comparison libswscale/rgb2rgb_template.c @ 29481:c080f1f5c07e

Cosmetics: - Place curly brackets in the same line as while/for/if/switch/else/do; - Place curly brackets at column 0 in the next line starting a function.

29480:a4d8dee13834

    end = s + src_size;
    #if HAVE_MMX
    __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");
    mm_end = end - 23;
    __asm__ volatile("movq        %0, %%mm7"::"m"(mask32a):"memory");
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "punpckldq    3%1, %%mm0    \n\t"
            "movd         6%1, %%mm1    \n\t"

        s += 24;
    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
    #endif
    while (s < end)
    {
    #if HAVE_BIGENDIAN
        /* RGB24 (= R,G,B) -> RGB32 (= A,B,G,R) */
        *dest++ = 255;
        *dest++ = s[2];
        *dest++ = s[1];

#endif
    end = s + src_size;
#if HAVE_MMX
    __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");
    mm_end = end - 31;
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movq          %1, %%mm0    \n\t"
            "movq         8%1, %%mm1    \n\t"
            "movq        16%1, %%mm4    \n\t"

        s += 32;
    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end)
    {
#if HAVE_BIGENDIAN
        /* RGB32 (= A,B,G,R) -> RGB24 (= R,G,B) */
        s++;
        dest[2] = *s++;
        dest[1] = *s++;

    end = s + src_size;
#if 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%1         \n\t"
            "movq        %1, %%mm0  \n\t"
            "movq       8%1, %%mm2  \n\t"
            "movq     %%mm0, %%mm1  \n\t"

    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    mm_end = end - 3;
    while (s < mm_end)
    {
        register unsigned x= *((const uint32_t *)s);
        *((uint32_t *)d) = (x&0x7FFF7FFF) + (x&0x7FE07FE0);
        d+=4;
        s+=4;
    }
    if (s < end)
    {
        register unsigned short x= *((const uint16_t *)s);
        *((uint16_t *)d) = (x&0x7FFF) + (x&0x7FE0);
    }
}

#if 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%1         \n\t"
            "movq        %1, %%mm0  \n\t"
            "movq       8%1, %%mm2  \n\t"
            "movq     %%mm0, %%mm1  \n\t"

    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    mm_end = end - 3;
    while (s < mm_end)
    {
        register uint32_t x= *((const uint32_t*)s);
        *((uint32_t *)d) = ((x>>1)&0x7FE07FE0) | (x&0x001F001F);
        s+=4;
        d+=4;
    }
    if (s < end)
    {
        register uint16_t x= *((const uint16_t*)s);
        *((uint16_t *)d) = ((x>>1)&0x7FE0) | (x&0x001F);
    }
}

    __asm__ volatile(PREFETCH"    %0"::"m"(*src):"memory");
    __asm__ volatile(
        "movq    %0, %%mm7    \n\t"
        "movq    %1, %%mm6    \n\t"
        ::"m"(red_16mask),"m"(green_16mask));
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         4%1, %%mm3    \n\t"
            "punpckldq    8%1, %%mm0    \n\t"

    }
#endif
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end)
    {
        register int rgb = *(const uint32_t*)s; s += 4;
        *d++ = ((rgb&0xFF)>>3) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>8);
    }
}

    __asm__ volatile(
        "movq          %0, %%mm7    \n\t"
        "movq          %1, %%mm6    \n\t"
        ::"m"(red_16mask),"m"(green_16mask));
    mm_end = end - 15;
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         4%1, %%mm3    \n\t"
            "punpckldq    8%1, %%mm0    \n\t"

        s += 16;
    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end)
    {
        register int rgb = *(const uint32_t*)s; s += 4;
        *d++ = ((rgb&0xF8)<<8) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>19);
    }
}

