rgb2rgb_template.c
上传用户:riyaled888
上传日期:2009-03-27
资源大小:7338k
文件大小:66k
- /*
- *
- * rgb2rgb.c, Software RGB to RGB convertor
- * pluralize by Software PAL8 to RGB convertor
- * Software YUV to YUV convertor
- * Software YUV to RGB convertor
- * Written by Nick Kurshev.
- * palette & yuv & runtime cpu stuff by Michael (michaelni@gmx.at) (under GPL)
- */
- #include <stddef.h>
- #include <inttypes.h> /* for __WORDSIZE */
- #ifndef __WORDSIZE
- // #warning You have misconfigured system and probably will lose performance!
- #define __WORDSIZE MP_WORDSIZE
- #endif
- #undef PREFETCH
- #undef MOVNTQ
- #undef EMMS
- #undef SFENCE
- #undef MMREG_SIZE
- #undef PREFETCHW
- #undef PAVGB
- #ifdef HAVE_SSE2
- #define MMREG_SIZE 16
- #else
- #define MMREG_SIZE 8
- #endif
- #ifdef HAVE_3DNOW
- #define PREFETCH "prefetch"
- #define PREFETCHW "prefetchw"
- #define PAVGB "pavgusb"
- #elif defined ( HAVE_MMX2 )
- #define PREFETCH "prefetchnta"
- #define PREFETCHW "prefetcht0"
- #define PAVGB "pavgb"
- #else
- #define PREFETCH "/nop"
- #define PREFETCHW "/nop"
- #endif
- #ifdef HAVE_3DNOW
- /* On K6 femms is faster of emms. On K7 femms is directly mapped on emms. */
- #define EMMS "femms"
- #else
- #define EMMS "emms"
- #endif
- #ifdef HAVE_MMX2
- #define MOVNTQ "movntq"
- #define SFENCE "sfence"
- #else
- #define MOVNTQ "movq"
- #define SFENCE "/nop"
- #endif
- static inline void RENAME(rgb24to32)(const uint8_t *src,uint8_t *dst,unsigned src_size)
- {
- uint8_t *dest = dst;
- const uint8_t *s = src;
- const uint8_t *end;
- #ifdef HAVE_MMX
- const uint8_t *mm_end;
- #endif
- end = s + src_size;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*s):"memory");
- mm_end = end - 23;
- __asm __volatile("movq %0, %%mm7"::"m"(mask32):"memory");
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movd %1, %%mm0nt"
- "punpckldq 3%1, %%mm0nt"
- "movd 6%1, %%mm1nt"
- "punpckldq 9%1, %%mm1nt"
- "movd 12%1, %%mm2nt"
- "punpckldq 15%1, %%mm2nt"
- "movd 18%1, %%mm3nt"
- "punpckldq 21%1, %%mm3nt"
- "pand %%mm7, %%mm0nt"
- "pand %%mm7, %%mm1nt"
- "pand %%mm7, %%mm2nt"
- "pand %%mm7, %%mm3nt"
- MOVNTQ" %%mm0, %0nt"
- MOVNTQ" %%mm1, 8%0nt"
- MOVNTQ" %%mm2, 16%0nt"
- MOVNTQ" %%mm3, 24%0"
- :"=m"(*dest)
- :"m"(*s)
- :"memory");
- dest += 32;
- s += 24;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- *dest++ = *s++;
- *dest++ = *s++;
- *dest++ = *s++;
- *dest++ = 0;
- }
- }
- static inline void RENAME(rgb32to24)(const uint8_t *src,uint8_t *dst,unsigned src_size)
- {
- uint8_t *dest = dst;
- const uint8_t *s = src;
- const uint8_t *end;
- #ifdef HAVE_MMX
- const uint8_t *mm_end;
- #endif
- end = s + src_size;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*s):"memory");
- mm_end = end - 31;
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movq %1, %%mm0nt"
- "movq 8%1, %%mm1nt"
- "movq 16%1, %%mm4nt"
- "movq 24%1, %%mm5nt"
- "movq %%mm0, %%mm2nt"
- "movq %%mm1, %%mm3nt"
- "movq %%mm4, %%mm6nt"
- "movq %%mm5, %%mm7nt"
- "psrlq $8, %%mm2nt"
- "psrlq $8, %%mm3nt"
- "psrlq $8, %%mm6nt"
- "psrlq $8, %%mm7nt"
- "pand %2, %%mm0nt"
- "pand %2, %%mm1nt"
- "pand %2, %%mm4nt"
- "pand %2, %%mm5nt"
- "pand %3, %%mm2nt"
- "pand %3, %%mm3nt"
- "pand %3, %%mm6nt"
- "pand %3, %%mm7nt"
- "por %%mm2, %%mm0nt"
- "por %%mm3, %%mm1nt"
- "por %%mm6, %%mm4nt"
- "por %%mm7, %%mm5nt"
- "movq %%mm1, %%mm2nt"
- "movq %%mm4, %%mm3nt"
- "psllq $48, %%mm2nt"
- "psllq $32, %%mm3nt"
- "pand %4, %%mm2nt"
- "pand %5, %%mm3nt"
- "por %%mm2, %%mm0nt"
- "psrlq $16, %%mm1nt"
- "psrlq $32, %%mm4nt"
- "psllq $16, %%mm5nt"
- "por %%mm3, %%mm1nt"
- "pand %6, %%mm5nt"
- "por %%mm5, %%mm4nt"
- MOVNTQ" %%mm0, %0nt"
- MOVNTQ" %%mm1, 8%0nt"
- MOVNTQ" %%mm4, 16%0"
- :"=m"(*dest)
- :"m"(*s),"m"(mask24l),
- "m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)
- :"memory");
- dest += 24;
- s += 32;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- *dest++ = *s++;
- *dest++ = *s++;
- *dest++ = *s++;
- s++;
- }
- }
- /*
- Original by Strepto/Astral
- ported to gcc & bugfixed : A'rpi
- MMX2, 3DNOW optimization by Nick Kurshev
- 32bit c version, and and&add trick by Michael Niedermayer
- */
- static inline void RENAME(rgb15to16)(const uint8_t *src,uint8_t *dst,unsigned src_size)
- {
- register const uint8_t* s=src;
- register uint8_t* d=dst;
- register const uint8_t *end;
- const uint8_t *mm_end;
- end = s + src_size;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*s));
- __asm __volatile("movq %0, %%mm4"::"m"(mask15s));
- mm_end = end - 15;
- while(s<mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movq %1, %%mm0nt"
- "movq 8%1, %%mm2nt"
- "movq %%mm0, %%mm1nt"
- "movq %%mm2, %%mm3nt"
- "pand %%mm4, %%mm0nt"
- "pand %%mm4, %%mm2nt"
- "paddw %%mm1, %%mm0nt"
- "paddw %%mm3, %%mm2nt"
- MOVNTQ" %%mm0, %0nt"
- MOVNTQ" %%mm2, 8%0"
- :"=m"(*d)
- :"m"(*s)
- );
- d+=16;
- s+=16;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- mm_end = end - 3;
- while(s < mm_end)
- {
- register unsigned x= *((uint32_t *)s);
- *((uint32_t *)d) = (x&0x7FFF7FFF) + (x&0x7FE07FE0);
- d+=4;
- s+=4;
- }
- if(s < end)
- {
- register unsigned short x= *((uint16_t *)s);
- *((uint16_t *)d) = (x&0x7FFF) + (x&0x7FE0);
- }
- }
- static inline void RENAME(rgb16to15)(const uint8_t *src,uint8_t *dst,unsigned src_size)
- {
- register const uint8_t* s=src;
- register uint8_t* d=dst;
- register const uint8_t *end;
- const uint8_t *mm_end;
- end = s + src_size;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*s));
- __asm __volatile("movq %0, %%mm7"::"m"(mask15rg));
- __asm __volatile("movq %0, %%mm6"::"m"(mask15b));
- mm_end = end - 15;
- while(s<mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movq %1, %%mm0nt"
- "movq 8%1, %%mm2nt"
- "movq %%mm0, %%mm1nt"
- "movq %%mm2, %%mm3nt"
- "psrlq $1, %%mm0nt"
- "psrlq $1, %%mm2nt"
- "pand %%mm7, %%mm0nt"
- "pand %%mm7, %%mm2nt"
- "pand %%mm6, %%mm1nt"
- "pand %%mm6, %%mm3nt"
- "por %%mm1, %%mm0nt"
- "por %%mm3, %%mm2nt"
- MOVNTQ" %%mm0, %0nt"
- MOVNTQ" %%mm2, 8%0"
- :"=m"(*d)
- :"m"(*s)
- );
- d+=16;
- s+=16;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- mm_end = end - 3;
- while(s < mm_end)
- {
- register uint32_t x= *((uint32_t *)s);
- *((uint32_t *)d) = ((x>>1)&0x7FE07FE0) | (x&0x001F001F);
- s+=4;
- d+=4;
- }
- if(s < end)
- {
- register uint16_t x= *((uint16_t *)s);
- *((uint16_t *)d) = ((x>>1)&0x7FE0) | (x&0x001F);
- s+=2;
- d+=2;
- }
- }
- static inline void RENAME(rgb32to16)(const uint8_t *src, uint8_t *dst, unsigned src_size)
- {
- const uint8_t *s = src;
- const uint8_t *end;
- #ifdef HAVE_MMX
- const uint8_t *mm_end;
- #endif
- uint16_t *d = (uint16_t *)dst;
- end = s + src_size;
- #ifdef HAVE_MMX
- mm_end = end - 15;
- #if 1 //is faster only if multiplies are reasonable fast (FIXME figure out on which cpus this is faster, on Athlon its slightly faster)
- asm volatile(
- "movq %3, %%mm5 nt"
- "movq %4, %%mm6 nt"
- "movq %5, %%mm7 nt"
- ".balign 16 nt"
- "1: nt"
- PREFETCH" 32(%1) nt"
- "movd (%1), %%mm0 nt"
- "movd 4(%1), %%mm3 nt"
- "punpckldq 8(%1), %%mm0 nt"
- "punpckldq 12(%1), %%mm3 nt"
- "movq %%mm0, %%mm1 nt"
- "movq %%mm3, %%mm4 nt"
- "pand %%mm6, %%mm0 nt"
- "pand %%mm6, %%mm3 nt"
- "pmaddwd %%mm7, %%mm0 nt"
- "pmaddwd %%mm7, %%mm3 nt"
- "pand %%mm5, %%mm1 nt"
- "pand %%mm5, %%mm4 nt"
- "por %%mm1, %%mm0 nt"
- "por %%mm4, %%mm3 nt"
- "psrld $5, %%mm0 nt"
- "pslld $11, %%mm3 nt"
- "por %%mm3, %%mm0 nt"
- MOVNTQ" %%mm0, (%0) nt"
- "addl $16, %1 nt"
- "addl $8, %0 nt"
- "cmpl %2, %1 nt"
- " jb 1b nt"
- : "+r" (d), "+r"(s)
- : "r" (mm_end), "m" (mask3216g), "m" (mask3216br), "m" (mul3216)
- );
- #else
- __asm __volatile(PREFETCH" %0"::"m"(*src):"memory");
- __asm __volatile(
- "movq %0, %%mm7nt"
- "movq %1, %%mm6nt"
- ::"m"(red_16mask),"m"(green_16mask));
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movd %1, %%mm0nt"
- "movd 4%1, %%mm3nt"
- "punpckldq 8%1, %%mm0nt"
- "punpckldq 12%1, %%mm3nt"
- "movq %%mm0, %%mm1nt"
- "movq %%mm0, %%mm2nt"
- "movq %%mm3, %%mm4nt"
- "movq %%mm3, %%mm5nt"
- "psrlq $3, %%mm0nt"
- "psrlq $3, %%mm3nt"
- "pand %2, %%mm0nt"
- "pand %2, %%mm3nt"
- "psrlq $5, %%mm1nt"
- "psrlq $5, %%mm4nt"
- "pand %%mm6, %%mm1nt"
- "pand %%mm6, %%mm4nt"
- "psrlq $8, %%mm2nt"
- "psrlq $8, %%mm5nt"
- "pand %%mm7, %%mm2nt"
- "pand %%mm7, %%mm5nt"
- "por %%mm1, %%mm0nt"
- "por %%mm4, %%mm3nt"
- "por %%mm2, %%mm0nt"
- "por %%mm5, %%mm3nt"
- "psllq $16, %%mm3nt"
- "por %%mm3, %%mm0nt"
- MOVNTQ" %%mm0, %0nt"
- :"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
- d += 4;
- s += 16;
- }
- #endif
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- const int src= *s; s += 4;
- *d++ = ((src&0xFF)>>3) + ((src&0xFC00)>>5) + ((src&0xF80000)>>8);
- // *d++ = ((src>>3)&0x1F) + ((src>>5)&0x7E0) + ((src>>8)&0xF800);
- }
- }
- static inline void RENAME(rgb32tobgr16)(const uint8_t *src, uint8_t *dst, unsigned int src_size)
- {
- const uint8_t *s = src;
- const uint8_t *end;
- #ifdef HAVE_MMX
- const uint8_t *mm_end;
- #endif
- uint16_t *d = (uint16_t *)dst;
- end = s + src_size;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*src):"memory");
- __asm __volatile(
- "movq %0, %%mm7nt"
- "movq %1, %%mm6nt"
- ::"m"(red_16mask),"m"(green_16mask));
- mm_end = end - 15;
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movd %1, %%mm0nt"
- "movd 4%1, %%mm3nt"
- "punpckldq 8%1, %%mm0nt"
- "punpckldq 12%1, %%mm3nt"
- "movq %%mm0, %%mm1nt"
- "movq %%mm0, %%mm2nt"
- "movq %%mm3, %%mm4nt"
- "movq %%mm3, %%mm5nt"
- "psllq $8, %%mm0nt"
- "psllq $8, %%mm3nt"
- "pand %%mm7, %%mm0nt"
- "pand %%mm7, %%mm3nt"
- "psrlq $5, %%mm1nt"
- "psrlq $5, %%mm4nt"
- "pand %%mm6, %%mm1nt"
- "pand %%mm6, %%mm4nt"
- "psrlq $19, %%mm2nt"
- "psrlq $19, %%mm5nt"
- "pand %2, %%mm2nt"
- "pand %2, %%mm5nt"
- "por %%mm1, %%mm0nt"
- "por %%mm4, %%mm3nt"
- "por %%mm2, %%mm0nt"
- "por %%mm5, %%mm3nt"
- "psllq $16, %%mm3nt"
