FFmpeg
rgb2rgb_template.c
Go to the documentation of this file.
1 /*
2  * software RGB to RGB converter
3  * pluralize by software PAL8 to RGB converter
4  * software YUV to YUV converter
5  * software YUV to RGB converter
6  * Written by Nick Kurshev.
7  * palette & YUV & runtime CPU stuff by Michael (michaelni@gmx.at)
8  * lot of big-endian byte order fixes by Alex Beregszaszi
9  *
10  * This file is part of FFmpeg.
11  *
12  * FFmpeg is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU Lesser General Public
14  * License as published by the Free Software Foundation; either
15  * version 2.1 of the License, or (at your option) any later version.
16  *
17  * FFmpeg is distributed in the hope that it will be useful,
18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20  * Lesser General Public License for more details.
21  *
22  * You should have received a copy of the GNU Lesser General Public
23  * License along with FFmpeg; if not, write to the Free Software
24  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25  */
26 
27 #include <stddef.h>
28 #include <stdint.h>
29 
30 #include "libavutil/attributes.h"
31 #include "libavutil/x86/asm.h"
32 
33 #undef PREFETCH
34 #undef MOVNTQ
35 #undef EMMS
36 #undef SFENCE
37 #undef PAVGB
38 
39 #if COMPILE_TEMPLATE_AMD3DNOW
40 #define PREFETCH "prefetch"
41 #define PAVGB "pavgusb"
42 #elif COMPILE_TEMPLATE_MMXEXT
43 #define PREFETCH "prefetchnta"
44 #define PAVGB "pavgb"
45 #else
46 #define PREFETCH " # nop"
47 #endif
48 
49 #if COMPILE_TEMPLATE_AMD3DNOW
50 /* On K6 femms is faster than emms. On K7 femms is directly mapped to emms. */
51 #define EMMS "femms"
52 #else
53 #define EMMS "emms"
54 #endif
55 
56 #if COMPILE_TEMPLATE_MMXEXT
57 #define MOVNTQ "movntq"
58 #define SFENCE "sfence"
59 #else
60 #define MOVNTQ "movq"
61 #define SFENCE " # nop"
62 #endif
63 
64 #if !COMPILE_TEMPLATE_SSE2
65 
66 #if !COMPILE_TEMPLATE_AMD3DNOW
67 
68 static inline void RENAME(rgb24tobgr32)(const uint8_t *src, uint8_t *dst, int src_size)
69 {
70  uint8_t *dest = dst;
71  const uint8_t *s = src;
72  const uint8_t *end;
73  const uint8_t *mm_end;
74  end = s + src_size;
75  __asm__ volatile(PREFETCH" %0"::"m"(*s):"memory");
76  mm_end = end - 23;
77  __asm__ volatile("movq %0, %%mm7"::"m"(mask32a):"memory");
78  while (s < mm_end) {
79  __asm__ volatile(
80  PREFETCH" 32(%1) \n\t"
81  "movd (%1), %%mm0 \n\t"
82  "punpckldq 3(%1), %%mm0 \n\t"
83  "movd 6(%1), %%mm1 \n\t"
84  "punpckldq 9(%1), %%mm1 \n\t"
85  "movd 12(%1), %%mm2 \n\t"
86  "punpckldq 15(%1), %%mm2 \n\t"
87  "movd 18(%1), %%mm3 \n\t"
88  "punpckldq 21(%1), %%mm3 \n\t"
89  "por %%mm7, %%mm0 \n\t"
90  "por %%mm7, %%mm1 \n\t"
91  "por %%mm7, %%mm2 \n\t"
92  "por %%mm7, %%mm3 \n\t"
93  MOVNTQ" %%mm0, (%0) \n\t"
94  MOVNTQ" %%mm1, 8(%0) \n\t"
95  MOVNTQ" %%mm2, 16(%0) \n\t"
96  MOVNTQ" %%mm3, 24(%0)"
97  :: "r"(dest), "r"(s)
98  :"memory");
99  dest += 32;
100  s += 24;
101  }
102  __asm__ volatile(SFENCE:::"memory");
103  __asm__ volatile(EMMS:::"memory");
104  while (s < end) {
105  *dest++ = *s++;
106  *dest++ = *s++;
107  *dest++ = *s++;
108  *dest++ = 255;
109  }
110 }
111 
112 #define STORE_BGR24_MMX \
113  "psrlq $8, %%mm2 \n\t" \
114  "psrlq $8, %%mm3 \n\t" \
115  "psrlq $8, %%mm6 \n\t" \
116  "psrlq $8, %%mm7 \n\t" \
117  "pand "MANGLE(mask24l)", %%mm0\n\t" \
118  "pand "MANGLE(mask24l)", %%mm1\n\t" \
119  "pand "MANGLE(mask24l)", %%mm4\n\t" \
120  "pand "MANGLE(mask24l)", %%mm5\n\t" \
121  "pand "MANGLE(mask24h)", %%mm2\n\t" \
122  "pand "MANGLE(mask24h)", %%mm3\n\t" \
123  "pand "MANGLE(mask24h)", %%mm6\n\t" \
124  "pand "MANGLE(mask24h)", %%mm7\n\t" \
125  "por %%mm2, %%mm0 \n\t" \
126  "por %%mm3, %%mm1 \n\t" \
127  "por %%mm6, %%mm4 \n\t" \
128  "por %%mm7, %%mm5 \n\t" \
129  \
130  "movq %%mm1, %%mm2 \n\t" \
131  "movq %%mm4, %%mm3 \n\t" \
132  "psllq $48, %%mm2 \n\t" \
133  "psllq $32, %%mm3 \n\t" \
134  "por %%mm2, %%mm0 \n\t" \
135  "psrlq $16, %%mm1 \n\t" \
136  "psrlq $32, %%mm4 \n\t" \
137  "psllq $16, %%mm5 \n\t" \
138  "por %%mm3, %%mm1 \n\t" \
139  "por %%mm5, %%mm4 \n\t" \
140  \
141  MOVNTQ" %%mm0, (%0) \n\t" \
142  MOVNTQ" %%mm1, 8(%0) \n\t" \
143  MOVNTQ" %%mm4, 16(%0)"
144 
145 
146 static inline void RENAME(rgb32tobgr24)(const uint8_t *src, uint8_t *dst, int src_size)
147 {
148  uint8_t *dest = dst;
149  const uint8_t *s = src;
150  const uint8_t *end;
151  const uint8_t *mm_end;
152  end = s + src_size;
153  __asm__ volatile(PREFETCH" %0"::"m"(*s):"memory");
154  mm_end = end - 31;
155  while (s < mm_end) {
156  __asm__ volatile(
157  PREFETCH" 32(%1) \n\t"
158  "movq (%1), %%mm0 \n\t"
159  "movq 8(%1), %%mm1 \n\t"
160  "movq 16(%1), %%mm4 \n\t"
161  "movq 24(%1), %%mm5 \n\t"
162  "movq %%mm0, %%mm2 \n\t"
163  "movq %%mm1, %%mm3 \n\t"
164  "movq %%mm4, %%mm6 \n\t"
165  "movq %%mm5, %%mm7 \n\t"
166  STORE_BGR24_MMX
167  :: "r"(dest), "r"(s)
168  NAMED_CONSTRAINTS_ADD(mask24l,mask24h)
169  :"memory");
170  dest += 24;
171  s += 32;
172  }
173  __asm__ volatile(SFENCE:::"memory");
174  __asm__ volatile(EMMS:::"memory");
175  while (s < end) {
176  *dest++ = *s++;
177  *dest++ = *s++;
178  *dest++ = *s++;
179  s++;
180  }
181 }
182 
183 /*
184  original by Strepto/Astral
185  ported to gcc & bugfixed: A'rpi
186  MMXEXT, 3DNOW optimization by Nick Kurshev
187  32-bit C version, and and&add trick by Michael Niedermayer
188 */
189 static inline void RENAME(rgb15to16)(const uint8_t *src, uint8_t *dst, int src_size)
190 {
191  register const uint8_t* s=src;
192  register uint8_t* d=dst;
193  register const uint8_t *end;
194  const uint8_t *mm_end;
195  end = s + src_size;
196  __asm__ volatile(PREFETCH" %0"::"m"(*s));
197  __asm__ volatile("movq %0, %%mm4"::"m"(mask15s));
198  mm_end = end - 15;
199  while (s<mm_end) {
200  __asm__ volatile(
201  PREFETCH" 32(%1) \n\t"
202  "movq (%1), %%mm0 \n\t"
203  "movq 8(%1), %%mm2 \n\t"
204  "movq %%mm0, %%mm1 \n\t"
205  "movq %%mm2, %%mm3 \n\t"
206  "pand %%mm4, %%mm0 \n\t"
207  "pand %%mm4, %%mm2 \n\t"
208  "paddw %%mm1, %%mm0 \n\t"
209  "paddw %%mm3, %%mm2 \n\t"
210  MOVNTQ" %%mm0, (%0) \n\t"
211  MOVNTQ" %%mm2, 8(%0)"
212  :: "r"(d), "r"(s)
213  );
214  d+=16;
215  s+=16;
216  }
217  __asm__ volatile(SFENCE:::"memory");
218  __asm__ volatile(EMMS:::"memory");
219  mm_end = end - 3;
220  while (s < mm_end) {
221  register unsigned x= *((const uint32_t *)s);
222  *((uint32_t *)d) = (x&0x7FFF7FFF) + (x&0x7FE07FE0);
223  d+=4;
224  s+=4;
225  }
226  if (s < end) {
227  register unsigned short x= *((const uint16_t *)s);
228  *((uint16_t *)d) = (x&0x7FFF) + (x&0x7FE0);
229  }
230 }
231 
232 static inline void RENAME(rgb16to15)(const uint8_t *src, uint8_t *dst, int src_size)
233 {
234  register const uint8_t* s=src;
235  register uint8_t* d=dst;
236  register const uint8_t *end;
237  const uint8_t *mm_end;
238  end = s + src_size;
239  __asm__ volatile(PREFETCH" %0"::"m"(*s));
240  __asm__ volatile("movq %0, %%mm7"::"m"(mask15rg));
241  __asm__ volatile("movq %0, %%mm6"::"m"(mask15b));
242  mm_end = end - 15;
243  while (s<mm_end) {
244  __asm__ volatile(
245  PREFETCH" 32(%1) \n\t"
246  "movq (%1), %%mm0 \n\t"
247  "movq 8(%1), %%mm2 \n\t"
248  "movq %%mm0, %%mm1 \n\t"
249  "movq %%mm2, %%mm3 \n\t"
250  "psrlq $1, %%mm0 \n\t"
251  "psrlq $1, %%mm2 \n\t"
252  "pand %%mm7, %%mm0 \n\t"
253  "pand %%mm7, %%mm2 \n\t"
254  "pand %%mm6, %%mm1 \n\t"
255  "pand %%mm6, %%mm3 \n\t"
256  "por %%mm1, %%mm0 \n\t"
257  "por %%mm3, %%mm2 \n\t"
258  MOVNTQ" %%mm0, (%0) \n\t"
259  MOVNTQ" %%mm2, 8(%0)"
260  :: "r"(d), "r"(s)
261  );
262  d+=16;
263  s+=16;
264  }
265  __asm__ volatile(SFENCE:::"memory");
266  __asm__ volatile(EMMS:::"memory");
267  mm_end = end - 3;
268  while (s < mm_end) {
269  register uint32_t x= *((const uint32_t*)s);
270  *((uint32_t *)d) = ((x>>1)&0x7FE07FE0) | (x&0x001F001F);
271  s+=4;
272  d+=4;
273  }
274  if (s < end) {
275  register uint16_t x= *((const uint16_t*)s);
276  *((uint16_t *)d) = ((x>>1)&0x7FE0) | (x&0x001F);
277  }
278 }
279 
280 static inline void RENAME(rgb32to16)(const uint8_t *src, uint8_t *dst, int src_size)
281 {
282  const uint8_t *s = src;
283  const uint8_t *end;
284  const uint8_t *mm_end;
285  uint16_t *d = (uint16_t *)dst;
286  end = s + src_size;
287  mm_end = end - 15;
288  __asm__ volatile(
289  "movq %3, %%mm5 \n\t"
290  "movq %4, %%mm6 \n\t"
291  "movq %5, %%mm7 \n\t"
292  "jmp 2f \n\t"
293  ".p2align 4 \n\t"
294  "1: \n\t"
295  PREFETCH" 32(%1) \n\t"
296  "movd (%1), %%mm0 \n\t"
297  "movd 4(%1), %%mm3 \n\t"
298  "punpckldq 8(%1), %%mm0 \n\t"
299  "punpckldq 12(%1), %%mm3 \n\t"
300  "movq %%mm0, %%mm1 \n\t"
301  "movq %%mm3, %%mm4 \n\t"
302  "pand %%mm6, %%mm0 \n\t"
303  "pand %%mm6, %%mm3 \n\t"
304  "pmaddwd %%mm7, %%mm0 \n\t"
305  "pmaddwd %%mm7, %%mm3 \n\t"
306  "pand %%mm5, %%mm1 \n\t"
307  "pand %%mm5, %%mm4 \n\t"
308  "por %%mm1, %%mm0 \n\t"
309  "por %%mm4, %%mm3 \n\t"
310  "psrld $5, %%mm0 \n\t"
311  "pslld $11, %%mm3 \n\t"
312  "por %%mm3, %%mm0 \n\t"
313  MOVNTQ" %%mm0, (%0) \n\t"
314  "add $16, %1 \n\t"
315  "add $8, %0 \n\t"
316  "2: \n\t"
317  "cmp %2, %1 \n\t"
318  " jb 1b \n\t"
319  : "+r" (d), "+r"(s)
320  : "r" (mm_end), "m" (mask3216g), "m" (mask3216br), "m" (mul3216)
321  );
322  __asm__ volatile(SFENCE:::"memory");
323  __asm__ volatile(EMMS:::"memory");
324  while (s < end) {
325  register int rgb = *(const uint32_t*)s; s += 4;
326  *d++ = ((rgb&0xFF)>>3) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>8);
327  }
328 }
329 
330 static inline void RENAME(rgb32tobgr16)(const uint8_t *src, uint8_t *dst, int src_size)
331 {
332  const uint8_t *s = src;
333  const uint8_t *end;
334  const uint8_t *mm_end;
335  uint16_t *d = (uint16_t *)dst;
336  end = s + src_size;
337  __asm__ volatile(PREFETCH" %0"::"m"(*src):"memory");
338  __asm__ volatile(
339  "movq %0, %%mm7 \n\t"
340  "movq %1, %%mm6 \n\t"
341  ::"m"(red_16mask),"m"(green_16mask));
342  mm_end = end - 15;
343  while (s < mm_end) {
344  __asm__ volatile(
345  PREFETCH" 32(%1) \n\t"
346  "movd (%1), %%mm0 \n\t"
347  "movd 4(%1), %%mm3 \n\t"
348  "punpckldq 8(%1), %%mm0 \n\t"
349  "punpckldq 12(%1), %%mm3 \n\t"
350  "movq %%mm0, %%mm1 \n\t"
351  "movq %%mm0, %%mm2 \n\t"
352  "movq %%mm3, %%mm4 \n\t"
353  "movq %%mm3, %%mm5 \n\t"
354  "psllq $8, %%mm0 \n\t"
355  "psllq $8, %%mm3 \n\t"
356  "pand %%mm7, %%mm0 \n\t"
357  "pand %%mm7, %%mm3 \n\t"
358  "psrlq $5, %%mm1 \n\t"
359  "psrlq $5, %%mm4 \n\t"
360  "pand %%mm6, %%mm1 \n\t"
361  "pand %%mm6, %%mm4 \n\t"
362  "psrlq $19, %%mm2 \n\t"
363  "psrlq $19, %%mm5 \n\t"
364  "pand %2, %%mm2 \n\t"
365  "pand %2, %%mm5 \n\t"
366  "por %%mm1, %%mm0 \n\t"
367  "por %%mm4, %%mm3 \n\t"
368  "por %%mm2, %%mm0 \n\t"
369  "por %%mm5, %%mm3 \n\t"
370  "psllq $16, %%mm3 \n\t"
371  "por %%mm3, %%mm0 \n\t"
372  MOVNTQ" %%mm0, (%0) \n\t"
373  :: "r"(d),"r"(s),"m"(blue_16mask):"memory");
374  d += 4;
375  s += 16;
376  }
377  __asm__ volatile(SFENCE:::"memory");
378  __asm__ volatile(EMMS:::"memory");
379  while (s < end) {
380  register int rgb = *(const uint32_t*)s; s += 4;
381  *d++ = ((rgb&0xF8)<<8) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>19);
382  }
383 }
384 
385 static inline void RENAME(rgb32to15)(const uint8_t *src, uint8_t *dst, int src_size)
386 {
387  const uint8_t *s = src;
388  const uint8_t *end;
389  const uint8_t *mm_end;
390  uint16_t *d = (uint16_t *)dst;
391  end = s + src_size;
392  mm_end = end - 15;
393  __asm__ volatile(
394  "movq %3, %%mm5 \n\t"
395  "movq %4, %%mm6 \n\t"
396  "movq %5, %%mm7 \n\t"
397  "jmp 2f \n\t"
398  ".p2align 4 \n\t"
399  "1: \n\t"
400  PREFETCH" 32(%1) \n\t"
401  "movd (%1), %%mm0 \n\t"
402  "movd 4(%1), %%mm3 \n\t"
403  "punpckldq 8(%1), %%mm0 \n\t"
404  "punpckldq 12(%1), %%mm3 \n\t"
