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"
25 static const __attribute__ ((aligned(8))) uint64_t round_tab[3]={
26 0x0000000000000000ULL,
27 0x0001000100010001ULL,
28 0x0002000200020002ULL,
31 static attribute_used __attribute__ ((aligned(8))) uint64_t bone= 0x0101010101010101LL;
33 static inline void sad8_1_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
35 long len= -(stride*h);
39 "movq (%1, %%"REG_a"), %%mm0 \n\t"
40 "movq (%2, %%"REG_a"), %%mm2 \n\t"
41 "movq (%2, %%"REG_a"), %%mm4 \n\t"
42 "add %3, %%"REG_a" \n\t"
43 "psubusb %%mm0, %%mm2 \n\t"
44 "psubusb %%mm4, %%mm0 \n\t"
45 "movq (%1, %%"REG_a"), %%mm1 \n\t"
46 "movq (%2, %%"REG_a"), %%mm3 \n\t"
47 "movq (%2, %%"REG_a"), %%mm5 \n\t"
48 "psubusb %%mm1, %%mm3 \n\t"
49 "psubusb %%mm5, %%mm1 \n\t"
50 "por %%mm2, %%mm0 \n\t"
51 "por %%mm1, %%mm3 \n\t"
52 "movq %%mm0, %%mm1 \n\t"
53 "movq %%mm3, %%mm2 \n\t"
54 "punpcklbw %%mm7, %%mm0 \n\t"
55 "punpckhbw %%mm7, %%mm1 \n\t"
56 "punpcklbw %%mm7, %%mm3 \n\t"
57 "punpckhbw %%mm7, %%mm2 \n\t"
58 "paddw %%mm1, %%mm0 \n\t"
59 "paddw %%mm3, %%mm2 \n\t"
60 "paddw %%mm2, %%mm0 \n\t"
61 "paddw %%mm0, %%mm6 \n\t"
62 "add %3, %%"REG_a" \n\t"
65 : "r" (blk1 - len), "r" (blk2 - len), "r" ((long)stride)
69 static inline void sad8_1_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h)
71 long len= -(stride*h);
75 "movq (%1, %%"REG_a"), %%mm0 \n\t"
76 "movq (%2, %%"REG_a"), %%mm2 \n\t"
77 "psadbw %%mm2, %%mm0 \n\t"
78 "add %3, %%"REG_a" \n\t"
79 "movq (%1, %%"REG_a"), %%mm1 \n\t"
80 "movq (%2, %%"REG_a"), %%mm3 \n\t"
81 "psadbw %%mm1, %%mm3 \n\t"
82 "paddw %%mm3, %%mm0 \n\t"
83 "paddw %%mm0, %%mm6 \n\t"
84 "add %3, %%"REG_a" \n\t"
87 : "r" (blk1 - len), "r" (blk2 - len), "r" ((long)stride)
91 static inline void sad8_2_mmx2(uint8_t *blk1a, uint8_t *blk1b, uint8_t *blk2, int stride, int h)
93 long len= -(stride*h);
97 "movq (%1, %%"REG_a"), %%mm0 \n\t"
98 "movq (%2, %%"REG_a"), %%mm2 \n\t"
99 "pavgb %%mm2, %%mm0 \n\t"
100 "movq (%3, %%"REG_a"), %%mm2 \n\t"
101 "psadbw %%mm2, %%mm0 \n\t"
102 "add %4, %%"REG_a" \n\t"
103 "movq (%1, %%"REG_a"), %%mm1 \n\t"
104 "movq (%2, %%"REG_a"), %%mm3 \n\t"
105 "pavgb %%mm1, %%mm3 \n\t"
106 "movq (%3, %%"REG_a"), %%mm1 \n\t"
107 "psadbw %%mm1, %%mm3 \n\t"
108 "paddw %%mm3, %%mm0 \n\t"
109 "paddw %%mm0, %%mm6 \n\t"
110 "add %4, %%"REG_a" \n\t"
113 : "r" (blk1a - len), "r" (blk1b -len), "r" (blk2 - len), "r" ((long)stride)