- "por %%mm3, %%mm0nt"
- MOVNTQ" %%mm0, %0nt"
- :"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
- d += 4;
- s += 16;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- const int src= *s; s += 4;
- *d++ = ((src&0xF8)<<8) + ((src&0xFC00)>>5) + ((src&0xF80000)>>19);
- }
- }
- static inline void RENAME(rgb32to15)(const uint8_t *src, uint8_t *dst, unsigned src_size)
- {
- const uint8_t *s = src;
- const uint8_t *end;
- #ifdef HAVE_MMX
- const uint8_t *mm_end;
- #endif
- uint16_t *d = (uint16_t *)dst;
- end = s + src_size;
- #ifdef HAVE_MMX
- mm_end = end - 15;
- #if 1 //is faster only if multiplies are reasonable fast (FIXME figure out on which cpus this is faster, on Athlon its slightly faster)
- asm volatile(
- "movq %3, %%mm5 nt"
- "movq %4, %%mm6 nt"
- "movq %5, %%mm7 nt"
- ".balign 16 nt"
- "1: nt"
- PREFETCH" 32(%1) nt"
- "movd (%1), %%mm0 nt"
- "movd 4(%1), %%mm3 nt"
- "punpckldq 8(%1), %%mm0 nt"
- "punpckldq 12(%1), %%mm3 nt"
- "movq %%mm0, %%mm1 nt"
- "movq %%mm3, %%mm4 nt"
- "pand %%mm6, %%mm0 nt"
- "pand %%mm6, %%mm3 nt"
- "pmaddwd %%mm7, %%mm0 nt"
- "pmaddwd %%mm7, %%mm3 nt"
- "pand %%mm5, %%mm1 nt"
- "pand %%mm5, %%mm4 nt"
- "por %%mm1, %%mm0 nt"
- "por %%mm4, %%mm3 nt"
- "psrld $6, %%mm0 nt"
- "pslld $10, %%mm3 nt"
- "por %%mm3, %%mm0 nt"
- MOVNTQ" %%mm0, (%0) nt"
- "addl $16, %1 nt"
- "addl $8, %0 nt"
- "cmpl %2, %1 nt"
- " jb 1b nt"
- : "+r" (d), "+r"(s)
- : "r" (mm_end), "m" (mask3215g), "m" (mask3216br), "m" (mul3215)
- );
- #else
- __asm __volatile(PREFETCH" %0"::"m"(*src):"memory");
- __asm __volatile(
- "movq %0, %%mm7nt"
- "movq %1, %%mm6nt"
- ::"m"(red_15mask),"m"(green_15mask));
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movd %1, %%mm0nt"
- "movd 4%1, %%mm3nt"
- "punpckldq 8%1, %%mm0nt"
- "punpckldq 12%1, %%mm3nt"
- "movq %%mm0, %%mm1nt"
- "movq %%mm0, %%mm2nt"
- "movq %%mm3, %%mm4nt"
- "movq %%mm3, %%mm5nt"
- "psrlq $3, %%mm0nt"
- "psrlq $3, %%mm3nt"
- "pand %2, %%mm0nt"
- "pand %2, %%mm3nt"
- "psrlq $6, %%mm1nt"
- "psrlq $6, %%mm4nt"
- "pand %%mm6, %%mm1nt"
- "pand %%mm6, %%mm4nt"
- "psrlq $9, %%mm2nt"
- "psrlq $9, %%mm5nt"
- "pand %%mm7, %%mm2nt"
- "pand %%mm7, %%mm5nt"
- "por %%mm1, %%mm0nt"
- "por %%mm4, %%mm3nt"
- "por %%mm2, %%mm0nt"
- "por %%mm5, %%mm3nt"
- "psllq $16, %%mm3nt"
- "por %%mm3, %%mm0nt"
- MOVNTQ" %%mm0, %0nt"
- :"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
- d += 4;
- s += 16;
- }
- #endif
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- const int src= *s; s += 4;
- *d++ = ((src&0xFF)>>3) + ((src&0xF800)>>6) + ((src&0xF80000)>>9);
- }
- }
- static inline void RENAME(rgb32tobgr15)(const uint8_t *src, uint8_t *dst, unsigned src_size)
- {
- const uint8_t *s = src;
- const uint8_t *end;
- #ifdef HAVE_MMX
- const uint8_t *mm_end;
- #endif
- uint16_t *d = (uint16_t *)dst;
- end = s + src_size;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*src):"memory");
- __asm __volatile(
- "movq %0, %%mm7nt"
- "movq %1, %%mm6nt"
- ::"m"(red_15mask),"m"(green_15mask));
- mm_end = end - 15;
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movd %1, %%mm0nt"
- "movd 4%1, %%mm3nt"
- "punpckldq 8%1, %%mm0nt"
- "punpckldq 12%1, %%mm3nt"
- "movq %%mm0, %%mm1nt"
- "movq %%mm0, %%mm2nt"
- "movq %%mm3, %%mm4nt"
- "movq %%mm3, %%mm5nt"
- "psllq $7, %%mm0nt"
- "psllq $7, %%mm3nt"
- "pand %%mm7, %%mm0nt"
- "pand %%mm7, %%mm3nt"
- "psrlq $6, %%mm1nt"
- "psrlq $6, %%mm4nt"
- "pand %%mm6, %%mm1nt"
- "pand %%mm6, %%mm4nt"
- "psrlq $19, %%mm2nt"
- "psrlq $19, %%mm5nt"
- "pand %2, %%mm2nt"
- "pand %2, %%mm5nt"
- "por %%mm1, %%mm0nt"
- "por %%mm4, %%mm3nt"
- "por %%mm2, %%mm0nt"
- "por %%mm5, %%mm3nt"
- "psllq $16, %%mm3nt"
- "por %%mm3, %%mm0nt"
- MOVNTQ" %%mm0, %0nt"
- :"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
- d += 4;
- s += 16;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- const int src= *s; s += 4;
- *d++ = ((src&0xF8)<<7) + ((src&0xF800)>>6) + ((src&0xF80000)>>19);
- }
- }
- static inline void RENAME(rgb24to16)(const uint8_t *src, uint8_t *dst, unsigned src_size)
- {
- const uint8_t *s = src;
- const uint8_t *end;
- #ifdef HAVE_MMX
- const uint8_t *mm_end;
- #endif
- uint16_t *d = (uint16_t *)dst;
- end = s + src_size;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*src):"memory");
- __asm __volatile(
- "movq %0, %%mm7nt"
- "movq %1, %%mm6nt"
- ::"m"(red_16mask),"m"(green_16mask));
- mm_end = end - 11;
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movd %1, %%mm0nt"
- "movd 3%1, %%mm3nt"
- "punpckldq 6%1, %%mm0nt"
- "punpckldq 9%1, %%mm3nt"
- "movq %%mm0, %%mm1nt"
- "movq %%mm0, %%mm2nt"
- "movq %%mm3, %%mm4nt"
- "movq %%mm3, %%mm5nt"
- "psrlq $3, %%mm0nt"
- "psrlq $3, %%mm3nt"
- "pand %2, %%mm0nt"
- "pand %2, %%mm3nt"
- "psrlq $5, %%mm1nt"
- "psrlq $5, %%mm4nt"
- "pand %%mm6, %%mm1nt"
- "pand %%mm6, %%mm4nt"
- "psrlq $8, %%mm2nt"
- "psrlq $8, %%mm5nt"
- "pand %%mm7, %%mm2nt"
- "pand %%mm7, %%mm5nt"
- "por %%mm1, %%mm0nt"
- "por %%mm4, %%mm3nt"
- "por %%mm2, %%mm0nt"
- "por %%mm5, %%mm3nt"
- "psllq $16, %%mm3nt"
- "por %%mm3, %%mm0nt"
- MOVNTQ" %%mm0, %0nt"
- :"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
- d += 4;
- s += 12;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- const int b= *s++;
- const int g= *s++;
- const int r= *s++;
- *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
- }
- }
- static inline void RENAME(rgb24tobgr16)(const uint8_t *src, uint8_t *dst, unsigned int src_size)
- {
- const uint8_t *s = src;
- const uint8_t *end;
- #ifdef HAVE_MMX
- const uint8_t *mm_end;
- #endif
- uint16_t *d = (uint16_t *)dst;
- end = s + src_size;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*src):"memory");
- __asm __volatile(
- "movq %0, %%mm7nt"
- "movq %1, %%mm6nt"
- ::"m"(red_16mask),"m"(green_16mask));
- mm_end = end - 15;
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movd %1, %%mm0nt"
- "movd 3%1, %%mm3nt"
- "punpckldq 6%1, %%mm0nt"
- "punpckldq 9%1, %%mm3nt"
- "movq %%mm0, %%mm1nt"
- "movq %%mm0, %%mm2nt"
- "movq %%mm3, %%mm4nt"
- "movq %%mm3, %%mm5nt"
- "psllq $8, %%mm0nt"
- "psllq $8, %%mm3nt"
- "pand %%mm7, %%mm0nt"
- "pand %%mm7, %%mm3nt"
- "psrlq $5, %%mm1nt"
- "psrlq $5, %%mm4nt"
- "pand %%mm6, %%mm1nt"
- "pand %%mm6, %%mm4nt"
- "psrlq $19, %%mm2nt"
- "psrlq $19, %%mm5nt"
- "pand %2, %%mm2nt"
- "pand %2, %%mm5nt"
- "por %%mm1, %%mm0nt"
- "por %%mm4, %%mm3nt"
- "por %%mm2, %%mm0nt"
- "por %%mm5, %%mm3nt"
- "psllq $16, %%mm3nt"
- "por %%mm3, %%mm0nt"
- MOVNTQ" %%mm0, %0nt"
- :"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
- d += 4;
- s += 12;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- const int r= *s++;
- const int g= *s++;
- const int b= *s++;
- *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
- }
- }
- static inline void RENAME(rgb24to15)(const uint8_t *src, uint8_t *dst, unsigned src_size)
- {
- const uint8_t *s = src;
- const uint8_t *end;
- #ifdef HAVE_MMX
- const uint8_t *mm_end;
- #endif
- uint16_t *d = (uint16_t *)dst;
- end = s + src_size;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*src):"memory");
- __asm __volatile(
- "movq %0, %%mm7nt"
- "movq %1, %%mm6nt"
- ::"m"(red_15mask),"m"(green_15mask));
- mm_end = end - 11;
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movd %1, %%mm0nt"
- "movd 3%1, %%mm3nt"
- "punpckldq 6%1, %%mm0nt"
- "punpckldq 9%1, %%mm3nt"
- "movq %%mm0, %%mm1nt"
- "movq %%mm0, %%mm2nt"
- "movq %%mm3, %%mm4nt"
- "movq %%mm3, %%mm5nt"
- "psrlq $3, %%mm0nt"
- "psrlq $3, %%mm3nt"
- "pand %2, %%mm0nt"
- "pand %2, %%mm3nt"
- "psrlq $6, %%mm1nt"
- "psrlq $6, %%mm4nt"
- "pand %%mm6, %%mm1nt"
- "pand %%mm6, %%mm4nt"
- "psrlq $9, %%mm2nt"
- "psrlq $9, %%mm5nt"
- "pand %%mm7, %%mm2nt"
- "pand %%mm7, %%mm5nt"
- "por %%mm1, %%mm0nt"
- "por %%mm4, %%mm3nt"
- "por %%mm2, %%mm0nt"
- "por %%mm5, %%mm3nt"
- "psllq $16, %%mm3nt"
- "por %%mm3, %%mm0nt"
- MOVNTQ" %%mm0, %0nt"
- :"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
- d += 4;
- s += 12;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- const int b= *s++;
- const int g= *s++;
- const int r= *s++;
- *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
- }
- }
- static inline void RENAME(rgb24tobgr15)(const uint8_t *src, uint8_t *dst, unsigned src_size)
- {
- const uint8_t *s = src;
- const uint8_t *end;
- #ifdef HAVE_MMX
- const uint8_t *mm_end;
- #endif
- uint16_t *d = (uint16_t *)dst;
- end = s + src_size;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*src):"memory");
- __asm __volatile(
- "movq %0, %%mm7nt"
- "movq %1, %%mm6nt"
- ::"m"(red_15mask),"m"(green_15mask));
- mm_end = end - 15;
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movd %1, %%mm0nt"
- "movd 3%1, %%mm3nt"
- "punpckldq 6%1, %%mm0nt"
- "punpckldq 9%1, %%mm3nt"
- "movq %%mm0, %%mm1nt"
- "movq %%mm0, %%mm2nt"
- "movq %%mm3, %%mm4nt"
- "movq %%mm3, %%mm5nt"
- "psllq $7, %%mm0nt"
- "psllq $7, %%mm3nt"
- "pand %%mm7, %%mm0nt"
- "pand %%mm7, %%mm3nt"
- "psrlq $6, %%mm1nt"
- "psrlq $6, %%mm4nt"
- "pand %%mm6, %%mm1nt"
- "pand %%mm6, %%mm4nt"
- "psrlq $19, %%mm2nt"
- "psrlq $19, %%mm5nt"
- "pand %2, %%mm2nt"
- "pand %2, %%mm5nt"
- "por %%mm1, %%mm0nt"
- "por %%mm4, %%mm3nt"
- "por %%mm2, %%mm0nt"
- "por %%mm5, %%mm3nt"
- "psllq $16, %%mm3nt"
- "por %%mm3, %%mm0nt"
- MOVNTQ" %%mm0, %0nt"
- :"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
- d += 4;
- s += 12;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- const int r= *s++;
- const int g= *s++;
- const int b= *s++;
- *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
- }
- }
- /*
- I use here less accurate approximation by simply
- left-shifting the input
- value and filling the low order bits with
- zeroes. This method improves png's
- compression but this scheme cannot reproduce white exactly, since it does not
- generate an all-ones maximum value; the net effect is to darken the
- image slightly.
