dsputilenc_mmx.c
Go to the documentation of this file.
1 /*
2  * MMX optimized DSP utils
3  * Copyright (c) 2000, 2001 Fabrice Bellard
4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5  *
6  * This file is part of Libav.
7  *
8  * Libav is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * Libav is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with Libav; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  *
22  * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
23  */
24 
25 #include "libavutil/cpu.h"
26 #include "libavutil/x86_cpu.h"
27 #include "libavcodec/dsputil.h"
28 #include "libavcodec/mpegvideo.h"
29 #include "libavcodec/mathops.h"
30 #include "dsputil_mmx.h"
31 
32 
33 static void get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int line_size)
34 {
35  __asm__ volatile(
36  "mov $-128, %%"REG_a" \n\t"
37  "pxor %%mm7, %%mm7 \n\t"
38  ".p2align 4 \n\t"
39  "1: \n\t"
40  "movq (%0), %%mm0 \n\t"
41  "movq (%0, %2), %%mm2 \n\t"
42  "movq %%mm0, %%mm1 \n\t"
43  "movq %%mm2, %%mm3 \n\t"
44  "punpcklbw %%mm7, %%mm0 \n\t"
45  "punpckhbw %%mm7, %%mm1 \n\t"
46  "punpcklbw %%mm7, %%mm2 \n\t"
47  "punpckhbw %%mm7, %%mm3 \n\t"
48  "movq %%mm0, (%1, %%"REG_a") \n\t"
49  "movq %%mm1, 8(%1, %%"REG_a") \n\t"
50  "movq %%mm2, 16(%1, %%"REG_a") \n\t"
51  "movq %%mm3, 24(%1, %%"REG_a") \n\t"
52  "add %3, %0 \n\t"
53  "add $32, %%"REG_a" \n\t"
54  "js 1b \n\t"
55  : "+r" (pixels)
56  : "r" (block+64), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*2)
57  : "%"REG_a
58  );
59 }
60 
61 static void get_pixels_sse2(DCTELEM *block, const uint8_t *pixels, int line_size)
62 {
63  __asm__ volatile(
64  "pxor %%xmm4, %%xmm4 \n\t"
65  "movq (%0), %%xmm0 \n\t"
66  "movq (%0, %2), %%xmm1 \n\t"
67  "movq (%0, %2,2), %%xmm2 \n\t"
68  "movq (%0, %3), %%xmm3 \n\t"
69  "lea (%0,%2,4), %0 \n\t"
70  "punpcklbw %%xmm4, %%xmm0 \n\t"
71  "punpcklbw %%xmm4, %%xmm1 \n\t"
72  "punpcklbw %%xmm4, %%xmm2 \n\t"
73  "punpcklbw %%xmm4, %%xmm3 \n\t"
74  "movdqa %%xmm0, (%1) \n\t"
75  "movdqa %%xmm1, 16(%1) \n\t"
76  "movdqa %%xmm2, 32(%1) \n\t"
77  "movdqa %%xmm3, 48(%1) \n\t"
78  "movq (%0), %%xmm0 \n\t"
79  "movq (%0, %2), %%xmm1 \n\t"
80  "movq (%0, %2,2), %%xmm2 \n\t"
81  "movq (%0, %3), %%xmm3 \n\t"
82  "punpcklbw %%xmm4, %%xmm0 \n\t"
83  "punpcklbw %%xmm4, %%xmm1 \n\t"
84  "punpcklbw %%xmm4, %%xmm2 \n\t"
85  "punpcklbw %%xmm4, %%xmm3 \n\t"
86  "movdqa %%xmm0, 64(%1) \n\t"
87  "movdqa %%xmm1, 80(%1) \n\t"
88  "movdqa %%xmm2, 96(%1) \n\t"
89  "movdqa %%xmm3, 112(%1) \n\t"
90  : "+r" (pixels)
91  : "r" (block), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3)
92  );
93 }
94 
95 static inline void diff_pixels_mmx(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride)
96 {
97  __asm__ volatile(
98  "pxor %%mm7, %%mm7 \n\t"
99  "mov $-128, %%"REG_a" \n\t"
100  ".p2align 4 \n\t"
101  "1: \n\t"
102  "movq (%0), %%mm0 \n\t"
103  "movq (%1), %%mm2 \n\t"
104  "movq %%mm0, %%mm1 \n\t"
105  "movq %%mm2, %%mm3 \n\t"
106  "punpcklbw %%mm7, %%mm0 \n\t"
107  "punpckhbw %%mm7, %%mm1 \n\t"
108  "punpcklbw %%mm7, %%mm2 \n\t"
109  "punpckhbw %%mm7, %%mm3 \n\t"
110  "psubw %%mm2, %%mm0 \n\t"
111  "psubw %%mm3, %%mm1 \n\t"
112  "movq %%mm0, (%2, %%"REG_a") \n\t"
113  "movq %%mm1, 8(%2, %%"REG_a") \n\t"
114  "add %3, %0 \n\t"
115  "add %3, %1 \n\t"
116  "add $16, %%"REG_a" \n\t"
117  "jnz 1b \n\t"
118  : "+r" (s1), "+r" (s2)
119  : "r" (block+64), "r" ((x86_reg)stride)
120  : "%"REG_a
121  );
122 }
123 
124 static int pix_sum16_mmx(uint8_t * pix, int line_size){
125  const int h=16;
126  int sum;
127  x86_reg index= -line_size*h;
128 
129  __asm__ volatile(
130  "pxor %%mm7, %%mm7 \n\t"
131  "pxor %%mm6, %%mm6 \n\t"
132  "1: \n\t"
133  "movq (%2, %1), %%mm0 \n\t"
134  "movq (%2, %1), %%mm1 \n\t"
135  "movq 8(%2, %1), %%mm2 \n\t"
136  "movq 8(%2, %1), %%mm3 \n\t"
137  "punpcklbw %%mm7, %%mm0 \n\t"
138  "punpckhbw %%mm7, %%mm1 \n\t"
139  "punpcklbw %%mm7, %%mm2 \n\t"
140  "punpckhbw %%mm7, %%mm3 \n\t"
141  "paddw %%mm0, %%mm1 \n\t"
142  "paddw %%mm2, %%mm3 \n\t"
143  "paddw %%mm1, %%mm3 \n\t"
