgb %%mm3, %%mm0 \n\t" |
130 "movq (%3, %%eax), %%mm2 \n\t" | |
131 "psadbw %%mm2, %%mm0 \n\t" | |
132 "addl %4, %%eax \n\t" | |
133 "movq (%1, %%eax), %%mm1 \n\t" | |
134 "movq (%2, %%eax), %%mm3 \n\t" | |
135 "movq 1(%1, %%eax), %%mm2 \n\t" | |
136 "movq 1(%2, %%eax), %%mm4 \n\t" | |
137 "pavgb %%mm3, %%mm1 \n\t" | |
138 "pavgb %%mm4, %%mm2 \n\t" | |
330 | 139 "psubusb %%mm5, %%mm2 \n\t" |
294 | 140 "pavgb %%mm1, %%mm2 \n\t" |
141 "movq (%3, %%eax), %%mm1 \n\t" | |
142 "psadbw %%mm1, %%mm2 \n\t" | |
143 "paddw %%mm2, %%mm0 \n\t" | |
144 "paddw %%mm0, %%mm6 \n\t" | |
145 "addl %4, %%eax \n\t" | |
146 " js 1b \n\t" | |
147 : "+a" (len) | |
148 : "r" (blk1 - len), "r" (blk1 - len + stride), "r" (blk2 - len), "r" (stride) | |
149 ); | |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
150 } |
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
151 |
1064 | 152 static inline void sad8_2_mmx(uint8_t *blk1a, uint8_t *blk1b, uint8_t *blk2, int stride, int h) |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
153 { |
294 | 154 int len= -(stride<<h); |
155 asm volatile( | |
156 ".balign 16 \n\t" | |
157 "1: \n\t" | |
158 "movq (%1, %%eax), %%mm0 \n\t" | |
159 "movq (%2, %%eax), %%mm1 \n\t" | |
160 "movq (%1, %%eax), %%mm2 \n\t" | |
161 "movq (%2, %%eax), %%mm3 \n\t" | |
162 "punpcklbw %%mm7, %%mm0 \n\t" | |
163 "punpcklbw %%mm7, %%mm1 \n\t" | |
164 "punpckhbw %%mm7, %%mm2 \n\t" | |
165 "punpckhbw %%mm7, %%mm3 \n\t" | |
166 "paddw %%mm0, %%mm1 \n\t" | |
167 "paddw %%mm2, %%mm3 \n\t" | |
1057 | 168 "movq (%3, %%eax), %%mm4 \n\t" |
294 | 169 "movq (%3, %%eax), %%mm2 \n\t" |
170 "paddw %%mm5, %%mm1 \n\t" | |
171 "paddw %%mm5, %%mm3 \n\t" | |
172 "psrlw $1, %%mm1 \n\t" | |
173 "psrlw $1, %%mm3 \n\t" | |
174 "packuswb %%mm3, %%mm1 \n\t" | |
175 "psubusb %%mm1, %%mm4 \n\t" | |
176 "psubusb %%mm2, %%mm1 \n\t" | |
177 "por %%mm4, %%mm1 \n\t" | |
178 "movq %%mm1, %%mm0 \n\t" | |
179 "punpcklbw %%mm7, %%mm0 \n\t" | |
180 "punpckhbw %%mm7, %%mm1 \n\t" | |
181 "paddw %%mm1, %%mm0 \n\t" | |
182 "paddw %%mm0, %%mm6 \n\t" | |
183 "addl %4, %%eax \n\t" | |
184 " js 1b \n\t" | |
185 : "+a" (len) | |
186 : "r" (blk1a - len), "r" (blk1b -len), "r" (blk2 - len), "r" (stride) | |
187 ); | |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
188 } |
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
189 |
1064 | 190 static inline void sad8_4_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h) |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
191 { |
294 | 192 int len= -(stride<<h); |
193 asm volatile( | |
194 ".balign 16 \n\t" | |