- The better method should be "left bit replication":
- 4 3 2 1 0
- ---------
- 1 1 0 1 1
- 7 6 5 4 3 2 1 0
- ----------------
- 1 1 0 1 1 1 1 0
- |=======| |===|
- | Leftmost Bits Repeated to Fill Open Bits
- |
- Original Bits
- */
- static inline void RENAME(rgb15to24)(const uint8_t *src, uint8_t *dst, unsigned src_size)
- {
- const uint16_t *end;
- #ifdef HAVE_MMX
- const uint16_t *mm_end;
- #endif
- uint8_t *d = (uint8_t *)dst;
- const uint16_t *s = (uint16_t *)src;
- end = s + src_size/2;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*s):"memory");
- mm_end = end - 7;
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movq %1, %%mm0nt"
- "movq %1, %%mm1nt"
- "movq %1, %%mm2nt"
- "pand %2, %%mm0nt"
- "pand %3, %%mm1nt"
- "pand %4, %%mm2nt"
- "psllq $3, %%mm0nt"
- "psrlq $2, %%mm1nt"
- "psrlq $7, %%mm2nt"
- "movq %%mm0, %%mm3nt"
- "movq %%mm1, %%mm4nt"
- "movq %%mm2, %%mm5nt"
- "punpcklwd %5, %%mm0nt"
- "punpcklwd %5, %%mm1nt"
- "punpcklwd %5, %%mm2nt"
- "punpckhwd %5, %%mm3nt"
- "punpckhwd %5, %%mm4nt"
- "punpckhwd %5, %%mm5nt"
- "psllq $8, %%mm1nt"
- "psllq $16, %%mm2nt"
- "por %%mm1, %%mm0nt"
- "por %%mm2, %%mm0nt"
- "psllq $8, %%mm4nt"
- "psllq $16, %%mm5nt"
- "por %%mm4, %%mm3nt"
- "por %%mm5, %%mm3nt"
- "movq %%mm0, %%mm6nt"
- "movq %%mm3, %%mm7nt"
-
- "movq 8%1, %%mm0nt"
- "movq 8%1, %%mm1nt"
- "movq 8%1, %%mm2nt"
- "pand %2, %%mm0nt"
- "pand %3, %%mm1nt"
- "pand %4, %%mm2nt"
- "psllq $3, %%mm0nt"
- "psrlq $2, %%mm1nt"
- "psrlq $7, %%mm2nt"
- "movq %%mm0, %%mm3nt"
- "movq %%mm1, %%mm4nt"
- "movq %%mm2, %%mm5nt"
- "punpcklwd %5, %%mm0nt"
- "punpcklwd %5, %%mm1nt"
- "punpcklwd %5, %%mm2nt"
- "punpckhwd %5, %%mm3nt"
- "punpckhwd %5, %%mm4nt"
- "punpckhwd %5, %%mm5nt"
- "psllq $8, %%mm1nt"
- "psllq $16, %%mm2nt"
- "por %%mm1, %%mm0nt"
- "por %%mm2, %%mm0nt"
- "psllq $8, %%mm4nt"
- "psllq $16, %%mm5nt"
- "por %%mm4, %%mm3nt"
- "por %%mm5, %%mm3nt"
- :"=m"(*d)
- :"m"(*s),"m"(mask15b),"m"(mask15g),"m"(mask15r), "m"(mmx_null)
- :"memory");
- /* Borrowed 32 to 24 */
- __asm __volatile(
- "movq %%mm0, %%mm4nt"
- "movq %%mm3, %%mm5nt"
- "movq %%mm6, %%mm0nt"
- "movq %%mm7, %%mm1nt"
-
- "movq %%mm4, %%mm6nt"
- "movq %%mm5, %%mm7nt"
- "movq %%mm0, %%mm2nt"
- "movq %%mm1, %%mm3nt"
- "psrlq $8, %%mm2nt"
- "psrlq $8, %%mm3nt"
- "psrlq $8, %%mm6nt"
- "psrlq $8, %%mm7nt"
- "pand %2, %%mm0nt"
- "pand %2, %%mm1nt"
- "pand %2, %%mm4nt"
- "pand %2, %%mm5nt"
- "pand %3, %%mm2nt"
- "pand %3, %%mm3nt"
- "pand %3, %%mm6nt"
- "pand %3, %%mm7nt"
- "por %%mm2, %%mm0nt"
- "por %%mm3, %%mm1nt"
- "por %%mm6, %%mm4nt"
- "por %%mm7, %%mm5nt"
- "movq %%mm1, %%mm2nt"
- "movq %%mm4, %%mm3nt"
- "psllq $48, %%mm2nt"
- "psllq $32, %%mm3nt"
- "pand %4, %%mm2nt"
- "pand %5, %%mm3nt"
- "por %%mm2, %%mm0nt"
- "psrlq $16, %%mm1nt"
- "psrlq $32, %%mm4nt"
- "psllq $16, %%mm5nt"
- "por %%mm3, %%mm1nt"
- "pand %6, %%mm5nt"
- "por %%mm5, %%mm4nt"
- MOVNTQ" %%mm0, %0nt"
- MOVNTQ" %%mm1, 8%0nt"
- MOVNTQ" %%mm4, 16%0"
- :"=m"(*d)
- :"m"(*s),"m"(mask24l),"m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)
- :"memory");
- d += 24;
- s += 8;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- register uint16_t bgr;
- bgr = *s++;
- *d++ = (bgr&0x1F)<<3;
- *d++ = (bgr&0x3E0)>>2;
- *d++ = (bgr&0x7C00)>>7;
- }
- }
- static inline void RENAME(rgb16to24)(const uint8_t *src, uint8_t *dst, unsigned src_size)
- {
- const uint16_t *end;
- #ifdef HAVE_MMX
- const uint16_t *mm_end;
- #endif
- uint8_t *d = (uint8_t *)dst;
- const uint16_t *s = (const uint16_t *)src;
- end = s + src_size/2;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*s):"memory");
- mm_end = end - 7;
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movq %1, %%mm0nt"
- "movq %1, %%mm1nt"
- "movq %1, %%mm2nt"
- "pand %2, %%mm0nt"
- "pand %3, %%mm1nt"
- "pand %4, %%mm2nt"
- "psllq $3, %%mm0nt"
- "psrlq $3, %%mm1nt"
- "psrlq $8, %%mm2nt"
- "movq %%mm0, %%mm3nt"
- "movq %%mm1, %%mm4nt"
- "movq %%mm2, %%mm5nt"
- "punpcklwd %5, %%mm0nt"
- "punpcklwd %5, %%mm1nt"
- "punpcklwd %5, %%mm2nt"
- "punpckhwd %5, %%mm3nt"
- "punpckhwd %5, %%mm4nt"
- "punpckhwd %5, %%mm5nt"
- "psllq $8, %%mm1nt"
- "psllq $16, %%mm2nt"
- "por %%mm1, %%mm0nt"
- "por %%mm2, %%mm0nt"
- "psllq $8, %%mm4nt"
- "psllq $16, %%mm5nt"
- "por %%mm4, %%mm3nt"
- "por %%mm5, %%mm3nt"
-
- "movq %%mm0, %%mm6nt"
- "movq %%mm3, %%mm7nt"
- "movq 8%1, %%mm0nt"
- "movq 8%1, %%mm1nt"
- "movq 8%1, %%mm2nt"
- "pand %2, %%mm0nt"
- "pand %3, %%mm1nt"
- "pand %4, %%mm2nt"
- "psllq $3, %%mm0nt"
- "psrlq $3, %%mm1nt"
- "psrlq $8, %%mm2nt"
- "movq %%mm0, %%mm3nt"
- "movq %%mm1, %%mm4nt"
- "movq %%mm2, %%mm5nt"
- "punpcklwd %5, %%mm0nt"
- "punpcklwd %5, %%mm1nt"
- "punpcklwd %5, %%mm2nt"
- "punpckhwd %5, %%mm3nt"
- "punpckhwd %5, %%mm4nt"
- "punpckhwd %5, %%mm5nt"
- "psllq $8, %%mm1nt"
- "psllq $16, %%mm2nt"
- "por %%mm1, %%mm0nt"
- "por %%mm2, %%mm0nt"
- "psllq $8, %%mm4nt"
- "psllq $16, %%mm5nt"
- "por %%mm4, %%mm3nt"
- "por %%mm5, %%mm3nt"
- :"=m"(*d)
- :"m"(*s),"m"(mask16b),"m"(mask16g),"m"(mask16r),"m"(mmx_null)
- :"memory");
- /* Borrowed 32 to 24 */
- __asm __volatile(
- "movq %%mm0, %%mm4nt"
- "movq %%mm3, %%mm5nt"
- "movq %%mm6, %%mm0nt"
- "movq %%mm7, %%mm1nt"
-
- "movq %%mm4, %%mm6nt"
- "movq %%mm5, %%mm7nt"
- "movq %%mm0, %%mm2nt"
- "movq %%mm1, %%mm3nt"
- "psrlq $8, %%mm2nt"
- "psrlq $8, %%mm3nt"
- "psrlq $8, %%mm6nt"
- "psrlq $8, %%mm7nt"
- "pand %2, %%mm0nt"
- "pand %2, %%mm1nt"
- "pand %2, %%mm4nt"
- "pand %2, %%mm5nt"
- "pand %3, %%mm2nt"
- "pand %3, %%mm3nt"
- "pand %3, %%mm6nt"
- "pand %3, %%mm7nt"
- "por %%mm2, %%mm0nt"
- "por %%mm3, %%mm1nt"
- "por %%mm6, %%mm4nt"
- "por %%mm7, %%mm5nt"
- "movq %%mm1, %%mm2nt"
- "movq %%mm4, %%mm3nt"
- "psllq $48, %%mm2nt"
- "psllq $32, %%mm3nt"
- "pand %4, %%mm2nt"
- "pand %5, %%mm3nt"
- "por %%mm2, %%mm0nt"
- "psrlq $16, %%mm1nt"
- "psrlq $32, %%mm4nt"
- "psllq $16, %%mm5nt"
- "por %%mm3, %%mm1nt"
- "pand %6, %%mm5nt"
- "por %%mm5, %%mm4nt"
- MOVNTQ" %%mm0, %0nt"
- MOVNTQ" %%mm1, 8%0nt"
- MOVNTQ" %%mm4, 16%0"
- :"=m"(*d)
- :"m"(*s),"m"(mask24l),"m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)
- :"memory");
- d += 24;
- s += 8;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- register uint16_t bgr;
- bgr = *s++;
- *d++ = (bgr&0x1F)<<3;
- *d++ = (bgr&0x7E0)>>3;
- *d++ = (bgr&0xF800)>>8;
- }
- }
- static inline void RENAME(rgb15to32)(const uint8_t *src, uint8_t *dst, unsigned src_size)
- {
- const uint16_t *end;
- #ifdef HAVE_MMX
- const uint16_t *mm_end;
- #endif
- uint8_t *d = (uint8_t *)dst;
- const uint16_t *s = (const uint16_t *)src;
- end = s + src_size/2;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*s):"memory");
- __asm __volatile("pxor %%mm7,%%mm7nt":::"memory");