117 static inline void sad8_4_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h)
119 long len= -(stride*h);
122 "movq "MANGLE(bone)", %%mm5 \n\t"
124 "movq (%1, %%"REG_a"), %%mm0 \n\t"
125 "movq (%2, %%"REG_a"), %%mm2 \n\t"
126 "movq 1(%1, %%"REG_a"), %%mm1 \n\t"
127 "movq 1(%2, %%"REG_a"), %%mm3 \n\t"
128 "pavgb %%mm2, %%mm0 \n\t"
129 "pavgb %%mm1, %%mm3 \n\t"
130 "psubusb %%mm5, %%mm3 \n\t"
131 "pavgb %%mm3, %%mm0 \n\t"
132 "movq (%3, %%"REG_a"), %%mm2 \n\t"
133 "psadbw %%mm2, %%mm0 \n\t"
134 "add %4, %%"REG_a" \n\t"
135 "movq (%1, %%"REG_a"), %%mm1 \n\t"
136 "movq (%2, %%"REG_a"), %%mm3 \n\t"
137 "movq 1(%1, %%"REG_a"), %%mm2 \n\t"
138 "movq 1(%2, %%"REG_a"), %%mm4 \n\t"
139 "pavgb %%mm3, %%mm1 \n\t"
140 "pavgb %%mm4, %%mm2 \n\t"
141 "psubusb %%mm5, %%mm2 \n\t"
142 "pavgb %%mm1, %%mm2 \n\t"
143 "movq (%3, %%"REG_a"), %%mm1 \n\t"
144 "psadbw %%mm1, %%mm2 \n\t"
145 "paddw %%mm2, %%mm0 \n\t"
146 "paddw %%mm0, %%mm6 \n\t"
147 "add %4, %%"REG_a" \n\t"
150 : "r" (blk1 - len), "r" (blk1 - len + stride), "r" (blk2 - len), "r" ((long)stride)
154 static inline void sad8_2_mmx(uint8_t *blk1a, uint8_t *blk1b, uint8_t *blk2, int stride, int h)
156 long len= -(stride*h);
160 "movq (%1, %%"REG_a"), %%mm0 \n\t"
161 "movq (%2, %%"REG_a"), %%mm1 \n\t"
162 "movq (%1, %%"REG_a"), %%mm2 \n\t"
163 "movq (%2, %%"REG_a"), %%mm3 \n\t"
164 "punpcklbw %%mm7, %%mm0 \n\t"
165 "punpcklbw %%mm7, %%mm1 \n\t"
166 "punpckhbw %%mm7, %%mm2 \n\t"
167 "punpckhbw %%mm7, %%mm3 \n\t"
168 "paddw %%mm0, %%mm1 \n\t"
169 "paddw %%mm2, %%mm3 \n\t"
170 "movq (%3, %%"REG_a"), %%mm4 \n\t"
171 "movq (%3, %%"REG_a"), %%mm2 \n\t"
172 "paddw %%mm5, %%mm1 \n\t"
173 "paddw %%mm5, %%mm3 \n\t"
174 "psrlw $1, %%mm1 \n\t"
175 "psrlw $1, %%mm3 \n\t"
176 "packuswb %%mm3, %%mm1 \n\t"
177 "psubusb %%mm1, %%mm4 \n\t"
178 "psubusb %%mm2, %%mm1 \n\t"
179 "por %%mm4, %%mm1 \n\t"
180 "movq %%mm1, %%mm0 \n\t"
181 "punpcklbw %%mm7, %%mm0 \n\t"
182 "punpckhbw %%mm7, %%mm1 \n\t"
183 "paddw %%mm1, %%mm0 \n\t"
184 "paddw %%mm0, %%mm6 \n\t"
185 "add %4, %%"REG_a" \n\t"
188 : "r" (blk1a - len), "r" (blk1b -len), "r" (blk2 - len), "r" ((long)stride)
192 static inline void sad8_4_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
194 long len= -(stride*h);
198 "movq (%1, %%"REG_a"), %%mm0 \n\t"
199 "movq (%2, %%"REG_a"), %%mm1 \n\t"
200 "movq %%mm0, %%mm4 \n\t"