144  "paddw %%mm3, %%mm6 \n\t"
145  "add %3, %1 \n\t"
146  " js 1b \n\t"
147  "movq %%mm6, %%mm5 \n\t"
148  "psrlq $32, %%mm6 \n\t"
149  "paddw %%mm5, %%mm6 \n\t"
150  "movq %%mm6, %%mm5 \n\t"
151  "psrlq $16, %%mm6 \n\t"
152  "paddw %%mm5, %%mm6 \n\t"
153  "movd %%mm6, %0 \n\t"
154  "andl $0xFFFF, %0 \n\t"
155  : "=&r" (sum), "+r" (index)
156  : "r" (pix - index), "r" ((x86_reg)line_size)
157  );
158 
159  return sum;
160 }
161 
162 static int pix_norm1_mmx(uint8_t *pix, int line_size) {
163  int tmp;
164  __asm__ volatile (
165  "movl $16,%%ecx\n"
166  "pxor %%mm0,%%mm0\n"
167  "pxor %%mm7,%%mm7\n"
168  "1:\n"
169  "movq (%0),%%mm2\n" /* mm2 = pix[0-7] */
170  "movq 8(%0),%%mm3\n" /* mm3 = pix[8-15] */
171 
172  "movq %%mm2,%%mm1\n" /* mm1 = mm2 = pix[0-7] */
173 
174  "punpckhbw %%mm0,%%mm1\n" /* mm1 = [pix4-7] */
175  "punpcklbw %%mm0,%%mm2\n" /* mm2 = [pix0-3] */
176 
177  "movq %%mm3,%%mm4\n" /* mm4 = mm3 = pix[8-15] */
178  "punpckhbw %%mm0,%%mm3\n" /* mm3 = [pix12-15] */
179  "punpcklbw %%mm0,%%mm4\n" /* mm4 = [pix8-11] */
180 
181  "pmaddwd %%mm1,%%mm1\n" /* mm1 = (pix0^2+pix1^2,pix2^2+pix3^2) */
182  "pmaddwd %%mm2,%%mm2\n" /* mm2 = (pix4^2+pix5^2,pix6^2+pix7^2) */
183 
184  "pmaddwd %%mm3,%%mm3\n"
185  "pmaddwd %%mm4,%%mm4\n"
186 
187  "paddd %%mm1,%%mm2\n" /* mm2 = (pix0^2+pix1^2+pix4^2+pix5^2,
188  pix2^2+pix3^2+pix6^2+pix7^2) */
189  "paddd %%mm3,%%mm4\n"
190  "paddd %%mm2,%%mm7\n"
191 
192  "add %2, %0\n"
193  "paddd %%mm4,%%mm7\n"
194  "dec %%ecx\n"
195  "jnz 1b\n"
196 
197  "movq %%mm7,%%mm1\n"
198  "psrlq $32, %%mm7\n" /* shift hi dword to lo */
199  "paddd %%mm7,%%mm1\n"
200  "movd %%mm1,%1\n"
201  : "+r" (pix), "=r"(tmp) : "r" ((x86_reg)line_size) : "%ecx" );
202  return tmp;
203 }
204 
205 static int sse8_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
206  int tmp;
207  __asm__ volatile (
208  "movl %4,%%ecx\n"
209  "shr $1,%%ecx\n"
210  "pxor %%mm0,%%mm0\n" /* mm0 = 0 */
211  "pxor %%mm7,%%mm7\n" /* mm7 holds the sum */
212  "1:\n"
213  "movq (%0),%%mm1\n" /* mm1 = pix1[0][0-7] */
214  "movq (%1),%%mm2\n" /* mm2 = pix2[0][0-7] */
215  "movq (%0,%3),%%mm3\n" /* mm3 = pix1[1][0-7] */
216  "movq (%1,%3),%%mm4\n" /* mm4 = pix2[1][0-7] */
217 
218  /* todo: mm1-mm2, mm3-mm4 */
219  /* algo: subtract mm1 from mm2 with saturation and vice versa */
220  /* OR the results to get absolute difference */
221  "movq %%mm1,%%mm5\n"
222  "movq %%mm3,%%mm6\n"
223  "psubusb %%mm2,%%mm1\n"
224  "psubusb %%mm4,%%mm3\n"
225  "psubusb %%mm5,%%mm2\n"
226  "psubusb %%mm6,%%mm4\n"
227 
228  "por %%mm1,%%mm2\n"
229  "por %%mm3,%%mm4\n"
230 
231  /* now convert to 16-bit vectors so we can square them */
232  "movq %%mm2,%%mm1\n"
233  "movq %%mm4,%%mm3\n"
234 
235  "punpckhbw %%mm0,%%mm2\n"
236  "punpckhbw %%mm0,%%mm4\n"
237  "punpcklbw %%mm0,%%mm1\n" /* mm1 now spread over (mm1,mm2) */
238  "punpcklbw %%mm0,%%mm3\n" /* mm4 now spread over (mm3,mm4) */
239 
240  "pmaddwd %%mm2,%%mm2\n"
241  "pmaddwd %%mm4,%%mm4\n"
242  "pmaddwd %%mm1,%%mm1\n"
243  "pmaddwd %%mm3,%%mm3\n"
244 
245  "lea (%0,%3,2), %0\n" /* pix1 += 2*line_size */
246  "lea (%1,%3,2), %1\n" /* pix2 += 2*line_size */
247 
248  "paddd %%mm2,%%mm1\n"
249  "paddd %%mm4,%%mm3\n"
250  "paddd %%mm1,%%mm7\n"
251  "paddd %%mm3,%%mm7\n"
252 
253  "decl %%ecx\n"
254  "jnz 1b\n"
255 
256  "movq %%mm7,%%mm1\n"
257  "psrlq $32, %%mm7\n" /* shift hi dword to lo */
258  "paddd %%mm7,%%mm1\n"
259  "movd %%mm1,%2\n"
260  : "+r" (pix1), "+r" (pix2), "=r"(tmp)
261  : "r" ((x86_reg)line_size) , "m" (h)
262  : "%ecx");
263  return tmp;
264 }
265 
266 static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
267  int tmp;
268  __asm__ volatile (
269  "movl %4,%%ecx\n"
270  "pxor %%mm0,%%mm0\n" /* mm0 = 0 */
271  "pxor %%mm7,%%mm7\n" /* mm7 holds the sum */
272  "1:\n"
273  "movq (%0),%%mm1\n" /* mm1 = pix1[0-7] */
274  "movq (%1),%%mm2\n" /* mm2 = pix2[0-7] */
275  "movq 8(%0),%%mm3\n" /* mm3 = pix1[8-15] */
276  "movq 8(%1),%%mm4\n" /* mm4 = pix2[8-15] */
277 
278  /* todo: mm1-mm2, mm3-mm4 */
279  /* algo: subtract mm1 from mm2 with saturation and vice versa */