195 "1: \n\t" | |
196 "movq (%1, %%eax), %%mm0 \n\t" | |
197 "movq (%2, %%eax), %%mm1 \n\t" | |
198 "movq %%mm0, %%mm4 \n\t" | |
199 "movq %%mm1, %%mm2 \n\t" | |
200 "punpcklbw %%mm7, %%mm0 \n\t" | |
201 "punpcklbw %%mm7, %%mm1 \n\t" | |
202 "punpckhbw %%mm7, %%mm4 \n\t" | |
203 "punpckhbw %%mm7, %%mm2 \n\t" | |
204 "paddw %%mm1, %%mm0 \n\t" | |
205 "paddw %%mm2, %%mm4 \n\t" | |
206 "movq 1(%1, %%eax), %%mm2 \n\t" | |
207 "movq 1(%2, %%eax), %%mm3 \n\t" | |
208 "movq %%mm2, %%mm1 \n\t" | |
209 "punpcklbw %%mm7, %%mm2 \n\t" | |
210 "punpckhbw %%mm7, %%mm1 \n\t" | |
211 "paddw %%mm0, %%mm2 \n\t" | |
212 "paddw %%mm4, %%mm1 \n\t" | |
213 "movq %%mm3, %%mm4 \n\t" | |
214 "punpcklbw %%mm7, %%mm3 \n\t" | |
215 "punpckhbw %%mm7, %%mm4 \n\t" | |
216 "paddw %%mm3, %%mm2 \n\t" | |
217 "paddw %%mm4, %%mm1 \n\t" | |
1057 | 218 "movq (%3, %%eax), %%mm3 \n\t" |
219 "movq (%3, %%eax), %%mm4 \n\t" | |
294 | 220 "paddw %%mm5, %%mm2 \n\t" |
221 "paddw %%mm5, %%mm1 \n\t" | |
222 "psrlw $2, %%mm2 \n\t" | |
223 "psrlw $2, %%mm1 \n\t" | |
224 "packuswb %%mm1, %%mm2 \n\t" | |
225 "psubusb %%mm2, %%mm3 \n\t" | |
226 "psubusb %%mm4, %%mm2 \n\t" | |
227 "por %%mm3, %%mm2 \n\t" | |
228 "movq %%mm2, %%mm0 \n\t" | |
229 "punpcklbw %%mm7, %%mm0 \n\t" | |
230 "punpckhbw %%mm7, %%mm2 \n\t" | |
231 "paddw %%mm2, %%mm0 \n\t" | |
232 "paddw %%mm0, %%mm6 \n\t" | |
233 "addl %4, %%eax \n\t" | |
234 " js 1b \n\t" | |
235 : "+a" (len) | |
236 : "r" (blk1 - len), "r" (blk1 -len + stride), "r" (blk2 - len), "r" (stride) | |
237 ); | |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
238 } |
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
239 |
1057 | 240 static inline int sum_mmx(void) |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
241 { |
294 | 242 int ret; |
243 asm volatile( | |
244 "movq %%mm6, %%mm0 \n\t" | |
245 "psrlq $32, %%mm6 \n\t" | |
246 "paddw %%mm0, %%mm6 \n\t" | |
247 "movq %%mm6, %%mm0 \n\t" | |
248 "psrlq $16, %%mm6 \n\t" | |
249 "paddw %%mm0, %%mm6 \n\t" | |
250 "movd %%mm6, %0 \n\t" | |
251 : "=r" (ret) | |
252 ); | |
253 return ret&0xFFFF; | |
254 } | |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
255 |
1057 | 256 static inline int sum_mmx2(void) |
294 | 257 { |
258 int ret; | |
259 asm volatile( | |
260 "movd %%mm6, %0 \n\t" | |
261 : "=r" (ret) | |
262 ); | |
263 return ret; | |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
264 } |
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
265 |
900 | 266 |
294 | 267 #define PIX_SAD(suf)\ |
1064 | 268 static int pix_abs8x8_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\ |