- mm_end = end - 3;
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movq %1, %%mm0nt"
- "movq %1, %%mm1nt"
- "movq %1, %%mm2nt"
- "pand %2, %%mm0nt"
- "pand %3, %%mm1nt"
- "pand %4, %%mm2nt"
- "psllq $3, %%mm0nt"
- "psrlq $2, %%mm1nt"
- "psrlq $7, %%mm2nt"
- "movq %%mm0, %%mm3nt"
- "movq %%mm1, %%mm4nt"
- "movq %%mm2, %%mm5nt"
- "punpcklwd %%mm7, %%mm0nt"
- "punpcklwd %%mm7, %%mm1nt"
- "punpcklwd %%mm7, %%mm2nt"
- "punpckhwd %%mm7, %%mm3nt"
- "punpckhwd %%mm7, %%mm4nt"
- "punpckhwd %%mm7, %%mm5nt"
- "psllq $8, %%mm1nt"
- "psllq $16, %%mm2nt"
- "por %%mm1, %%mm0nt"
- "por %%mm2, %%mm0nt"
- "psllq $8, %%mm4nt"
- "psllq $16, %%mm5nt"
- "por %%mm4, %%mm3nt"
- "por %%mm5, %%mm3nt"
- MOVNTQ" %%mm0, %0nt"
- MOVNTQ" %%mm3, 8%0nt"
- :"=m"(*d)
- :"m"(*s),"m"(mask15b),"m"(mask15g),"m"(mask15r)
- :"memory");
- d += 16;
- s += 4;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- #if 0 //slightly slower on athlon
- int bgr= *s++;
- *((uint32_t*)d)++ = ((bgr&0x1F)<<3) + ((bgr&0x3E0)<<6) + ((bgr&0x7C00)<<9);
- #else
- //FIXME this is very likely wrong for bigendian (and the following converters too)
- register uint16_t bgr;
- bgr = *s++;
- *d++ = (bgr&0x1F)<<3;
- *d++ = (bgr&0x3E0)>>2;
- *d++ = (bgr&0x7C00)>>7;
- *d++ = 0;
- #endif
- }
- }
- static inline void RENAME(rgb16to32)(const uint8_t *src, uint8_t *dst, unsigned src_size)
- {
- const uint16_t *end;
- #ifdef HAVE_MMX
- const uint16_t *mm_end;
- #endif
- uint8_t *d = (uint8_t *)dst;
- const uint16_t *s = (uint16_t *)src;
- end = s + src_size/2;
- #ifdef HAVE_MMX
- __asm __volatile(PREFETCH" %0"::"m"(*s):"memory");
- __asm __volatile("pxor %%mm7,%%mm7nt":::"memory");
- mm_end = end - 3;
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1nt"
- "movq %1, %%mm0nt"
- "movq %1, %%mm1nt"
- "movq %1, %%mm2nt"
- "pand %2, %%mm0nt"
- "pand %3, %%mm1nt"
- "pand %4, %%mm2nt"
- "psllq $3, %%mm0nt"
- "psrlq $3, %%mm1nt"
- "psrlq $8, %%mm2nt"
- "movq %%mm0, %%mm3nt"
- "movq %%mm1, %%mm4nt"
- "movq %%mm2, %%mm5nt"
- "punpcklwd %%mm7, %%mm0nt"
- "punpcklwd %%mm7, %%mm1nt"
- "punpcklwd %%mm7, %%mm2nt"
- "punpckhwd %%mm7, %%mm3nt"
- "punpckhwd %%mm7, %%mm4nt"
- "punpckhwd %%mm7, %%mm5nt"
- "psllq $8, %%mm1nt"
- "psllq $16, %%mm2nt"
- "por %%mm1, %%mm0nt"
- "por %%mm2, %%mm0nt"
- "psllq $8, %%mm4nt"
- "psllq $16, %%mm5nt"
- "por %%mm4, %%mm3nt"
- "por %%mm5, %%mm3nt"
- MOVNTQ" %%mm0, %0nt"
- MOVNTQ" %%mm3, 8%0nt"
- :"=m"(*d)
- :"m"(*s),"m"(mask16b),"m"(mask16g),"m"(mask16r)
- :"memory");
- d += 16;
- s += 4;
- }
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #endif
- while(s < end)
- {
- register uint16_t bgr;
- bgr = *s++;
- *d++ = (bgr&0x1F)<<3;
- *d++ = (bgr&0x7E0)>>3;
- *d++ = (bgr&0xF800)>>8;
- *d++ = 0;
- }
- }
- static inline void RENAME(rgb32tobgr32)(const uint8_t *src, uint8_t *dst, unsigned int src_size)
- {
- #ifdef HAVE_MMX
- /* TODO: unroll this loop */
- asm volatile (
- "xorl %%eax, %%eax nt"
- ".balign 16 nt"
- "1: nt"
- PREFETCH" 32(%0, %%eax) nt"
- "movq (%0, %%eax), %%mm0 nt"
- "movq %%mm0, %%mm1 nt"
- "movq %%mm0, %%mm2 nt"
- "pslld $16, %%mm0 nt"
- "psrld $16, %%mm1 nt"
- "pand "MANGLE(mask32r)", %%mm0 nt"
- "pand "MANGLE(mask32g)", %%mm2 nt"
- "pand "MANGLE(mask32b)", %%mm1 nt"
- "por %%mm0, %%mm2 nt"
- "por %%mm1, %%mm2 nt"
- MOVNTQ" %%mm2, (%1, %%eax) nt"
- "addl $8, %%eax nt"
- "cmpl %2, %%eax nt"
- " jb 1b nt"
- :: "r" (src), "r"(dst), "r" (src_size-7)
- : "%eax"
- );
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- #else
- unsigned i;
- unsigned num_pixels = src_size >> 2;
- for(i=0; i<num_pixels; i++)
- {
- #ifdef WORDS_BIGENDIAN
- dst[4*i + 1] = src[4*i + 3];
- dst[4*i + 2] = src[4*i + 2];
- dst[4*i + 3] = src[4*i + 1];
- #else
- dst[4*i + 0] = src[4*i + 2];
- dst[4*i + 1] = src[4*i + 1];
- dst[4*i + 2] = src[4*i + 0];
- #endif
- }
- #endif
- }
- static inline void RENAME(rgb24tobgr24)(const uint8_t *src, uint8_t *dst, unsigned int src_size)
- {
- unsigned i;
- #ifdef HAVE_MMX
- int mmx_size= 23 - src_size;
- asm volatile (
- "movq "MANGLE(mask24r)", %%mm5 nt"
- "movq "MANGLE(mask24g)", %%mm6 nt"
- "movq "MANGLE(mask24b)", %%mm7 nt"
- ".balign 16 nt"
- "1: nt"
- PREFETCH" 32(%1, %%eax) nt"
- "movq (%1, %%eax), %%mm0 nt" // BGR BGR BG
- "movq (%1, %%eax), %%mm1 nt" // BGR BGR BG
- "movq 2(%1, %%eax), %%mm2 nt" // R BGR BGR B
- "psllq $16, %%mm0 nt" // 00 BGR BGR
- "pand %%mm5, %%mm0 nt"
- "pand %%mm6, %%mm1 nt"
- "pand %%mm7, %%mm2 nt"
- "por %%mm0, %%mm1 nt"
- "por %%mm2, %%mm1 nt"
- "movq 6(%1, %%eax), %%mm0 nt" // BGR BGR BG
- MOVNTQ" %%mm1, (%2, %%eax) nt" // RGB RGB RG
- "movq 8(%1, %%eax), %%mm1 nt" // R BGR BGR B
- "movq 10(%1, %%eax), %%mm2 nt" // GR BGR BGR
- "pand %%mm7, %%mm0 nt"
- "pand %%mm5, %%mm1 nt"
- "pand %%mm6, %%mm2 nt"
- "por %%mm0, %%mm1 nt"
- "por %%mm2, %%mm1 nt"
- "movq 14(%1, %%eax), %%mm0 nt" // R BGR BGR B
- MOVNTQ" %%mm1, 8(%2, %%eax) nt" // B RGB RGB R
- "movq 16(%1, %%eax), %%mm1 nt" // GR BGR BGR
- "movq 18(%1, %%eax), %%mm2 nt" // BGR BGR BG
- "pand %%mm6, %%mm0 nt"
- "pand %%mm7, %%mm1 nt"
- "pand %%mm5, %%mm2 nt"
- "por %%mm0, %%mm1 nt"
- "por %%mm2, %%mm1 nt"
- MOVNTQ" %%mm1, 16(%2, %%eax) nt"
- "addl $24, %%eax nt"
- " js 1b nt"
- : "+a" (mmx_size)
- : "r" (src-mmx_size), "r"(dst-mmx_size)
- );
- __asm __volatile(SFENCE:::"memory");
- __asm __volatile(EMMS:::"memory");
- if(mmx_size==23) return; //finihsed, was multiple of 8
- src+= src_size;
- dst+= src_size;
- src_size= 23-mmx_size;
- src-= src_size;
- dst-= src_size;
- #endif
- for(i=0; i<src_size; i+=3)
- {
- register uint8_t x;
- x = src[i + 2];
- dst[i + 1] = src[i + 1];
- dst[i + 2] = src[i + 0];
- dst[i + 0] = x;
- }
- }
- static inline void RENAME(yuvPlanartoyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
- unsigned int width, unsigned int height,
- int lumStride, int chromStride, int dstStride, int vertLumPerChroma)
- {
- unsigned y;
- const unsigned chromWidth= width>>1;
- for(y=0; y<height; y++)
- {
- #ifdef HAVE_MMX
- //FIXME handle 2 lines a once (fewer prefetch, reuse some chrom, but very likely limited by mem anyway)
- asm volatile(
- "xorl %%eax, %%eax nt"
- ".balign 16 nt"
- "1: nt"
- PREFETCH" 32(%1, %%eax, 2) nt"
- PREFETCH" 32(%2, %%eax) nt"
- PREFETCH" 32(%3, %%eax) nt"
- "movq (%2, %%eax), %%mm0 nt" // U(0)
- "movq %%mm0, %%mm2 nt" // U(0)
- "movq (%3, %%eax), %%mm1 nt" // V(0)
- "punpcklbw %%mm1, %%mm0 nt" // UVUV UVUV(0)
- "punpckhbw %%mm1, %%mm2 nt" // UVUV UVUV(8)
- "movq (%1, %%eax,2), %%mm3 nt" // Y(0)
- "movq 8(%1, %%eax,2), %%mm5 nt" // Y(8)
- "movq %%mm3, %%mm4 nt" // Y(0)
- "movq %%mm5, %%mm6 nt" // Y(8)
- "punpcklbw %%mm0, %%mm3 nt" // YUYV YUYV(0)
- "punpckhbw %%mm0, %%mm4 nt" // YUYV YUYV(4)