280  /* OR the results to get absolute difference */
281  "movq %%mm1,%%mm5\n"
282  "movq %%mm3,%%mm6\n"
283  "psubusb %%mm2,%%mm1\n"
284  "psubusb %%mm4,%%mm3\n"
285  "psubusb %%mm5,%%mm2\n"
286  "psubusb %%mm6,%%mm4\n"
287 
288  "por %%mm1,%%mm2\n"
289  "por %%mm3,%%mm4\n"
290 
291  /* now convert to 16-bit vectors so we can square them */
292  "movq %%mm2,%%mm1\n"
293  "movq %%mm4,%%mm3\n"
294 
295  "punpckhbw %%mm0,%%mm2\n"
296  "punpckhbw %%mm0,%%mm4\n"
297  "punpcklbw %%mm0,%%mm1\n" /* mm1 now spread over (mm1,mm2) */
298  "punpcklbw %%mm0,%%mm3\n" /* mm4 now spread over (mm3,mm4) */
299 
300  "pmaddwd %%mm2,%%mm2\n"
301  "pmaddwd %%mm4,%%mm4\n"
302  "pmaddwd %%mm1,%%mm1\n"
303  "pmaddwd %%mm3,%%mm3\n"
304 
305  "add %3,%0\n"
306  "add %3,%1\n"
307 
308  "paddd %%mm2,%%mm1\n"
309  "paddd %%mm4,%%mm3\n"
310  "paddd %%mm1,%%mm7\n"
311  "paddd %%mm3,%%mm7\n"
312 
313  "decl %%ecx\n"
314  "jnz 1b\n"
315 
316  "movq %%mm7,%%mm1\n"
317  "psrlq $32, %%mm7\n" /* shift hi dword to lo */
318  "paddd %%mm7,%%mm1\n"
319  "movd %%mm1,%2\n"
320  : "+r" (pix1), "+r" (pix2), "=r"(tmp)
321  : "r" ((x86_reg)line_size) , "m" (h)
322  : "%ecx");
323  return tmp;
324 }
325 
326 int ff_sse16_sse2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h);
327 
328 static int hf_noise8_mmx(uint8_t * pix1, int line_size, int h) {
329  int tmp;
330  __asm__ volatile (
331  "movl %3,%%ecx\n"
332  "pxor %%mm7,%%mm7\n"
333  "pxor %%mm6,%%mm6\n"
334 
335  "movq (%0),%%mm0\n"
336  "movq %%mm0, %%mm1\n"
337  "psllq $8, %%mm0\n"
338  "psrlq $8, %%mm1\n"
339  "psrlq $8, %%mm0\n"
340  "movq %%mm0, %%mm2\n"
341  "movq %%mm1, %%mm3\n"
342  "punpcklbw %%mm7,%%mm0\n"
343  "punpcklbw %%mm7,%%mm1\n"
344  "punpckhbw %%mm7,%%mm2\n"
345  "punpckhbw %%mm7,%%mm3\n"
346  "psubw %%mm1, %%mm0\n"
347  "psubw %%mm3, %%mm2\n"
348 
349  "add %2,%0\n"
350 
351  "movq (%0),%%mm4\n"
352  "movq %%mm4, %%mm1\n"
353  "psllq $8, %%mm4\n"
354  "psrlq $8, %%mm1\n"
355  "psrlq $8, %%mm4\n"
356  "movq %%mm4, %%mm5\n"
357  "movq %%mm1, %%mm3\n"
358  "punpcklbw %%mm7,%%mm4\n"
359  "punpcklbw %%mm7,%%mm1\n"
360  "punpckhbw %%mm7,%%mm5\n"
361  "punpckhbw %%mm7,%%mm3\n"
362  "psubw %%mm1, %%mm4\n"
363  "psubw %%mm3, %%mm5\n"
364  "psubw %%mm4, %%mm0\n"
365  "psubw %%mm5, %%mm2\n"
366  "pxor %%mm3, %%mm3\n"
367  "pxor %%mm1, %%mm1\n"
368  "pcmpgtw %%mm0, %%mm3\n\t"
369  "pcmpgtw %%mm2, %%mm1\n\t"
370  "pxor %%mm3, %%mm0\n"
371  "pxor %%mm1, %%mm2\n"
372  "psubw %%mm3, %%mm0\n"
373  "psubw %%mm1, %%mm2\n"
374  "paddw %%mm0, %%mm2\n"
375  "paddw %%mm2, %%mm6\n"
376 
377  "add %2,%0\n"
378  "1:\n"
379 
380  "movq (%0),%%mm0\n"
381  "movq %%mm0, %%mm1\n"
382  "psllq $8, %%mm0\n"
383  "psrlq $8, %%mm1\n"
384  "psrlq $8, %%mm0\n"
385  "movq %%mm0, %%mm2\n"
386  "movq %%mm1, %%mm3\n"
387  "punpcklbw %%mm7,%%mm0\n"
388  "punpcklbw %%mm7,%%mm1\n"
389  "punpckhbw %%mm7,%%mm2\n"
390  "punpckhbw %%mm7,%%mm3\n"
391  "psubw %%mm1, %%mm0\n"
392  "psubw %%mm3, %%mm2\n"
393  "psubw %%mm0, %%mm4\n"
394  "psubw %%mm2, %%mm5\n"
395  "pxor %%mm3, %%mm3\n"
396  "pxor %%mm1, %%mm1\n"
397  "pcmpgtw %%mm4, %%mm3\n\t"
398  "pcmpgtw %%mm5, %%mm1\n\t"
399  "pxor %%mm3, %%mm4\n"
400  "pxor %%mm1, %%mm5\n"
401  "psubw %%mm3, %%mm4\n"
402  "psubw %%mm1, %%mm5\n"
403  "paddw %%mm4, %%mm5\n"
404  "paddw %%mm5, %%mm6\n"
405 
406  "add %2,%0\n"
407 
408  "movq (%0),%%mm4\n"
409  "movq %%mm4, %%mm1\n"
410  "psllq $8, %%mm4\n"
411  "psrlq $8, %%mm1\n"
412  "psrlq $8, %%mm4\n"
413  "movq %%mm4, %%mm5\n"
414  "movq %%mm1, %%mm3\n"
415  "punpcklbw %%mm7,%%mm4\n"
416  "punpcklbw %%mm7,%%mm1\n"
417  "punpckhbw %%mm7,%%mm5\n"
418  "punpckhbw %%mm7,%%mm3\n"
419  "psubw %%mm1, %%mm4\n"
420  "psubw %%mm3, %%mm5\n"
421  "psubw %%mm4, %%mm0\n"
422  "psubw %%mm5, %%mm2\n"
423  "pxor %%mm3, %%mm3\n"
424  "pxor %%mm1, %%mm1\n"
425  "pcmpgtw %%mm0, %%mm3\n\t"
426  "pcmpgtw %%mm2, %%mm1\n\t"
427  "pxor %%mm3, %%mm0\n"
428  "pxor %%mm1, %%mm2\n"
429  "psubw %%mm3, %%mm0\n"
430  "psubw %%mm1, %%mm2\n"
431  "paddw %%mm0, %%mm2\n"
432  "paddw %%mm2, %%mm6\n"
433 
434  "add %2,%0\n"
435  "subl $2, %%ecx\n"