294 | 269 {\ |
270 asm volatile("pxor %%mm7, %%mm7 \n\t"\ | |
271 "pxor %%mm6, %%mm6 \n\t":);\ | |
272 \ | |
273 sad8_ ## suf(blk1, blk2, stride, 3);\ | |
274 \ | |
275 return sum_ ## suf();\ | |
276 }\ | |
1064 | 277 static int sad8x8_ ## suf(void *s, uint8_t *blk2, uint8_t *blk1, int stride)\ |
936 | 278 {\ |
279 asm volatile("pxor %%mm7, %%mm7 \n\t"\ | |
280 "pxor %%mm6, %%mm6 \n\t":);\ | |
281 \ | |
282 sad8_ ## suf(blk1, blk2, stride, 3);\ | |
283 \ | |
284 return sum_ ## suf();\ | |
285 }\ | |
294 | 286 \ |
1064 | 287 static int pix_abs8x8_x2_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\ |
294 | 288 {\ |
289 asm volatile("pxor %%mm7, %%mm7 \n\t"\ | |
290 "pxor %%mm6, %%mm6 \n\t"\ | |
291 "movq %0, %%mm5 \n\t"\ | |
292 :: "m"(round_tab[1]) \ | |
293 );\ | |
294 \ | |
900 | 295 sad8_2_ ## suf(blk1, blk1+1, blk2, stride, 3);\ |
294 | 296 \ |
297 return sum_ ## suf();\ | |
298 }\ | |
299 \ | |
1064 | 300 static int pix_abs8x8_y2_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\ |
294 | 301 {\ |
302 asm volatile("pxor %%mm7, %%mm7 \n\t"\ | |
303 "pxor %%mm6, %%mm6 \n\t"\ | |
304 "movq %0, %%mm5 \n\t"\ | |
305 :: "m"(round_tab[1]) \ | |
306 );\ | |
307 \ | |
308 sad8_2_ ## suf(blk1, blk1+stride, blk2, stride, 3);\ | |
309 \ | |
310 return sum_ ## suf();\ | |
311 }\ | |
312 \ | |
1064 | 313 static int pix_abs8x8_xy2_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\ |
294 | 314 {\ |
315 asm volatile("pxor %%mm7, %%mm7 \n\t"\ | |
316 "pxor %%mm6, %%mm6 \n\t"\ | |
317 "movq %0, %%mm5 \n\t"\ | |
318 :: "m"(round_tab[2]) \ | |
319 );\ | |
320 \ | |
321 sad8_4_ ## suf(blk1, blk2, stride, 3);\ | |
322 \ | |
323 return sum_ ## suf();\ | |
324 }\ | |
325 \ | |
1064 | 326 static int pix_abs16x16_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\ |
294 | 327 {\ |
328 asm volatile("pxor %%mm7, %%mm7 \n\t"\ | |
329 "pxor %%mm6, %%mm6 \n\t":);\ | |
330 \ | |
331 sad8_ ## suf(blk1 , blk2 , stride, 4);\ | |
332 sad8_ ## suf(blk1+8, blk2+8, stride, 4);\ | |
333 \ | |
334 return sum_ ## suf();\ | |
335 }\ | |
1064 | 336 static int sad16x16_ ## suf(void *s, uint8_t *blk2, uint8_t *blk1, int stride)\ |
936 | 337 {\ |
338 asm volatile("pxor %%mm7, %%mm7 \n\t"\ | |
339 "pxor %%mm6, %%mm6 \n\t":);\ | |
340 \ | |
341 sad8_ ## suf(blk1 , blk2 , stride, 4);\ | |
342 sad8_ ## suf(blk1+8, blk2+8, stride, 4);\ | |
343 \ | |
344 return sum_ ## suf();\ | |
345 }\ | |
1064 | 346 static int pix_abs16x16_x2_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\ |
294 | 347 {\ |
348 asm volatile("pxor %%mm7, %%mm7 \n\t"\ | |
