Mercurial > mplayer.hg

comparison postproc/rgb2rgb_template.c @ 6096:f38c7228a094

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
fixing end overwrite bugs (some at least) this needs testing, untested parts should be reversed before release i tested 24->32, 15->16, bgr24->rgb, bgr32->rgb (retesting them isnt a bad idea either ...)
author michael
date Wed, 15 May 2002 02:27:41 +0000
parents f4f3cfcd0d64
children c5cf988c6d6f
comparison
equal deleted inserted replaced
6095:891cff8aba60 6096:f38c7228a094
54 { 54 {
55 uint8_t *dest = dst; 55 uint8_t *dest = dst;
56 const uint8_t *s = src; 56 const uint8_t *s = src;
57 const uint8_t *end; 57 const uint8_t *end;
58 #ifdef HAVE_MMX 58 #ifdef HAVE_MMX
59 uint8_t *mm_end; 59 const uint8_t *mm_end;
60 #endif 60 #endif
61 end = s + src_size; 61 end = s + src_size;
62 #ifdef HAVE_MMX 62 #ifdef HAVE_MMX
63 __asm __volatile(PREFETCH" %0"::"m"(*s):"memory"); 63 __asm __volatile(PREFETCH" %0"::"m"(*s):"memory");
64 mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*4))*(MMREG_SIZE*4)); 64 mm_end = end - 23;
65 __asm __volatile("movq %0, %%mm7"::"m"(mask32):"memory"); 65 __asm __volatile("movq %0, %%mm7"::"m"(mask32):"memory");
66 if(mm_end == end) mm_end -= MMREG_SIZE*4;
67 while(s < mm_end) 66 while(s < mm_end)
68 { 67 {
69 __asm __volatile( 68 __asm __volatile(
70 PREFETCH" 32%1\n\t" 69 PREFETCH" 32%1\n\t"
71 "movd %1, %%mm0\n\t" 70 "movd %1, %%mm0\n\t"
106 { 105 {
107 uint8_t *dest = dst; 106 uint8_t *dest = dst;
108 const uint8_t *s = src; 107 const uint8_t *s = src;
109 const uint8_t *end; 108 const uint8_t *end;
110 #ifdef HAVE_MMX 109 #ifdef HAVE_MMX
111 uint8_t *mm_end; 110 const uint8_t *mm_end;
112 #endif 111 #endif
113 end = s + src_size; 112 end = s + src_size;
114 #ifdef HAVE_MMX 113 #ifdef HAVE_MMX
115 __asm __volatile(PREFETCH" %0"::"m"(*s):"memory"); 114 __asm __volatile(PREFETCH" %0"::"m"(*s):"memory");
116 mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*4))*(MMREG_SIZE*4)); 115 mm_end = end - 31;
117 while(s < mm_end) 116 while(s < mm_end)
118 { 117 {
119 __asm __volatile( 118 __asm __volatile(
120 PREFETCH" 32%1\n\t" 119 PREFETCH" 32%1\n\t"
121 "movq %1, %%mm0\n\t" 120 "movq %1, %%mm0\n\t"
186 32bit c version, and and&add trick by Michael Niedermayer 185 32bit c version, and and&add trick by Michael Niedermayer
187 */ 186 */
188 static inline void RENAME(rgb15to16)(const uint8_t *src,uint8_t *dst,unsigned src_size) 187 static inline void RENAME(rgb15to16)(const uint8_t *src,uint8_t *dst,unsigned src_size)
189 { 188 {
190 #ifdef HAVE_MMX 189 #ifdef HAVE_MMX
191 register const char* s=src+src_size; 190 register int offs=15-src_size;
192 register char* d=dst+src_size; 191 register const char* s=src-offs;
193 register int offs=-src_size; 192 register char* d=dst-offs;
194 __asm __volatile(PREFETCH" %0"::"m"(*(s+offs))); 193 __asm __volatile(PREFETCH" %0"::"m"(*(s+offs)));