201 "movq %%mm1, %%mm2 \n\t"
202 "punpcklbw %%mm7, %%mm0 \n\t"
203 "punpcklbw %%mm7, %%mm1 \n\t"
204 "punpckhbw %%mm7, %%mm4 \n\t"
205 "punpckhbw %%mm7, %%mm2 \n\t"
206 "paddw %%mm1, %%mm0 \n\t"
207 "paddw %%mm2, %%mm4 \n\t"
208 "movq 1(%1, %%"REG_a"), %%mm2 \n\t"
209 "movq 1(%2, %%"REG_a"), %%mm3 \n\t"
210 "movq %%mm2, %%mm1 \n\t"
211 "punpcklbw %%mm7, %%mm2 \n\t"
212 "punpckhbw %%mm7, %%mm1 \n\t"
213 "paddw %%mm0, %%mm2 \n\t"
214 "paddw %%mm4, %%mm1 \n\t"
215 "movq %%mm3, %%mm4 \n\t"
216 "punpcklbw %%mm7, %%mm3 \n\t"
217 "punpckhbw %%mm7, %%mm4 \n\t"
218 "paddw %%mm3, %%mm2 \n\t"
219 "paddw %%mm4, %%mm1 \n\t"
220 "movq (%3, %%"REG_a"), %%mm3 \n\t"
221 "movq (%3, %%"REG_a"), %%mm4 \n\t"
222 "paddw %%mm5, %%mm2 \n\t"
223 "paddw %%mm5, %%mm1 \n\t"
224 "psrlw $2, %%mm2 \n\t"
225 "psrlw $2, %%mm1 \n\t"
226 "packuswb %%mm1, %%mm2 \n\t"
227 "psubusb %%mm2, %%mm3 \n\t"
228 "psubusb %%mm4, %%mm2 \n\t"
229 "por %%mm3, %%mm2 \n\t"
230 "movq %%mm2, %%mm0 \n\t"
231 "punpcklbw %%mm7, %%mm0 \n\t"
232 "punpckhbw %%mm7, %%mm2 \n\t"
233 "paddw %%mm2, %%mm0 \n\t"
234 "paddw %%mm0, %%mm6 \n\t"
235 "add %4, %%"REG_a" \n\t"
238 : "r" (blk1 - len), "r" (blk1 -len + stride), "r" (blk2 - len), "r" ((long)stride)
242 static inline int sum_mmx(void)
246 "movq %%mm6, %%mm0 \n\t"
247 "psrlq $32, %%mm6 \n\t"
248 "paddw %%mm0, %%mm6 \n\t"
249 "movq %%mm6, %%mm0 \n\t"
250 "psrlq $16, %%mm6 \n\t"
251 "paddw %%mm0, %%mm6 \n\t"
252 "movd %%mm6, %0 \n\t"
258 static inline int sum_mmx2(void)
262 "movd %%mm6, %0 \n\t"
269 #define PIX_SAD(suf)\
270 static int sad8_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
273 asm volatile("pxor %%mm7, %%mm7 \n\t"\
274 "pxor %%mm6, %%mm6 \n\t":);\
276 sad8_1_ ## suf(blk1, blk2, stride, 8);\
278 return sum_ ## suf();\
280 static int sad8_x2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
283 asm volatile("pxor %%mm7, %%mm7 \n\t"\
284 "pxor %%mm6, %%mm6 \n\t"\
285 "movq %0, %%mm5 \n\t"\
286 :: "m"(round_tab[1]) \
289 sad8_2_ ## suf(blk1, blk1+1, blk2, stride, 8);\
291 return sum_ ## suf();\
294 static int sad8_y2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
297 asm volatile("pxor %%mm7, %%mm7 \n\t"\
298 "pxor %%mm6, %%mm6 \n\t"\
299 "movq %0, %%mm5 \n\t"\
300 :: "m"(round_tab[1]) \
303 sad8_2_ ## suf(blk1, blk1+stride, blk2, stride, 8);\
305 return sum_ ## suf();\
308 static int sad8_xy2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