349 "pxor %%mm6, %%mm6 \n\t"\ | |
350 "movq %0, %%mm5 \n\t"\ | |
351 :: "m"(round_tab[1]) \ | |
352 );\ | |
353 \ | |
354 sad8_2_ ## suf(blk1 , blk1+1, blk2 , stride, 4);\ | |
355 sad8_2_ ## suf(blk1+8, blk1+9, blk2+8, stride, 4);\ | |
356 \ | |
357 return sum_ ## suf();\ | |
358 }\ | |
1064 | 359 static int pix_abs16x16_y2_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\ |
294 | 360 {\ |
361 asm volatile("pxor %%mm7, %%mm7 \n\t"\ | |
362 "pxor %%mm6, %%mm6 \n\t"\ | |
363 "movq %0, %%mm5 \n\t"\ | |
364 :: "m"(round_tab[1]) \ | |
365 );\ | |
366 \ | |
367 sad8_2_ ## suf(blk1 , blk1+stride, blk2 , stride, 4);\ | |
368 sad8_2_ ## suf(blk1+8, blk1+stride+8,blk2+8, stride, 4);\ | |
369 \ | |
370 return sum_ ## suf();\ | |
371 }\ | |
1064 | 372 static int pix_abs16x16_xy2_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\ |
294 | 373 {\ |
374 asm volatile("pxor %%mm7, %%mm7 \n\t"\ | |
375 "pxor %%mm6, %%mm6 \n\t"\ | |
376 "movq %0, %%mm5 \n\t"\ | |
377 :: "m"(round_tab[2]) \ | |
378 );\ | |
379 \ | |
380 sad8_4_ ## suf(blk1 , blk2 , stride, 4);\ | |
381 sad8_4_ ## suf(blk1+8, blk2+8, stride, 4);\ | |
382 \ | |
383 return sum_ ## suf();\ | |
384 }\ | |
72
3049d6d452a3
suppressed nasm dependancy - rewrote forward DCT and motion estimation code
glantau
parents:
diff
changeset
|
385 |
294 | 386 PIX_SAD(mmx) |
387 PIX_SAD(mmx2) | |
1057 | 388 |
1092 | 389 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx) |
1057 | 390 { |
391 if (mm_flags & MM_MMX) { | |
392 c->pix_abs16x16 = pix_abs16x16_mmx; | |
393 c->pix_abs16x16_x2 = pix_abs16x16_x2_mmx; | |
394 c->pix_abs16x16_y2 = pix_abs16x16_y2_mmx; | |
395 c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx; | |
396 c->pix_abs8x8 = pix_abs8x8_mmx; | |
397 c->pix_abs8x8_x2 = pix_abs8x8_x2_mmx; | |
398 c->pix_abs8x8_y2 = pix_abs8x8_y2_mmx; | |
399 c->pix_abs8x8_xy2 = pix_abs8x8_xy2_mmx; | |
400 | |
401 c->sad[0]= sad16x16_mmx; | |
402 c->sad[1]= sad8x8_mmx; | |
403 } | |
404 if (mm_flags & MM_MMXEXT) { | |
405 c->pix_abs16x16 = pix_abs16x16_mmx2; | |
406 c->pix_abs8x8 = pix_abs8x8_mmx2; | |
407 | |
408 c->sad[0]= sad16x16_mmx2; | |
409 c->sad[1]= sad8x8_mmx2; | |
1092 | 410 |
411 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){ | |
412 c->pix_abs16x16_x2 = pix_abs16x16_x2_mmx2; | |
413 c->pix_abs16x16_y2 = pix_abs16x16_y2_mmx2; | |
414 c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx2; | |
415 c->pix_abs8x8_x2 = pix_abs8x8_x2_mmx2; | |
416 c->pix_abs8x8_y2 = pix_abs8x8_y2_mmx2; | |
417 c->pix_abs8x8_xy2 = pix_abs8x8_xy2_mmx2; | |
418 } | |
1057 | 419 } |
420 } |