- "punpcklbw %%mm2, %%mm5 nt" // YUYV YUYV(8)
- "punpckhbw %%mm2, %%mm6 nt" // YUYV YUYV(12)
- MOVNTQ" %%mm3, (%0, %%eax, 4) nt"
- MOVNTQ" %%mm4, 8(%0, %%eax, 4) nt"
- MOVNTQ" %%mm5, 16(%0, %%eax, 4) nt"
- MOVNTQ" %%mm6, 24(%0, %%eax, 4) nt"
- "addl $8, %%eax nt"
- "cmpl %4, %%eax nt"
- " jb 1b nt"
- ::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "g" (chromWidth)
- : "%eax"
- );
- #else
- #if defined ARCH_ALPHA && defined HAVE_MVI
- #define pl2yuy2(n)
- y1 = yc[n];
- y2 = yc2[n];
- u = uc[n];
- v = vc[n];
- asm("unpkbw %1, %0" : "=r"(y1) : "r"(y1));
- asm("unpkbw %1, %0" : "=r"(y2) : "r"(y2));
- asm("unpkbl %1, %0" : "=r"(u) : "r"(u));
- asm("unpkbl %1, %0" : "=r"(v) : "r"(v));
- yuv1 = (u << 8) + (v << 24);
- yuv2 = yuv1 + y2;
- yuv1 += y1;
- qdst[n] = yuv1;
- qdst2[n] = yuv2;
- int i;
- uint64_t *qdst = (uint64_t *) dst;
- uint64_t *qdst2 = (uint64_t *) (dst + dstStride);
- const uint32_t *yc = (uint32_t *) ysrc;
- const uint32_t *yc2 = (uint32_t *) (ysrc + lumStride);
- const uint16_t *uc = (uint16_t*) usrc, *vc = (uint16_t*) vsrc;
- for(i = 0; i < chromWidth; i += 8){
- uint64_t y1, y2, yuv1, yuv2;
- uint64_t u, v;
- /* Prefetch */
- asm("ldq $31,64(%0)" :: "r"(yc));
- asm("ldq $31,64(%0)" :: "r"(yc2));
- asm("ldq $31,64(%0)" :: "r"(uc));
- asm("ldq $31,64(%0)" :: "r"(vc));
- pl2yuy2(0);
- pl2yuy2(1);
- pl2yuy2(2);
- pl2yuy2(3);
- yc += 4;
- yc2 += 4;
- uc += 4;
- vc += 4;
- qdst += 4;
- qdst2 += 4;
- }
- y++;
- ysrc += lumStride;
- dst += dstStride;
- #elif __WORDSIZE >= 64
- int i;
- uint64_t *ldst = (uint64_t *) dst;
- const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;
- for(i = 0; i < chromWidth; i += 2){
- uint64_t k, l;
- k = yc[0] + (uc[0] << 8) +
- (yc[1] << 16) + (vc[0] << 24);
- l = yc[2] + (uc[1] << 8) +
- (yc[3] << 16) + (vc[1] << 24);
- *ldst++ = k + (l << 32);
- yc += 4;
- uc += 2;
- vc += 2;
- }
- #else
- int i, *idst = (int32_t *) dst;
- const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;
- for(i = 0; i < chromWidth; i++){
- #ifdef WORDS_BIGENDIAN
- *idst++ = (yc[0] << 24)+ (uc[0] << 16) +
- (yc[1] << 8) + (vc[0] << 0);
- #else
- *idst++ = yc[0] + (uc[0] << 8) +
- (yc[1] << 16) + (vc[0] << 24);
- #endif
- yc += 2;
- uc++;
- vc++;
- }
- #endif
- #endif
- if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )
- {
- usrc += chromStride;
- vsrc += chromStride;
- }
- ysrc += lumStride;
- dst += dstStride;
- }
- #ifdef HAVE_MMX
- asm( EMMS" nt"
- SFENCE" nt"
- :::"memory");
- #endif
- }
- /**
- *
- * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
- * problem for anyone then tell me, and ill fix it)
- */
- static inline void RENAME(yv12toyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
- unsigned int width, unsigned int height,
- int lumStride, int chromStride, int dstStride)
- {
- //FIXME interpolate chroma
- RENAME(yuvPlanartoyuy2)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 2);
- }
- static inline void RENAME(yuvPlanartouyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
- unsigned int width, unsigned int height,
- int lumStride, int chromStride, int dstStride, int vertLumPerChroma)
- {
- unsigned y;
- const unsigned chromWidth= width>>1;
- for(y=0; y<height; y++)
- {
- #ifdef HAVE_MMX
- //FIXME handle 2 lines a once (fewer prefetch, reuse some chrom, but very likely limited by mem anyway)
- asm volatile(
- "xorl %%eax, %%eax nt"
- ".balign 16 nt"
- "1: nt"
- PREFETCH" 32(%1, %%eax, 2) nt"
- PREFETCH" 32(%2, %%eax) nt"
- PREFETCH" 32(%3, %%eax) nt"
- "movq (%2, %%eax), %%mm0 nt" // U(0)
- "movq %%mm0, %%mm2 nt" // U(0)
- "movq (%3, %%eax), %%mm1 nt" // V(0)
- "punpcklbw %%mm1, %%mm0 nt" // UVUV UVUV(0)
- "punpckhbw %%mm1, %%mm2 nt" // UVUV UVUV(8)
- "movq (%1, %%eax,2), %%mm3 nt" // Y(0)
- "movq 8(%1, %%eax,2), %%mm5 nt" // Y(8)
- "movq %%mm0, %%mm4 nt" // Y(0)
- "movq %%mm2, %%mm6 nt" // Y(8)
- "punpcklbw %%mm3, %%mm0 nt" // YUYV YUYV(0)
- "punpckhbw %%mm3, %%mm4 nt" // YUYV YUYV(4)
- "punpcklbw %%mm5, %%mm2 nt" // YUYV YUYV(8)
- "punpckhbw %%mm5, %%mm6 nt" // YUYV YUYV(12)
- MOVNTQ" %%mm0, (%0, %%eax, 4) nt"
- MOVNTQ" %%mm4, 8(%0, %%eax, 4) nt"
- MOVNTQ" %%mm2, 16(%0, %%eax, 4) nt"
- MOVNTQ" %%mm6, 24(%0, %%eax, 4) nt"
- "addl $8, %%eax nt"
- "cmpl %4, %%eax nt"
- " jb 1b nt"
- ::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "g" (chromWidth)
- : "%eax"
- );
- #else
- //FIXME adapt the alpha asm code from yv12->yuy2
- #if __WORDSIZE >= 64
- int i;
- uint64_t *ldst = (uint64_t *) dst;
- const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;
- for(i = 0; i < chromWidth; i += 2){
- uint64_t k, l;
- k = uc[0] + (yc[0] << 8) +
- (vc[0] << 16) + (yc[1] << 24);
- l = uc[1] + (yc[2] << 8) +
- (vc[1] << 16) + (yc[3] << 24);
- *ldst++ = k + (l << 32);
- yc += 4;
- uc += 2;
- vc += 2;
- }
- #else
- int i, *idst = (int32_t *) dst;
- const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;
- for(i = 0; i < chromWidth; i++){
- #ifdef WORDS_BIGENDIAN
- *idst++ = (uc[0] << 24)+ (yc[0] << 16) +
- (vc[0] << 8) + (yc[1] << 0);
- #else
- *idst++ = uc[0] + (yc[0] << 8) +
- (vc[0] << 16) + (yc[1] << 24);
- #endif
- yc += 2;
- uc++;
- vc++;
- }
- #endif
- #endif
- if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )
- {
- usrc += chromStride;
- vsrc += chromStride;
- }
- ysrc += lumStride;
- dst += dstStride;
- }
- #ifdef HAVE_MMX
- asm( EMMS" nt"
- SFENCE" nt"
- :::"memory");
- #endif
- }
- /**
- *
- * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
- * problem for anyone then tell me, and ill fix it)
- */
- static inline void RENAME(yv12touyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
- unsigned int width, unsigned int height,
- int lumStride, int chromStride, int dstStride)
- {
- //FIXME interpolate chroma
- RENAME(yuvPlanartouyvy)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 2);
- }
- /**
- *
- * width should be a multiple of 16
- */
- static inline void RENAME(yuv422ptoyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
- unsigned int width, unsigned int height,
- int lumStride, int chromStride, int dstStride)
- {
- RENAME(yuvPlanartoyuy2)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 1);
- }
- /**
- *
- * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
- * problem for anyone then tell me, and ill fix it)
- */
- static inline void RENAME(yuy2toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
- unsigned int width, unsigned int height,
- int lumStride, int chromStride, int srcStride)
- {
- unsigned y;
- const unsigned chromWidth= width>>1;
- for(y=0; y<height; y+=2)
- {
- #ifdef HAVE_MMX
- asm volatile(
- "xorl %%eax, %%eax nt"
- "pcmpeqw %%mm7, %%mm7 nt"
- "psrlw $8, %%mm7 nt" // FF,00,FF,00...