436  " jnz 1b\n"
437 
438  "movq %%mm6, %%mm0\n"
439  "punpcklwd %%mm7,%%mm0\n"
440  "punpckhwd %%mm7,%%mm6\n"
441  "paddd %%mm0, %%mm6\n"
442 
443  "movq %%mm6,%%mm0\n"
444  "psrlq $32, %%mm6\n"
445  "paddd %%mm6,%%mm0\n"
446  "movd %%mm0,%1\n"
447  : "+r" (pix1), "=r"(tmp)
448  : "r" ((x86_reg)line_size) , "g" (h-2)
449  : "%ecx");
450  return tmp;
451 }
452 
453 static int hf_noise16_mmx(uint8_t * pix1, int line_size, int h) {
454  int tmp;
455  uint8_t * pix= pix1;
456  __asm__ volatile (
457  "movl %3,%%ecx\n"
458  "pxor %%mm7,%%mm7\n"
459  "pxor %%mm6,%%mm6\n"
460 
461  "movq (%0),%%mm0\n"
462  "movq 1(%0),%%mm1\n"
463  "movq %%mm0, %%mm2\n"
464  "movq %%mm1, %%mm3\n"
465  "punpcklbw %%mm7,%%mm0\n"
466  "punpcklbw %%mm7,%%mm1\n"
467  "punpckhbw %%mm7,%%mm2\n"
468  "punpckhbw %%mm7,%%mm3\n"
469  "psubw %%mm1, %%mm0\n"
470  "psubw %%mm3, %%mm2\n"
471 
472  "add %2,%0\n"
473 
474  "movq (%0),%%mm4\n"
475  "movq 1(%0),%%mm1\n"
476  "movq %%mm4, %%mm5\n"
477  "movq %%mm1, %%mm3\n"
478  "punpcklbw %%mm7,%%mm4\n"
479  "punpcklbw %%mm7,%%mm1\n"
480  "punpckhbw %%mm7,%%mm5\n"
481  "punpckhbw %%mm7,%%mm3\n"
482  "psubw %%mm1, %%mm4\n"
483  "psubw %%mm3, %%mm5\n"
484  "psubw %%mm4, %%mm0\n"
485  "psubw %%mm5, %%mm2\n"
486  "pxor %%mm3, %%mm3\n"
487  "pxor %%mm1, %%mm1\n"
488  "pcmpgtw %%mm0, %%mm3\n\t"
489  "pcmpgtw %%mm2, %%mm1\n\t"
490  "pxor %%mm3, %%mm0\n"
491  "pxor %%mm1, %%mm2\n"
492  "psubw %%mm3, %%mm0\n"
493  "psubw %%mm1, %%mm2\n"
494  "paddw %%mm0, %%mm2\n"
495  "paddw %%mm2, %%mm6\n"
496 
497  "add %2,%0\n"
498  "1:\n"
499 
500  "movq (%0),%%mm0\n"
501  "movq 1(%0),%%mm1\n"
502  "movq %%mm0, %%mm2\n"
503  "movq %%mm1, %%mm3\n"
504  "punpcklbw %%mm7,%%mm0\n"
505  "punpcklbw %%mm7,%%mm1\n"
506  "punpckhbw %%mm7,%%mm2\n"
507  "punpckhbw %%mm7,%%mm3\n"
508  "psubw %%mm1, %%mm0\n"
509  "psubw %%mm3, %%mm2\n"
510  "psubw %%mm0, %%mm4\n"
511  "psubw %%mm2, %%mm5\n"
512  "pxor %%mm3, %%mm3\n"
513  "pxor %%mm1, %%mm1\n"
514  "pcmpgtw %%mm4, %%mm3\n\t"
515  "pcmpgtw %%mm5, %%mm1\n\t"
516  "pxor %%mm3, %%mm4\n"
517  "pxor %%mm1, %%mm5\n"
518  "psubw %%mm3, %%mm4\n"
519  "psubw %%mm1, %%mm5\n"
520  "paddw %%mm4, %%mm5\n"
521  "paddw %%mm5, %%mm6\n"
522 
523  "add %2,%0\n"
524 
525  "movq (%0),%%mm4\n"
526  "movq 1(%0),%%mm1\n"
527  "movq %%mm4, %%mm5\n"
528  "movq %%mm1, %%mm3\n"
529  "punpcklbw %%mm7,%%mm4\n"
530  "punpcklbw %%mm7,%%mm1\n"
531  "punpckhbw %%mm7,%%mm5\n"
532  "punpckhbw %%mm7,%%mm3\n"
533  "psubw %%mm1, %%mm4\n"
534  "psubw %%mm3, %%mm5\n"
535  "psubw %%mm4, %%mm0\n"
536  "psubw %%mm5, %%mm2\n"
537  "pxor %%mm3, %%mm3\n"
538  "pxor %%mm1, %%mm1\n"
539  "pcmpgtw %%mm0, %%mm3\n\t"
540  "pcmpgtw %%mm2, %%mm1\n\t"
541  "pxor %%mm3, %%mm0\n"
542  "pxor %%mm1, %%mm2\n"
543  "psubw %%mm3, %%mm0\n"
544  "psubw %%mm1, %%mm2\n"
545  "paddw %%mm0, %%mm2\n"
546  "paddw %%mm2, %%mm6\n"
547 
548  "add %2,%0\n"
549  "subl $2, %%ecx\n"
550  " jnz 1b\n"
551 
552  "movq %%mm6, %%mm0\n"
553  "punpcklwd %%mm7,%%mm0\n"
554  "punpckhwd %%mm7,%%mm6\n"
555  "paddd %%mm0, %%mm6\n"
556 
557  "movq %%mm6,%%mm0\n"
558  "psrlq $32, %%mm6\n"
559  "paddd %%mm6,%%mm0\n"
560  "movd %%mm0,%1\n"
561  : "+r" (pix1), "=r"(tmp)
562  : "r" ((x86_reg)line_size) , "g" (h-2)
563  : "%ecx");
564  return tmp + hf_noise8_mmx(pix+8, line_size, h);
565 }
566 
567 static int nsse16_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
568  MpegEncContext *c = p;
569  int score1, score2;
570 
571  if(c) score1 = c->dsp.sse[0](c, pix1, pix2, line_size, h);
572  else score1 = sse16_mmx(c, pix1, pix2, line_size, h);
573  score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);
574 
575  if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
576  else return score1 + FFABS(score2)*8;
577 }
578 
579 static int nsse8_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
580  MpegEncContext *c = p;
581  int score1= sse8_mmx(c, pix1, pix2, line_size, h);
582  int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);
583 
584  if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
585  else return score1 + FFABS(score2)*8;
586 }