405  "movq %%mm0, %%mm1 \n\t"
406  "movq %%mm3, %%mm4 \n\t"
407  "pand %%mm6, %%mm0 \n\t"
408  "pand %%mm6, %%mm3 \n\t"
409  "pmaddwd %%mm7, %%mm0 \n\t"
410  "pmaddwd %%mm7, %%mm3 \n\t"
411  "pand %%mm5, %%mm1 \n\t"
412  "pand %%mm5, %%mm4 \n\t"
413  "por %%mm1, %%mm0 \n\t"
414  "por %%mm4, %%mm3 \n\t"
415  "psrld $6, %%mm0 \n\t"
416  "pslld $10, %%mm3 \n\t"
417  "por %%mm3, %%mm0 \n\t"
418  MOVNTQ" %%mm0, (%0) \n\t"
419  "add $16, %1 \n\t"
420  "add $8, %0 \n\t"
421  "2: \n\t"
422  "cmp %2, %1 \n\t"
423  " jb 1b \n\t"
424  : "+r" (d), "+r"(s)
425  : "r" (mm_end), "m" (mask3215g), "m" (mask3216br), "m" (mul3215)
426  );
427  __asm__ volatile(SFENCE:::"memory");
428  __asm__ volatile(EMMS:::"memory");
429  while (s < end) {
430  register int rgb = *(const uint32_t*)s; s += 4;
431  *d++ = ((rgb&0xFF)>>3) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>9);
432  }
433 }
434 
435 static inline void RENAME(rgb32tobgr15)(const uint8_t *src, uint8_t *dst, int src_size)
436 {
437  const uint8_t *s = src;
438  const uint8_t *end;
439  const uint8_t *mm_end;
440  uint16_t *d = (uint16_t *)dst;
441  end = s + src_size;
442  __asm__ volatile(PREFETCH" %0"::"m"(*src):"memory");
443  __asm__ volatile(
444  "movq %0, %%mm7 \n\t"
445  "movq %1, %%mm6 \n\t"
446  ::"m"(red_15mask),"m"(green_15mask));
447  mm_end = end - 15;
448  while (s < mm_end) {
449  __asm__ volatile(
450  PREFETCH" 32(%1) \n\t"
451  "movd (%1), %%mm0 \n\t"
452  "movd 4(%1), %%mm3 \n\t"
453  "punpckldq 8(%1), %%mm0 \n\t"
454  "punpckldq 12(%1), %%mm3 \n\t"
455  "movq %%mm0, %%mm1 \n\t"
456  "movq %%mm0, %%mm2 \n\t"
457  "movq %%mm3, %%mm4 \n\t"
458  "movq %%mm3, %%mm5 \n\t"
459  "psllq $7, %%mm0 \n\t"
460  "psllq $7, %%mm3 \n\t"
461  "pand %%mm7, %%mm0 \n\t"
462  "pand %%mm7, %%mm3 \n\t"
463  "psrlq $6, %%mm1 \n\t"
464  "psrlq $6, %%mm4 \n\t"
465  "pand %%mm6, %%mm1 \n\t"
466  "pand %%mm6, %%mm4 \n\t"
467  "psrlq $19, %%mm2 \n\t"
468  "psrlq $19, %%mm5 \n\t"
469  "pand %2, %%mm2 \n\t"
470  "pand %2, %%mm5 \n\t"
471  "por %%mm1, %%mm0 \n\t"
472  "por %%mm4, %%mm3 \n\t"
473  "por %%mm2, %%mm0 \n\t"
474  "por %%mm5, %%mm3 \n\t"
475  "psllq $16, %%mm3 \n\t"
476  "por %%mm3, %%mm0 \n\t"
477  MOVNTQ" %%mm0, (%0) \n\t"
478  ::"r"(d),"r"(s),"m"(blue_15mask):"memory");
479  d += 4;
480  s += 16;
481  }
482  __asm__ volatile(SFENCE:::"memory");
483  __asm__ volatile(EMMS:::"memory");
484  while (s < end) {
485  register int rgb = *(const uint32_t*)s; s += 4;
486  *d++ = ((rgb&0xF8)<<7) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>19);
487  }
488 }
489 
490 static inline void RENAME(rgb24tobgr16)(const uint8_t *src, uint8_t *dst, int src_size)
491 {
492  const uint8_t *s = src;
493  const uint8_t *end;
494  const uint8_t *mm_end;
495  uint16_t *d = (uint16_t *)dst;
496  end = s + src_size;
497  __asm__ volatile(PREFETCH" %0"::"m"(*src):"memory");
498  __asm__ volatile(
499  "movq %0, %%mm7 \n\t"
500  "movq %1, %%mm6 \n\t"
501  ::"m"(red_16mask),"m"(green_16mask));
502  mm_end = end - 11;
503  while (s < mm_end) {
504  __asm__ volatile(
505  PREFETCH" 32(%1) \n\t"
506  "movd (%1), %%mm0 \n\t"
507  "movd 3(%1), %%mm3 \n\t"
508  "punpckldq 6(%1), %%mm0 \n\t"
509  "punpckldq 9(%1), %%mm3 \n\t"
510  "movq %%mm0, %%mm1 \n\t"
511  "movq %%mm0, %%mm2 \n\t"
512  "movq %%mm3, %%mm4 \n\t"
513  "movq %%mm3, %%mm5 \n\t"
514  "psrlq $3, %%mm0 \n\t"
515  "psrlq $3, %%mm3 \n\t"
516  "pand %2, %%mm0 \n\t"
517  "pand %2, %%mm3 \n\t"
518  "psrlq $5, %%mm1 \n\t"
519  "psrlq $5, %%mm4 \n\t"
520  "pand %%mm6, %%mm1 \n\t"
521  "pand %%mm6, %%mm4 \n\t"
522  "psrlq $8, %%mm2 \n\t"
523  "psrlq $8, %%mm5 \n\t"
524  "pand %%mm7, %%mm2 \n\t"
525  "pand %%mm7, %%mm5 \n\t"
526  "por %%mm1, %%mm0 \n\t"
527  "por %%mm4, %%mm3 \n\t"
528  "por %%mm2, %%mm0 \n\t"
529  "por %%mm5, %%mm3 \n\t"
530  "psllq $16, %%mm3 \n\t"
531  "por %%mm3, %%mm0 \n\t"
532  MOVNTQ" %%mm0, (%0) \n\t"
533  ::"r"(d),"r"(s),"m"(blue_16mask):"memory");
534  d += 4;
535  s += 12;
536  }
537  __asm__ volatile(SFENCE:::"memory");
538  __asm__ volatile(EMMS:::"memory");
539  while (s < end) {
540  const int b = *s++;
541  const int g = *s++;
542  const int r = *s++;
543  *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
544  }
545 }
546 
547 static inline void RENAME(rgb24to16)(const uint8_t *src, uint8_t *dst, int src_size)
548 {
549  const uint8_t *s = src;
550  const uint8_t *end;
551  const uint8_t *mm_end;
552  uint16_t *d = (uint16_t *)dst;
553  end = s + src_size;
554  __asm__ volatile(PREFETCH" %0"::"m"(*src):"memory");
555  __asm__ volatile(
556  "movq %0, %%mm7 \n\t"
557  "movq %1, %%mm6 \n\t"
558  ::"m"(red_16mask),"m"(green_16mask));
559  mm_end = end - 15;
560  while (s < mm_end) {
561  __asm__ volatile(
562  PREFETCH" 32(%1) \n\t"
563  "movd (%1), %%mm0 \n\t"
564  "movd 3(%1), %%mm3 \n\t"
565  "punpckldq 6(%1), %%mm0 \n\t"
566  "punpckldq 9(%1), %%mm3 \n\t"
567  "movq %%mm0, %%mm1 \n\t"
568  "movq %%mm0, %%mm2 \n\t"
569  "movq %%mm3, %%mm4 \n\t"
570  "movq %%mm3, %%mm5 \n\t"
571  "psllq $8, %%mm0 \n\t"
572  "psllq $8, %%mm3 \n\t"
573  "pand %%mm7, %%mm0 \n\t"
574  "pand %%mm7, %%mm3 \n\t"
575  "psrlq $5, %%mm1 \n\t"
576  "psrlq $5, %%mm4 \n\t"
577  "pand %%mm6, %%mm1 \n\t"
578  "pand %%mm6, %%mm4 \n\t"
579  "psrlq $19, %%mm2 \n\t"
580  "psrlq $19, %%mm5 \n\t"
581  "pand %2, %%mm2 \n\t"
582  "pand %2, %%mm5 \n\t"
583  "por %%mm1, %%mm0 \n\t"
584  "por %%mm4, %%mm3 \n\t"
585  "por %%mm2, %%mm0 \n\t"
586  "por %%mm5, %%mm3 \n\t"
587  "psllq $16, %%mm3 \n\t"
588  "por %%mm3, %%mm0 \n\t"
589  MOVNTQ" %%mm0, (%0) \n\t"
590  ::"r"(d),"r"(s),"m"(blue_16mask):"memory");
591  d += 4;
592  s += 12;
593  }
594  __asm__ volatile(SFENCE:::"memory");
595  __asm__ volatile(EMMS:::"memory");
596  while (s < end) {
597  const int r = *s++;
598  const int g = *s++;
599  const int b = *s++;
600  *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
601  }
602 }
603 
604 static inline void RENAME(rgb24tobgr15)(const uint8_t *src, uint8_t *dst, int src_size)
605 {
606  const uint8_t *s = src;
607  const uint8_t *end;
608  const uint8_t *mm_end;
609  uint16_t *d = (uint16_t *)dst;
610  end = s + src_size;
611  __asm__ volatile(PREFETCH" %0"::"m"(*src):"memory");
612  __asm__ volatile(
613  "movq %0, %%mm7 \n\t"
614  "movq %1, %%mm6 \n\t"
615  ::"m"(red_15mask),"m"(green_15mask));
616  mm_end = end - 11;
617  while (s < mm_end) {
618  __asm__ volatile(
619  PREFETCH" 32(%1) \n\t"
620  "movd (%1), %%mm0 \n\t"
621  "movd 3(%1), %%mm3 \n\t"
622  "punpckldq 6(%1), %%mm0 \n\t"
623  "punpckldq 9(%1), %%mm3 \n\t"
624  "movq %%mm0, %%mm1 \n\t"
625  "movq %%mm0, %%mm2 \n\t"
626  "movq %%mm3, %%mm4 \n\t"
627  "movq %%mm3, %%mm5 \n\t"
628  "psrlq $3, %%mm0 \n\t"
629  "psrlq $3, %%mm3 \n\t"
630  "pand %2, %%mm0 \n\t"
631  "pand %2, %%mm3 \n\t"
632  "psrlq $6, %%mm1 \n\t"
633  "psrlq $6, %%mm4 \n\t"
634  "pand %%mm6, %%mm1 \n\t"
635  "pand %%mm6, %%mm4 \n\t"
636  "psrlq $9, %%mm2 \n\t"
637  "psrlq $9, %%mm5 \n\t"
638  "pand %%mm7, %%mm2 \n\t"
639  "pand %%mm7, %%mm5 \n\t"
640  "por %%mm1, %%mm0 \n\t"
641  "por %%mm4, %%mm3 \n\t"
642  "por %%mm2, %%mm0 \n\t"
643  "por %%mm5, %%mm3 \n\t"
644  "psllq $16, %%mm3 \n\t"
645  "por %%mm3, %%mm0 \n\t"
646  MOVNTQ" %%mm0, (%0) \n\t"
647  ::"r"(d),"r"(s),"m"(blue_15mask):"memory");
648  d += 4;
649  s += 12;
650  }
651  __asm__ volatile(SFENCE:::"memory");
652  __asm__ volatile(EMMS:::"memory");
653  while (s < end) {
654  const int b = *s++;
655  const int g = *s++;
656  const int r = *s++;
657  *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
658  }
659 }
660 
661 static inline void RENAME(rgb24to15)(const uint8_t *src, uint8_t *dst, int src_size)
662 {
663  const uint8_t *s = src;
664  const uint8_t *end;
665  const uint8_t *mm_end;
666  uint16_t *d = (uint16_t *)dst;
667  end = s + src_size;
668  __asm__ volatile(PREFETCH" %0"::"m"(*src):"memory");
669  __asm__ volatile(
670  "movq %0, %%mm7 \n\t"
671  "movq %1, %%mm6 \n\t"
672  ::"m"(red_15mask),"m"(green_15mask));
673  mm_end = end - 15;
674  while (s < mm_end) {
675  __asm__ volatile(
676  PREFETCH" 32(%1) \n\t"
677  "movd (%1), %%mm0 \n\t"
678  "movd 3(%1), %%mm3 \n\t"
679  "punpckldq 6(%1), %%mm0 \n\t"
680  "punpckldq 9(%1), %%mm3 \n\t"
681  "movq %%mm0, %%mm1 \n\t"
682  "movq %%mm0, %%mm2 \n\t"
683  "movq %%mm3, %%mm4 \n\t"
684  "movq %%mm3, %%mm5 \n\t"
685  "psllq $7, %%mm0 \n\t"
686  "psllq $7, %%mm3 \n\t"
687  "pand %%mm7, %%mm0 \n\t"
688  "pand %%mm7, %%mm3 \n\t"
689  "psrlq $6, %%mm1 \n\t"
690  "psrlq $6, %%mm4 \n\t"
691  "pand %%mm6, %%mm1 \n\t"
692  "pand %%mm6, %%mm4 \n\t"
693  "psrlq $19, %%mm2 \n\t"
694  "psrlq $19, %%mm5 \n\t"
695  "pand %2, %%mm2 \n\t"
696  "pand %2, %%mm5 \n\t"
697  "por %%mm1, %%mm0 \n\t"
698  "por %%mm4, %%mm3 \n\t"
699  "por %%mm2, %%mm0 \n\t"
700  "por %%mm5, %%mm3 \n\t"
701  "psllq $16, %%mm3 \n\t"
702  "por %%mm3, %%mm0 \n\t"
703  MOVNTQ" %%mm0, (%0) \n\t"
704  ::"r"(d),"r"(s),"m"(blue_15mask):"memory");
705  d += 4;
706  s += 12;
707  }
708  __asm__ volatile(SFENCE:::"memory");
709  __asm__ volatile(EMMS:::"memory");
710  while (s < end) {
711  const int r = *s++;
712  const int g = *s++;
713  const int b = *s++;
714  *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
715  }
716 }
717 
718 static inline void RENAME(rgb15tobgr24)(const uint8_t *src, uint8_t *dst, int src_size)
719 {
720  const uint16_t *end;
721  const uint16_t *mm_end;
722  uint8_t *d = dst;
723  const uint16_t *s = (const uint16_t*)src;
724  end = s + src_size/2;
725  __asm__ volatile(PREFETCH" %0"::"m"(*s):"memory");
726  mm_end = end - 7;
727  while (s < mm_end) {
728  __asm__ volatile(
729  PREFETCH" 32(%1) \n\t"
730  "movq (%1), %%mm0 \n\t"
731  "movq (%1), %%mm1 \n\t"
732  "movq (%1), %%mm2 \n\t"
733  "pand %2, %%mm0 \n\t"
734  "pand %3, %%mm1 \n\t"
735  "pand %4, %%mm2 \n\t"
736  "psllq $5, %%mm0 \n\t"
737  "pmulhw "MANGLE(mul15_mid)", %%mm0 \n\t"
738  "pmulhw "MANGLE(mul15_mid)", %%mm1 \n\t"
739  "pmulhw "MANGLE(mul15_hi)", %%mm2 \n\t"
740  "movq %%mm0, %%mm3 \n\t"
741  "movq %%mm1, %%mm4 \n\t"
742  "movq %%mm2, %%mm5 \n\t"
743  "punpcklwd %5, %%mm0 \n\t"
744  "punpcklwd %5, %%mm1 \n\t"
745  "punpcklwd %5, %%mm2 \n\t"
746  "punpckhwd %5, %%mm3 \n\t"
747  "punpckhwd %5, %%mm4 \n\t"
748  "punpckhwd %5, %%mm5 \n\t"
749  "psllq $8, %%mm1 \n\t"
750  "psllq $16, %%mm2 \n\t"
751  "por %%mm1, %%mm0 \n\t"
752  "por %%mm2, %%mm0 \n\t"
753  "psllq $8, %%mm4 \n\t"
754  "psllq $16, %%mm5 \n\t"
755  "por %%mm4, %%mm3 \n\t"
756  "por %%mm5, %%mm3 \n\t"
757 
758  "movq %%mm0, %%mm6 \n\t"
759  "movq %%mm3, %%mm7 \n\t"
760 
761  "movq 8(%1), %%mm0 \n\t"
762  "movq 8(%1), %%mm1 \n\t"
763  "movq 8(%1), %%mm2 \n\t"
764  "pand %2, %%mm0 \n\t"
765  "pand %3, %%mm1 \n\t"
766  "pand %4, %%mm2 \n\t"
767  "psllq $5, %%mm0 \n\t"
768  "pmulhw "MANGLE(mul15_mid)", %%mm0 \n\t"
769  "pmulhw "MANGLE(mul15_mid)", %%mm1 \n\t"
770  "pmulhw "MANGLE(mul15_hi)", %%mm2 \n\t"
771  "movq %%mm0, %%mm3 \n\t"
772  "movq %%mm1, %%mm4 \n\t"
773  "movq %%mm2, %%mm5 \n\t"
774  "punpcklwd %5, %%mm0 \n\t"
775  "punpcklwd %5, %%mm1 \n\t"
776  "punpcklwd %5, %%mm2 \n\t"
777  "punpckhwd %5, %%mm3 \n\t"
778  "punpckhwd %5, %%mm4 \n\t"
779  "punpckhwd %5, %%mm5 \n\t"
780  "psllq $8, %%mm1 \n\t"
781  "psllq $16, %%mm2 \n\t"
782  "por %%mm1, %%mm0 \n\t"
783  "por %%mm2, %%mm0 \n\t"
784  "psllq $8, %%mm4 \n\t"
785  "psllq $16, %%mm5 \n\t"
786  "por %%mm4, %%mm3 \n\t"
787  "por %%mm5, %%mm3 \n\t"
788 
789  :"=m"(*d)
790  :"r"(s),"m"(mask15b),"m"(mask15g),"m"(mask15r), "m"(mmx_null)
791  NAMED_CONSTRAINTS_ADD(mul15_mid,mul15_hi)
792  :"memory");
793  /* borrowed 32 to 24 */
794  __asm__ volatile(
795  "movq %%mm0, %%mm4 \n\t"
796  "movq %%mm3, %%mm5 \n\t"
797  "movq %%mm6, %%mm0 \n\t"
798  "movq %%mm7, %%mm1 \n\t"
799 
800  "movq %%mm4, %%mm6 \n\t"