- ".balign 16 nt"
- "1: nt"
- PREFETCH" 64(%0, %%eax, 4) nt"
- "movq (%0, %%eax, 4), %%mm0 nt" // YUYV YUYV(0)
- "movq 8(%0, %%eax, 4), %%mm1 nt" // YUYV YUYV(4)
- "movq %%mm0, %%mm2 nt" // YUYV YUYV(0)
- "movq %%mm1, %%mm3 nt" // YUYV YUYV(4)
- "psrlw $8, %%mm0 nt" // U0V0 U0V0(0)
- "psrlw $8, %%mm1 nt" // U0V0 U0V0(4)
- "pand %%mm7, %%mm2 nt" // Y0Y0 Y0Y0(0)
- "pand %%mm7, %%mm3 nt" // Y0Y0 Y0Y0(4)
- "packuswb %%mm1, %%mm0 nt" // UVUV UVUV(0)
- "packuswb %%mm3, %%mm2 nt" // YYYY YYYY(0)
- MOVNTQ" %%mm2, (%1, %%eax, 2) nt"
- "movq 16(%0, %%eax, 4), %%mm1 nt" // YUYV YUYV(8)
- "movq 24(%0, %%eax, 4), %%mm2 nt" // YUYV YUYV(12)
- "movq %%mm1, %%mm3 nt" // YUYV YUYV(8)
- "movq %%mm2, %%mm4 nt" // YUYV YUYV(12)
- "psrlw $8, %%mm1 nt" // U0V0 U0V0(8)
- "psrlw $8, %%mm2 nt" // U0V0 U0V0(12)
- "pand %%mm7, %%mm3 nt" // Y0Y0 Y0Y0(8)
- "pand %%mm7, %%mm4 nt" // Y0Y0 Y0Y0(12)
- "packuswb %%mm2, %%mm1 nt" // UVUV UVUV(8)
- "packuswb %%mm4, %%mm3 nt" // YYYY YYYY(8)
- MOVNTQ" %%mm3, 8(%1, %%eax, 2) nt"
- "movq %%mm0, %%mm2 nt" // UVUV UVUV(0)
- "movq %%mm1, %%mm3 nt" // UVUV UVUV(8)
- "psrlw $8, %%mm0 nt" // V0V0 V0V0(0)
- "psrlw $8, %%mm1 nt" // V0V0 V0V0(8)
- "pand %%mm7, %%mm2 nt" // U0U0 U0U0(0)
- "pand %%mm7, %%mm3 nt" // U0U0 U0U0(8)
- "packuswb %%mm1, %%mm0 nt" // VVVV VVVV(0)
- "packuswb %%mm3, %%mm2 nt" // UUUU UUUU(0)
- MOVNTQ" %%mm0, (%3, %%eax) nt"
- MOVNTQ" %%mm2, (%2, %%eax) nt"
- "addl $8, %%eax nt"
- "cmpl %4, %%eax nt"
- " jb 1b nt"
- ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
- : "memory", "%eax"
- );
- ydst += lumStride;
- src += srcStride;
- asm volatile(
- "xorl %%eax, %%eax nt"
- ".balign 16 nt"
- "1: nt"
- PREFETCH" 64(%0, %%eax, 4) nt"
- "movq (%0, %%eax, 4), %%mm0 nt" // YUYV YUYV(0)
- "movq 8(%0, %%eax, 4), %%mm1 nt" // YUYV YUYV(4)
- "movq 16(%0, %%eax, 4), %%mm2 nt" // YUYV YUYV(8)
- "movq 24(%0, %%eax, 4), %%mm3 nt" // YUYV YUYV(12)
- "pand %%mm7, %%mm0 nt" // Y0Y0 Y0Y0(0)
- "pand %%mm7, %%mm1 nt" // Y0Y0 Y0Y0(4)
- "pand %%mm7, %%mm2 nt" // Y0Y0 Y0Y0(8)
- "pand %%mm7, %%mm3 nt" // Y0Y0 Y0Y0(12)
- "packuswb %%mm1, %%mm0 nt" // YYYY YYYY(0)
- "packuswb %%mm3, %%mm2 nt" // YYYY YYYY(8)
- MOVNTQ" %%mm0, (%1, %%eax, 2) nt"
- MOVNTQ" %%mm2, 8(%1, %%eax, 2) nt"
- "addl $8, %%eax nt"
- "cmpl %4, %%eax nt"
- " jb 1b nt"
- ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
- : "memory", "%eax"
- );
- #else
- unsigned i;
- for(i=0; i<chromWidth; i++)
- {
- ydst[2*i+0] = src[4*i+0];
- udst[i] = src[4*i+1];
- ydst[2*i+1] = src[4*i+2];
- vdst[i] = src[4*i+3];
- }
- ydst += lumStride;
- src += srcStride;
- for(i=0; i<chromWidth; i++)
- {
- ydst[2*i+0] = src[4*i+0];
- ydst[2*i+1] = src[4*i+2];
- }
- #endif
- udst += chromStride;
- vdst += chromStride;
- ydst += lumStride;
- src += srcStride;
- }
- #ifdef HAVE_MMX
- asm volatile( EMMS" nt"
- SFENCE" nt"
- :::"memory");
- #endif
- }
- static inline void RENAME(yvu9toyv12)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc,
- uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
- unsigned int width, unsigned int height, int lumStride, int chromStride)
- {
- /* Y Plane */
- memcpy(ydst, ysrc, width*height);
- /* XXX: implement upscaling for U,V */
- }
- static inline void RENAME(planar2x)(const uint8_t *src, uint8_t *dst, int srcWidth, int srcHeight, int srcStride, int dstStride)
- {
- int x,y;
-
- dst[0]= src[0];
-
- // first line
- for(x=0; x<srcWidth-1; x++){
- dst[2*x+1]= (3*src[x] + src[x+1])>>2;
- dst[2*x+2]= ( src[x] + 3*src[x+1])>>2;
- }
- dst[2*srcWidth-1]= src[srcWidth-1];
-
- dst+= dstStride;
- for(y=1; y<srcHeight; y++){
- #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
- const int mmxSize= srcWidth&~15;
- asm volatile(
- "movl %4, %%eax nt"
- "1: nt"
- "movq (%0, %%eax), %%mm0 nt"
- "movq (%1, %%eax), %%mm1 nt"
- "movq 1(%0, %%eax), %%mm2 nt"
- "movq 1(%1, %%eax), %%mm3 nt"
- "movq -1(%0, %%eax), %%mm4 nt"
- "movq -1(%1, %%eax), %%mm5 nt"
- PAVGB" %%mm0, %%mm5 nt"
- PAVGB" %%mm0, %%mm3 nt"
- PAVGB" %%mm0, %%mm5 nt"
- PAVGB" %%mm0, %%mm3 nt"
- PAVGB" %%mm1, %%mm4 nt"
- PAVGB" %%mm1, %%mm2 nt"
- PAVGB" %%mm1, %%mm4 nt"
- PAVGB" %%mm1, %%mm2 nt"
- "movq %%mm5, %%mm7 nt"
- "movq %%mm4, %%mm6 nt"
- "punpcklbw %%mm3, %%mm5 nt"
- "punpckhbw %%mm3, %%mm7 nt"
- "punpcklbw %%mm2, %%mm4 nt"
- "punpckhbw %%mm2, %%mm6 nt"
- #if 1
- MOVNTQ" %%mm5, (%2, %%eax, 2) nt"
- MOVNTQ" %%mm7, 8(%2, %%eax, 2) nt"
- MOVNTQ" %%mm4, (%3, %%eax, 2) nt"
- MOVNTQ" %%mm6, 8(%3, %%eax, 2) nt"
- #else
- "movq %%mm5, (%2, %%eax, 2) nt"
- "movq %%mm7, 8(%2, %%eax, 2) nt"
- "movq %%mm4, (%3, %%eax, 2) nt"
- "movq %%mm6, 8(%3, %%eax, 2) nt"
- #endif
- "addl $8, %%eax nt"
- " js 1b nt"
- :: "r" (src + mmxSize ), "r" (src + srcStride + mmxSize ),
- "r" (dst + mmxSize*2), "r" (dst + dstStride + mmxSize*2),
- "g" (-mmxSize)
- : "%eax"
- );
- #else
- const int mmxSize=1;
- #endif
- dst[0 ]= (3*src[0] + src[srcStride])>>2;
- dst[dstStride]= ( src[0] + 3*src[srcStride])>>2;
- for(x=mmxSize-1; x<srcWidth-1; x++){
- dst[2*x +1]= (3*src[x+0] + src[x+srcStride+1])>>2;
- dst[2*x+dstStride+2]= ( src[x+0] + 3*src[x+srcStride+1])>>2;
- dst[2*x+dstStride+1]= ( src[x+1] + 3*src[x+srcStride ])>>2;
- dst[2*x +2]= (3*src[x+1] + src[x+srcStride ])>>2;
- }
- dst[srcWidth*2 -1 ]= (3*src[srcWidth-1] + src[srcWidth-1 + srcStride])>>2;
- dst[srcWidth*2 -1 + dstStride]= ( src[srcWidth-1] + 3*src[srcWidth-1 + srcStride])>>2;
- dst+=dstStride*2;
- src+=srcStride;
- }
-
- // last line
- #if 1
- dst[0]= src[0];
-
- for(x=0; x<srcWidth-1; x++){
- dst[2*x+1]= (3*src[x] + src[x+1])>>2;
- dst[2*x+2]= ( src[x] + 3*src[x+1])>>2;
- }
- dst[2*srcWidth-1]= src[srcWidth-1];
- #else
- for(x=0; x<srcWidth; x++){
- dst[2*x+0]=
- dst[2*x+1]= src[x];
- }
- #endif
- #ifdef HAVE_MMX
- asm volatile( EMMS" nt"
- SFENCE" nt"
- :::"memory");
- #endif
- }
- /**
- *
- * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
- * problem for anyone then tell me, and ill fix it)
- * chrominance data is only taken from every secound line others are ignored FIXME write HQ version
- */
- static inline void RENAME(uyvytoyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
- unsigned int width, unsigned int height,
- int lumStride, int chromStride, int srcStride)
- {
- unsigned y;
- const unsigned chromWidth= width>>1;
- for(y=0; y<height; y+=2)
- {
- #ifdef HAVE_MMX
- asm volatile(
- "xorl %%eax, %%eax nt"
- "pcmpeqw %%mm7, %%mm7 nt"
- "psrlw $8, %%mm7 nt" // FF,00,FF,00...
- ".balign 16 nt"
- "1: nt"
- PREFETCH" 64(%0, %%eax, 4) nt"
- "movq (%0, %%eax, 4), %%mm0 nt" // UYVY UYVY(0)
- "movq 8(%0, %%eax, 4), %%mm1 nt" // UYVY UYVY(4)
- "movq %%mm0, %%mm2 nt" // UYVY UYVY(0)
- "movq %%mm1, %%mm3 nt" // UYVY UYVY(4)
- "pand %%mm7, %%mm0 nt" // U0V0 U0V0(0)
- "pand %%mm7, %%mm1 nt" // U0V0 U0V0(4)
- "psrlw $8, %%mm2 nt" // Y0Y0 Y0Y0(0)
- "psrlw $8, %%mm3 nt" // Y0Y0 Y0Y0(4)
- "packuswb %%mm1, %%mm0 nt" // UVUV UVUV(0)
- "packuswb %%mm3, %%mm2 nt" // YYYY YYYY(0)
- MOVNTQ" %%mm2, (%1, %%eax, 2) nt"
- "movq 16(%0, %%eax, 4), %%mm1 nt" // UYVY UYVY(8)
- "movq 24(%0, %%eax, 4), %%mm2 nt" // UYVY UYVY(12)
- "movq %%mm1, %%mm3 nt" // UYVY UYVY(8)
- "movq %%mm2, %%mm4 nt" // UYVY UYVY(12)
- "pand %%mm7, %%mm1 nt" // U0V0 U0V0(8)
- "pand %%mm7, %%mm2 nt" // U0V0 U0V0(12)
- "psrlw $8, %%mm3 nt" // Y0Y0 Y0Y0(8)
- "psrlw $8, %%mm4 nt" // Y0Y0 Y0Y0(12)
- "packuswb %%mm2, %%mm1 nt" // UVUV UVUV(8)
- "packuswb %%mm4, %%mm3 nt" // YYYY YYYY(8)
- MOVNTQ" %%mm3, 8(%1, %%eax, 2) nt"
- "movq %%mm0, %%mm2 nt" // UVUV UVUV(0)
- "movq %%mm1, %%mm3 nt" // UVUV UVUV(8)
- "psrlw $8, %%mm0 nt" // V0V0 V0V0(0)
- "psrlw $8, %%mm1 nt" // V0V0 V0V0(8)
- "pand %%mm7, %%mm2 nt" // U0U0 U0U0(0)
- "pand %%mm7, %%mm3 nt" // U0U0 U0U0(8)
- "packuswb %%mm1, %%mm0 nt" // VVVV VVVV(0)
- "packuswb %%mm3, %%mm2 nt" // UUUU UUUU(0)
- MOVNTQ" %%mm0, (%3, %%eax) nt"
- MOVNTQ" %%mm2, (%2, %%eax) nt"
- "addl $8, %%eax nt"
- "cmpl %4, %%eax nt"
- " jb 1b nt"
- ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
- : "memory", "%eax"
- );
- ydst += lumStride;
- src += srcStride;
- asm volatile(
- "xorl %%eax, %%eax nt"
- ".balign 16 nt"
- "1: nt"
- PREFETCH" 64(%0, %%eax, 4) nt"
- "movq (%0, %%eax, 4), %%mm0 nt" // YUYV YUYV(0)
- "movq 8(%0, %%eax, 4), %%mm1 nt" // YUYV YUYV(4)
- "movq 16(%0, %%eax, 4), %%mm2 nt" // YUYV YUYV(8)
- "movq 24(%0, %%eax, 4), %%mm3 nt" // YUYV YUYV(12)
- "psrlw $8, %%mm0 nt" // Y0Y0 Y0Y0(0)
- "psrlw $8, %%mm1 nt" // Y0Y0 Y0Y0(4)
- "psrlw $8, %%mm2 nt" // Y0Y0 Y0Y0(8)
- "psrlw $8, %%mm3 nt" // Y0Y0 Y0Y0(12)
- "packuswb %%mm1, %%mm0 nt" // YYYY YYYY(0)
- "packuswb %%mm3, %%mm2 nt" // YYYY YYYY(8)
- MOVNTQ" %%mm0, (%1, %%eax, 2) nt"
- MOVNTQ" %%mm2, 8(%1, %%eax, 2) nt"
- "addl $8, %%eax nt"
- "cmpl %4, %%eax nt"
- " jb 1b nt"
- ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
- : "memory", "%eax"
- );
- #else
- unsigned i;
- for(i=0; i<chromWidth; i++)
- {
- udst[i] = src[4*i+0];
- ydst[2*i+0] = src[4*i+1];
- vdst[i] = src[4*i+2];
- ydst[2*i+1] = src[4*i+3];
- }
- ydst += lumStride;
- src += srcStride;
- for(i=0; i<chromWidth; i++)
- {
- ydst[2*i+0] = src[4*i+1];
- ydst[2*i+1] = src[4*i+3];
- }
- #endif
- udst += chromStride;
- vdst += chromStride;
- ydst += lumStride;
- src += srcStride;
- }
- #ifdef HAVE_MMX
- asm volatile( EMMS" nt"
- SFENCE" nt"
- :::"memory");
- #endif
- }
- /**
- *
- * height should be a multiple of 2 and width should be a multiple of 2 (if this is a
- * problem for anyone then tell me, and ill fix it)
- * chrominance data is only taken from every secound line others are ignored in the C version FIXME write HQ version
- */
- static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
- unsigned int width, unsigned int height,
- int lumStride, int chromStride, int srcStride)
- {
- unsigned y;
- const unsigned chromWidth= width>>1;
- #ifdef HAVE_MMX
- for(y=0; y<height-2; y+=2)
- {
- unsigned i;
- for(i=0; i<2; i++)
- {
- asm volatile(
- "movl %2, %%eax nt"
- "movq "MANGLE(bgr2YCoeff)", %%mm6 nt"
- "movq "MANGLE(w1111)", %%mm5 nt"
- "pxor %%mm7, %%mm7 nt"
- "leal (%%eax, %%eax, 2), %%ebx nt"
- ".balign 16 nt"
- "1: nt"
- PREFETCH" 64(%0, %%ebx) nt"
- "movd (%0, %%ebx), %%mm0 nt"
- "movd 3(%0, %%ebx), %%mm1 nt"
- "punpcklbw %%mm7, %%mm0 nt"
- "punpcklbw %%mm7, %%mm1 nt"
- "movd 6(%0, %%ebx), %%mm2 nt"
- "movd 9(%0, %%ebx), %%mm3 nt"
- "punpcklbw %%mm7, %%mm2 nt"
- "punpcklbw %%mm7, %%mm3 nt"
- "pmaddwd %%mm6, %%mm0 nt"
- "pmaddwd %%mm6, %%mm1 nt"
- "pmaddwd %%mm6, %%mm2 nt"
- "pmaddwd %%mm6, %%mm3 nt"
- #ifndef FAST_BGR2YV12
- "psrad $8, %%mm0 nt"
- "psrad $8, %%mm1 nt"
- "psrad $8, %%mm2 nt"
- "psrad $8, %%mm3 nt"
- #endif
- "packssdw %%mm1, %%mm0 nt"
- "packssdw %%mm3, %%mm2 nt"
- "pmaddwd %%mm5, %%mm0 nt"
- "pmaddwd %%mm5, %%mm2 nt"
- "packssdw %%mm2, %%mm0 nt"
- "psraw $7, %%mm0 nt"
- "movd 12(%0, %%ebx), %%mm4 nt"
- "movd 15(%0, %%ebx), %%mm1 nt"
- "punpcklbw %%mm7, %%mm4 nt"
- "punpcklbw %%mm7, %%mm1 nt"
- "movd 18(%0, %%ebx), %%mm2 nt"
- "movd 21(%0, %%ebx), %%mm3 nt"
- "punpcklbw %%mm7, %%mm2 nt"
- "punpcklbw %%mm7, %%mm3 nt"
- "pmaddwd %%mm6, %%mm4 nt"
- "pmaddwd %%mm6, %%mm1 nt"
- "pmaddwd %%mm6, %%mm2 nt"
- "pmaddwd %%mm6, %%mm3 nt"
- #ifndef FAST_BGR2YV12
- "psrad $8, %%mm4 nt"
- "psrad $8, %%mm1 nt"
- "psrad $8, %%mm2 nt"
- "psrad $8, %%mm3 nt"
- #endif
- "packssdw %%mm1, %%mm4 nt"
- "packssdw %%mm3, %%mm2 nt"
- "pmaddwd %%mm5, %%mm4 nt"
- "pmaddwd %%mm5, %%mm2 nt"
- "addl $24, %%ebx nt"
- "packssdw %%mm2, %%mm4 nt"
- "psraw $7, %%mm4 nt"
- "packuswb %%mm4, %%mm0 nt"
- "paddusb "MANGLE(bgr2YOffset)", %%mm0 nt"
- MOVNTQ" %%mm0, (%1, %%eax) nt"
- "addl $8, %%eax nt"
- " js 1b nt"
- : : "r" (src+width*3), "r" (ydst+width), "g" (-width)
- : "%eax", "%ebx"
- );
- ydst += lumStride;
- src += srcStride;
- }
- src -= srcStride*2;
- asm volatile(
- "movl %4, %%eax nt"
- "movq "MANGLE(w1111)", %%mm5 nt"
- "movq "MANGLE(bgr2UCoeff)", %%mm6 nt"
- "pxor %%mm7, %%mm7 nt"
- "leal (%%eax, %%eax, 2), %%ebx nt"
- "addl %%ebx, %%ebx nt"
- ".balign 16 nt"
- "1: nt"
- PREFETCH" 64(%0, %%ebx) nt"
- PREFETCH" 64(%1, %%ebx) nt"
- #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
- "movq (%0, %%ebx), %%mm0 nt"
- "movq (%1, %%ebx), %%mm1 nt"
- "movq 6(%0, %%ebx), %%mm2 nt"
- "movq 6(%1, %%ebx), %%mm3 nt"
- PAVGB" %%mm1, %%mm0 nt"
- PAVGB" %%mm3, %%mm2 nt"
- "movq %%mm0, %%mm1 nt"
- "movq %%mm2, %%mm3 nt"
- "psrlq $24, %%mm0 nt"
- "psrlq $24, %%mm2 nt"
- PAVGB" %%mm1, %%mm0 nt"
- PAVGB" %%mm3, %%mm2 nt"
- "punpcklbw %%mm7, %%mm0 nt"
- "punpcklbw %%mm7, %%mm2 nt"
- #else
- "movd (%0, %%ebx), %%mm0 nt"
- "movd (%1, %%ebx), %%mm1 nt"
- "movd 3(%0, %%ebx), %%mm2 nt"
- "movd 3(%1, %%ebx), %%mm3 nt"
- "punpcklbw %%mm7, %%mm0 nt"
- "punpcklbw %%mm7, %%mm1 nt"
- "punpcklbw %%mm7, %%mm2 nt"
- "punpcklbw %%mm7, %%mm3 nt"
- "paddw %%mm1, %%mm0 nt"
- "paddw %%mm3, %%mm2 nt"
- "paddw %%mm2, %%mm0 nt"
- "movd 6(%0, %%ebx), %%mm4 nt"
- "movd 6(%1, %%ebx), %%mm1 nt"
- "movd 9(%0, %%ebx), %%mm2 nt"
- "movd 9(%1, %%ebx), %%mm3 nt"
- "punpcklbw %%mm7, %%mm4 nt"
- "punpcklbw %%mm7, %%mm1 nt"
- "punpcklbw %%mm7, %%mm2 nt"
- "punpcklbw %%mm7, %%mm3 nt"
- "paddw %%mm1, %%mm4 nt"
- "paddw %%mm3, %%mm2 nt"
- "paddw %%mm4, %%mm2 nt"
- "psrlw $2, %%mm0 nt"
- "psrlw $2, %%mm2 nt"
- #endif
- "movq "MANGLE(bgr2VCoeff)", %%mm1 nt"
- "movq "MANGLE(bgr2VCoeff)", %%mm3 nt"
- "pmaddwd %%mm0, %%mm1 nt"
- "pmaddwd %%mm2, %%mm3 nt"
- "pmaddwd %%mm6, %%mm0 nt"
- "pmaddwd %%mm6, %%mm2 nt"
- #ifndef FAST_BGR2YV12
- "psrad $8, %%mm0 nt"
- "psrad $8, %%mm1 nt"
- "psrad $8, %%mm2 nt"
- "psrad $8, %%mm3 nt"
- #endif
- "packssdw %%mm2, %%mm0 nt"
- "packssdw %%mm3, %%mm1 nt"
- "pmaddwd %%mm5, %%mm0 nt"
- "pmaddwd %%mm5, %%mm1 nt"
- "packssdw %%mm1, %%mm0 nt" // V1 V0 U1 U0
- "psraw $7, %%mm0 nt"
- #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
- "movq 12(%0, %%ebx), %%mm4 nt"
- "movq 12(%1, %%ebx), %%mm1 nt"
- "movq 18(%0, %%ebx), %%mm2 nt"
- "movq 18(%1, %%ebx), %%mm3 nt"
- PAVGB" %%mm1, %%mm4 nt"
- PAVGB" %%mm3, %%mm2 nt"
- "movq %%mm4, %%mm1 nt"
- "movq %%mm2, %%mm3 nt"
- "psrlq $24, %%mm4 nt"
- "psrlq $24, %%mm2 nt"
- PAVGB" %%mm1, %%mm4 nt"
- PAVGB" %%mm3, %%mm2 nt"
- "punpcklbw %%mm7, %%mm4 nt"
- "punpcklbw %%mm7, %%mm2 nt"
- #else
- "movd 12(%0, %%ebx), %%mm4 nt"
- "movd 12(%1, %%ebx), %%mm1 nt"
- "movd 15(%0, %%ebx), %%mm2 nt"
- "movd 15(%1, %%ebx), %%mm3 nt"
- "punpcklbw %%mm7, %%mm4 nt"
- "punpcklbw %%mm7, %%mm1 nt"
- "punpcklbw %%mm7, %%mm2 nt"
- "punpcklbw %%mm7, %%mm3 nt"
- "paddw %%mm1, %%mm4 nt"
- "paddw %%mm3, %%mm2 nt"
- "paddw %%mm2, %%mm4 nt"
- "movd 18(%0, %%ebx), %%mm5 nt"
- "movd 18(%1, %%ebx), %%mm1 nt"
- "movd 21(%0, %%ebx), %%mm2 nt"
- "movd 21(%1, %%ebx), %%mm3 nt"
- "punpcklbw %%mm7, %%mm5 nt"
- "punpcklbw %%mm7, %%mm1 nt"
- "punpcklbw %%mm7, %%mm2 nt"
- "punpcklbw %%mm7, %%mm3 nt"
- "paddw %%mm1, %%mm5 nt"
- "paddw %%mm3, %%mm2 nt"
- "paddw %%mm5, %%mm2 nt"
- "movq "MANGLE(w1111)", %%mm5 nt"
- "psrlw $2, %%mm4 nt"
- "psrlw $2, %%mm2 nt"
- #endif
- "movq "MANGLE(bgr2VCoeff)", %%mm1 nt"
- "movq "MANGLE(bgr2VCoeff)", %%mm3 nt"
- "pmaddwd %%mm4, %%mm1 nt"
- "pmaddwd %%mm2, %%mm3 nt"
- "pmaddwd %%mm6, %%mm4 nt"
- "pmaddwd %%mm6, %%mm2 nt"
- #ifndef FAST_BGR2YV12
- "psrad $8, %%mm4 nt"
- "psrad $8, %%mm1 nt"
- "psrad $8, %%mm2 nt"
- "psrad $8, %%mm3 nt"
- #endif
- "packssdw %%mm2, %%mm4 nt"
- "packssdw %%mm3, %%mm1 nt"
- "pmaddwd %%mm5, %%mm4 nt"
- "pmaddwd %%mm5, %%mm1 nt"
- "addl $24, %%ebx nt"
- "packssdw %%mm1, %%mm4 nt" // V3 V2 U3 U2
- "psraw $7, %%mm4 nt"
- "movq %%mm0, %%mm1 nt"
- "punpckldq %%mm4, %%mm0 nt"
- "punpckhdq %%mm4, %%mm1 nt"
- "packsswb %%mm1, %%mm0 nt"
- "paddb "MANGLE(bgr2UVOffset)", %%mm0 nt"
- "movd %%mm0, (%2, %%eax) nt"
- "punpckhdq %%mm0, %%mm0 nt"
- "movd %%mm0, (%3, %%eax) nt"
- "addl $4, %%eax nt"
- " js 1b nt"
- : : "r" (src+chromWidth*6), "r" (src+srcStride+chromWidth*6), "r" (udst+chromWidth), "r" (vdst+chromWidth), "g" (-chromWidth)