587 
588 static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {
589  int tmp;
590 
591  assert( (((int)pix) & 7) == 0);
592  assert((line_size &7) ==0);
593 
594 #define SUM(in0, in1, out0, out1) \
595  "movq (%0), %%mm2\n"\
596  "movq 8(%0), %%mm3\n"\
597  "add %2,%0\n"\
598  "movq %%mm2, " #out0 "\n"\
599  "movq %%mm3, " #out1 "\n"\
600  "psubusb " #in0 ", %%mm2\n"\
601  "psubusb " #in1 ", %%mm3\n"\
602  "psubusb " #out0 ", " #in0 "\n"\
603  "psubusb " #out1 ", " #in1 "\n"\
604  "por %%mm2, " #in0 "\n"\
605  "por %%mm3, " #in1 "\n"\
606  "movq " #in0 ", %%mm2\n"\
607  "movq " #in1 ", %%mm3\n"\
608  "punpcklbw %%mm7, " #in0 "\n"\
609  "punpcklbw %%mm7, " #in1 "\n"\
610  "punpckhbw %%mm7, %%mm2\n"\
611  "punpckhbw %%mm7, %%mm3\n"\
612  "paddw " #in1 ", " #in0 "\n"\
613  "paddw %%mm3, %%mm2\n"\
614  "paddw %%mm2, " #in0 "\n"\
615  "paddw " #in0 ", %%mm6\n"
616 
617 
618  __asm__ volatile (
619  "movl %3,%%ecx\n"
620  "pxor %%mm6,%%mm6\n"
621  "pxor %%mm7,%%mm7\n"
622  "movq (%0),%%mm0\n"
623  "movq 8(%0),%%mm1\n"
624  "add %2,%0\n"
625  "jmp 2f\n"
626  "1:\n"
627 
628  SUM(%%mm4, %%mm5, %%mm0, %%mm1)
629  "2:\n"
630  SUM(%%mm0, %%mm1, %%mm4, %%mm5)
631 
632  "subl $2, %%ecx\n"
633  "jnz 1b\n"
634 
635  "movq %%mm6,%%mm0\n"
636  "psrlq $32, %%mm6\n"
637  "paddw %%mm6,%%mm0\n"
638  "movq %%mm0,%%mm6\n"
639  "psrlq $16, %%mm0\n"
640  "paddw %%mm6,%%mm0\n"
641  "movd %%mm0,%1\n"
642  : "+r" (pix), "=r"(tmp)
643  : "r" ((x86_reg)line_size) , "m" (h)
644  : "%ecx");
645  return tmp & 0xFFFF;
646 }
647 #undef SUM
648 
649 static int vsad_intra16_mmx2(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {
650  int tmp;
651 
652  assert( (((int)pix) & 7) == 0);
653  assert((line_size &7) ==0);
654 
655 #define SUM(in0, in1, out0, out1) \
656  "movq (%0), " #out0 "\n"\
657  "movq 8(%0), " #out1 "\n"\
658  "add %2,%0\n"\
659  "psadbw " #out0 ", " #in0 "\n"\
660  "psadbw " #out1 ", " #in1 "\n"\
661  "paddw " #in1 ", " #in0 "\n"\
662  "paddw " #in0 ", %%mm6\n"
663 
664  __asm__ volatile (
665  "movl %3,%%ecx\n"
666  "pxor %%mm6,%%mm6\n"
667  "pxor %%mm7,%%mm7\n"
668  "movq (%0),%%mm0\n"
669  "movq 8(%0),%%mm1\n"
670  "add %2,%0\n"
671  "jmp 2f\n"
672  "1:\n"
673 
674  SUM(%%mm4, %%mm5, %%mm0, %%mm1)
675  "2:\n"
676  SUM(%%mm0, %%mm1, %%mm4, %%mm5)
677 
678  "subl $2, %%ecx\n"
679  "jnz 1b\n"
680 
681  "movd %%mm6,%1\n"
682  : "+r" (pix), "=r"(tmp)
683  : "r" ((x86_reg)line_size) , "m" (h)
684  : "%ecx");
685  return tmp;
686 }
687 #undef SUM
688 
689 static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
690  int tmp;
691 
692  assert( (((int)pix1) & 7) == 0);
693  assert( (((int)pix2) & 7) == 0);
694  assert((line_size &7) ==0);
695 
696 #define SUM(in0, in1, out0, out1) \
697  "movq (%0),%%mm2\n"\
698  "movq (%1)," #out0 "\n"\
699  "movq 8(%0),%%mm3\n"\
700  "movq 8(%1)," #out1 "\n"\
701  "add %3,%0\n"\
702  "add %3,%1\n"\
703  "psubb " #out0 ", %%mm2\n"\
704  "psubb " #out1 ", %%mm3\n"\
705  "pxor %%mm7, %%mm2\n"\
706  "pxor %%mm7, %%mm3\n"\
707  "movq %%mm2, " #out0 "\n"\
708  "movq %%mm3, " #out1 "\n"\
709  "psubusb " #in0 ", %%mm2\n"\
710  "psubusb " #in1 ", %%mm3\n"\
711  "psubusb " #out0 ", " #in0 "\n"\
712  "psubusb " #out1 ", " #in1 "\n"\
713  "por %%mm2, " #in0 "\n"\
714  "por %%mm3, " #in1 "\n"\
715  "movq " #in0 ", %%mm2\n"\
716  "movq " #in1 ", %%mm3\n"\
717  "punpcklbw %%mm7, " #in0 "\n"\
718  "punpcklbw %%mm7, " #in1 "\n"\
719  "punpckhbw %%mm7, %%mm2\n"\
720  "punpckhbw %%mm7, %%mm3\n"\
721  "paddw " #in1 ", " #in0 "\n"\
722  "paddw %%mm3, %%mm2\n"\
723  "paddw %%mm2, " #in0 "\n"\
724  "paddw " #in0 ", %%mm6\n"
725 
726 
727  __asm__ volatile (
728  "movl %4,%%ecx\n"
729  "pxor %%mm6,%%mm6\n"
730  "pcmpeqw %%mm7,%%mm7\n"
731  "psllw $15, %%mm7\n"
732  "packsswb %%mm7, %%mm7\n"
733  "movq (%0),%%mm0\n"
734  "movq (%1),%%mm2\n"
735  "movq 8(%0),%%mm1\n"
736  "movq 8(%1),%%mm3\n"
737  "add %3,%0\n"
738  "add %3,%1\n"
739  "psubb %%mm2, %%mm0\n"
740  "psubb %%mm3, %%mm1\n"
741  "pxor %%mm7, %%mm0\n"
742  "pxor %%mm7, %%mm1\n"
743  "jmp 2f\n"
744  "1:\n"
745 
746  SUM(%%mm4, %%mm5, %%mm0, %%mm1)