    __asm__ volatile(PREFETCH"    %0"::"m"(*src):"memory");
    __asm__ volatile(
        "movq          %0, %%mm7    \n\t"
        "movq          %1, %%mm6    \n\t"
        ::"m"(red_15mask),"m"(green_15mask));
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         4%1, %%mm3    \n\t"
            "punpckldq    8%1, %%mm0    \n\t"

    }
#endif
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end)
    {
        register int rgb = *(const uint32_t*)s; s += 4;
        *d++ = ((rgb&0xFF)>>3) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>9);
    }
}

    __asm__ volatile(
        "movq          %0, %%mm7    \n\t"
        "movq          %1, %%mm6    \n\t"
        ::"m"(red_15mask),"m"(green_15mask));
    mm_end = end - 15;
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         4%1, %%mm3    \n\t"
            "punpckldq    8%1, %%mm0    \n\t"

        s += 16;
    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end)
    {
        register int rgb = *(const uint32_t*)s; s += 4;
        *d++ = ((rgb&0xF8)<<7) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>19);
    }
}

    __asm__ volatile(
        "movq         %0, %%mm7     \n\t"
        "movq         %1, %%mm6     \n\t"
        ::"m"(red_16mask),"m"(green_16mask));
    mm_end = end - 11;
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         3%1, %%mm3    \n\t"
            "punpckldq    6%1, %%mm0    \n\t"

        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);
    }

    __asm__ volatile(
        "movq         %0, %%mm7     \n\t"
        "movq         %1, %%mm6     \n\t"
        ::"m"(red_16mask),"m"(green_16mask));
    mm_end = end - 15;
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         3%1, %%mm3    \n\t"
            "punpckldq    6%1, %%mm0    \n\t"

        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);
    }

    __asm__ volatile(
        "movq          %0, %%mm7    \n\t"
        "movq          %1, %%mm6    \n\t"
        ::"m"(red_15mask),"m"(green_15mask));
    mm_end = end - 11;
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         3%1, %%mm3    \n\t"
            "punpckldq    6%1, %%mm0    \n\t"

        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);
    }

    __asm__ volatile(
        "movq         %0, %%mm7     \n\t"
        "movq         %1, %%mm6     \n\t"
        ::"m"(red_15mask),"m"(green_15mask));
    mm_end = end - 15;
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"   32%1            \n\t"
            "movd         %1, %%mm0     \n\t"
            "movd        3%1, %%mm3     \n\t"
            "punpckldq   6%1, %%mm0     \n\t"

        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);
    }

    const uint16_t *s = (const uint16_t*)src;
    end = s + src_size/2;
#if HAVE_MMX
    __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");
    mm_end = end - 7;
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movq          %1, %%mm0    \n\t"
            "movq          %1, %%mm1    \n\t"
            "movq          %1, %%mm2    \n\t"

        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;

    const uint16_t *s = (const uint16_t *)src;
    end = s + src_size/2;
#if HAVE_MMX
    __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");
    mm_end = end - 7;
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movq          %1, %%mm0    \n\t"
            "movq          %1, %%mm1    \n\t"
            "movq          %1, %%mm2    \n\t"

        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;

#if HAVE_MMX
    __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");
    __asm__ volatile("pxor    %%mm7,%%mm7    \n\t":::"memory");
    __asm__ volatile("pcmpeqd %%mm6,%%mm6    \n\t":::"memory");
    mm_end = end - 3;
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movq          %1, %%mm0    \n\t"
            "movq          %1, %%mm1    \n\t"
            "movq          %1, %%mm2    \n\t"

        s += 4;
    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end)
    {
        register uint16_t bgr;
        bgr = *s++;
#if HAVE_BIGENDIAN
        *d++ = 255;
        *d++ = (bgr&0x7C00)>>7;