195 __asm __volatile( 194 __asm __volatile(
196 "movq %0, %%mm4\n\t" 195 "movq %0, %%mm4\n\t"
197 ::"m"(mask15s)); 196 ::"m"(mask15s));
198 while(offs<0) 197 while(offs<0)
250 #ifdef HAVE_MMX 249 #ifdef HAVE_MMX
251 const uint8_t *s = src; 250 const uint8_t *s = src;
252 const uint8_t *end,*mm_end; 251 const uint8_t *end,*mm_end;
253 uint16_t *d = (uint16_t *)dst; 252 uint16_t *d = (uint16_t *)dst;
254 end = s + src_size; 253 end = s + src_size;
255 mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2)); 254 mm_end = end - 15;
256 __asm __volatile(PREFETCH" %0"::"m"(*src):"memory"); 255 __asm __volatile(PREFETCH" %0"::"m"(*src):"memory");
257 __asm __volatile( 256 __asm __volatile(
258 "movq %0, %%mm7\n\t" 257 "movq %0, %%mm7\n\t"
259 "movq %1, %%mm6\n\t" 258 "movq %1, %%mm6\n\t"
260 ::"m"(red_16mask),"m"(green_16mask)); 259 ::"m"(red_16mask),"m"(green_16mask));
296 while(s < end) 295 while(s < end)
297 { 296 {
298 const int b= *s++; 297 const int b= *s++;
299 const int g= *s++; 298 const int g= *s++;
300 const int r= *s++; 299 const int r= *s++;
300 s++;
301 *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8); 301 *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
302 } 302 }
303 __asm __volatile(SFENCE:::"memory"); 303 __asm __volatile(SFENCE:::"memory");
304 __asm __volatile(EMMS:::"memory"); 304 __asm __volatile(EMMS:::"memory");
305 #else 305 #else
321 #ifdef HAVE_MMX 321 #ifdef HAVE_MMX
322 const uint8_t *s = src; 322 const uint8_t *s = src;
323 const uint8_t *end,*mm_end; 323 const uint8_t *end,*mm_end;
324 uint16_t *d = (uint16_t *)dst; 324 uint16_t *d = (uint16_t *)dst;
325 end = s + src_size; 325 end = s + src_size;
326 mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2)); 326 mm_end = end - 15;
327 __asm __volatile(PREFETCH" %0"::"m"(*src):"memory"); 327 __asm __volatile(PREFETCH" %0"::"m"(*src):"memory");
328 __asm __volatile( 328 __asm __volatile(
329 "movq %0, %%mm7\n\t" 329 "movq %0, %%mm7\n\t"
330 "movq %1, %%mm6\n\t" 330 "movq %1, %%mm6\n\t"
331 ::"m"(red_15mask),"m"(green_15mask)); 331 ::"m"(red_15mask),"m"(green_15mask));
367 while(s < end) 367 while(s < end)
368 { 368 {
369 const int b= *s++; 369 const int b= *s++;
370 const int g= *s++; 370 const int g= *s++;
371 const int r= *s++; 371 const int r= *s++;
372 s++;
372 *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7); 373 *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
373 } 374 }
374 __asm __volatile(SFENCE:::"memory"); 375 __asm __volatile(SFENCE:::"memory");
375 __asm __volatile(EMMS:::"memory"); 376 __asm __volatile(EMMS:::"memory");
376 #else 377 #else
392 #ifdef HAVE_MMX 393 #ifdef HAVE_MMX
393 const uint8_t *s = src; 394 const uint8_t *s = src;
394 const uint8_t *end,*mm_end; 395 const uint8_t *end,*mm_end;
395 uint16_t *d = (uint16_t *)dst; 396 uint16_t *d = (uint16_t *)dst;
396 end = s + src_size; 397 end = s + src_size;