801  "movq %%mm5, %%mm7 \n\t"
802  "movq %%mm0, %%mm2 \n\t"
803  "movq %%mm1, %%mm3 \n\t"
804 
805  STORE_BGR24_MMX
806 
807  :: "r"(d), "m"(*s)
808  NAMED_CONSTRAINTS_ADD(mask24l,mask24h)
809  :"memory");
810  d += 24;
811  s += 8;
812  }
813  __asm__ volatile(SFENCE:::"memory");
814  __asm__ volatile(EMMS:::"memory");
815  while (s < end) {
816  register uint16_t bgr;
817  bgr = *s++;
818  *d++ = ((bgr&0x1F)<<3) | ((bgr&0x1F)>>2);
819  *d++ = ((bgr&0x3E0)>>2) | ((bgr&0x3E0)>>7);
820  *d++ = ((bgr&0x7C00)>>7) | ((bgr&0x7C00)>>12);
821  }
822 }
823 
824 static inline void RENAME(rgb16tobgr24)(const uint8_t *src, uint8_t *dst, int src_size)
825 {
826  const uint16_t *end;
827  const uint16_t *mm_end;
828  uint8_t *d = (uint8_t *)dst;
829  const uint16_t *s = (const uint16_t *)src;
830  end = s + src_size/2;
831  __asm__ volatile(PREFETCH" %0"::"m"(*s):"memory");
832  mm_end = end - 7;
833  while (s < mm_end) {
834  __asm__ volatile(
835  PREFETCH" 32(%1) \n\t"
836  "movq (%1), %%mm0 \n\t"
837  "movq (%1), %%mm1 \n\t"
838  "movq (%1), %%mm2 \n\t"
839  "pand %2, %%mm0 \n\t"
840  "pand %3, %%mm1 \n\t"
841  "pand %4, %%mm2 \n\t"
842  "psllq $5, %%mm0 \n\t"
843  "psrlq $1, %%mm2 \n\t"
844  "pmulhw "MANGLE(mul15_mid)", %%mm0 \n\t"
845  "pmulhw "MANGLE(mul16_mid)", %%mm1 \n\t"
846  "pmulhw "MANGLE(mul15_hi)", %%mm2 \n\t"
847  "movq %%mm0, %%mm3 \n\t"
848  "movq %%mm1, %%mm4 \n\t"
849  "movq %%mm2, %%mm5 \n\t"
850  "punpcklwd %5, %%mm0 \n\t"
851  "punpcklwd %5, %%mm1 \n\t"
852  "punpcklwd %5, %%mm2 \n\t"
853  "punpckhwd %5, %%mm3 \n\t"
854  "punpckhwd %5, %%mm4 \n\t"
855  "punpckhwd %5, %%mm5 \n\t"
856  "psllq $8, %%mm1 \n\t"
857  "psllq $16, %%mm2 \n\t"
858  "por %%mm1, %%mm0 \n\t"
859  "por %%mm2, %%mm0 \n\t"
860  "psllq $8, %%mm4 \n\t"
861  "psllq $16, %%mm5 \n\t"
862  "por %%mm4, %%mm3 \n\t"
863  "por %%mm5, %%mm3 \n\t"
864 
865  "movq %%mm0, %%mm6 \n\t"
866  "movq %%mm3, %%mm7 \n\t"
867 
868  "movq 8(%1), %%mm0 \n\t"
869  "movq 8(%1), %%mm1 \n\t"
870  "movq 8(%1), %%mm2 \n\t"
871  "pand %2, %%mm0 \n\t"
872  "pand %3, %%mm1 \n\t"
873  "pand %4, %%mm2 \n\t"
874  "psllq $5, %%mm0 \n\t"
875  "psrlq $1, %%mm2 \n\t"
876  "pmulhw "MANGLE(mul15_mid)", %%mm0 \n\t"
877  "pmulhw "MANGLE(mul16_mid)", %%mm1 \n\t"
878  "pmulhw "MANGLE(mul15_hi)", %%mm2 \n\t"
879  "movq %%mm0, %%mm3 \n\t"
880  "movq %%mm1, %%mm4 \n\t"
881  "movq %%mm2, %%mm5 \n\t"
882  "punpcklwd %5, %%mm0 \n\t"
883  "punpcklwd %5, %%mm1 \n\t"
884  "punpcklwd %5, %%mm2 \n\t"
885  "punpckhwd %5, %%mm3 \n\t"
886  "punpckhwd %5, %%mm4 \n\t"
887  "punpckhwd %5, %%mm5 \n\t"
888  "psllq $8, %%mm1 \n\t"
889  "psllq $16, %%mm2 \n\t"
890  "por %%mm1, %%mm0 \n\t"
891  "por %%mm2, %%mm0 \n\t"
892  "psllq $8, %%mm4 \n\t"
893  "psllq $16, %%mm5 \n\t"
894  "por %%mm4, %%mm3 \n\t"
895  "por %%mm5, %%mm3 \n\t"
896  :"=m"(*d)
897  :"r"(s),"m"(mask16b),"m"(mask16g),"m"(mask16r),"m"(mmx_null)
898  NAMED_CONSTRAINTS_ADD(mul15_mid,mul16_mid,mul15_hi)
899  :"memory");
900  /* borrowed 32 to 24 */
901  __asm__ volatile(
902  "movq %%mm0, %%mm4 \n\t"
903  "movq %%mm3, %%mm5 \n\t"
904  "movq %%mm6, %%mm0 \n\t"
905  "movq %%mm7, %%mm1 \n\t"
906 
907  "movq %%mm4, %%mm6 \n\t"
908  "movq %%mm5, %%mm7 \n\t"
909  "movq %%mm0, %%mm2 \n\t"
910  "movq %%mm1, %%mm3 \n\t"
911 
912  STORE_BGR24_MMX
913 
914  :: "r"(d), "m"(*s)
915  NAMED_CONSTRAINTS_ADD(mask24l,mask24h)
916  :"memory");
917  d += 24;
918  s += 8;
919  }
920  __asm__ volatile(SFENCE:::"memory");
921  __asm__ volatile(EMMS:::"memory");
922  while (s < end) {
923  register uint16_t bgr;
924  bgr = *s++;
925  *d++ = ((bgr&0x1F)<<3) | ((bgr&0x1F)>>2);
926  *d++ = ((bgr&0x7E0)>>3) | ((bgr&0x7E0)>>9);
927  *d++ = ((bgr&0xF800)>>8) | ((bgr&0xF800)>>13);
928  }
929 }
930 
931 /*
932  * mm0 = 00 B3 00 B2 00 B1 00 B0
933  * mm1 = 00 G3 00 G2 00 G1 00 G0
934  * mm2 = 00 R3 00 R2 00 R1 00 R0
935  * mm6 = FF FF FF FF FF FF FF FF
936  * mm7 = 00 00 00 00 00 00 00 00
937  */
938 #define PACK_RGB32 \
939  "packuswb %%mm7, %%mm0 \n\t" /* 00 00 00 00 B3 B2 B1 B0 */ \
940  "packuswb %%mm7, %%mm1 \n\t" /* 00 00 00 00 G3 G2 G1 G0 */ \
941  "packuswb %%mm7, %%mm2 \n\t" /* 00 00 00 00 R3 R2 R1 R0 */ \
942  "punpcklbw %%mm1, %%mm0 \n\t" /* G3 B3 G2 B2 G1 B1 G0 B0 */ \
943  "punpcklbw %%mm6, %%mm2 \n\t" /* FF R3 FF R2 FF R1 FF R0 */ \
944  "movq %%mm0, %%mm3 \n\t" \
945  "punpcklwd %%mm2, %%mm0 \n\t" /* FF R1 G1 B1 FF R0 G0 B0 */ \
946  "punpckhwd %%mm2, %%mm3 \n\t" /* FF R3 G3 B3 FF R2 G2 B2 */ \
947  MOVNTQ" %%mm0, (%0) \n\t" \
948  MOVNTQ" %%mm3, 8(%0) \n\t" \
949 
950 static inline void RENAME(rgb15to32)(const uint8_t *src, uint8_t *dst, int src_size)
951 {
952  const uint16_t *end;
953  const uint16_t *mm_end;
954  uint8_t *d = dst;
955  const uint16_t *s = (const uint16_t *)src;
956  end = s + src_size/2;
957  __asm__ volatile(PREFETCH" %0"::"m"(*s):"memory");
958  __asm__ volatile("pxor %%mm7,%%mm7 \n\t":::"memory");
959  __asm__ volatile("pcmpeqd %%mm6,%%mm6 \n\t":::"memory");
960  mm_end = end - 3;
961  while (s < mm_end) {
962  __asm__ volatile(
963  PREFETCH" 32(%1) \n\t"
964  "movq (%1), %%mm0 \n\t"
965  "movq (%1), %%mm1 \n\t"
966  "movq (%1), %%mm2 \n\t"
967  "pand %2, %%mm0 \n\t"
968  "pand %3, %%mm1 \n\t"
969  "pand %4, %%mm2 \n\t"
970  "psllq $5, %%mm0 \n\t"
971  "pmulhw %5, %%mm0 \n\t"
972  "pmulhw %5, %%mm1 \n\t"
973  "pmulhw "MANGLE(mul15_hi)", %%mm2 \n\t"
974  PACK_RGB32
975  ::"r"(d),"r"(s),"m"(mask15b),"m"(mask15g),"m"(mask15r) ,"m"(mul15_mid)
976  NAMED_CONSTRAINTS_ADD(mul15_hi)
977  :"memory");
978  d += 16;
979  s += 4;
980  }
981  __asm__ volatile(SFENCE:::"memory");
982  __asm__ volatile(EMMS:::"memory");
983  while (s < end) {
984  register uint16_t bgr;
985  bgr = *s++;
986  *d++ = ((bgr&0x1F)<<3) | ((bgr&0x1F)>>2);
987  *d++ = ((bgr&0x3E0)>>2) | ((bgr&0x3E0)>>7);
988  *d++ = ((bgr&0x7C00)>>7) | ((bgr&0x7C00)>>12);
989  *d++ = 255;
990  }
991 }
992 
993 static inline void RENAME(rgb16to32)(const uint8_t *src, uint8_t *dst, int src_size)
994 {
995  const uint16_t *end;
996  const uint16_t *mm_end;
997  uint8_t *d = dst;
998  const uint16_t *s = (const uint16_t*)src;
999  end = s + src_size/2;
1000  __asm__ volatile(PREFETCH" %0"::"m"(*s):"memory");
1001  __asm__ volatile("pxor %%mm7,%%mm7 \n\t":::"memory");
1002  __asm__ volatile("pcmpeqd %%mm6,%%mm6 \n\t":::"memory");
1003  mm_end = end - 3;
1004  while (s < mm_end) {
1005  __asm__ volatile(
1006  PREFETCH" 32(%1) \n\t"
1007  "movq (%1), %%mm0 \n\t"
1008  "movq (%1), %%mm1 \n\t"
1009  "movq (%1), %%mm2 \n\t"
1010  "pand %2, %%mm0 \n\t"
1011  "pand %3, %%mm1 \n\t"
1012  "pand %4, %%mm2 \n\t"
1013  "psllq $5, %%mm0 \n\t"
1014  "psrlq $1, %%mm2 \n\t"
1015  "pmulhw %5, %%mm0 \n\t"
1016  "pmulhw "MANGLE(mul16_mid)", %%mm1 \n\t"
1017  "pmulhw "MANGLE(mul15_hi)", %%mm2 \n\t"
1018  PACK_RGB32
1019  ::"r"(d),"r"(s),"m"(mask16b),"m"(mask16g),"m"(mask16r),"m"(mul15_mid)
1020  NAMED_CONSTRAINTS_ADD(mul16_mid,mul15_hi)
1021  :"memory");
1022  d += 16;
1023  s += 4;
1024  }
1025  __asm__ volatile(SFENCE:::"memory");
1026  __asm__ volatile(EMMS:::"memory");
1027  while (s < end) {
1028  register uint16_t bgr;
1029  bgr = *s++;
1030  *d++ = ((bgr&0x1F)<<3) | ((bgr&0x1F)>>2);
1031  *d++ = ((bgr&0x7E0)>>3) | ((bgr&0x7E0)>>9);
1032  *d++ = ((bgr&0xF800)>>8) | ((bgr&0xF800)>>13);
1033  *d++ = 255;
1034  }
1035 }
1036 
1037 static inline void RENAME(rgb24tobgr24)(const uint8_t *src, uint8_t *dst, int src_size)
1038 {
1039  unsigned i;
1040  x86_reg mmx_size= 23 - src_size;
1041  __asm__ volatile (
1042  "test %%"FF_REG_a", %%"FF_REG_a" \n\t"
1043  "jns 2f \n\t"
1044  "movq "MANGLE(mask24r)", %%mm5 \n\t"
1045  "movq "MANGLE(mask24g)", %%mm6 \n\t"
1046  "movq "MANGLE(mask24b)", %%mm7 \n\t"
1047  ".p2align 4 \n\t"
1048  "1: \n\t"
1049  PREFETCH" 32(%1, %%"FF_REG_a") \n\t"
1050  "movq (%1, %%"FF_REG_a"), %%mm0 \n\t" // BGR BGR BG
1051  "movq (%1, %%"FF_REG_a"), %%mm1 \n\t" // BGR BGR BG
1052  "movq 2(%1, %%"FF_REG_a"), %%mm2 \n\t" // R BGR BGR B
1053  "psllq $16, %%mm0 \n\t" // 00 BGR BGR
1054  "pand %%mm5, %%mm0 \n\t"
1055  "pand %%mm6, %%mm1 \n\t"
1056  "pand %%mm7, %%mm2 \n\t"
1057  "por %%mm0, %%mm1 \n\t"
1058  "por %%mm2, %%mm1 \n\t"
1059  "movq 6(%1, %%"FF_REG_a"), %%mm0 \n\t" // BGR BGR BG
1060  MOVNTQ" %%mm1,(%2, %%"FF_REG_a") \n\t" // RGB RGB RG
1061  "movq 8(%1, %%"FF_REG_a"), %%mm1 \n\t" // R BGR BGR B
1062  "movq 10(%1, %%"FF_REG_a"), %%mm2 \n\t" // GR BGR BGR
1063  "pand %%mm7, %%mm0 \n\t"
1064  "pand %%mm5, %%mm1 \n\t"
1065  "pand %%mm6, %%mm2 \n\t"
1066  "por %%mm0, %%mm1 \n\t"
1067  "por %%mm2, %%mm1 \n\t"
1068  "movq 14(%1, %%"FF_REG_a"), %%mm0 \n\t" // R BGR BGR B
1069  MOVNTQ" %%mm1, 8(%2, %%"FF_REG_a")\n\t" // B RGB RGB R
1070  "movq 16(%1, %%"FF_REG_a"), %%mm1 \n\t" // GR BGR BGR
1071  "movq 18(%1, %%"FF_REG_a"), %%mm2 \n\t" // BGR BGR BG
1072  "pand %%mm6, %%mm0 \n\t"
1073  "pand %%mm7, %%mm1 \n\t"
1074  "pand %%mm5, %%mm2 \n\t"
1075  "por %%mm0, %%mm1 \n\t"
1076  "por %%mm2, %%mm1 \n\t"
1077  MOVNTQ" %%mm1, 16(%2, %%"FF_REG_a") \n\t"
1078  "add $24, %%"FF_REG_a" \n\t"
1079  " js 1b \n\t"
1080  "2: \n\t"
1081  : "+a" (mmx_size)
1082  : "r" (src-mmx_size), "r"(dst-mmx_size)
1083  NAMED_CONSTRAINTS_ADD(mask24r,mask24g,mask24b)
1084  );
1085 
1086  __asm__ volatile(SFENCE:::"memory");
1087  __asm__ volatile(EMMS:::"memory");
1088 
1089  if (mmx_size==23) return; //finished, was multiple of 8
1090 
1091  src+= src_size;
1092  dst+= src_size;
1093  src_size= 23-mmx_size;
1094  src-= src_size;
1095  dst-= src_size;
1096  for (i=0; i<src_size; i+=3) {
1097  register uint8_t x;
1098  x = src[i + 2];
1099  dst[i + 1] = src[i + 1];
1100  dst[i + 2] = src[i + 0];
1101  dst[i + 0] = x;
1102  }
1103 }
1104 
1105 static inline void RENAME(yuvPlanartoyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
1106  int width, int height,
1107  int lumStride, int chromStride, int dstStride, int vertLumPerChroma)
1108 {
1109  int y;
1110  const x86_reg chromWidth= width>>1;
1111  for (y=0; y<height; y++) {
1112  //FIXME handle 2 lines at once (fewer prefetches, reuse some chroma, but very likely memory-limited anyway)
1113  __asm__ volatile(
1114  "xor %%"FF_REG_a", %%"FF_REG_a" \n\t"
1115  ".p2align 4 \n\t"
1116  "1: \n\t"
1117  PREFETCH" 32(%1, %%"FF_REG_a", 2) \n\t"
1118  PREFETCH" 32(%2, %%"FF_REG_a") \n\t"
1119  PREFETCH" 32(%3, %%"FF_REG_a") \n\t"
1120  "movq (%2, %%"FF_REG_a"), %%mm0 \n\t" // U(0)
1121  "movq %%mm0, %%mm2 \n\t" // U(0)
1122  "movq (%3, %%"FF_REG_a"), %%mm1 \n\t" // V(0)
1123  "punpcklbw %%mm1, %%mm0 \n\t" // UVUV UVUV(0)
1124  "punpckhbw %%mm1, %%mm2 \n\t" // UVUV UVUV(8)
1125 
1126  "movq (%1, %%"FF_REG_a",2), %%mm3 \n\t" // Y(0)
1127  "movq 8(%1, %%"FF_REG_a",2), %%mm5 \n\t" // Y(8)
1128  "movq %%mm3, %%mm4 \n\t" // Y(0)
1129  "movq %%mm5, %%mm6 \n\t" // Y(8)
1130  "punpcklbw %%mm0, %%mm3 \n\t" // YUYV YUYV(0)
1131  "punpckhbw %%mm0, %%mm4 \n\t" // YUYV YUYV(4)
1132  "punpcklbw %%mm2, %%mm5 \n\t" // YUYV YUYV(8)
1133  "punpckhbw %%mm2, %%mm6 \n\t" // YUYV YUYV(12)
1134 
1135  MOVNTQ" %%mm3, (%0, %%"FF_REG_a", 4) \n\t"
1136  MOVNTQ" %%mm4, 8(%0, %%"FF_REG_a", 4) \n\t"
1137  MOVNTQ" %%mm5, 16(%0, %%"FF_REG_a", 4) \n\t"
1138  MOVNTQ" %%mm6, 24(%0, %%"FF_REG_a", 4) \n\t"
1139 
1140  "add $8, %%"FF_REG_a" \n\t"
1141  "cmp %4, %%"FF_REG_a" \n\t"
1142  " jb 1b \n\t"
1143  ::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "g" (chromWidth)
1144  : "%"FF_REG_a
1145  );
1146  if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) {
1147  usrc += chromStride;
1148  vsrc += chromStride;
1149  }
1150  ysrc += lumStride;
1151  dst += dstStride;
1152  }
1153  __asm__(EMMS" \n\t"
1154  SFENCE" \n\t"
1155  :::"memory");
1156 }
1157 
1158 /**
1159  * Height should be a multiple of 2 and width should be a multiple of 16.