747  "2:\n"
748  SUM(%%mm0, %%mm1, %%mm4, %%mm5)
749 
750  "subl $2, %%ecx\n"
751  "jnz 1b\n"
752 
753  "movq %%mm6,%%mm0\n"
754  "psrlq $32, %%mm6\n"
755  "paddw %%mm6,%%mm0\n"
756  "movq %%mm0,%%mm6\n"
757  "psrlq $16, %%mm0\n"
758  "paddw %%mm6,%%mm0\n"
759  "movd %%mm0,%2\n"
760  : "+r" (pix1), "+r" (pix2), "=r"(tmp)
761  : "r" ((x86_reg)line_size) , "m" (h)
762  : "%ecx");
763  return tmp & 0x7FFF;
764 }
765 #undef SUM
766 
767 static int vsad16_mmx2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
768  int tmp;
769 
770  assert( (((int)pix1) & 7) == 0);
771  assert( (((int)pix2) & 7) == 0);
772  assert((line_size &7) ==0);
773 
774 #define SUM(in0, in1, out0, out1) \
775  "movq (%0)," #out0 "\n"\
776  "movq (%1),%%mm2\n"\
777  "movq 8(%0)," #out1 "\n"\
778  "movq 8(%1),%%mm3\n"\
779  "add %3,%0\n"\
780  "add %3,%1\n"\
781  "psubb %%mm2, " #out0 "\n"\
782  "psubb %%mm3, " #out1 "\n"\
783  "pxor %%mm7, " #out0 "\n"\
784  "pxor %%mm7, " #out1 "\n"\
785  "psadbw " #out0 ", " #in0 "\n"\
786  "psadbw " #out1 ", " #in1 "\n"\
787  "paddw " #in1 ", " #in0 "\n"\
788  "paddw " #in0 ", %%mm6\n"
789 
790  __asm__ volatile (
791  "movl %4,%%ecx\n"
792  "pxor %%mm6,%%mm6\n"
793  "pcmpeqw %%mm7,%%mm7\n"
794  "psllw $15, %%mm7\n"
795  "packsswb %%mm7, %%mm7\n"
796  "movq (%0),%%mm0\n"
797  "movq (%1),%%mm2\n"
798  "movq 8(%0),%%mm1\n"
799  "movq 8(%1),%%mm3\n"
800  "add %3,%0\n"
801  "add %3,%1\n"
802  "psubb %%mm2, %%mm0\n"
803  "psubb %%mm3, %%mm1\n"
804  "pxor %%mm7, %%mm0\n"
805  "pxor %%mm7, %%mm1\n"
806  "jmp 2f\n"
807  "1:\n"
808 
809  SUM(%%mm4, %%mm5, %%mm0, %%mm1)
810  "2:\n"
811  SUM(%%mm0, %%mm1, %%mm4, %%mm5)
812 
813  "subl $2, %%ecx\n"
814  "jnz 1b\n"
815 
816  "movd %%mm6,%2\n"
817  : "+r" (pix1), "+r" (pix2), "=r"(tmp)
818  : "r" ((x86_reg)line_size) , "m" (h)
819  : "%ecx");
820  return tmp;
821 }
822 #undef SUM
823 
824 static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
825  x86_reg i=0;
826  __asm__ volatile(
827  "1: \n\t"
828  "movq (%2, %0), %%mm0 \n\t"
829  "movq (%1, %0), %%mm1 \n\t"
830  "psubb %%mm0, %%mm1 \n\t"
831  "movq %%mm1, (%3, %0) \n\t"
832  "movq 8(%2, %0), %%mm0 \n\t"
833  "movq 8(%1, %0), %%mm1 \n\t"
834  "psubb %%mm0, %%mm1 \n\t"
835  "movq %%mm1, 8(%3, %0) \n\t"
836  "add $16, %0 \n\t"
837  "cmp %4, %0 \n\t"
838  " jb 1b \n\t"
839  : "+r" (i)
840  : "r"(src1), "r"(src2), "r"(dst), "r"((x86_reg)w-15)
841  );
842  for(; i<w; i++)
843  dst[i+0] = src1[i+0]-src2[i+0];
844 }
845 
846 static void sub_hfyu_median_prediction_mmx2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top){
847  x86_reg i=0;
848  uint8_t l, lt;
849 
850  __asm__ volatile(
851  "1: \n\t"
852  "movq -1(%1, %0), %%mm0 \n\t" // LT
853  "movq (%1, %0), %%mm1 \n\t" // T
854  "movq -1(%2, %0), %%mm2 \n\t" // L
855  "movq (%2, %0), %%mm3 \n\t" // X
856  "movq %%mm2, %%mm4 \n\t" // L
857  "psubb %%mm0, %%mm2 \n\t"
858  "paddb %%mm1, %%mm2 \n\t" // L + T - LT
859  "movq %%mm4, %%mm5 \n\t" // L
860  "pmaxub %%mm1, %%mm4 \n\t" // max(T, L)
861  "pminub %%mm5, %%mm1 \n\t" // min(T, L)
862  "pminub %%mm2, %%mm4 \n\t"
863  "pmaxub %%mm1, %%mm4 \n\t"
864  "psubb %%mm4, %%mm3 \n\t" // dst - pred
865  "movq %%mm3, (%3, %0) \n\t"
866  "add $8, %0 \n\t"
867  "cmp %4, %0 \n\t"
868  " jb 1b \n\t"
869  : "+r" (i)
870  : "r"(src1), "r"(src2), "r"(dst), "r"((x86_reg)w)
871  );
872 
873  l= *left;
874  lt= *left_top;
875 
876  dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);
877 
878  *left_top= src1[w-1];
879  *left = src2[w-1];
880 }
881 
882 #define MMABS_MMX(a,z)\
883  "pxor " #z ", " #z " \n\t"\
884  "pcmpgtw " #a ", " #z " \n\t"\
885  "pxor " #z ", " #a " \n\t"\
886  "psubw " #z ", " #a " \n\t"
887 
888 #define MMABS_MMX2(a,z)\
889  "pxor " #z ", " #z " \n\t"\
890  "psubw " #a ", " #z " \n\t"\
891  "pmaxsw " #z ", " #a " \n\t"
892 
893 #define MMABS_SSSE3(a,z)\
894  "pabsw " #a ", " #a " \n\t"
895 
896 #define MMABS_SUM(a,z, sum)\
897  MMABS(a,z)\
898  "paddusw " #a ", " #sum " \n\t"
899 
900 /* FIXME: HSUM_* saturates at 64k, while an 8x8 hadamard or dct block can get up to