397 mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2)); 398 mm_end = end - 11;
398 __asm __volatile(PREFETCH" %0"::"m"(*src):"memory"); 399 __asm __volatile(PREFETCH" %0"::"m"(*src):"memory");
399 __asm __volatile( 400 __asm __volatile(
400 "movq %0, %%mm7\n\t" 401 "movq %0, %%mm7\n\t"
401 "movq %1, %%mm6\n\t" 402 "movq %1, %%mm6\n\t"
402 ::"m"(red_16mask),"m"(green_16mask)); 403 ::"m"(red_16mask),"m"(green_16mask));
403 if(mm_end == end) mm_end -= MMREG_SIZE*2;
404 while(s < mm_end) 404 while(s < mm_end)
405 { 405 {
406 __asm __volatile( 406 __asm __volatile(
407 PREFETCH" 32%1\n\t" 407 PREFETCH" 32%1\n\t"
408 "movd %1, %%mm0\n\t" 408 "movd %1, %%mm0\n\t"
464 #ifdef HAVE_MMX 464 #ifdef HAVE_MMX
465 const uint8_t *s = src; 465 const uint8_t *s = src;
466 const uint8_t *end,*mm_end; 466 const uint8_t *end,*mm_end;
467 uint16_t *d = (uint16_t *)dst; 467 uint16_t *d = (uint16_t *)dst;
468 end = s + src_size; 468 end = s + src_size;
469 mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2)); 469 mm_end = end -11;
470 __asm __volatile(PREFETCH" %0"::"m"(*src):"memory"); 470 __asm __volatile(PREFETCH" %0"::"m"(*src):"memory");
471 __asm __volatile( 471 __asm __volatile(
472 "movq %0, %%mm7\n\t" 472 "movq %0, %%mm7\n\t"
473 "movq %1, %%mm6\n\t" 473 "movq %1, %%mm6\n\t"
474 ::"m"(red_15mask),"m"(green_15mask)); 474 ::"m"(red_15mask),"m"(green_15mask));
475 if(mm_end == end) mm_end -= MMREG_SIZE*2;
476 while(s < mm_end) 475 while(s < mm_end)
477 { 476 {
478 __asm __volatile( 477 __asm __volatile(
479 PREFETCH" 32%1\n\t" 478 PREFETCH" 32%1\n\t"
480 "movd %1, %%mm0\n\t" 479 "movd %1, %%mm0\n\t"
548 "pand "MANGLE(mask32g)", %%mm2 \n\t" 547 "pand "MANGLE(mask32g)", %%mm2 \n\t"
549 "pand "MANGLE(mask32b)", %%mm1 \n\t" 548 "pand "MANGLE(mask32b)", %%mm1 \n\t"
550 "por %%mm0, %%mm2 \n\t" 549 "por %%mm0, %%mm2 \n\t"
551 "por %%mm1, %%mm2 \n\t" 550 "por %%mm1, %%mm2 \n\t"
552 MOVNTQ" %%mm2, (%1, %%eax) \n\t" 551 MOVNTQ" %%mm2, (%1, %%eax) \n\t"
553 "addl $2, %%eax \n\t" 552 "addl $8, %%eax \n\t"
554 "cmpl %2, %%eax \n\t" 553 "cmpl %2, %%eax \n\t"
555 " jb 1b \n\t" 554 " jb 1b \n\t"
556 :: "r" (src), "r"(dst), "r" (src_size) 555 :: "r" (src), "r"(dst), "r" (src_size)
557 : "%eax" 556 : "%eax"
558 ); 557 );
618 ); 617 );
619 618
620 __asm __volatile(SFENCE:::"memory"); 619 __asm __volatile(SFENCE:::"memory");
621 __asm __volatile(EMMS:::"memory"); 620 __asm __volatile(EMMS:::"memory");
622 621
623 if(!mmx_size) return; //finihsed, was multiple of 8 622 if(mmx_size==23) return; //finihsed, was multiple of 8
624
625 src+= src_size; 623 src+= src_size;
626 dst+= src_size; 624 dst+= src_size;
627 src_size= 24-mmx_size; 625 src_size= 23 - mmx_size;
628 src-= src_size; 626 src-= src_size;
629 dst-= src_size; 627 dst-= src_size;
630 #endif 628 #endif
631 for(i=0; i<src_size; i+=3) 629 for(i=0; i<src_size; i+=3)
632 { 630 {