1160  * (If this is a problem for anyone then tell me, and I will fix it.)
1161  */
1162 static inline void RENAME(yv12toyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
1163  int width, int height,
1164  int lumStride, int chromStride, int dstStride)
1165 {
1166  //FIXME interpolate chroma
1167  RENAME(yuvPlanartoyuy2)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 2);
1168 }
1169 
1170 static inline void RENAME(yuvPlanartouyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
1171  int width, int height,
1172  int lumStride, int chromStride, int dstStride, int vertLumPerChroma)
1173 {
1174  int y;
1175  const x86_reg chromWidth= width>>1;
1176  for (y=0; y<height; y++) {
1177  //FIXME handle 2 lines at once (fewer prefetches, reuse some chroma, but very likely memory-limited anyway)
1178  __asm__ volatile(
1179  "xor %%"FF_REG_a", %%"FF_REG_a" \n\t"
1180  ".p2align 4 \n\t"
1181  "1: \n\t"
1182  PREFETCH" 32(%1, %%"FF_REG_a", 2) \n\t"
1183  PREFETCH" 32(%2, %%"FF_REG_a") \n\t"
1184  PREFETCH" 32(%3, %%"FF_REG_a") \n\t"
1185  "movq (%2, %%"FF_REG_a"), %%mm0 \n\t" // U(0)
1186  "movq %%mm0, %%mm2 \n\t" // U(0)
1187  "movq (%3, %%"FF_REG_a"), %%mm1 \n\t" // V(0)
1188  "punpcklbw %%mm1, %%mm0 \n\t" // UVUV UVUV(0)
1189  "punpckhbw %%mm1, %%mm2 \n\t" // UVUV UVUV(8)
1190 
1191  "movq (%1, %%"FF_REG_a",2), %%mm3 \n\t" // Y(0)
1192  "movq 8(%1, %%"FF_REG_a",2), %%mm5 \n\t" // Y(8)
1193  "movq %%mm0, %%mm4 \n\t" // Y(0)
1194  "movq %%mm2, %%mm6 \n\t" // Y(8)
1195  "punpcklbw %%mm3, %%mm0 \n\t" // YUYV YUYV(0)
1196  "punpckhbw %%mm3, %%mm4 \n\t" // YUYV YUYV(4)
1197  "punpcklbw %%mm5, %%mm2 \n\t" // YUYV YUYV(8)
1198  "punpckhbw %%mm5, %%mm6 \n\t" // YUYV YUYV(12)
1199 
1200  MOVNTQ" %%mm0, (%0, %%"FF_REG_a", 4) \n\t"
1201  MOVNTQ" %%mm4, 8(%0, %%"FF_REG_a", 4) \n\t"
1202  MOVNTQ" %%mm2, 16(%0, %%"FF_REG_a", 4) \n\t"
1203  MOVNTQ" %%mm6, 24(%0, %%"FF_REG_a", 4) \n\t"
1204 
1205  "add $8, %%"FF_REG_a" \n\t"
1206  "cmp %4, %%"FF_REG_a" \n\t"
1207  " jb 1b \n\t"
1208  ::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "g" (chromWidth)
1209  : "%"FF_REG_a
1210  );
1211  if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) {
1212  usrc += chromStride;
1213  vsrc += chromStride;
1214  }
1215  ysrc += lumStride;
1216  dst += dstStride;
1217  }
1218  __asm__(EMMS" \n\t"
1219  SFENCE" \n\t"
1220  :::"memory");
1221 }
1222 
1223 /**
1224  * Height should be a multiple of 2 and width should be a multiple of 16
1225  * (If this is a problem for anyone then tell me, and I will fix it.)
1226  */
1227 static inline void RENAME(yv12touyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
1228  int width, int height,
1229  int lumStride, int chromStride, int dstStride)
1230 {
1231  //FIXME interpolate chroma
1232  RENAME(yuvPlanartouyvy)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 2);
1233 }
1234 
1235 /**
1236  * Width should be a multiple of 16.
1237  */
1238 static inline void RENAME(yuv422ptouyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
1239  int width, int height,
1240  int lumStride, int chromStride, int dstStride)
1241 {
1242  RENAME(yuvPlanartouyvy)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 1);
1243 }
1244 
1245 /**
1246  * Width should be a multiple of 16.
1247  */
1248 static inline void RENAME(yuv422ptoyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
1249  int width, int height,
1250  int lumStride, int chromStride, int dstStride)
1251 {
1252  RENAME(yuvPlanartoyuy2)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 1);
1253 }
1254 
1255 /**
1256  * Height should be a multiple of 2 and width should be a multiple of 16.
1257  * (If this is a problem for anyone then tell me, and I will fix it.)
1258  */
1259 static inline void RENAME(yuy2toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
1260  int width, int height,
1261  int lumStride, int chromStride, int srcStride)
1262 {
1263  int y;
1264  const x86_reg chromWidth= width>>1;
1265  for (y=0; y<height; y+=2) {
1266  __asm__ volatile(
1267  "xor %%"FF_REG_a", %%"FF_REG_a"\n\t"
1268  "pcmpeqw %%mm7, %%mm7 \n\t"
1269  "psrlw $8, %%mm7 \n\t" // FF,00,FF,00...
1270  ".p2align 4 \n\t"
1271  "1: \n\t"
1272  PREFETCH" 64(%0, %%"FF_REG_a", 4) \n\t"
1273  "movq (%0, %%"FF_REG_a", 4), %%mm0 \n\t" // YUYV YUYV(0)
1274  "movq 8(%0, %%"FF_REG_a", 4), %%mm1 \n\t" // YUYV YUYV(4)
1275  "movq %%mm0, %%mm2 \n\t" // YUYV YUYV(0)
1276  "movq %%mm1, %%mm3 \n\t" // YUYV YUYV(4)
1277  "psrlw $8, %%mm0 \n\t" // U0V0 U0V0(0)
1278  "psrlw $8, %%mm1 \n\t" // U0V0 U0V0(4)
1279  "pand %%mm7, %%mm2 \n\t" // Y0Y0 Y0Y0(0)
1280  "pand %%mm7, %%mm3 \n\t" // Y0Y0 Y0Y0(4)
1281  "packuswb %%mm1, %%mm0 \n\t" // UVUV UVUV(0)
1282  "packuswb %%mm3, %%mm2 \n\t" // YYYY YYYY(0)
1283 
1284  MOVNTQ" %%mm2, (%1, %%"FF_REG_a", 2) \n\t"
1285 
1286  "movq 16(%0, %%"FF_REG_a", 4), %%mm1 \n\t" // YUYV YUYV(8)
1287  "movq 24(%0, %%"FF_REG_a", 4), %%mm2 \n\t" // YUYV YUYV(12)
1288  "movq %%mm1, %%mm3 \n\t" // YUYV YUYV(8)
1289  "movq %%mm2, %%mm4 \n\t" // YUYV YUYV(12)
1290  "psrlw $8, %%mm1 \n\t" // U0V0 U0V0(8)
1291  "psrlw $8, %%mm2 \n\t" // U0V0 U0V0(12)
1292  "pand %%mm7, %%mm3 \n\t" // Y0Y0 Y0Y0(8)
1293  "pand %%mm7, %%mm4 \n\t" // Y0Y0 Y0Y0(12)
1294  "packuswb %%mm2, %%mm1 \n\t" // UVUV UVUV(8)
1295  "packuswb %%mm4, %%mm3 \n\t" // YYYY YYYY(8)
1296 
1297  MOVNTQ" %%mm3, 8(%1, %%"FF_REG_a", 2) \n\t"
1298 
1299  "movq %%mm0, %%mm2 \n\t" // UVUV UVUV(0)
1300  "movq %%mm1, %%mm3 \n\t" // UVUV UVUV(8)
1301  "psrlw $8, %%mm0 \n\t" // V0V0 V0V0(0)
1302  "psrlw $8, %%mm1 \n\t" // V0V0 V0V0(8)
1303  "pand %%mm7, %%mm2 \n\t" // U0U0 U0U0(0)
1304  "pand %%mm7, %%mm3 \n\t" // U0U0 U0U0(8)
1305  "packuswb %%mm1, %%mm0 \n\t" // VVVV VVVV(0)
1306  "packuswb %%mm3, %%mm2 \n\t" // UUUU UUUU(0)
1307 
1308  MOVNTQ" %%mm0, (%3, %%"FF_REG_a") \n\t"
1309  MOVNTQ" %%mm2, (%2, %%"FF_REG_a") \n\t"
1310 
1311  "add $8, %%"FF_REG_a" \n\t"
1312  "cmp %4, %%"FF_REG_a" \n\t"
1313  " jb 1b \n\t"
1314  ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
1315  : "memory", "%"FF_REG_a
1316  );
1317 
1318  ydst += lumStride;
1319  src += srcStride;
1320 
1321  __asm__ volatile(
1322  "xor %%"FF_REG_a", %%"FF_REG_a"\n\t"
1323  ".p2align 4 \n\t"
1324  "1: \n\t"
1325  PREFETCH" 64(%0, %%"FF_REG_a", 4) \n\t"
1326  "movq (%0, %%"FF_REG_a", 4), %%mm0 \n\t" // YUYV YUYV(0)
1327  "movq 8(%0, %%"FF_REG_a", 4), %%mm1 \n\t" // YUYV YUYV(4)
1328  "movq 16(%0, %%"FF_REG_a", 4), %%mm2 \n\t" // YUYV YUYV(8)
1329  "movq 24(%0, %%"FF_REG_a", 4), %%mm3 \n\t" // YUYV YUYV(12)
1330  "pand %%mm7, %%mm0 \n\t" // Y0Y0 Y0Y0(0)
1331  "pand %%mm7, %%mm1 \n\t" // Y0Y0 Y0Y0(4)
1332  "pand %%mm7, %%mm2 \n\t" // Y0Y0 Y0Y0(8)
1333  "pand %%mm7, %%mm3 \n\t" // Y0Y0 Y0Y0(12)
1334  "packuswb %%mm1, %%mm0 \n\t" // YYYY YYYY(0)
1335  "packuswb %%mm3, %%mm2 \n\t" // YYYY YYYY(8)
1336 
1337  MOVNTQ" %%mm0, (%1, %%"FF_REG_a", 2) \n\t"
1338  MOVNTQ" %%mm2, 8(%1, %%"FF_REG_a", 2) \n\t"
1339 
1340  "add $8, %%"FF_REG_a"\n\t"
1341  "cmp %4, %%"FF_REG_a"\n\t"
1342  " jb 1b \n\t"
1343 
1344  ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
1345  : "memory", "%"FF_REG_a
1346  );
1347  udst += chromStride;
1348  vdst += chromStride;
1349  ydst += lumStride;
1350  src += srcStride;
1351  }
1352  __asm__ volatile(EMMS" \n\t"
1353  SFENCE" \n\t"
1354  :::"memory");
1355 }
1356 #endif /* !COMPILE_TEMPLATE_AMD3DNOW */
1357 
1358 #if COMPILE_TEMPLATE_MMXEXT || COMPILE_TEMPLATE_AMD3DNOW
1359 static inline void RENAME(planar2x)(const uint8_t *src, uint8_t *dst, int srcWidth, int srcHeight, int srcStride, int dstStride)
1360 {
1361  int x,y;
1362 
1363  dst[0]= src[0];
1364 
1365  // first line
1366  for (x=0; x<srcWidth-1; x++) {
1367  dst[2*x+1]= (3*src[x] + src[x+1])>>2;
1368  dst[2*x+2]= ( src[x] + 3*src[x+1])>>2;
1369  }
1370  dst[2*srcWidth-1]= src[srcWidth-1];
1371 
1372  dst+= dstStride;
1373 
1374  for (y=1; y<srcHeight; y++) {
1375  x86_reg mmxSize= srcWidth&~15;
1376 
1377  if (mmxSize) {
1378  __asm__ volatile(
1379  "mov %4, %%"FF_REG_a" \n\t"
1380  "movq "MANGLE(mmx_ff)", %%mm0 \n\t"
1381  "movq (%0, %%"FF_REG_a"), %%mm4 \n\t"
1382  "movq %%mm4, %%mm2 \n\t"
1383  "psllq $8, %%mm4 \n\t"
1384  "pand %%mm0, %%mm2 \n\t"
1385  "por %%mm2, %%mm4 \n\t"
1386  "movq (%1, %%"FF_REG_a"), %%mm5 \n\t"
1387  "movq %%mm5, %%mm3 \n\t"
1388  "psllq $8, %%mm5 \n\t"
1389  "pand %%mm0, %%mm3 \n\t"
1390  "por %%mm3, %%mm5 \n\t"
1391  "1: \n\t"
1392  "movq (%0, %%"FF_REG_a"), %%mm0 \n\t"
1393  "movq (%1, %%"FF_REG_a"), %%mm1 \n\t"
1394  "movq 1(%0, %%"FF_REG_a"), %%mm2 \n\t"
1395  "movq 1(%1, %%"FF_REG_a"), %%mm3 \n\t"
1396  PAVGB" %%mm0, %%mm5 \n\t"
1397  PAVGB" %%mm0, %%mm3 \n\t"
1398  PAVGB" %%mm0, %%mm5 \n\t"
1399  PAVGB" %%mm0, %%mm3 \n\t"
1400  PAVGB" %%mm1, %%mm4 \n\t"
1401  PAVGB" %%mm1, %%mm2 \n\t"
1402  PAVGB" %%mm1, %%mm4 \n\t"
1403  PAVGB" %%mm1, %%mm2 \n\t"
1404  "movq %%mm5, %%mm7 \n\t"
1405  "movq %%mm4, %%mm6 \n\t"
1406  "punpcklbw %%mm3, %%mm5 \n\t"
1407  "punpckhbw %%mm3, %%mm7 \n\t"
1408  "punpcklbw %%mm2, %%mm4 \n\t"
1409  "punpckhbw %%mm2, %%mm6 \n\t"
1410  MOVNTQ" %%mm5, (%2, %%"FF_REG_a", 2) \n\t"
1411  MOVNTQ" %%mm7, 8(%2, %%"FF_REG_a", 2) \n\t"
1412  MOVNTQ" %%mm4, (%3, %%"FF_REG_a", 2) \n\t"
1413  MOVNTQ" %%mm6, 8(%3, %%"FF_REG_a", 2) \n\t"
1414  "add $8, %%"FF_REG_a" \n\t"
1415  "movq -1(%0, %%"FF_REG_a"), %%mm4 \n\t"
1416  "movq -1(%1, %%"FF_REG_a"), %%mm5 \n\t"
1417  " js 1b \n\t"
1418  :: "r" (src + mmxSize ), "r" (src + srcStride + mmxSize ),
1419  "r" (dst + mmxSize*2), "r" (dst + dstStride + mmxSize*2),
1420  "g" (-mmxSize)
1421  NAMED_CONSTRAINTS_ADD(mmx_ff)
1422  : "%"FF_REG_a
1423  );
1424  } else {
1425  mmxSize = 1;
1426  dst[0] = (src[0] * 3 + src[srcStride]) >> 2;
1427  dst[dstStride] = (src[0] + 3 * src[srcStride]) >> 2;
1428  }
1429 
1430  for (x=mmxSize-1; x<srcWidth-1; x++) {
1431  dst[2*x +1]= (3*src[x+0] + src[x+srcStride+1])>>2;
1432  dst[2*x+dstStride+2]= ( src[x+0] + 3*src[x+srcStride+1])>>2;
1433  dst[2*x+dstStride+1]= ( src[x+1] + 3*src[x+srcStride ])>>2;
1434  dst[2*x +2]= (3*src[x+1] + src[x+srcStride ])>>2;
1435  }
1436  dst[srcWidth*2 -1 ]= (3*src[srcWidth-1] + src[srcWidth-1 + srcStride])>>2;
1437  dst[srcWidth*2 -1 + dstStride]= ( src[srcWidth-1] + 3*src[srcWidth-1 + srcStride])>>2;
1438 
1439  dst+=dstStride*2;
1440  src+=srcStride;
1441  }
1442 
1443  // last line
1444  dst[0]= src[0];
1445 
1446  for (x=0; x<srcWidth-1; x++) {
1447  dst[2*x+1]= (3*src[x] + src[x+1])>>2;
1448  dst[2*x+2]= ( src[x] + 3*src[x+1])>>2;
1449  }
1450  dst[2*srcWidth-1]= src[srcWidth-1];
1451 
1452  __asm__ volatile(EMMS" \n\t"
1453  SFENCE" \n\t"
1454  :::"memory");
1455 }
1456 #endif /* COMPILE_TEMPLATE_MMXEXT || COMPILE_TEMPLATE_AMD3DNOW */
1457 
1458 #if !COMPILE_TEMPLATE_AMD3DNOW
1459 /**
1460  * Height should be a multiple of 2 and width should be a multiple of 16.