311 asm volatile("pxor %%mm7, %%mm7 \n\t"\
312 "pxor %%mm6, %%mm6 \n\t"\
313 "movq %0, %%mm5 \n\t"\
314 :: "m"(round_tab[2]) \
317 sad8_4_ ## suf(blk1, blk2, stride, 8);\
319 return sum_ ## suf();\
322 static int sad16_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
324 asm volatile("pxor %%mm7, %%mm7 \n\t"\
325 "pxor %%mm6, %%mm6 \n\t":);\
327 sad8_1_ ## suf(blk1 , blk2 , stride, h);\
328 sad8_1_ ## suf(blk1+8, blk2+8, stride, h);\
330 return sum_ ## suf();\
332 static int sad16_x2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
334 asm volatile("pxor %%mm7, %%mm7 \n\t"\
335 "pxor %%mm6, %%mm6 \n\t"\
336 "movq %0, %%mm5 \n\t"\
337 :: "m"(round_tab[1]) \
340 sad8_2_ ## suf(blk1 , blk1+1, blk2 , stride, h);\
341 sad8_2_ ## suf(blk1+8, blk1+9, blk2+8, stride, h);\
343 return sum_ ## suf();\
345 static int sad16_y2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
347 asm volatile("pxor %%mm7, %%mm7 \n\t"\
348 "pxor %%mm6, %%mm6 \n\t"\
349 "movq %0, %%mm5 \n\t"\
350 :: "m"(round_tab[1]) \
353 sad8_2_ ## suf(blk1 , blk1+stride, blk2 , stride, h);\
354 sad8_2_ ## suf(blk1+8, blk1+stride+8,blk2+8, stride, h);\
356 return sum_ ## suf();\
358 static int sad16_xy2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
360 asm volatile("pxor %%mm7, %%mm7 \n\t"\
361 "pxor %%mm6, %%mm6 \n\t"\
362 "movq %0, %%mm5 \n\t"\
363 :: "m"(round_tab[2]) \
366 sad8_4_ ## suf(blk1 , blk2 , stride, h);\
367 sad8_4_ ## suf(blk1+8, blk2+8, stride, h);\
369 return sum_ ## suf();\
375 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx)
377 if (mm_flags & MM_MMX) {
378 c->pix_abs[0][0] = sad16_mmx;
379 c->pix_abs[0][1] = sad16_x2_mmx;
380 c->pix_abs[0][2] = sad16_y2_mmx;
381 c->pix_abs[0][3] = sad16_xy2_mmx;
382 c->pix_abs[1][0] = sad8_mmx;
383 c->pix_abs[1][1] = sad8_x2_mmx;
384 c->pix_abs[1][2] = sad8_y2_mmx;
385 c->pix_abs[1][3] = sad8_xy2_mmx;
387 c->sad[0]= sad16_mmx;
390 if (mm_flags & MM_MMXEXT) {
391 c->pix_abs[0][0] = sad16_mmx2;
392 c->pix_abs[1][0] = sad8_mmx2;
394 c->sad[0]= sad16_mmx2;
395 c->sad[1]= sad8_mmx2;
397 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
398 c->pix_abs[0][1] = sad16_x2_mmx2;
399 c->pix_abs[0][2] = sad16_y2_mmx2;
400 c->pix_abs[0][3] = sad16_xy2_mmx2;
401 c->pix_abs[1][1] = sad8_x2_mmx2;
402 c->pix_abs[1][2] = sad8_y2_mmx2;
403 c->pix_abs[1][3] = sad8_xy2_mmx2;