2 * MMX optimized motion estimation
3 * Copyright (c) 2001 Fabrice Bellard.
4 * Copyright (c) 2002-2004 Michael Niedermayer
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.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
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
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 * mostly by Michael Niedermayer <michaelni@gmx.at>
22 #include "../dsputil.h"
24 static const __attribute__ ((aligned(8))) uint64_t round_tab[3]={
25 0x0000000000000000ULL,
26 0x0001000100010001ULL,
27 0x0002000200020002ULL,
30 static __attribute__ ((aligned(8), unused)) uint64_t bone= 0x0101010101010101LL;
32 static inline void sad8_1_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
38 "movq (%1, %%eax), %%mm0 \n\t"
39 "movq (%2, %%eax), %%mm2 \n\t"
40 "movq (%2, %%eax), %%mm4 \n\t"
42 "psubusb %%mm0, %%mm2 \n\t"
43 "psubusb %%mm4, %%mm0 \n\t"
44 "movq (%1, %%eax), %%mm1 \n\t"
45 "movq (%2, %%eax), %%mm3 \n\t"
46 "movq (%2, %%eax), %%mm5 \n\t"
47 "psubusb %%mm1, %%mm3 \n\t"
48 "psubusb %%mm5, %%mm1 \n\t"
49 "por %%mm2, %%mm0 \n\t"
50 "por %%mm1, %%mm3 \n\t"
51 "movq %%mm0, %%mm1 \n\t"
52 "movq %%mm3, %%mm2 \n\t"
53 "punpcklbw %%mm7, %%mm0 \n\t"
54 "punpckhbw %%mm7, %%mm1 \n\t"
55 "punpcklbw %%mm7, %%mm3 \n\t"
56 "punpckhbw %%mm7, %%mm2 \n\t"
57 "paddw %%mm1, %%mm0 \n\t"
58 "paddw %%mm3, %%mm2 \n\t"
59 "paddw %%mm2, %%mm0 \n\t"
60 "paddw %%mm0, %%mm6 \n\t"
64 : "r" (blk1 - len), "r" (blk2 - len), "r" (stride)
68 static inline void sad8_1_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h)
74 "movq (%1, %%eax), %%mm0 \n\t"
75 "movq (%2, %%eax), %%mm2 \n\t"
76 "psadbw %%mm2, %%mm0 \n\t"
78 "movq (%1, %%eax), %%mm1 \n\t"
79 "movq (%2, %%eax), %%mm3 \n\t"
80 "psadbw %%mm1, %%mm3 \n\t"
81 "paddw %%mm3, %%mm0 \n\t"
82 "paddw %%mm0, %%mm6 \n\t"
86 : "r" (blk1 - len), "r" (blk2 - len), "r" (stride)
90 static inline void sad8_2_mmx2(uint8_t *blk1a, uint8_t *blk1b, uint8_t *blk2, int stride, int h)
96 "movq (%1, %%eax), %%mm0 \n\t"
97 "movq (%2, %%eax), %%mm2 \n\t"
98 "pavgb %%mm2, %%mm0 \n\t"
99 "movq (%3, %%eax), %%mm2 \n\t"
100 "psadbw %%mm2, %%mm0 \n\t"
101 "addl %4, %%eax \n\t"
102 "movq (%1, %%eax), %%mm1 \n\t"
103 "movq (%2, %%eax), %%mm3 \n\t"
104 "pavgb %%mm1, %%mm3 \n\t"
105 "movq (%3, %%eax), %%mm1 \n\t"
106 "psadbw %%mm1, %%mm3 \n\t"
107 "paddw %%mm3, %%mm0 \n\t"
108 "paddw %%mm0, %%mm6 \n\t"
109 "addl %4, %%eax \n\t"
112 : "r" (blk1a - len), "r" (blk1b -len), "r" (blk2 - len), "r" (stride)
116 static inline void sad8_4_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h)
118 int len= -(stride*h);
121 "movq "MANGLE(bone)", %%mm5 \n\t"
123 "movq (%1, %%eax), %%mm0 \n\t"
124 "movq (%2, %%eax), %%mm2 \n\t"
125 "movq 1(%1, %%eax), %%mm1 \n\t"
126 "movq 1(%2, %%eax), %%mm3 \n\t"
127 "pavgb %%mm2, %%mm0 \n\t"
128 "pavgb %%mm1, %%mm3 \n\t"
129 "psubusb %%mm5, %%mm3 \n\t"
130 "pavgb %%mm3, %%mm0 \n\t"
131 "movq (%3, %%eax), %%mm2 \n\t"
132 "psadbw %%mm2, %%mm0 \n\t"
133 "addl %4, %%eax \n\t"
134 "movq (%1, %%eax), %%mm1 \n\t"
135 "movq (%2, %%eax), %%mm3 \n\t"
136 "movq 1(%1, %%eax), %%mm2 \n\t"
137 "movq 1(%2, %%eax), %%mm4 \n\t"