1461  * (If this is a problem for anyone then tell me, and I will fix it.)
1462  * Chrominance data is only taken from every second line, others are ignored.
1463  * FIXME: Write HQ version.
1464  */
1465 static inline void RENAME(uyvytoyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
1466  int width, int height,
1467  int lumStride, int chromStride, int srcStride)
1468 {
1469  int y;
1470  const x86_reg chromWidth= width>>1;
1471  for (y=0; y<height; y+=2) {
1472  __asm__ volatile(
1473  "xor %%"FF_REG_a", %%"FF_REG_a" \n\t"
1474  "pcmpeqw %%mm7, %%mm7 \n\t"
1475  "psrlw $8, %%mm7 \n\t" // FF,00,FF,00...
1476  ".p2align 4 \n\t"
1477  "1: \n\t"
1478  PREFETCH" 64(%0, %%"FF_REG_a", 4) \n\t"
1479  "movq (%0, %%"FF_REG_a", 4), %%mm0 \n\t" // UYVY UYVY(0)
1480  "movq 8(%0, %%"FF_REG_a", 4), %%mm1 \n\t" // UYVY UYVY(4)
1481  "movq %%mm0, %%mm2 \n\t" // UYVY UYVY(0)
1482  "movq %%mm1, %%mm3 \n\t" // UYVY UYVY(4)
1483  "pand %%mm7, %%mm0 \n\t" // U0V0 U0V0(0)
1484  "pand %%mm7, %%mm1 \n\t" // U0V0 U0V0(4)
1485  "psrlw $8, %%mm2 \n\t" // Y0Y0 Y0Y0(0)
1486  "psrlw $8, %%mm3 \n\t" // Y0Y0 Y0Y0(4)
1487  "packuswb %%mm1, %%mm0 \n\t" // UVUV UVUV(0)
1488  "packuswb %%mm3, %%mm2 \n\t" // YYYY YYYY(0)
1489 
1490  MOVNTQ" %%mm2, (%1, %%"FF_REG_a", 2) \n\t"
1491 
1492  "movq 16(%0, %%"FF_REG_a", 4), %%mm1 \n\t" // UYVY UYVY(8)
1493  "movq 24(%0, %%"FF_REG_a", 4), %%mm2 \n\t" // UYVY UYVY(12)
1494  "movq %%mm1, %%mm3 \n\t" // UYVY UYVY(8)
1495  "movq %%mm2, %%mm4 \n\t" // UYVY UYVY(12)
1496  "pand %%mm7, %%mm1 \n\t" // U0V0 U0V0(8)
1497  "pand %%mm7, %%mm2 \n\t" // U0V0 U0V0(12)
1498  "psrlw $8, %%mm3 \n\t" // Y0Y0 Y0Y0(8)
1499  "psrlw $8, %%mm4 \n\t" // Y0Y0 Y0Y0(12)
1500  "packuswb %%mm2, %%mm1 \n\t" // UVUV UVUV(8)
1501  "packuswb %%mm4, %%mm3 \n\t" // YYYY YYYY(8)
1502 
1503  MOVNTQ" %%mm3, 8(%1, %%"FF_REG_a", 2) \n\t"
1504 
1505  "movq %%mm0, %%mm2 \n\t" // UVUV UVUV(0)
1506  "movq %%mm1, %%mm3 \n\t" // UVUV UVUV(8)
1507  "psrlw $8, %%mm0 \n\t" // V0V0 V0V0(0)
1508  "psrlw $8, %%mm1 \n\t" // V0V0 V0V0(8)
1509  "pand %%mm7, %%mm2 \n\t" // U0U0 U0U0(0)
1510  "pand %%mm7, %%mm3 \n\t" // U0U0 U0U0(8)
1511  "packuswb %%mm1, %%mm0 \n\t" // VVVV VVVV(0)
1512  "packuswb %%mm3, %%mm2 \n\t" // UUUU UUUU(0)
1513 
1514  MOVNTQ" %%mm0, (%3, %%"FF_REG_a") \n\t"
1515  MOVNTQ" %%mm2, (%2, %%"FF_REG_a") \n\t"
1516 
1517  "add $8, %%"FF_REG_a" \n\t"
1518  "cmp %4, %%"FF_REG_a" \n\t"
1519  " jb 1b \n\t"
1520  ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
1521  : "memory", "%"FF_REG_a
1522  );
1523 
1524  ydst += lumStride;
1525  src += srcStride;
1526 
1527  __asm__ volatile(
1528  "xor %%"FF_REG_a", %%"FF_REG_a" \n\t"
1529  ".p2align 4 \n\t"
1530  "1: \n\t"
1531  PREFETCH" 64(%0, %%"FF_REG_a", 4) \n\t"
1532  "movq (%0, %%"FF_REG_a", 4), %%mm0 \n\t" // YUYV YUYV(0)
1533  "movq 8(%0, %%"FF_REG_a", 4), %%mm1 \n\t" // YUYV YUYV(4)
1534  "movq 16(%0, %%"FF_REG_a", 4), %%mm2 \n\t" // YUYV YUYV(8)
1535  "movq 24(%0, %%"FF_REG_a", 4), %%mm3 \n\t" // YUYV YUYV(12)
1536  "psrlw $8, %%mm0 \n\t" // Y0Y0 Y0Y0(0)
1537  "psrlw $8, %%mm1 \n\t" // Y0Y0 Y0Y0(4)
1538  "psrlw $8, %%mm2 \n\t" // Y0Y0 Y0Y0(8)
1539  "psrlw $8, %%mm3 \n\t" // Y0Y0 Y0Y0(12)
1540  "packuswb %%mm1, %%mm0 \n\t" // YYYY YYYY(0)
1541  "packuswb %%mm3, %%mm2 \n\t" // YYYY YYYY(8)
1542 
1543  MOVNTQ" %%mm0, (%1, %%"FF_REG_a", 2) \n\t"
1544  MOVNTQ" %%mm2, 8(%1, %%"FF_REG_a", 2) \n\t"
1545 
1546  "add $8, %%"FF_REG_a" \n\t"
1547  "cmp %4, %%"FF_REG_a" \n\t"
1548  " jb 1b \n\t"
1549 
1550  ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
1551  : "memory", "%"FF_REG_a
1552  );
1553  udst += chromStride;
1554  vdst += chromStride;
1555  ydst += lumStride;
1556  src += srcStride;
1557  }
1558  __asm__ volatile(EMMS" \n\t"
1559  SFENCE" \n\t"
1560  :::"memory");
1561 }
1562 #endif /* !COMPILE_TEMPLATE_AMD3DNOW */
1563 
1564 /**
1565  * Height should be a multiple of 2 and width should be a multiple of 2.
1566  * (If this is a problem for anyone then tell me, and I will fix it.)
1567  * Chrominance data is only taken from every second line,
1568  * others are ignored in the C version.
1569  * FIXME: Write HQ version.
1570  */
1571 #if HAVE_7REGS
1572 static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
1573  int width, int height,
1574  int lumStride, int chromStride, int srcStride,
1575  int32_t *rgb2yuv)
1576 {
1577 #define BGR2Y_IDX "16*4+16*32"
1578 #define BGR2U_IDX "16*4+16*33"
1579 #define BGR2V_IDX "16*4+16*34"
1580  int y;
1581  const x86_reg chromWidth= width>>1;
1582 
1583  if (height > 2) {
1584  ff_rgb24toyv12_c(src, ydst, udst, vdst, width, 2, lumStride, chromStride, srcStride, rgb2yuv);
1585  src += 2*srcStride;
1586  ydst += 2*lumStride;
1587  udst += chromStride;
1588  vdst += chromStride;
1589  height -= 2;
1590  }
1591 
1592  for (y=0; y<height-2; y+=2) {
1593  int i;
1594  for (i=0; i<2; i++) {
1595  __asm__ volatile(
1596  "mov %2, %%"FF_REG_a"\n\t"
1597  "movq "BGR2Y_IDX"(%3), %%mm6 \n\t"
1598  "movq "MANGLE(ff_w1111)", %%mm5 \n\t"
1599  "pxor %%mm7, %%mm7 \n\t"
1600  "lea (%%"FF_REG_a", %%"FF_REG_a", 2), %%"FF_REG_d" \n\t"
1601  ".p2align 4 \n\t"
1602  "1: \n\t"
1603  PREFETCH" 64(%0, %%"FF_REG_d") \n\t"
1604  "movd (%0, %%"FF_REG_d"), %%mm0 \n\t"
1605  "movd 3(%0, %%"FF_REG_d"), %%mm1 \n\t"
1606  "punpcklbw %%mm7, %%mm0 \n\t"
1607  "punpcklbw %%mm7, %%mm1 \n\t"
1608  "movd 6(%0, %%"FF_REG_d"), %%mm2 \n\t"
1609  "movd 9(%0, %%"FF_REG_d"), %%mm3 \n\t"
1610  "punpcklbw %%mm7, %%mm2 \n\t"
1611  "punpcklbw %%mm7, %%mm3 \n\t"
1612  "pmaddwd %%mm6, %%mm0 \n\t"
1613  "pmaddwd %%mm6, %%mm1 \n\t"
1614  "pmaddwd %%mm6, %%mm2 \n\t"
1615  "pmaddwd %%mm6, %%mm3 \n\t"
1616  "psrad $8, %%mm0 \n\t"
1617  "psrad $8, %%mm1 \n\t"
1618  "psrad $8, %%mm2 \n\t"
1619  "psrad $8, %%mm3 \n\t"
1620  "packssdw %%mm1, %%mm0 \n\t"
1621  "packssdw %%mm3, %%mm2 \n\t"
1622  "pmaddwd %%mm5, %%mm0 \n\t"
1623  "pmaddwd %%mm5, %%mm2 \n\t"
1624  "packssdw %%mm2, %%mm0 \n\t"
1625  "psraw $7, %%mm0 \n\t"
1626 
1627  "movd 12(%0, %%"FF_REG_d"), %%mm4 \n\t"
1628  "movd 15(%0, %%"FF_REG_d"), %%mm1 \n\t"
1629  "punpcklbw %%mm7, %%mm4 \n\t"
1630  "punpcklbw %%mm7, %%mm1 \n\t"
1631  "movd 18(%0, %%"FF_REG_d"), %%mm2 \n\t"
1632  "movd 21(%0, %%"FF_REG_d"), %%mm3 \n\t"
1633  "punpcklbw %%mm7, %%mm2 \n\t"
1634  "punpcklbw %%mm7, %%mm3 \n\t"
1635  "pmaddwd %%mm6, %%mm4 \n\t"
1636  "pmaddwd %%mm6, %%mm1 \n\t"
1637  "pmaddwd %%mm6, %%mm2 \n\t"
1638  "pmaddwd %%mm6, %%mm3 \n\t"
1639  "psrad $8, %%mm4 \n\t"
1640  "psrad $8, %%mm1 \n\t"
1641  "psrad $8, %%mm2 \n\t"
1642  "psrad $8, %%mm3 \n\t"
1643  "packssdw %%mm1, %%mm4 \n\t"
1644  "packssdw %%mm3, %%mm2 \n\t"
1645  "pmaddwd %%mm5, %%mm4 \n\t"
1646  "pmaddwd %%mm5, %%mm2 \n\t"
1647  "add $24, %%"FF_REG_d"\n\t"
1648  "packssdw %%mm2, %%mm4 \n\t"
1649  "psraw $7, %%mm4 \n\t"
1650 
1651  "packuswb %%mm4, %%mm0 \n\t"
1652  "paddusb "MANGLE(ff_bgr2YOffset)", %%mm0 \n\t"
1653 
1654  MOVNTQ" %%mm0, (%1, %%"FF_REG_a") \n\t"
1655  "add $8, %%"FF_REG_a" \n\t"
1656  " js 1b \n\t"
1657  : : "r" (src+width*3), "r" (ydst+width), "g" ((x86_reg)-width), "r"(rgb2yuv)
1658  NAMED_CONSTRAINTS_ADD(ff_w1111,ff_bgr2YOffset)
1659  : "%"FF_REG_a, "%"FF_REG_d
1660  );
1661  ydst += lumStride;
1662  src += srcStride;
1663  }
1664  src -= srcStride*2;
1665  __asm__ volatile(
1666  "mov %4, %%"FF_REG_a"\n\t"
1667  "movq "MANGLE(ff_w1111)", %%mm5 \n\t"
1668  "movq "BGR2U_IDX"(%5), %%mm6 \n\t"
1669  "pxor %%mm7, %%mm7 \n\t"
1670  "lea (%%"FF_REG_a", %%"FF_REG_a", 2), %%"FF_REG_d" \n\t"
1671  "add %%"FF_REG_d", %%"FF_REG_d"\n\t"
1672  ".p2align 4 \n\t"
1673  "1: \n\t"
1674  PREFETCH" 64(%0, %%"FF_REG_d") \n\t"
1675  PREFETCH" 64(%1, %%"FF_REG_d") \n\t"
1676 #if COMPILE_TEMPLATE_MMXEXT || COMPILE_TEMPLATE_AMD3DNOW
1677  "movq (%0, %%"FF_REG_d"), %%mm0 \n\t"
1678  "movq (%1, %%"FF_REG_d"), %%mm1 \n\t"
1679  "movq 6(%0, %%"FF_REG_d"), %%mm2 \n\t"
1680  "movq 6(%1, %%"FF_REG_d"), %%mm3 \n\t"
1681  PAVGB" %%mm1, %%mm0 \n\t"
1682  PAVGB" %%mm3, %%mm2 \n\t"
1683  "movq %%mm0, %%mm1 \n\t"
1684  "movq %%mm2, %%mm3 \n\t"
1685  "psrlq $24, %%mm0 \n\t"
1686  "psrlq $24, %%mm2 \n\t"
1687  PAVGB" %%mm1, %%mm0 \n\t"
1688  PAVGB" %%mm3, %%mm2 \n\t"
1689  "punpcklbw %%mm7, %%mm0 \n\t"
1690  "punpcklbw %%mm7, %%mm2 \n\t"
1691 #else
1692  "movd (%0, %%"FF_REG_d"), %%mm0 \n\t"
1693  "movd (%1, %%"FF_REG_d"), %%mm1 \n\t"
1694  "movd 3(%0, %%"FF_REG_d"), %%mm2 \n\t"
1695  "movd 3(%1, %%"FF_REG_d"), %%mm3 \n\t"
1696  "punpcklbw %%mm7, %%mm0 \n\t"
1697  "punpcklbw %%mm7, %%mm1 \n\t"
1698  "punpcklbw %%mm7, %%mm2 \n\t"
1699  "punpcklbw %%mm7, %%mm3 \n\t"
1700  "paddw %%mm1, %%mm0 \n\t"
1701  "paddw %%mm3, %%mm2 \n\t"
1702  "paddw %%mm2, %%mm0 \n\t"
1703  "movd 6(%0, %%"FF_REG_d"), %%mm4 \n\t"
1704  "movd 6(%1, %%"FF_REG_d"), %%mm1 \n\t"
1705  "movd 9(%0, %%"FF_REG_d"), %%mm2 \n\t"
1706  "movd 9(%1, %%"FF_REG_d"), %%mm3 \n\t"
1707  "punpcklbw %%mm7, %%mm4 \n\t"
1708  "punpcklbw %%mm7, %%mm1 \n\t"
1709  "punpcklbw %%mm7, %%mm2 \n\t"
1710  "punpcklbw %%mm7, %%mm3 \n\t"
1711  "paddw %%mm1, %%mm4 \n\t"
1712  "paddw %%mm3, %%mm2 \n\t"
1713  "paddw %%mm4, %%mm2 \n\t"
1714  "psrlw $2, %%mm0 \n\t"
1715  "psrlw $2, %%mm2 \n\t"
1716 #endif
1717  "movq "BGR2V_IDX"(%5), %%mm1 \n\t"
1718  "movq "BGR2V_IDX"(%5), %%mm3 \n\t"
1719 
1720  "pmaddwd %%mm0, %%mm1 \n\t"
1721  "pmaddwd %%mm2, %%mm3 \n\t"
1722  "pmaddwd %%mm6, %%mm0 \n\t"
1723  "pmaddwd %%mm6, %%mm2 \n\t"
1724  "psrad $8, %%mm0 \n\t"
1725  "psrad $8, %%mm1 \n\t"
1726  "psrad $8, %%mm2 \n\t"
1727  "psrad $8, %%mm3 \n\t"
1728  "packssdw %%mm2, %%mm0 \n\t"
1729  "packssdw %%mm3, %%mm1 \n\t"
1730  "pmaddwd %%mm5, %%mm0 \n\t"
1731  "pmaddwd %%mm5, %%mm1 \n\t"
1732  "packssdw %%mm1, %%mm0 \n\t" // V1 V0 U1 U0
1733  "psraw $7, %%mm0 \n\t"
1734 
1735 #if COMPILE_TEMPLATE_MMXEXT || COMPILE_TEMPLATE_AMD3DNOW
1736  "movq 12(%0, %%"FF_REG_d"), %%mm4 \n\t"
1737  "movq 12(%1, %%"FF_REG_d"), %%mm1 \n\t"
1738  "movq 18(%0, %%"FF_REG_d"), %%mm2 \n\t"
1739  "movq 18(%1, %%"FF_REG_d"), %%mm3 \n\t"
1740  PAVGB" %%mm1, %%mm4 \n\t"
1741  PAVGB" %%mm3, %%mm2 \n\t"
1742  "movq %%mm4, %%mm1 \n\t"
1743  "movq %%mm2, %%mm3 \n\t"
1744  "psrlq $24, %%mm4 \n\t"
1745  "psrlq $24, %%mm2 \n\t"
1746  PAVGB" %%mm1, %%mm4 \n\t"
1747  PAVGB" %%mm3, %%mm2 \n\t"
1748  "punpcklbw %%mm7, %%mm4 \n\t"
1749  "punpcklbw %%mm7, %%mm2 \n\t"
1750 #else
1751  "movd 12(%0, %%"FF_REG_d"), %%mm4 \n\t"
1752  "movd 12(%1, %%"FF_REG_d"), %%mm1 \n\t"
1753  "movd 15(%0, %%"FF_REG_d"), %%mm2 \n\t"
1754  "movd 15(%1, %%"FF_REG_d"), %%mm3 \n\t"
1755  "punpcklbw %%mm7, %%mm4 \n\t"
1756  "punpcklbw %%mm7, %%mm1 \n\t"
1757  "punpcklbw %%mm7, %%mm2 \n\t"
1758  "punpcklbw %%mm7, %%mm3 \n\t"