901  * about 100k on extreme inputs. But that's very unlikely to occur in natural video,
902  * and it's even more unlikely to not have any alternative mvs/modes with lower cost. */
903 #define HSUM_MMX(a, t, dst)\
904  "movq "#a", "#t" \n\t"\
905  "psrlq $32, "#a" \n\t"\
906  "paddusw "#t", "#a" \n\t"\
907  "movq "#a", "#t" \n\t"\
908  "psrlq $16, "#a" \n\t"\
909  "paddusw "#t", "#a" \n\t"\
910  "movd "#a", "#dst" \n\t"\
911 
912 #define HSUM_MMX2(a, t, dst)\
913  "pshufw $0x0E, "#a", "#t" \n\t"\
914  "paddusw "#t", "#a" \n\t"\
915  "pshufw $0x01, "#a", "#t" \n\t"\
916  "paddusw "#t", "#a" \n\t"\
917  "movd "#a", "#dst" \n\t"\
918 
919 #define HSUM_SSE2(a, t, dst)\
920  "movhlps "#a", "#t" \n\t"\
921  "paddusw "#t", "#a" \n\t"\
922  "pshuflw $0x0E, "#a", "#t" \n\t"\
923  "paddusw "#t", "#a" \n\t"\
924  "pshuflw $0x01, "#a", "#t" \n\t"\
925  "paddusw "#t", "#a" \n\t"\
926  "movd "#a", "#dst" \n\t"\
927 
928 #define hadamard_func(cpu) \
929 int ff_hadamard8_diff_##cpu (void *s, uint8_t *src1, uint8_t *src2, \
930  int stride, int h); \
931 int ff_hadamard8_diff16_##cpu(void *s, uint8_t *src1, uint8_t *src2, \
932  int stride, int h);
933 
935 hadamard_func(mmx2)
936 hadamard_func(sse2)
937 hadamard_func(ssse3)
938 
939 #define DCT_SAD4(m,mm,o)\
940  "mov"#m" "#o"+ 0(%1), "#mm"2 \n\t"\
941  "mov"#m" "#o"+16(%1), "#mm"3 \n\t"\
942  "mov"#m" "#o"+32(%1), "#mm"4 \n\t"\
943  "mov"#m" "#o"+48(%1), "#mm"5 \n\t"\
944  MMABS_SUM(mm##2, mm##6, mm##0)\
945  MMABS_SUM(mm##3, mm##7, mm##1)\
946  MMABS_SUM(mm##4, mm##6, mm##0)\
947  MMABS_SUM(mm##5, mm##7, mm##1)\
948 
949 #define DCT_SAD_MMX\
950  "pxor %%mm0, %%mm0 \n\t"\
951  "pxor %%mm1, %%mm1 \n\t"\
952  DCT_SAD4(q, %%mm, 0)\
953  DCT_SAD4(q, %%mm, 8)\
954  DCT_SAD4(q, %%mm, 64)\
955  DCT_SAD4(q, %%mm, 72)\
956  "paddusw %%mm1, %%mm0 \n\t"\
957  HSUM(%%mm0, %%mm1, %0)
958 
959 #define DCT_SAD_SSE2\
960  "pxor %%xmm0, %%xmm0 \n\t"\
961  "pxor %%xmm1, %%xmm1 \n\t"\
962  DCT_SAD4(dqa, %%xmm, 0)\
963  DCT_SAD4(dqa, %%xmm, 64)\
964  "paddusw %%xmm1, %%xmm0 \n\t"\
965  HSUM(%%xmm0, %%xmm1, %0)
966 
967 #define DCT_SAD_FUNC(cpu) \
968 static int sum_abs_dctelem_##cpu(DCTELEM *block){\
969  int sum;\
970  __asm__ volatile(\
971  DCT_SAD\
972  :"=r"(sum)\
973  :"r"(block)\
974  );\
975  return sum&0xFFFF;\
976 }
977 
978 #define DCT_SAD DCT_SAD_MMX
979 #define HSUM(a,t,dst) HSUM_MMX(a,t,dst)
980 #define MMABS(a,z) MMABS_MMX(a,z)
981 DCT_SAD_FUNC(mmx)
982 #undef MMABS
983 #undef HSUM
984 
985 #define HSUM(a,t,dst) HSUM_MMX2(a,t,dst)
986 #define MMABS(a,z) MMABS_MMX2(a,z)
987 DCT_SAD_FUNC(mmx2)
988 #undef HSUM
989 #undef DCT_SAD
990 
991 #define DCT_SAD DCT_SAD_SSE2
992 #define HSUM(a,t,dst) HSUM_SSE2(a,t,dst)
993 DCT_SAD_FUNC(sse2)
994 #undef MMABS
995 
996 #if HAVE_SSSE3
997 #define MMABS(a,z) MMABS_SSSE3(a,z)
998 DCT_SAD_FUNC(ssse3)
999 #undef MMABS
1000 #endif
1001 #undef HSUM
1002 #undef DCT_SAD
1003 
1004 static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2, int size){
1005  int sum;
1006  x86_reg i=size;
1007  __asm__ volatile(
1008  "pxor %%mm4, %%mm4 \n"
1009  "1: \n"
1010  "sub $8, %0 \n"
1011  "movq (%2,%0), %%mm2 \n"
1012  "movq (%3,%0,2), %%mm0 \n"
1013  "movq 8(%3,%0,2), %%mm1 \n"
1014  "punpckhbw %%mm2, %%mm3 \n"
1015  "punpcklbw %%mm2, %%mm2 \n"
1016  "psraw $8, %%mm3 \n"
1017  "psraw $8, %%mm2 \n"
1018  "psubw %%mm3, %%mm1 \n"
1019  "psubw %%mm2, %%mm0 \n"
1020  "pmaddwd %%mm1, %%mm1 \n"
1021  "pmaddwd %%mm0, %%mm0 \n"
1022  "paddd %%mm1, %%mm4 \n"
1023  "paddd %%mm0, %%mm4 \n"
1024  "jg 1b \n"
1025  "movq %%mm4, %%mm3 \n"
1026  "psrlq $32, %%mm3 \n"
1027  "paddd %%mm3, %%mm4 \n"
1028  "movd %%mm4, %1 \n"
1029  :"+r"(i), "=r"(sum)
1030  :"r"(pix1), "r"(pix2)
1031  );
1032  return sum;
1033 }
1034 
1035 #define PHADDD(a, t)\
1036  "movq "#a", "#t" \n\t"\
1037  "psrlq $32, "#a" \n\t"\
1038  "paddd "#t", "#a" \n\t"
1039 /*
1040  pmulhw: dst[0-15]=(src[0-15]*dst[0-15])[16-31]
1041  pmulhrw: dst[0-15]=(src[0-15]*dst[0-15] + 0x8000)[16-31]
1042  pmulhrsw: dst[0-15]=(src[0-15]*dst[0-15] + 0x4000)[15-30]
1043  */
1044 #define PMULHRW(x, y, s, o)\