138 "pavgb %%mm3, %%mm1 \n\t"
139 "pavgb %%mm4, %%mm2 \n\t"
140 "psubusb %%mm5, %%mm2 \n\t"
141 "pavgb %%mm1, %%mm2 \n\t"
142 "movq (%3, %%eax), %%mm1 \n\t"
143 "psadbw %%mm1, %%mm2 \n\t"
144 "paddw %%mm2, %%mm0 \n\t"
145 "paddw %%mm0, %%mm6 \n\t"
146 "addl %4, %%eax \n\t"
149 : "r" (blk1 - len), "r" (blk1 - len + stride), "r" (blk2 - len), "r" (stride)
153 static inline void sad8_2_mmx(uint8_t *blk1a, uint8_t *blk1b, uint8_t *blk2, int stride, int h)
155 int len= -(stride*h);
159 "movq (%1, %%eax), %%mm0 \n\t"
160 "movq (%2, %%eax), %%mm1 \n\t"
161 "movq (%1, %%eax), %%mm2 \n\t"
162 "movq (%2, %%eax), %%mm3 \n\t"
163 "punpcklbw %%mm7, %%mm0 \n\t"
164 "punpcklbw %%mm7, %%mm1 \n\t"
165 "punpckhbw %%mm7, %%mm2 \n\t"
166 "punpckhbw %%mm7, %%mm3 \n\t"
167 "paddw %%mm0, %%mm1 \n\t"
168 "paddw %%mm2, %%mm3 \n\t"
169 "movq (%3, %%eax), %%mm4 \n\t"
170 "movq (%3, %%eax), %%mm2 \n\t"
171 "paddw %%mm5, %%mm1 \n\t"
172 "paddw %%mm5, %%mm3 \n\t"
173 "psrlw $1, %%mm1 \n\t"
174 "psrlw $1, %%mm3 \n\t"
175 "packuswb %%mm3, %%mm1 \n\t"
176 "psubusb %%mm1, %%mm4 \n\t"
177 "psubusb %%mm2, %%mm1 \n\t"
178 "por %%mm4, %%mm1 \n\t"
179 "movq %%mm1, %%mm0 \n\t"
180 "punpcklbw %%mm7, %%mm0 \n\t"
181 "punpckhbw %%mm7, %%mm1 \n\t"
182 "paddw %%mm1, %%mm0 \n\t"
183 "paddw %%mm0, %%mm6 \n\t"
184 "addl %4, %%eax \n\t"
187 : "r" (blk1a - len), "r" (blk1b -len), "r" (blk2 - len), "r" (stride)
191 static inline void sad8_4_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
193 int len= -(stride*h);
197 "movq (%1, %%eax), %%mm0 \n\t"
198 "movq (%2, %%eax), %%mm1 \n\t"
199 "movq %%mm0, %%mm4 \n\t"
200 "movq %%mm1, %%mm2 \n\t"
201 "punpcklbw %%mm7, %%mm0 \n\t"
202 "punpcklbw %%mm7, %%mm1 \n\t"
203 "punpckhbw %%mm7, %%mm4 \n\t"
204 "punpckhbw %%mm7, %%mm2 \n\t"
205 "paddw %%mm1, %%mm0 \n\t"
206 "paddw %%mm2, %%mm4 \n\t"
207 "movq 1(%1, %%eax), %%mm2 \n\t"
208 "movq 1(%2, %%eax), %%mm3 \n\t"
209 "movq %%mm2, %%mm1 \n\t"
210 "punpcklbw %%mm7, %%mm2 \n\t"
211 "punpckhbw %%mm7, %%mm1 \n\t"
212 "paddw %%mm0, %%mm2 \n\t"
213 "paddw %%mm4, %%mm1 \n\t"
214 "movq %%mm3, %%mm4 \n\t"
215 "punpcklbw %%mm7, %%mm3 \n\t"
216 "punpckhbw %%mm7, %%mm4 \n\t"
217 "paddw %%mm3, %%mm2 \n\t"
218 "paddw %%mm4, %%mm1 \n\t"
219 "movq (%3, %%eax), %%mm3 \n\t"
220 "movq (%3, %%eax), %%mm4 \n\t"
221 "paddw %%mm5, %%mm2 \n\t"
222 "paddw %%mm5, %%mm1 \n\t"
223 "psrlw $2, %%mm2 \n\t"
224 "psrlw $2, %%mm1 \n\t"
225 "packuswb %%mm1, %%mm2 \n\t"
226 "psubusb %%mm2, %%mm3 \n\t"
227 "psubusb %%mm4, %%mm2 \n\t"
228 "por %%mm3, %%mm2 \n\t"
229 "movq %%mm2, %%mm0 \n\t"
230 "punpcklbw %%mm7, %%mm0 \n\t"
231 "punpckhbw %%mm7, %%mm2 \n\t"
232 "paddw %%mm2, %%mm0 \n\t"
233 "paddw %%mm0, %%mm6 \n\t"
234 "addl %4, %%eax \n\t"
237 : "r" (blk1 - len), "r" (blk1 -len + stride), "r" (blk2 - len), "r" (stride)
241 static inline int sum_mmx(void)
245 "movq %%mm6, %%mm0 \n\t"
246 "psrlq $32, %%mm6 \n\t"
247 "paddw %%mm0, %%mm6 \n\t"
248 "movq %%mm6, %%mm0 \n\t"
249 "psrlq $16, %%mm6 \n\t"
250 "paddw %%mm0, %%mm6 \n\t"
251 "movd %%mm6, %0 \n\t"
257 static inline int sum_mmx2(void)
261 "movd %%mm6, %0 \n\t"
268 #define PIX_SAD(suf)\
269 static int sad8_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