- : "%eax", "%ebx"
- );
- udst += chromStride;
- vdst += chromStride;
- src += srcStride*2;
- }
- asm volatile( EMMS" nt"
- SFENCE" nt"
- :::"memory");
- #else
- y=0;
- #endif
- for(; y<height; y+=2)
- {
- unsigned i;
- for(i=0; i<chromWidth; i++)
- {
- unsigned int b= src[6*i+0];
- unsigned int g= src[6*i+1];
- unsigned int r= src[6*i+2];
- unsigned int Y = ((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16;
- unsigned int V = ((RV*r + GV*g + BV*b)>>RGB2YUV_SHIFT) + 128;
- unsigned int U = ((RU*r + GU*g + BU*b)>>RGB2YUV_SHIFT) + 128;
- udst[i] = U;
- vdst[i] = V;
- ydst[2*i] = Y;
- b= src[6*i+3];
- g= src[6*i+4];
- r= src[6*i+5];
- Y = ((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16;
- ydst[2*i+1] = Y;
- }
- ydst += lumStride;
- src += srcStride;
- for(i=0; i<chromWidth; i++)
- {
- unsigned int b= src[6*i+0];
- unsigned int g= src[6*i+1];
- unsigned int r= src[6*i+2];
- unsigned int Y = ((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16;
- ydst[2*i] = Y;
- b= src[6*i+3];
- g= src[6*i+4];
- r= src[6*i+5];
- Y = ((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16;
- ydst[2*i+1] = Y;
- }
- udst += chromStride;
- vdst += chromStride;
- ydst += lumStride;
- src += srcStride;
- }
- }
- void RENAME(interleaveBytes)(uint8_t *src1, uint8_t *src2, uint8_t *dest,
- unsigned width, unsigned height, int src1Stride,
- int src2Stride, int dstStride){
- unsigned h;
- for(h=0; h < height; h++)
- {
- unsigned w;
- #ifdef HAVE_MMX
- #ifdef HAVE_SSE2
- asm(
- "xorl %%eax, %%eax nt"
- "1: nt"
- PREFETCH" 64(%1, %%eax) nt"
- PREFETCH" 64(%2, %%eax) nt"
- "movdqa (%1, %%eax), %%xmm0 nt"
- "movdqa (%1, %%eax), %%xmm1 nt"
- "movdqa (%2, %%eax), %%xmm2 nt"
- "punpcklbw %%xmm2, %%xmm0 nt"
- "punpckhbw %%xmm2, %%xmm1 nt"
- "movntdq %%xmm0, (%0, %%eax, 2) nt"
- "movntdq %%xmm1, 16(%0, %%eax, 2)nt"
- "addl $16, %%eax nt"
- "cmpl %3, %%eax nt"
- " jb 1b nt"
- ::"r"(dest), "r"(src1), "r"(src2), "r" (width-15)
- : "memory", "%eax"
- );
- #else
- asm(
- "xorl %%eax, %%eax nt"
- "1: nt"
- PREFETCH" 64(%1, %%eax) nt"
- PREFETCH" 64(%2, %%eax) nt"
- "movq (%1, %%eax), %%mm0 nt"
- "movq 8(%1, %%eax), %%mm2 nt"
- "movq %%mm0, %%mm1 nt"
- "movq %%mm2, %%mm3 nt"
- "movq (%2, %%eax), %%mm4 nt"
- "movq 8(%2, %%eax), %%mm5 nt"
- "punpcklbw %%mm4, %%mm0 nt"
- "punpckhbw %%mm4, %%mm1 nt"
- "punpcklbw %%mm5, %%mm2 nt"
- "punpckhbw %%mm5, %%mm3 nt"
- MOVNTQ" %%mm0, (%0, %%eax, 2) nt"
- MOVNTQ" %%mm1, 8(%0, %%eax, 2) nt"
- MOVNTQ" %%mm2, 16(%0, %%eax, 2) nt"
- MOVNTQ" %%mm3, 24(%0, %%eax, 2) nt"
- "addl $16, %%eax nt"
- "cmpl %3, %%eax nt"
- " jb 1b nt"
- ::"r"(dest), "r"(src1), "r"(src2), "r" (width-15)
- : "memory", "%eax"
- );
- #endif
- for(w= (width&(~15)); w < width; w++)
- {
- dest[2*w+0] = src1[w];
- dest[2*w+1] = src2[w];
- }
- #else
- for(w=0; w < width; w++)
- {
- dest[2*w+0] = src1[w];
- dest[2*w+1] = src2[w];
- }
- #endif
- dest += dstStride;
- src1 += src1Stride;
- src2 += src2Stride;
- }
- #ifdef HAVE_MMX
- asm(
- EMMS" nt"
- SFENCE" nt"
- ::: "memory"
- );
- #endif
- }
- static inline void RENAME(vu9_to_vu12)(const uint8_t *src1, const uint8_t *src2,
- uint8_t *dst1, uint8_t *dst2,
- unsigned width, unsigned height,
- int srcStride1, int srcStride2,
- int dstStride1, int dstStride2)
- {
- unsigned int y,x,h;
- int w;
- w=width/2; h=height/2;
- #ifdef HAVE_MMX
- asm volatile(
- PREFETCH" %0nt"
- PREFETCH" %1nt"
- ::"m"(*(src1+srcStride1)),"m"(*(src2+srcStride2)):"memory");
- #endif
- for(y=0;y<h;y++){
- const uint8_t* s1=src1+srcStride1*(y>>1);
- uint8_t* d=dst1+dstStride1*y;
- x=0;
- #ifdef HAVE_MMX
- for(;x<w-31;x+=32)
- {
- asm volatile(
- PREFETCH" 32%1nt"
- "movq %1, %%mm0nt"
- "movq 8%1, %%mm2nt"
- "movq 16%1, %%mm4nt"
- "movq 24%1, %%mm6nt"
- "movq %%mm0, %%mm1nt"
- "movq %%mm2, %%mm3nt"
- "movq %%mm4, %%mm5nt"
- "movq %%mm6, %%mm7nt"
- "punpcklbw %%mm0, %%mm0nt"
- "punpckhbw %%mm1, %%mm1nt"
- "punpcklbw %%mm2, %%mm2nt"
- "punpckhbw %%mm3, %%mm3nt"
- "punpcklbw %%mm4, %%mm4nt"
- "punpckhbw %%mm5, %%mm5nt"
- "punpcklbw %%mm6, %%mm6nt"
- "punpckhbw %%mm7, %%mm7nt"
- MOVNTQ" %%mm0, %0nt"
- MOVNTQ" %%mm1, 8%0nt"
- MOVNTQ" %%mm2, 16%0nt"
- MOVNTQ" %%mm3, 24%0nt"
- MOVNTQ" %%mm4, 32%0nt"
- MOVNTQ" %%mm5, 40%0nt"
- MOVNTQ" %%mm6, 48%0nt"
- MOVNTQ" %%mm7, 56%0"
- :"=m"(d[2*x])
- :"m"(s1[x])
- :"memory");
- }
- #endif
- for(;x<w;x++) d[2*x]=d[2*x+1]=s1[x];
- }
- for(y=0;y<h;y++){
- const uint8_t* s2=src2+srcStride2*(y>>1);
- uint8_t* d=dst2+dstStride2*y;
- x=0;
- #ifdef HAVE_MMX
- for(;x<w-31;x+=32)
- {
- asm volatile(
- PREFETCH" 32%1nt"
- "movq %1, %%mm0nt"
- "movq 8%1, %%mm2nt"
- "movq 16%1, %%mm4nt"
- "movq 24%1, %%mm6nt"
- "movq %%mm0, %%mm1nt"
- "movq %%mm2, %%mm3nt"
- "movq %%mm4, %%mm5nt"
- "movq %%mm6, %%mm7nt"
- "punpcklbw %%mm0, %%mm0nt"
- "punpckhbw %%mm1, %%mm1nt"
- "punpcklbw %%mm2, %%mm2nt"
- "punpckhbw %%mm3, %%mm3nt"
- "punpcklbw %%mm4, %%mm4nt"
- "punpckhbw %%mm5, %%mm5nt"
- "punpcklbw %%mm6, %%mm6nt"
- "punpckhbw %%mm7, %%mm7nt"
- MOVNTQ" %%mm0, %0nt"
- MOVNTQ" %%mm1, 8%0nt"
- MOVNTQ" %%mm2, 16%0nt"
- MOVNTQ" %%mm3, 24%0nt"
- MOVNTQ" %%mm4, 32%0nt"
- MOVNTQ" %%mm5, 40%0nt"
- MOVNTQ" %%mm6, 48%0nt"
- MOVNTQ" %%mm7, 56%0"
- :"=m"(d[2*x])
- :"m"(s2[x])
- :"memory");
- }
- #endif
- for(;x<w;x++) d[2*x]=d[2*x+1]=s2[x];
- }
- #ifdef HAVE_MMX
- asm(
- EMMS" nt"
- SFENCE" nt"
- ::: "memory"
- );
- #endif
- }
- static inline void RENAME(yvu9_to_yuy2)(const uint8_t *src1, const uint8_t *src2, const uint8_t *src3,
- uint8_t *dst,
- unsigned width, unsigned height,
- int srcStride1, int srcStride2,
- int srcStride3, int dstStride)
- {
- unsigned y,x,w,h;
- w=width/2; h=height;
- for(y=0;y<h;y++){
- const uint8_t* yp=src1+srcStride1*y;
- const uint8_t* up=src2+srcStride2*(y>>2);
- const uint8_t* vp=src3+srcStride3*(y>>2);
- uint8_t* d=dst+dstStride*y;
- x=0;
- #ifdef HAVE_MMX
- for(;x<w-7;x+=8)
- {
- asm volatile(
- PREFETCH" 32(%1, %0)nt"
- PREFETCH" 32(%2, %0)nt"
- PREFETCH" 32(%3, %0)nt"
- "movq (%1, %0, 4), %%mm0nt" /* Y0Y1Y2Y3Y4Y5Y6Y7 */
- "movq (%2, %0), %%mm1nt" /* U0U1U2U3U4U5U6U7 */
- "movq (%3, %0), %%mm2nt" /* V0V1V2V3V4V5V6V7 */
- "movq %%mm0, %%mm3nt" /* Y0Y1Y2Y3Y4Y5Y6Y7 */
- "movq %%mm1, %%mm4nt" /* U0U1U2U3U4U5U6U7 */
- "movq %%mm2, %%mm5nt" /* V0V1V2V3V4V5V6V7 */
- "punpcklbw %%mm1, %%mm1nt" /* U0U0 U1U1 U2U2 U3U3 */
- "punpcklbw %%mm2, %%mm2nt" /* V0V0 V1V1 V2V2 V3V3 */
- "punpckhbw %%mm4, %%mm4nt" /* U4U4 U5U5 U6U6 U7U7 */
- "punpckhbw %%mm5, %%mm5nt" /* V4V4 V5V5 V6V6 V7V7 */
- "movq %%mm1, %%mm6nt"
- "punpcklbw %%mm2, %%mm1nt" /* U0V0 U0V0 U1V1 U1V1*/
- "punpcklbw %%mm1, %%mm0nt" /* Y0U0 Y1V0 Y2U0 Y3V0*/
- "punpckhbw %%mm1, %%mm3nt" /* Y4U1 Y5V1 Y6U1 Y7V1*/
- MOVNTQ" %%mm0, (%4, %0, 8)nt"
- MOVNTQ" %%mm3, 8(%4, %0, 8)nt"
-
- "punpckhbw %%mm2, %%mm6nt" /* U2V2 U2V2 U3V3 U3V3*/
- "movq 8(%1, %0, 4), %%mm0nt"
- "movq %%mm0, %%mm3nt"
- "punpcklbw %%mm6, %%mm0nt" /* Y U2 Y V2 Y U2 Y V2*/
- "punpckhbw %%mm6, %%mm3nt" /* Y U3 Y V3 Y U3 Y V3*/
- MOVNTQ" %%mm0, 16(%4, %0, 8)nt"
- MOVNTQ" %%mm3, 24(%4, %0, 8)nt"
- "movq %%mm4, %%mm6nt"
- "movq 16(%1, %0, 4), %%mm0nt"
- "movq %%mm0, %%mm3nt"
- "punpcklbw %%mm5, %%mm4nt"
- "punpcklbw %%mm4, %%mm0nt" /* Y U4 Y V4 Y U4 Y V4*/
- "punpckhbw %%mm4, %%mm3nt" /* Y U5 Y V5 Y U5 Y V5*/
- MOVNTQ" %%mm0, 32(%4, %0, 8)nt"
- MOVNTQ" %%mm3, 40(%4, %0, 8)nt"
-
- "punpckhbw %%mm5, %%mm6nt"
- "movq 24(%1, %0, 4), %%mm0nt"
- "movq %%mm0, %%mm3nt"
- "punpcklbw %%mm6, %%mm0nt" /* Y U6 Y V6 Y U6 Y V6*/
- "punpckhbw %%mm6, %%mm3nt" /* Y U7 Y V7 Y U7 Y V7*/
- MOVNTQ" %%mm0, 48(%4, %0, 8)nt"
- MOVNTQ" %%mm3, 56(%4, %0, 8)nt"
- : "+r" (x)
- : "r"(yp), "r" (up), "r"(vp), "r"(d)
- :"memory");
- }
- #endif
- for(; x<w; x++)
- {
- const int x2= x<<2;
- d[8*x+0]=yp[x2];
- d[8*x+1]=up[x];
- d[8*x+2]=yp[x2+1];
- d[8*x+3]=vp[x];
- d[8*x+4]=yp[x2+2];
- d[8*x+5]=up[x];
- d[8*x+6]=yp[x2+3];
- d[8*x+7]=vp[x];
- }
- }
- #ifdef HAVE_MMX
- asm(
- EMMS" nt"
- SFENCE" nt"
- ::: "memory"
- );
- #endif
- }