1759  "paddw %%mm1, %%mm4 \n\t"
1760  "paddw %%mm3, %%mm2 \n\t"
1761  "paddw %%mm2, %%mm4 \n\t"
1762  "movd 18(%0, %%"FF_REG_d"), %%mm5 \n\t"
1763  "movd 18(%1, %%"FF_REG_d"), %%mm1 \n\t"
1764  "movd 21(%0, %%"FF_REG_d"), %%mm2 \n\t"
1765  "movd 21(%1, %%"FF_REG_d"), %%mm3 \n\t"
1766  "punpcklbw %%mm7, %%mm5 \n\t"
1767  "punpcklbw %%mm7, %%mm1 \n\t"
1768  "punpcklbw %%mm7, %%mm2 \n\t"
1769  "punpcklbw %%mm7, %%mm3 \n\t"
1770  "paddw %%mm1, %%mm5 \n\t"
1771  "paddw %%mm3, %%mm2 \n\t"
1772  "paddw %%mm5, %%mm2 \n\t"
1773  "movq "MANGLE(ff_w1111)", %%mm5 \n\t"
1774  "psrlw $2, %%mm4 \n\t"
1775  "psrlw $2, %%mm2 \n\t"
1776 #endif
1777  "movq "BGR2V_IDX"(%5), %%mm1 \n\t"
1778  "movq "BGR2V_IDX"(%5), %%mm3 \n\t"
1779 
1780  "pmaddwd %%mm4, %%mm1 \n\t"
1781  "pmaddwd %%mm2, %%mm3 \n\t"
1782  "pmaddwd %%mm6, %%mm4 \n\t"
1783  "pmaddwd %%mm6, %%mm2 \n\t"
1784  "psrad $8, %%mm4 \n\t"
1785  "psrad $8, %%mm1 \n\t"
1786  "psrad $8, %%mm2 \n\t"
1787  "psrad $8, %%mm3 \n\t"
1788  "packssdw %%mm2, %%mm4 \n\t"
1789  "packssdw %%mm3, %%mm1 \n\t"
1790  "pmaddwd %%mm5, %%mm4 \n\t"
1791  "pmaddwd %%mm5, %%mm1 \n\t"
1792  "add $24, %%"FF_REG_d"\n\t"
1793  "packssdw %%mm1, %%mm4 \n\t" // V3 V2 U3 U2
1794  "psraw $7, %%mm4 \n\t"
1795 
1796  "movq %%mm0, %%mm1 \n\t"
1797  "punpckldq %%mm4, %%mm0 \n\t"
1798  "punpckhdq %%mm4, %%mm1 \n\t"
1799  "packsswb %%mm1, %%mm0 \n\t"
1800  "paddb "MANGLE(ff_bgr2UVOffset)", %%mm0 \n\t"
1801  "movd %%mm0, (%2, %%"FF_REG_a") \n\t"
1802  "punpckhdq %%mm0, %%mm0 \n\t"
1803  "movd %%mm0, (%3, %%"FF_REG_a") \n\t"
1804  "add $4, %%"FF_REG_a" \n\t"
1805  " js 1b \n\t"
1806  : : "r" (src+chromWidth*6), "r" (src+srcStride+chromWidth*6), "r" (udst+chromWidth), "r" (vdst+chromWidth), "g" (-chromWidth), "r"(rgb2yuv)
1807  NAMED_CONSTRAINTS_ADD(ff_w1111,ff_bgr2UVOffset)
1808  : "%"FF_REG_a, "%"FF_REG_d
1809  );
1810 
1811  udst += chromStride;
1812  vdst += chromStride;
1813  src += srcStride*2;
1814  }
1815 
1816  __asm__ volatile(EMMS" \n\t"
1817  SFENCE" \n\t"
1818  :::"memory");
1819 
1820  ff_rgb24toyv12_c(src, ydst, udst, vdst, width, height-y, lumStride, chromStride, srcStride, rgb2yuv);
1821 }
1822 #endif /* HAVE_7REGS */
1823 #endif /* !COMPILE_TEMPLATE_SSE2 */
1824 
1825 #if !COMPILE_TEMPLATE_AMD3DNOW && !COMPILE_TEMPLATE_AVX
1826 static void RENAME(interleaveBytes)(const uint8_t *src1, const uint8_t *src2, uint8_t *dest,
1827  int width, int height, int src1Stride,
1828  int src2Stride, int dstStride)
1829 {
1830  int h;
1831 
1832  for (h=0; h < height; h++) {
1833  int w;
1834 
1835  if (width >= 16) {
1836 #if COMPILE_TEMPLATE_SSE2
1837  if (!((((intptr_t)src1) | ((intptr_t)src2) | ((intptr_t)dest))&15)) {
1838  __asm__(
1839  "xor %%"FF_REG_a", %%"FF_REG_a" \n\t"
1840  "1: \n\t"
1841  PREFETCH" 64(%1, %%"FF_REG_a") \n\t"
1842  PREFETCH" 64(%2, %%"FF_REG_a") \n\t"
1843  "movdqa (%1, %%"FF_REG_a"), %%xmm0 \n\t"
1844  "movdqa (%1, %%"FF_REG_a"), %%xmm1 \n\t"
1845  "movdqa (%2, %%"FF_REG_a"), %%xmm2 \n\t"
1846  "punpcklbw %%xmm2, %%xmm0 \n\t"
1847  "punpckhbw %%xmm2, %%xmm1 \n\t"
1848  "movntdq %%xmm0, (%0, %%"FF_REG_a", 2) \n\t"
1849  "movntdq %%xmm1, 16(%0, %%"FF_REG_a", 2) \n\t"
1850  "add $16, %%"FF_REG_a" \n\t"
1851  "cmp %3, %%"FF_REG_a" \n\t"
1852  " jb 1b \n\t"
1853  ::"r"(dest), "r"(src1), "r"(src2), "r" ((x86_reg)width-15)
1854  : "memory", XMM_CLOBBERS("xmm0", "xmm1", "xmm2",) "%"FF_REG_a
1855  );
1856  } else
1857 #endif
1858  __asm__(
1859  "xor %%"FF_REG_a", %%"FF_REG_a" \n\t"
1860  "1: \n\t"
1861  PREFETCH" 64(%1, %%"FF_REG_a") \n\t"
1862  PREFETCH" 64(%2, %%"FF_REG_a") \n\t"
1863  "movq (%1, %%"FF_REG_a"), %%mm0 \n\t"
1864  "movq 8(%1, %%"FF_REG_a"), %%mm2 \n\t"
1865  "movq %%mm0, %%mm1 \n\t"
1866  "movq %%mm2, %%mm3 \n\t"
1867  "movq (%2, %%"FF_REG_a"), %%mm4 \n\t"
1868  "movq 8(%2, %%"FF_REG_a"), %%mm5 \n\t"
1869  "punpcklbw %%mm4, %%mm0 \n\t"
1870  "punpckhbw %%mm4, %%mm1 \n\t"
1871  "punpcklbw %%mm5, %%mm2 \n\t"
1872  "punpckhbw %%mm5, %%mm3 \n\t"
1873  MOVNTQ" %%mm0, (%0, %%"FF_REG_a", 2) \n\t"
1874  MOVNTQ" %%mm1, 8(%0, %%"FF_REG_a", 2) \n\t"
1875  MOVNTQ" %%mm2, 16(%0, %%"FF_REG_a", 2) \n\t"
1876  MOVNTQ" %%mm3, 24(%0, %%"FF_REG_a", 2) \n\t"
1877  "add $16, %%"FF_REG_a" \n\t"
1878  "cmp %3, %%"FF_REG_a" \n\t"
1879  " jb 1b \n\t"
1880  ::"r"(dest), "r"(src1), "r"(src2), "r" ((x86_reg)width-15)
1881  : "memory", "%"FF_REG_a
1882  );
1883 
1884  }
1885  for (w= (width&(~15)); w < width; w++) {
1886  dest[2*w+0] = src1[w];
1887  dest[2*w+1] = src2[w];
1888  }
1889  dest += dstStride;
1890  src1 += src1Stride;
1891  src2 += src2Stride;
1892  }
1893  __asm__(
1894  EMMS" \n\t"
1895  SFENCE" \n\t"
1896  ::: "memory"
1897  );
1898 }
1899 #endif /* !COMPILE_TEMPLATE_AMD3DNOW && !COMPILE_TEMPLATE_AVX */
1900 
1901 #if !COMPILE_TEMPLATE_AVX || HAVE_AVX_EXTERNAL
1902 #if !COMPILE_TEMPLATE_AMD3DNOW && (ARCH_X86_32 || COMPILE_TEMPLATE_SSE2) && COMPILE_TEMPLATE_MMXEXT == COMPILE_TEMPLATE_SSE2 && HAVE_X86ASM
1903 void RENAME(ff_nv12ToUV)(uint8_t *dstU, uint8_t *dstV,
1904  const uint8_t *unused,
1905  const uint8_t *src1,
1906  const uint8_t *src2,
1907  int w,
1908  uint32_t *unused2);
1909 static void RENAME(deinterleaveBytes)(const uint8_t *src, uint8_t *dst1, uint8_t *dst2,
1910  int width, int height, int srcStride,
1911  int dst1Stride, int dst2Stride)
1912 {
1913  int h;
1914 
1915  for (h = 0; h < height; h++) {
1916  RENAME(ff_nv12ToUV)(dst1, dst2, NULL, src, NULL, width, NULL);
1917  src += srcStride;
1918  dst1 += dst1Stride;
1919  dst2 += dst2Stride;
1920  }
1921  __asm__(
1922 #if !COMPILE_TEMPLATE_SSE2
1923  EMMS" \n\t"
1924 #endif
1925  SFENCE" \n\t"
1926  ::: "memory"
1927  );
1928 }
1929 #endif /* !COMPILE_TEMPLATE_AMD3DNOW */
1930 #endif /* !COMPILE_TEMPLATE_AVX || HAVE_AVX_EXTERNAL */
1931 
1932 #if !COMPILE_TEMPLATE_SSE2
1933 #if !COMPILE_TEMPLATE_AMD3DNOW
1934 static inline void RENAME(vu9_to_vu12)(const uint8_t *src1, const uint8_t *src2,
1935  uint8_t *dst1, uint8_t *dst2,
1936  int width, int height,
1937  int srcStride1, int srcStride2,
1938  int dstStride1, int dstStride2)
1939 {
1940  x86_reg x, y;
1941  int w,h;
1942  w=width/2; h=height/2;
1943  __asm__ volatile(
1944  PREFETCH" %0 \n\t"
1945  PREFETCH" %1 \n\t"
1946  ::"m"(*(src1+srcStride1)),"m"(*(src2+srcStride2)):"memory");
1947  for (y=0;y<h;y++) {
1948  const uint8_t* s1=src1+srcStride1*(y>>1);
1949  uint8_t* d=dst1+dstStride1*y;
1950  x=0;
1951  for (;x<w-31;x+=32) {
1952  __asm__ volatile(
1953  PREFETCH" 32(%1,%2) \n\t"
1954  "movq (%1,%2), %%mm0 \n\t"
1955  "movq 8(%1,%2), %%mm2 \n\t"
1956  "movq 16(%1,%2), %%mm4 \n\t"
1957  "movq 24(%1,%2), %%mm6 \n\t"
1958  "movq %%mm0, %%mm1 \n\t"
1959  "movq %%mm2, %%mm3 \n\t"
1960  "movq %%mm4, %%mm5 \n\t"
1961  "movq %%mm6, %%mm7 \n\t"
1962  "punpcklbw %%mm0, %%mm0 \n\t"
1963  "punpckhbw %%mm1, %%mm1 \n\t"
1964  "punpcklbw %%mm2, %%mm2 \n\t"
1965  "punpckhbw %%mm3, %%mm3 \n\t"
1966  "punpcklbw %%mm4, %%mm4 \n\t"
1967  "punpckhbw %%mm5, %%mm5 \n\t"
1968  "punpcklbw %%mm6, %%mm6 \n\t"
1969  "punpckhbw %%mm7, %%mm7 \n\t"
1970  MOVNTQ" %%mm0, (%0,%2,2) \n\t"
1971  MOVNTQ" %%mm1, 8(%0,%2,2) \n\t"
1972  MOVNTQ" %%mm2, 16(%0,%2,2) \n\t"
1973  MOVNTQ" %%mm3, 24(%0,%2,2) \n\t"
1974  MOVNTQ" %%mm4, 32(%0,%2,2) \n\t"
1975  MOVNTQ" %%mm5, 40(%0,%2,2) \n\t"
1976  MOVNTQ" %%mm6, 48(%0,%2,2) \n\t"
1977  MOVNTQ" %%mm7, 56(%0,%2,2)"
1978  :: "r"(d), "r"(s1), "r"(x)
1979  :"memory");
1980  }
1981  for (;x<w;x++) d[2*x]=d[2*x+1]=s1[x];
1982  }
1983  for (y=0;y<h;y++) {
1984  const uint8_t* s2=src2+srcStride2*(y>>1);
1985  uint8_t* d=dst2+dstStride2*y;
1986  x=0;
1987  for (;x<w-31;x+=32) {
1988  __asm__ volatile(
1989  PREFETCH" 32(%1,%2) \n\t"
1990  "movq (%1,%2), %%mm0 \n\t"
1991  "movq 8(%1,%2), %%mm2 \n\t"
1992  "movq 16(%1,%2), %%mm4 \n\t"
1993  "movq 24(%1,%2), %%mm6 \n\t"
1994  "movq %%mm0, %%mm1 \n\t"
1995  "movq %%mm2, %%mm3 \n\t"
1996  "movq %%mm4, %%mm5 \n\t"
1997  "movq %%mm6, %%mm7 \n\t"
1998  "punpcklbw %%mm0, %%mm0 \n\t"
1999  "punpckhbw %%mm1, %%mm1 \n\t"
2000  "punpcklbw %%mm2, %%mm2 \n\t"
2001  "punpckhbw %%mm3, %%mm3 \n\t"
2002  "punpcklbw %%mm4, %%mm4 \n\t"
2003  "punpckhbw %%mm5, %%mm5 \n\t"
2004  "punpcklbw %%mm6, %%mm6 \n\t"
2005  "punpckhbw %%mm7, %%mm7 \n\t"
2006  MOVNTQ" %%mm0, (%0,%2,2) \n\t"
2007  MOVNTQ" %%mm1, 8(%0,%2,2) \n\t"
2008  MOVNTQ" %%mm2, 16(%0,%2,2) \n\t"
2009  MOVNTQ" %%mm3, 24(%0,%2,2) \n\t"
2010  MOVNTQ" %%mm4, 32(%0,%2,2) \n\t"
2011  MOVNTQ" %%mm5, 40(%0,%2,2) \n\t"
2012  MOVNTQ" %%mm6, 48(%0,%2,2) \n\t"
2013  MOVNTQ" %%mm7, 56(%0,%2,2)"
2014  :: "r"(d), "r"(s2), "r"(x)
2015  :"memory");
2016  }
2017  for (;x<w;x++) d[2*x]=d[2*x+1]=s2[x];
2018  }
2019  __asm__(
2020  EMMS" \n\t"
2021  SFENCE" \n\t"
2022  ::: "memory"
2023  );
2024 }
2025 
2026 static inline void RENAME(yvu9_to_yuy2)(const uint8_t *src1, const uint8_t *src2, const uint8_t *src3,
2027  uint8_t *dst,
2028  int width, int height,
2029  int srcStride1, int srcStride2,
2030  int srcStride3, int dstStride)
2031 {
2032  x86_reg x;
2033  int y,w,h;
2034  w=width/2; h=height;
2035  for (y=0;y<h;y++) {
2036  const uint8_t* yp=src1+srcStride1*y;
2037  const uint8_t* up=src2+srcStride2*(y>>2);
2038  const uint8_t* vp=src3+srcStride3*(y>>2);
2039  uint8_t* d=dst+dstStride*y;
2040  x=0;
2041  for (;x<w-7;x+=8) {
2042  __asm__ volatile(
2043  PREFETCH" 32(%1, %0) \n\t"
2044  PREFETCH" 32(%2, %0) \n\t"
2045  PREFETCH" 32(%3, %0) \n\t"
2046  "movq (%1, %0, 4), %%mm0 \n\t" /* Y0Y1Y2Y3Y4Y5Y6Y7 */
2047  "movq (%2, %0), %%mm1 \n\t" /* U0U1U2U3U4U5U6U7 */
2048  "movq (%3, %0), %%mm2 \n\t" /* V0V1V2V3V4V5V6V7 */
2049  "movq %%mm0, %%mm3 \n\t" /* Y0Y1Y2Y3Y4Y5Y6Y7 */
2050  "movq %%mm1, %%mm4 \n\t" /* U0U1U2U3U4U5U6U7 */
2051  "movq %%mm2, %%mm5 \n\t" /* V0V1V2V3V4V5V6V7 */
2052  "punpcklbw %%mm1, %%mm1 \n\t" /* U0U0 U1U1 U2U2 U3U3 */
2053  "punpcklbw %%mm2, %%mm2 \n\t" /* V0V0 V1V1 V2V2 V3V3 */
2054  "punpckhbw %%mm4, %%mm4 \n\t" /* U4U4 U5U5 U6U6 U7U7 */
2055  "punpckhbw %%mm5, %%mm5 \n\t" /* V4V4 V5V5 V6V6 V7V7 */
2056 
2057  "movq %%mm1, %%mm6 \n\t"
2058  "punpcklbw %%mm2, %%mm1 \n\t" /* U0V0 U0V0 U1V1 U1V1*/
2059  "punpcklbw %%mm1, %%mm0 \n\t" /* Y0U0 Y1V0 Y2U0 Y3V0*/
2060  "punpckhbw %%mm1, %%mm3 \n\t" /* Y4U1 Y5V1 Y6U1 Y7V1*/
2061  MOVNTQ" %%mm0, (%4, %0, 8) \n\t"
2062  MOVNTQ" %%mm3, 8(%4, %0, 8) \n\t"
2063 
2064  "punpckhbw %%mm2, %%mm6 \n\t" /* U2V2 U2V2 U3V3 U3V3*/
2065  "movq 8(%1, %0, 4), %%mm0 \n\t"
2066  "movq %%mm0, %%mm3 \n\t"
2067  "punpcklbw %%mm6, %%mm0 \n\t" /* Y U2 Y V2 Y U2 Y V2*/
2068  "punpckhbw %%mm6, %%mm3 \n\t" /* Y U3 Y V3 Y U3 Y V3*/
2069  MOVNTQ" %%mm0, 16(%4, %0, 8) \n\t"
2070  MOVNTQ" %%mm3, 24(%4, %0, 8) \n\t"
2071 
2072  "movq %%mm4, %%mm6 \n\t"