1045  "pmulhw " #s ", "#x " \n\t"\
1046  "pmulhw " #s ", "#y " \n\t"\
1047  "paddw " #o ", "#x " \n\t"\
1048  "paddw " #o ", "#y " \n\t"\
1049  "psraw $1, "#x " \n\t"\
1050  "psraw $1, "#y " \n\t"
1051 #define DEF(x) x ## _mmx
1052 #define SET_RND MOVQ_WONE
1053 #define SCALE_OFFSET 1
1054 
1055 #include "dsputil_mmx_qns_template.c"
1056 
1057 #undef DEF
1058 #undef SET_RND
1059 #undef SCALE_OFFSET
1060 #undef PMULHRW
1061 
1062 #define DEF(x) x ## _3dnow
1063 #define SET_RND(x)
1064 #define SCALE_OFFSET 0
1065 #define PMULHRW(x, y, s, o)\
1066  "pmulhrw " #s ", "#x " \n\t"\
1067  "pmulhrw " #s ", "#y " \n\t"
1068 
1069 #include "dsputil_mmx_qns_template.c"
1070 
1071 #undef DEF
1072 #undef SET_RND
1073 #undef SCALE_OFFSET
1074 #undef PMULHRW
1075 
1076 #if HAVE_SSSE3
1077 #undef PHADDD
1078 #define DEF(x) x ## _ssse3
1079 #define SET_RND(x)
1080 #define SCALE_OFFSET -1
1081 #define PHADDD(a, t)\
1082  "pshufw $0x0E, "#a", "#t" \n\t"\
1083  "paddd "#t", "#a" \n\t" /* faster than phaddd on core2 */
1084 #define PMULHRW(x, y, s, o)\
1085  "pmulhrsw " #s ", "#x " \n\t"\
1086  "pmulhrsw " #s ", "#y " \n\t"
1087 
1088 #include "dsputil_mmx_qns_template.c"
1089 
1090 #undef DEF
1091 #undef SET_RND
1092 #undef SCALE_OFFSET
1093 #undef PMULHRW
1094 #undef PHADDD
1095 #endif //HAVE_SSSE3
1096 
1097 
1099 {
1100  int mm_flags = av_get_cpu_flags();
1101  int bit_depth = avctx->bits_per_raw_sample;
1102 
1103  if (mm_flags & AV_CPU_FLAG_MMX) {
1104  const int dct_algo = avctx->dct_algo;
1105  if (avctx->bits_per_raw_sample <= 8 &&
1106  (dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX)) {
1107  if(mm_flags & AV_CPU_FLAG_SSE2){
1108  c->fdct = ff_fdct_sse2;
1109  }else if(mm_flags & AV_CPU_FLAG_MMX2){
1110  c->fdct = ff_fdct_mmx2;
1111  }else{
1112  c->fdct = ff_fdct_mmx;
1113  }
1114  }
1115 
1116  if (bit_depth <= 8)
1119  c->pix_sum = pix_sum16_mmx;
1120 
1122  c->sum_abs_dctelem= sum_abs_dctelem_mmx;
1123 
1124 #if HAVE_YASM
1125  c->hadamard8_diff[0]= ff_hadamard8_diff16_mmx;
1126  c->hadamard8_diff[1]= ff_hadamard8_diff_mmx;
1127 #endif
1128 
1129  c->pix_norm1 = pix_norm1_mmx;
1130  c->sse[0] = (HAVE_YASM && mm_flags & AV_CPU_FLAG_SSE2) ? ff_sse16_sse2 : sse16_mmx;
1131  c->sse[1] = sse8_mmx;
1132  c->vsad[4]= vsad_intra16_mmx;
1133 
1134  c->nsse[0] = nsse16_mmx;
1135  c->nsse[1] = nsse8_mmx;
1136  if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
1137  c->vsad[0] = vsad16_mmx;
1138  }
1139 
1140  if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
1141  c->try_8x8basis= try_8x8basis_mmx;
1142  }
1143  c->add_8x8basis= add_8x8basis_mmx;
1144 
1145  c->ssd_int8_vs_int16 = ssd_int8_vs_int16_mmx;
1146 
1147 
1148  if (mm_flags & AV_CPU_FLAG_MMX2) {
1149  c->sum_abs_dctelem= sum_abs_dctelem_mmx2;
1150 #if HAVE_YASM
1151  c->hadamard8_diff[0]= ff_hadamard8_diff16_mmx2;
1152  c->hadamard8_diff[1]= ff_hadamard8_diff_mmx2;
1153 #endif
1154  c->vsad[4]= vsad_intra16_mmx2;
1155 
1156  if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
1157  c->vsad[0] = vsad16_mmx2;
1158  }
1159 
1161  }
1162 
1163  if(mm_flags & AV_CPU_FLAG_SSE2){
1164  if (bit_depth <= 8)
1166  c->sum_abs_dctelem= sum_abs_dctelem_sse2;
1167 #if HAVE_YASM && HAVE_ALIGNED_STACK
1168  c->hadamard8_diff[0]= ff_hadamard8_diff16_sse2;
1169  c->hadamard8_diff[1]= ff_hadamard8_diff_sse2;
1170 #endif
1171  }
1172 
1173 #if HAVE_SSSE3
1174  if(mm_flags & AV_CPU_FLAG_SSSE3){
1175  if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
1176  c->try_8x8basis= try_8x8basis_ssse3;
1177  }
1178  c->add_8x8basis= add_8x8basis_ssse3;
1179  c->sum_abs_dctelem= sum_abs_dctelem_ssse3;
1180 #if HAVE_YASM && HAVE_ALIGNED_STACK
1181  c->hadamard8_diff[0]= ff_hadamard8_diff16_ssse3;
1182  c->hadamard8_diff[1]= ff_hadamard8_diff_ssse3;
1183 #endif
1184  }
1185 #endif
1186 
1187  if(mm_flags & AV_CPU_FLAG_3DNOW){
1188  if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
1189  c->try_8x8basis= try_8x8basis_3dnow;
1190  }
1191  c->add_8x8basis= add_8x8basis_3dnow;
1192  }
1193  }
1194 
1195  dsputil_init_pix_mmx(c, avctx);
1196 }