272 asm volatile("pxor %%mm7, %%mm7 \n\t"\
273 "pxor %%mm6, %%mm6 \n\t":);\
275 sad8_1_ ## suf(blk1, blk2, stride, 8);\
277 return sum_ ## suf();\
279 static int sad8_x2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
282 asm volatile("pxor %%mm7, %%mm7 \n\t"\
283 "pxor %%mm6, %%mm6 \n\t"\
284 "movq %0, %%mm5 \n\t"\
285 :: "m"(round_tab[1]) \
288 sad8_2_ ## suf(blk1, blk1+1, blk2, stride, 8);\
290 return sum_ ## suf();\
293 static int sad8_y2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
296 asm volatile("pxor %%mm7, %%mm7 \n\t"\
297 "pxor %%mm6, %%mm6 \n\t"\
298 "movq %0, %%mm5 \n\t"\
299 :: "m"(round_tab[1]) \
302 sad8_2_ ## suf(blk1, blk1+stride, blk2, stride, 8);\
304 return sum_ ## suf();\
307 static int sad8_xy2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
310 asm volatile("pxor %%mm7, %%mm7 \n\t"\
311 "pxor %%mm6, %%mm6 \n\t"\
312 "movq %0, %%mm5 \n\t"\
313 :: "m"(round_tab[2]) \
316 sad8_4_ ## suf(blk1, blk2, stride, 8);\
318 return sum_ ## suf();\
321 static int sad16_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
323 asm volatile("pxor %%mm7, %%mm7 \n\t"\
324 "pxor %%mm6, %%mm6 \n\t":);\
326 sad8_1_ ## suf(blk1 , blk2 , stride, h);\
327 sad8_1_ ## suf(blk1+8, blk2+8, stride, h);\
329 return sum_ ## suf();\
331 static int sad16_x2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
333 asm volatile("pxor %%mm7, %%mm7 \n\t"\
334 "pxor %%mm6, %%mm6 \n\t"\
335 "movq %0, %%mm5 \n\t"\
336 :: "m"(round_tab[1]) \
339 sad8_2_ ## suf(blk1 , blk1+1, blk2 , stride, h);\
340 sad8_2_ ## suf(blk1+8, blk1+9, blk2+8, stride, h);\
342 return sum_ ## suf();\
344 static int sad16_y2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
346 asm volatile("pxor %%mm7, %%mm7 \n\t"\
347 "pxor %%mm6, %%mm6 \n\t"\
348 "movq %0, %%mm5 \n\t"\
349 :: "m"(round_tab[1]) \
352 sad8_2_ ## suf(blk1 , blk1+stride, blk2 , stride, h);\
353 sad8_2_ ## suf(blk1+8, blk1+stride+8,blk2+8, stride, h);\
355 return sum_ ## suf();\
357 static int sad16_xy2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
359 asm volatile("pxor %%mm7, %%mm7 \n\t"\
360 "pxor %%mm6, %%mm6 \n\t"\
361 "movq %0, %%mm5 \n\t"\
362 :: "m"(round_tab[2]) \
365 sad8_4_ ## suf(blk1 , blk2 , stride, h);\
366 sad8_4_ ## suf(blk1+8, blk2+8, stride, h);\
368 return sum_ ## suf();\
374 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx)
376 if (mm_flags & MM_MMX) {
377 c->pix_abs[0][0] = sad16_mmx;
378 c->pix_abs[0][1] = sad16_x2_mmx;
379 c->pix_abs[0][2] = sad16_y2_mmx;
380 c->pix_abs[0][3] = sad16_xy2_mmx;
381 c->pix_abs[1][0] = sad8_mmx;
382 c->pix_abs[1][1] = sad8_x2_mmx;
383 c->pix_abs[1][2] = sad8_y2_mmx;
384 c->pix_abs[1][3] = sad8_xy2_mmx;
386 c->sad[0]= sad16_mmx;
389 if (mm_flags & MM_MMXEXT) {
390 c->pix_abs[0][0] = sad16_mmx2;
391 c->pix_abs[1][0] = sad8_mmx2;
393 c->sad[0]= sad16_mmx2;
394 c->sad[1]= sad8_mmx2;
396 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
397 c->pix_abs[0][1] = sad16_x2_mmx2;
398 c->pix_abs[0][2] = sad16_y2_mmx2;
399 c->pix_abs[0][3] = sad16_xy2_mmx2;
400 c->pix_abs[1][1] = sad8_x2_mmx2;
401 c->pix_abs[1][2] = sad8_y2_mmx2;
402 c->pix_abs[1][3] = sad8_xy2_mmx2;