2073  "movq 16(%1, %0, 4), %%mm0 \n\t"
2074  "movq %%mm0, %%mm3 \n\t"
2075  "punpcklbw %%mm5, %%mm4 \n\t"
2076  "punpcklbw %%mm4, %%mm0 \n\t" /* Y U4 Y V4 Y U4 Y V4*/
2077  "punpckhbw %%mm4, %%mm3 \n\t" /* Y U5 Y V5 Y U5 Y V5*/
2078  MOVNTQ" %%mm0, 32(%4, %0, 8) \n\t"
2079  MOVNTQ" %%mm3, 40(%4, %0, 8) \n\t"
2080 
2081  "punpckhbw %%mm5, %%mm6 \n\t"
2082  "movq 24(%1, %0, 4), %%mm0 \n\t"
2083  "movq %%mm0, %%mm3 \n\t"
2084  "punpcklbw %%mm6, %%mm0 \n\t" /* Y U6 Y V6 Y U6 Y V6*/
2085  "punpckhbw %%mm6, %%mm3 \n\t" /* Y U7 Y V7 Y U7 Y V7*/
2086  MOVNTQ" %%mm0, 48(%4, %0, 8) \n\t"
2087  MOVNTQ" %%mm3, 56(%4, %0, 8) \n\t"
2088 
2089  : "+r" (x)
2090  : "r"(yp), "r" (up), "r"(vp), "r"(d)
2091  :"memory");
2092  }
2093  for (; x<w; x++) {
2094  const int x2 = x<<2;
2095  d[8*x+0] = yp[x2];
2096  d[8*x+1] = up[x];
2097  d[8*x+2] = yp[x2+1];
2098  d[8*x+3] = vp[x];
2099  d[8*x+4] = yp[x2+2];
2100  d[8*x+5] = up[x];
2101  d[8*x+6] = yp[x2+3];
2102  d[8*x+7] = vp[x];
2103  }
2104  }
2105  __asm__(
2106  EMMS" \n\t"
2107  SFENCE" \n\t"
2108  ::: "memory"
2109  );
2110 }
2111 #endif /* !COMPILE_TEMPLATE_AMD3DNOW */
2112 
2113 static void RENAME(extract_even)(const uint8_t *src, uint8_t *dst, x86_reg count)
2114 {
2115  dst += count;
2116  src += 2*count;
2117  count= - count;
2118 
2119  if(count <= -16) {
2120  count += 15;
2121  __asm__ volatile(
2122  "pcmpeqw %%mm7, %%mm7 \n\t"
2123  "psrlw $8, %%mm7 \n\t"
2124  "1: \n\t"
2125  "movq -30(%1, %0, 2), %%mm0 \n\t"
2126  "movq -22(%1, %0, 2), %%mm1 \n\t"
2127  "movq -14(%1, %0, 2), %%mm2 \n\t"
2128  "movq -6(%1, %0, 2), %%mm3 \n\t"
2129  "pand %%mm7, %%mm0 \n\t"
2130  "pand %%mm7, %%mm1 \n\t"
2131  "pand %%mm7, %%mm2 \n\t"
2132  "pand %%mm7, %%mm3 \n\t"
2133  "packuswb %%mm1, %%mm0 \n\t"
2134  "packuswb %%mm3, %%mm2 \n\t"
2135  MOVNTQ" %%mm0,-15(%2, %0) \n\t"
2136  MOVNTQ" %%mm2,- 7(%2, %0) \n\t"
2137  "add $16, %0 \n\t"
2138  " js 1b \n\t"
2139  : "+r"(count)
2140  : "r"(src), "r"(dst)
2141  );
2142  count -= 15;
2143  }
2144  while(count<0) {
2145  dst[count]= src[2*count];
2146  count++;
2147  }
2148 }
2149 
2150 static void RENAME(extract_odd)(const uint8_t *src, uint8_t *dst, x86_reg count)
2151 {
2152  src ++;
2153  dst += count;
2154  src += 2*count;
2155  count= - count;
2156 
2157  if(count < -16) {
2158  count += 16;
2159  __asm__ volatile(
2160  "pcmpeqw %%mm7, %%mm7 \n\t"
2161  "psrlw $8, %%mm7 \n\t"
2162  "1: \n\t"
2163  "movq -32(%1, %0, 2), %%mm0 \n\t"
2164  "movq -24(%1, %0, 2), %%mm1 \n\t"
2165  "movq -16(%1, %0, 2), %%mm2 \n\t"
2166  "movq -8(%1, %0, 2), %%mm3 \n\t"
2167  "pand %%mm7, %%mm0 \n\t"
2168  "pand %%mm7, %%mm1 \n\t"
2169  "pand %%mm7, %%mm2 \n\t"
2170  "pand %%mm7, %%mm3 \n\t"
2171  "packuswb %%mm1, %%mm0 \n\t"
2172  "packuswb %%mm3, %%mm2 \n\t"
2173  MOVNTQ" %%mm0,-16(%2, %0) \n\t"
2174  MOVNTQ" %%mm2,- 8(%2, %0) \n\t"
2175  "add $16, %0 \n\t"
2176  " js 1b \n\t"
2177  : "+r"(count)
2178  : "r"(src), "r"(dst)
2179  );
2180  count -= 16;
2181  }
2182  while(count<0) {
2183  dst[count]= src[2*count];
2184  count++;
2185  }
2186 }
2187 
2188 #if !COMPILE_TEMPLATE_AMD3DNOW
2189 static void RENAME(extract_even2)(const uint8_t *src, uint8_t *dst0, uint8_t *dst1, x86_reg count)
2190 {
2191  dst0+= count;
2192  dst1+= count;
2193  src += 4*count;
2194  count= - count;
2195  if(count <= -8) {
2196  count += 7;
2197  __asm__ volatile(
2198  "pcmpeqw %%mm7, %%mm7 \n\t"
2199  "psrlw $8, %%mm7 \n\t"
2200  "1: \n\t"
2201  "movq -28(%1, %0, 4), %%mm0 \n\t"
2202  "movq -20(%1, %0, 4), %%mm1 \n\t"
2203  "movq -12(%1, %0, 4), %%mm2 \n\t"
2204  "movq -4(%1, %0, 4), %%mm3 \n\t"
2205  "pand %%mm7, %%mm0 \n\t"
2206  "pand %%mm7, %%mm1 \n\t"
2207  "pand %%mm7, %%mm2 \n\t"
2208  "pand %%mm7, %%mm3 \n\t"
2209  "packuswb %%mm1, %%mm0 \n\t"
2210  "packuswb %%mm3, %%mm2 \n\t"
2211  "movq %%mm0, %%mm1 \n\t"
2212  "movq %%mm2, %%mm3 \n\t"
2213  "psrlw $8, %%mm0 \n\t"
2214  "psrlw $8, %%mm2 \n\t"
2215  "pand %%mm7, %%mm1 \n\t"
2216  "pand %%mm7, %%mm3 \n\t"
2217  "packuswb %%mm2, %%mm0 \n\t"
2218  "packuswb %%mm3, %%mm1 \n\t"
2219  MOVNTQ" %%mm0,- 7(%3, %0) \n\t"
2220  MOVNTQ" %%mm1,- 7(%2, %0) \n\t"
2221  "add $8, %0 \n\t"
2222  " js 1b \n\t"
2223  : "+r"(count)
2224  : "r"(src), "r"(dst0), "r"(dst1)
2225  );
2226  count -= 7;
2227  }
2228  while(count<0) {
2229  dst0[count]= src[4*count+0];
2230  dst1[count]= src[4*count+2];
2231  count++;
2232  }
2233 }
2234 #endif /* !COMPILE_TEMPLATE_AMD3DNOW */
2235 
2236 static void RENAME(extract_even2avg)(const uint8_t *src0, const uint8_t *src1, uint8_t *dst0, uint8_t *dst1, x86_reg count)
2237 {
2238  dst0 += count;
2239  dst1 += count;
2240  src0 += 4*count;
2241  src1 += 4*count;
2242  count= - count;
2243 #ifdef PAVGB
2244  if(count <= -8) {
2245  count += 7;
2246  __asm__ volatile(
2247  "pcmpeqw %%mm7, %%mm7 \n\t"
2248  "psrlw $8, %%mm7 \n\t"
2249  "1: \n\t"
2250  "movq -28(%1, %0, 4), %%mm0 \n\t"
2251  "movq -20(%1, %0, 4), %%mm1 \n\t"
2252  "movq -12(%1, %0, 4), %%mm2 \n\t"
2253  "movq -4(%1, %0, 4), %%mm3 \n\t"
2254  PAVGB" -28(%2, %0, 4), %%mm0 \n\t"
2255  PAVGB" -20(%2, %0, 4), %%mm1 \n\t"
2256  PAVGB" -12(%2, %0, 4), %%mm2 \n\t"
2257  PAVGB" - 4(%2, %0, 4), %%mm3 \n\t"
2258  "pand %%mm7, %%mm0 \n\t"
2259  "pand %%mm7, %%mm1 \n\t"
2260  "pand %%mm7, %%mm2 \n\t"
2261  "pand %%mm7, %%mm3 \n\t"
2262  "packuswb %%mm1, %%mm0 \n\t"
2263  "packuswb %%mm3, %%mm2 \n\t"
2264  "movq %%mm0, %%mm1 \n\t"
2265  "movq %%mm2, %%mm3 \n\t"
2266  "psrlw $8, %%mm0 \n\t"
2267  "psrlw $8, %%mm2 \n\t"
2268  "pand %%mm7, %%mm1 \n\t"
2269  "pand %%mm7, %%mm3 \n\t"
2270  "packuswb %%mm2, %%mm0 \n\t"
2271  "packuswb %%mm3, %%mm1 \n\t"
2272  MOVNTQ" %%mm0,- 7(%4, %0) \n\t"
2273  MOVNTQ" %%mm1,- 7(%3, %0) \n\t"
2274  "add $8, %0 \n\t"
2275  " js 1b \n\t"
2276  : "+r"(count)
2277  : "r"(src0), "r"(src1), "r"(dst0), "r"(dst1)
2278  );
2279  count -= 7;
2280  }
2281 #endif
2282  while(count<0) {
2283  dst0[count]= (src0[4*count+0]+src1[4*count+0])>>1;
2284  dst1[count]= (src0[4*count+2]+src1[4*count+2])>>1;
2285  count++;
2286  }
2287 }
2288 
2289 #if !COMPILE_TEMPLATE_AMD3DNOW
2290 static void RENAME(extract_odd2)(const uint8_t *src, uint8_t *dst0, uint8_t *dst1, x86_reg count)
2291 {
2292  dst0+= count;
2293  dst1+= count;
2294  src += 4*count;
2295  count= - count;
2296  if(count <= -8) {
2297  count += 7;
2298  __asm__ volatile(
2299  "pcmpeqw %%mm7, %%mm7 \n\t"
2300  "psrlw $8, %%mm7 \n\t"
2301  "1: \n\t"
2302  "movq -28(%1, %0, 4), %%mm0 \n\t"
2303  "movq -20(%1, %0, 4), %%mm1 \n\t"
2304  "movq -12(%1, %0, 4), %%mm2 \n\t"
2305  "movq -4(%1, %0, 4), %%mm3 \n\t"
2306  "psrlw $8, %%mm0 \n\t"
2307  "psrlw $8, %%mm1 \n\t"
2308  "psrlw $8, %%mm2 \n\t"
2309  "psrlw $8, %%mm3 \n\t"
2310  "packuswb %%mm1, %%mm0 \n\t"
2311  "packuswb %%mm3, %%mm2 \n\t"
2312  "movq %%mm0, %%mm1 \n\t"
2313  "movq %%mm2, %%mm3 \n\t"
2314  "psrlw $8, %%mm0 \n\t"
2315  "psrlw $8, %%mm2 \n\t"
2316  "pand %%mm7, %%mm1 \n\t"
2317  "pand %%mm7, %%mm3 \n\t"
2318  "packuswb %%mm2, %%mm0 \n\t"
2319  "packuswb %%mm3, %%mm1 \n\t"
2320  MOVNTQ" %%mm0,- 7(%3, %0) \n\t"
2321  MOVNTQ" %%mm1,- 7(%2, %0) \n\t"
2322  "add $8, %0 \n\t"
2323  " js 1b \n\t"
2324  : "+r"(count)
2325  : "r"(src), "r"(dst0), "r"(dst1)
2326  );
2327  count -= 7;
2328  }
2329  src++;
2330  while(count<0) {
2331  dst0[count]= src[4*count+0];
2332  dst1[count]= src[4*count+2];
2333  count++;
2334  }
2335 }
2336 #endif /* !COMPILE_TEMPLATE_AMD3DNOW */
2337 
2338 static void RENAME(extract_odd2avg)(const uint8_t *src0, const uint8_t *src1, uint8_t *dst0, uint8_t *dst1, x86_reg count)
2339 {
2340  dst0 += count;
2341  dst1 += count;
2342  src0 += 4*count;
2343  src1 += 4*count;
2344  count= - count;
2345 #ifdef PAVGB
2346  if(count <= -8) {
2347  count += 7;
2348  __asm__ volatile(
2349  "pcmpeqw %%mm7, %%mm7 \n\t"
2350  "psrlw $8, %%mm7 \n\t"
2351  "1: \n\t"
2352  "movq -28(%1, %0, 4), %%mm0 \n\t"
2353  "movq -20(%1, %0, 4), %%mm1 \n\t"
2354  "movq -12(%1, %0, 4), %%mm2 \n\t"
2355  "movq -4(%1, %0, 4), %%mm3 \n\t"
2356  PAVGB" -28(%2, %0, 4), %%mm0 \n\t"
2357  PAVGB" -20(%2, %0, 4), %%mm1 \n\t"
2358  PAVGB" -12(%2, %0, 4), %%mm2 \n\t"
2359  PAVGB" - 4(%2, %0, 4), %%mm3 \n\t"
2360  "psrlw $8, %%mm0 \n\t"
2361  "psrlw $8, %%mm1 \n\t"
2362  "psrlw $8, %%mm2 \n\t"
2363  "psrlw $8, %%mm3 \n\t"
2364  "packuswb %%mm1, %%mm0 \n\t"
2365  "packuswb %%mm3, %%mm2 \n\t"
2366  "movq %%mm0, %%mm1 \n\t"
2367  "movq %%mm2, %%mm3 \n\t"
2368  "psrlw $8, %%mm0 \n\t"
2369  "psrlw $8, %%mm2 \n\t"
2370  "pand %%mm7, %%mm1 \n\t"
2371  "pand %%mm7, %%mm3 \n\t"
2372  "packuswb %%mm2, %%mm0 \n\t"
2373  "packuswb %%mm3, %%mm1 \n\t"
2374  MOVNTQ" %%mm0,- 7(%4, %0) \n\t"
2375  MOVNTQ" %%mm1,- 7(%3, %0) \n\t"
2376  "add $8, %0 \n\t"
2377  " js 1b \n\t"
2378  : "+r"(count)
2379  : "r"(src0), "r"(src1), "r"(dst0), "r"(dst1)
2380  );
2381  count -= 7;
2382  }
2383 #endif
2384  src0++;
2385  src1++;
2386  while(count<0) {
2387  dst0[count]= (src0[4*count+0]+src1[4*count+0])>>1;
2388  dst1[count]= (src0[4*count+2]+src1[4*count+2])>>1;
2389  count++;
2390  }
2391 }
2392 
2393 static void RENAME(yuyvtoyuv420)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src,
2394  int width, int height,
2395  int lumStride, int chromStride, int srcStride)
2396 {
2397  int y;
2398  const int chromWidth = AV_CEIL_RSHIFT(width, 1);
2399 
2400  for (y=0; y<height; y++) {
2401  RENAME(extract_even)(src, ydst, width);
2402  if(y&1) {
2403  RENAME(extract_odd2avg)(src-srcStride, src, udst, vdst, chromWidth);
2404  udst+= chromStride;
2405  vdst+= chromStride;
2406  }
2407 
2408  src += srcStride;
2409  ydst+= lumStride;
2410  }
2411  __asm__(
2412  EMMS" \n\t"
2413  SFENCE" \n\t"
2414  ::: "memory"
2415  );
2416 }
2417 
2418 #if !COMPILE_TEMPLATE_AMD3DNOW
2419 static void RENAME(yuyvtoyuv422)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src,
2420  int width, int height,
2421  int lumStride, int chromStride, int srcStride)
2422 {
2423  int y;
2424  const int chromWidth = AV_CEIL_RSHIFT(width, 1);
2425 
2426  for (y=0; y<height; y++) {
2427  RENAME(extract_even)(src, ydst, width);
2428  RENAME(extract_odd2)(src, udst, vdst, chromWidth);
2429 
2430  src += srcStride;
2431  ydst+= lumStride;
2432  udst+= chromStride;
2433  vdst+= chromStride;
2434  }
2435  __asm__(
2436  EMMS" \n\t"
2437  SFENCE" \n\t"
2438  ::: "memory"
2439  );
2440 }
2441 #endif /* !COMPILE_TEMPLATE_AMD3DNOW */
2442 
2443 static void RENAME(uyvytoyuv420)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src,
2444  int width, int height,
2445  int lumStride, int chromStride, int srcStride)
2446 {
2447  int y;
2448  const int chromWidth = AV_CEIL_RSHIFT(width, 1);
2449 
2450  for (y=0; y<height; y++) {
2451  RENAME(extract_odd)(src, ydst, width);
2452  if(y&1) {
2453  RENAME(extract_even2avg)(src-srcStride, src, udst, vdst, chromWidth);
2454  udst+= chromStride;
2455  vdst+= chromStride;
2456  }
2457 
2458  src += srcStride;
2459  ydst+= lumStride;
2460  }
2461  __asm__(
2462  EMMS" \n\t"
2463  SFENCE" \n\t"
2464  ::: "memory"
2465  );
2466 }
2467 
2468 #if !COMPILE_TEMPLATE_AMD3DNOW
2469 static void RENAME(uyvytoyuv422)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src,
2470  int width, int height,
2471  int lumStride, int chromStride, int srcStride)
2472 {
2473  int y;
2474  const int chromWidth = AV_CEIL_RSHIFT(width, 1);
2475 
2476  for (y=0; y<height; y++) {
2477  RENAME(extract_odd)(src, ydst, width);
2478  RENAME(extract_even2)(src, udst, vdst, chromWidth);
2479 
2480  src += srcStride;
2481  ydst+= lumStride;
2482  udst+= chromStride;
2483  vdst+= chromStride;
2484  }
2485  __asm__(
2486  EMMS" \n\t"
2487  SFENCE" \n\t"
2488  ::: "memory"
2489  );
2490 }
2491 #endif /* !COMPILE_TEMPLATE_AMD3DNOW */
2492 #endif /* !COMPILE_TEMPLATE_SSE2 */
2493 
2494 static av_cold void RENAME(rgb2rgb_init)(void)
2495 {
2496 #if !COMPILE_TEMPLATE_SSE2
2497 #if !COMPILE_TEMPLATE_AMD3DNOW
2498  rgb15to16 = RENAME(rgb15to16);
2499  rgb15tobgr24 = RENAME(rgb15tobgr24);
2500  rgb15to32 = RENAME(rgb15to32);
2501  rgb16tobgr24 = RENAME(rgb16tobgr24);
2502  rgb16to32 = RENAME(rgb16to32);
2503  rgb16to15 = RENAME(rgb16to15);
2504  rgb24tobgr16 = RENAME(rgb24tobgr16);
2505  rgb24tobgr15 = RENAME(rgb24tobgr15);
2506  rgb24tobgr32 = RENAME(rgb24tobgr32);
2507  rgb32to16 = RENAME(rgb32to16);
2508  rgb32to15 = RENAME(rgb32to15);
2509  rgb32tobgr24 = RENAME(rgb32tobgr24);
2510  rgb24to15 = RENAME(rgb24to15);
2511  rgb24to16 = RENAME(rgb24to16);
2512  rgb24tobgr24 = RENAME(rgb24tobgr24);
2513  rgb32tobgr16 = RENAME(rgb32tobgr16);
2514  rgb32tobgr15 = RENAME(rgb32tobgr15);
2515  yv12toyuy2 = RENAME(yv12toyuy2);
2516  yv12touyvy = RENAME(yv12touyvy);
2517  yuv422ptoyuy2 = RENAME(yuv422ptoyuy2);
2518  yuv422ptouyvy = RENAME(yuv422ptouyvy);
2519  yuy2toyv12 = RENAME(yuy2toyv12);
2520  vu9_to_vu12 = RENAME(vu9_to_vu12);
2521  yvu9_to_yuy2 = RENAME(yvu9_to_yuy2);
2522  uyvytoyuv422 = RENAME(uyvytoyuv422);
2523  yuyvtoyuv422 = RENAME(yuyvtoyuv422);
2524 #endif /* !COMPILE_TEMPLATE_AMD3DNOW */
2525 
2526 #if COMPILE_TEMPLATE_MMXEXT || COMPILE_TEMPLATE_AMD3DNOW
2527  planar2x = RENAME(planar2x);
2528 #endif /* COMPILE_TEMPLATE_MMXEXT || COMPILE_TEMPLATE_AMD3DNOW */
2529 #if HAVE_7REGS
2530  ff_rgb24toyv12 = RENAME(rgb24toyv12);
2531 #endif /* HAVE_7REGS */
2532 
2533  yuyvtoyuv420 = RENAME(yuyvtoyuv420);
2534  uyvytoyuv420 = RENAME(uyvytoyuv420);
2535 #endif /* !COMPILE_TEMPLATE_SSE2 */
2536 
2537 #if !COMPILE_TEMPLATE_AMD3DNOW && !COMPILE_TEMPLATE_AVX
2538  interleaveBytes = RENAME(interleaveBytes);
2539 #endif /* !COMPILE_TEMPLATE_AMD3DNOW && !COMPILE_TEMPLATE_AVX */
2540 #if !COMPILE_TEMPLATE_AVX || HAVE_AVX_EXTERNAL
2541 #if !COMPILE_TEMPLATE_AMD3DNOW && (ARCH_X86_32 || COMPILE_TEMPLATE_SSE2) && COMPILE_TEMPLATE_MMXEXT == COMPILE_TEMPLATE_SSE2 && HAVE_X86ASM
2542  deinterleaveBytes = RENAME(deinterleaveBytes);
2543 #endif
2544 #endif
2545 }
STORE_BGR24_MMX
#define STORE_BGR24_MMX
Definition: rgb2rgb_template.c:112
rgb32tobgr24
void(* rgb32tobgr24)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:35
r
const char * r
Definition: vf_curves.c:116
PACK_RGB32
#define PACK_RGB32
Definition: rgb2rgb_template.c:938
yv12toyuy2
void(* yv12toyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, int width, int height, int lumStride, int chromStride, int dstStride)
Height should be a multiple of 2 and width should be a multiple of 16.
Definition: rgb2rgb.c:61
w
uint8_t w
Definition: llviddspenc.c:38
yuy2toyv12
void(* yuy2toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst, int width, int height, int lumStride, int chromStride, int srcStride)
Height should be a multiple of 2 and width should be a multiple of 16.
Definition: rgb2rgb.c:77
b
#define b
Definition: input.c:40
NAMED_CONSTRAINTS_ADD
#define NAMED_CONSTRAINTS_ADD(...)
Definition: asm.h:145
SFENCE
#define SFENCE
Definition: rgb2rgb_template.c:61
rgb32tobgr16
void(* rgb32tobgr16)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:36
yuyvtoyuv422
void(* yuyvtoyuv422)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src, int width, int height, int lumStride, int chromStride, int srcStride)
Definition: rgb2rgb.c:113
rgb24tobgr16
void(* rgb24tobgr16)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:40
rgb15to32
void(* rgb15to32)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:52
yv12touyvy
void(* yv12touyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, int width, int height, int lumStride, int chromStride, int dstStride)
Height should be a multiple of 2 and width should be a multiple of 16.
Definition: rgb2rgb.c:65
rgb32to16
void(* rgb32to16)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:45
rgb
Definition: rpzaenc.c:59
ff_rgb24toyv12
void(* ff_rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst, int width, int height, int lumStride, int chromStride, int srcStride, int32_t *rgb2yuv)
Height should be a multiple of 2 and width should be a multiple of 2.
Definition: rgb2rgb.c:81
av_cold
#define av_cold
Definition: attributes.h:90
rgb16tobgr24
void(* rgb16tobgr24)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:42
width
#define width
s
#define s(width, name)
Definition: cbs_vp9.c:257
AV_CEIL_RSHIFT
#define AV_CEIL_RSHIFT(a, b)
Definition: common.h:51
g
const char * g
Definition: vf_curves.c:117
s1
#define s1
Definition: regdef.h:38
rgb15tobgr24
void(* rgb15tobgr24)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:43
yuv422ptoyuy2
void(* yuv422ptoyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, int width, int height, int lumStride, int chromStride, int dstStride)
Width should be a multiple of 16.
Definition: rgb2rgb.c:69
rgb32tobgr15
void(* rgb32tobgr15)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:37
XMM_CLOBBERS
#define XMM_CLOBBERS(...)
Definition: asm.h:98
interleaveBytes
void(* interleaveBytes)(const uint8_t *src1, const uint8_t *src2, uint8_t *dst, int width, int height, int src1Stride, int src2Stride, int dstStride)
Definition: rgb2rgb.c:88
rgb2yuv
RGB2YUV_SHIFT RGB2YUV_SHIFT RGB2YUV_SHIFT RGB2YUV_SHIFT RGB2YUV_SHIFT RGB2YUV_SHIFT RGB2YUV_SHIFT RGB2YUV_SHIFT uint8_t const uint8_t const uint8_t const uint8_t int uint32_t * rgb2yuv
Definition: input.c:400
yvu9_to_yuy2
void(* yvu9_to_yuy2)(const uint8_t *src1, const uint8_t *src2, const uint8_t *src3, uint8_t *dst, int width, int height, int srcStride1, int srcStride2, int srcStride3, int dstStride)
Definition: rgb2rgb.c:99
NULL
#define NULL
Definition: coverity.c:32
rgb16to15
void(* rgb16to15)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:50
src
#define src
Definition: vp8dsp.c:255
yuyvtoyuv420
void(* yuyvtoyuv420)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src, int width, int height, int lumStride, int chromStride, int srcStride)
Definition: rgb2rgb.c:110
rgb24tobgr32
void(* rgb24tobgr32)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:38
s2
#define s2
Definition: regdef.h:39
PAVGB
#define PAVGB(a, b)
Definition: postprocess_template.c:88
asm.h
height
#define height
ff_rgb24toyv12_c
void ff_rgb24toyv12_c(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst, int width, int height, int lumStride, int chromStride, int srcStride, int32_t *rgb2yuv)
Height should be a multiple of 2 and width should be a multiple of 2.
Definition: rgb2rgb_template.c:649
attributes.h
rgb24to16
void(* rgb24to16)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:47
EMMS
#define EMMS
Definition: rgb2rgb_template.c:53
src0
#define src0
Definition: h264pred.c:139
uyvytoyuv422
void(* uyvytoyuv422)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src, int width, int height, int lumStride, int chromStride, int srcStride)
Definition: rgb2rgb.c:107
src1
#define src1
Definition: h264pred.c:140
i
int i
Definition: input.c:406
rgb24to15
void(* rgb24to15)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:48
PREFETCH
#define PREFETCH
Definition: rgb2rgb_template.c:46
RENAME
#define RENAME(name)
Definition: ffv1.h:195
rgb32to15
void(* rgb32to15)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:46
__asm__
__asm__(".macro parse_r var r\n\t" "\\var = -1\n\t" _IFC_REG(0) _IFC_REG(1) _IFC_REG(2) _IFC_REG(3) _IFC_REG(4) _IFC_REG(5) _IFC_REG(6) _IFC_REG(7) _IFC_REG(8) _IFC_REG(9) _IFC_REG(10) _IFC_REG(11) _IFC_REG(12) _IFC_REG(13) _IFC_REG(14) _IFC_REG(15) _IFC_REG(16) _IFC_REG(17) _IFC_REG(18) _IFC_REG(19) _IFC_REG(20) _IFC_REG(21) _IFC_REG(22) _IFC_REG(23) _IFC_REG(24) _IFC_REG(25) _IFC_REG(26) _IFC_REG(27) _IFC_REG(28) _IFC_REG(29) _IFC_REG(30) _IFC_REG(31) ".iflt \\var\n\t" ".error \"Unable to parse register name \\r\"\n\t" ".endif\n\t" ".endm")
deinterleaveBytes
void(* deinterleaveBytes)(const uint8_t *src, uint8_t *dst1, uint8_t *dst2, int width, int height, int srcStride, int dst1Stride, int dst2Stride)
Definition: rgb2rgb.c:91
uyvytoyuv420
void(* uyvytoyuv420)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src, int width, int height, int lumStride, int chromStride, int srcStride)
Definition: rgb2rgb.c:104
rgb16to32
void(* rgb16to32)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:49
rgb24tobgr15
void(* rgb24tobgr15)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:41
dstV
uint16_t * dstV
Definition: input.c:402
rgb15to16
void(* rgb15to16)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:51
MOVNTQ
#define MOVNTQ
Definition: rgb2rgb_template.c:60
yuv422ptouyvy
void(* yuv422ptouyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, int width, int height, int lumStride, int chromStride, int dstStride)
Width should be a multiple of 16.
Definition: rgb2rgb.c:73
vu9_to_vu12
void(* vu9_to_vu12)(const uint8_t *src1, const uint8_t *src2, uint8_t *dst1, uint8_t *dst2, int width, int height, int srcStride1, int srcStride2, int dstStride1, int dstStride2)
Definition: rgb2rgb.c:94
rgb24tobgr24
void(* rgb24tobgr24)(const uint8_t *src, uint8_t *dst, int src_size)
Definition: rgb2rgb.c:39
MANGLE
#define MANGLE(a)
Definition: asm.h:127
x86_reg
int x86_reg
Definition: asm.h:72
d
d
Definition: ffmpeg_filter.c:156
int32_t
int32_t
Definition: audioconvert.c:56
planar2x
void(* planar2x)(const uint8_t *src, uint8_t *dst, int width, int height, int srcStride, int dstStride)
Definition: rgb2rgb.c:86
h
h
Definition: vp9dsp_template.c:2038
Generated on Sat Sep 25 2021 19:22:16 for FFmpeg by   doxygen 1.8.17