#if HAVE_MMX
    __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");
    __asm__ volatile("pxor    %%mm7,%%mm7    \n\t":::"memory");
    __asm__ volatile("pcmpeqd %%mm6,%%mm6    \n\t":::"memory");
    mm_end = end - 3;
    while (s < mm_end)
    {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movq          %1, %%mm0    \n\t"
            "movq          %1, %%mm1    \n\t"
            "movq          %1, %%mm2    \n\t"

        s += 4;
    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end)
    {
        register uint16_t bgr;
        bgr = *s++;
#if HAVE_BIGENDIAN
        *d++ = 255;
        *d++ = (bgr&0xF800)>>8;

    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;

                                           long width, long height,
                                           long lumStride, long chromStride, long dstStride, long vertLumPerChroma)
{
    long y;
    const x86_reg chromWidth= width>>1;
    for (y=0; y<height; y++)
    {
#if HAVE_MMX
//FIXME handle 2 lines at once (fewer prefetches, reuse some chroma, but very likely memory-limited anyway)
        __asm__ volatile(
            "xor                 %%"REG_a", %%"REG_a"   \n\t"
            ASMALIGN(4)

        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));

#elif HAVE_FAST_64BIT
        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);

        }

#else
        int i, *idst = (int32_t *) dst;
        const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;
        for (i = 0; i < chromWidth; i++){
#if HAVE_BIGENDIAN
            *idst++ = (yc[0] << 24)+ (uc[0] << 16) +
                (yc[1] << 8) + (vc[0] << 0);
#else
            *idst++ = yc[0] + (uc[0] << 8) +

            uc++;
            vc++;
        }
#endif
#endif
        if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1)
        {
            usrc += chromStride;
            vsrc += chromStride;
        }
        ysrc += lumStride;
        dst  += dstStride;

                                           long width, long height,
                                           long lumStride, long chromStride, long dstStride, long vertLumPerChroma)
{
    long y;
    const x86_reg chromWidth= width>>1;
    for (y=0; y<height; y++)
    {
#if HAVE_MMX
//FIXME handle 2 lines at once (fewer prefetches, reuse some chroma, but very likely memory-limited anyway)
        __asm__ volatile(
            "xor                %%"REG_a", %%"REG_a"    \n\t"
            ASMALIGN(4)

#if HAVE_FAST_64BIT
        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);

        }

#else
        int i, *idst = (int32_t *) dst;
        const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;
        for (i = 0; i < chromWidth; i++){
#if HAVE_BIGENDIAN
            *idst++ = (uc[0] << 24)+ (yc[0] << 16) +
                (vc[0] << 8) + (yc[1] << 0);
#else
            *idst++ = uc[0] + (yc[0] << 8) +

            uc++;
            vc++;
        }
#endif
#endif
        if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1)
        {
            usrc += chromStride;
            vsrc += chromStride;
        }
        ysrc += lumStride;
        dst += dstStride;

                                      long width, long height,
                                      long lumStride, long chromStride, long srcStride)
{
    long y;
    const x86_reg chromWidth= width>>1;
    for (y=0; y<height; y+=2)
    {
#if HAVE_MMX
        __asm__ volatile(
            "xor                 %%"REG_a", %%"REG_a"   \n\t"
            "pcmpeqw                 %%mm7, %%mm7       \n\t"
            "psrlw                      $8, %%mm7       \n\t" // FF,00,FF,00...

            ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
            : "memory", "%"REG_a
        );
#else
        long 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;

    long 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 HAVE_MMX2 || HAVE_AMD3DNOW
        const x86_reg mmxSize= srcWidth&~15;
        __asm__ volatile(
            "mov           %4, %%"REG_a"            \n\t"
            "1:                                     \n\t"

        const x86_reg 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;
        }

    // 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

                                      long width, long height,
                                      long lumStride, long chromStride, long srcStride)
{
    long y;
    const x86_reg chromWidth= width>>1;
    for (y=0; y<height; y+=2)
    {
#if HAVE_MMX
        __asm__ volatile(
            "xor                 %%"REG_a", %%"REG_a"   \n\t"
            "pcmpeqw             %%mm7, %%mm7   \n\t"
            "psrlw                  $8, %%mm7   \n\t" // FF,00,FF,00...

            ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
            : "memory", "%"REG_a
        );
#else
        long 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;

                                       long lumStride, long chromStride, long srcStride)
{
    long y;
    const x86_reg chromWidth= width>>1;
#if HAVE_MMX
    for (y=0; y<height-2; y+=2)
    {
        long i;
        for (i=0; i<2; i++)
        {
            __asm__ volatile(
                "mov                        %2, %%"REG_a"   \n\t"
                "movq  "MANGLE(ff_bgr2YCoeff)", %%mm6       \n\t"
                "movq       "MANGLE(ff_w1111)", %%mm5       \n\t"
                "pxor                    %%mm7, %%mm7       \n\t"

                     SFENCE"     \n\t"
                     :::"memory");
#else
    y=0;
#endif
    for (; y<height; y+=2)
    {
        long 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;

            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;

    }
}

static void RENAME(interleaveBytes)(uint8_t *src1, uint8_t *src2, uint8_t *dest,
                             long width, long height, long src1Stride,
                             long src2Stride, long dstStride){
    long h;

    for (h=0; h < height; h++)
    {
        long w;

#if HAVE_MMX
#if HAVE_SSE2
        __asm__(

            " jb                     1b             \n\t"
            ::"r"(dest), "r"(src1), "r"(src2), "r" ((x86_reg)width-15)
            : "memory", "%"REG_a
        );
#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;

    __asm__ volatile(
        PREFETCH" %0    \n\t"
        PREFETCH" %1    \n\t"
        ::"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;
#if HAVE_MMX
        for (;x<w-31;x+=32)
        {
            __asm__ volatile(
                PREFETCH"   32%1        \n\t"
                "movq         %1, %%mm0 \n\t"
                "movq        8%1, %%mm2 \n\t"
                "movq       16%1, %%mm4 \n\t"

                :"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;
#if HAVE_MMX
        for (;x<w-31;x+=32)
        {
            __asm__ volatile(
                PREFETCH"   32%1        \n\t"
                "movq         %1, %%mm0 \n\t"
                "movq        8%1, %%mm2 \n\t"
                "movq       16%1, %%mm4 \n\t"

                                        long srcStride3, long dstStride)
{
    x86_reg x;
    long y,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;
#if HAVE_MMX
        for (;x<w-7;x+=8)
        {
            __asm__ volatile(
                PREFETCH"   32(%1, %0)          \n\t"
                PREFETCH"   32(%2, %0)          \n\t"
                PREFETCH"   32(%3, %0)          \n\t"
                "movq      (%1, %0, 4), %%mm0   \n\t" /* Y0Y1Y2Y3Y4Y5Y6Y7 */

                : "+r" (x)
                : "r"(yp), "r" (up), "r"(vp), "r"(d)
                :"memory");
        }
#endif
        for (; x<w; x++)
        {
            const long 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];

    dst +=   count;
    src += 2*count;
    count= - count;

#if HAVE_MMX
    if(count <= -16){
        count += 15;
        __asm__ volatile(
            "pcmpeqw       %%mm7, %%mm7        \n\t"
            "psrlw            $8, %%mm7        \n\t"
            "1:                                \n\t"

            : "r"(src), "r"(dst)
        );
        count -= 15;
    }
#endif
    while(count<0){
        dst[count]= src[2*count];
        count++;
    }
}

    dst0+=   count;
    dst1+=   count;
    src += 4*count;
    count= - count;
#if HAVE_MMX
    if(count <= -8){
        count += 7;
        __asm__ volatile(
            "pcmpeqw       %%mm7, %%mm7        \n\t"
            "psrlw            $8, %%mm7        \n\t"
            "1:                                \n\t"

            : "r"(src), "r"(dst0), "r"(dst1)
        );
        count -= 7;
    }
#endif
    while(count<0){
        dst0[count]= src[4*count+0];
        dst1[count]= src[4*count+2];
        count++;
    }
}

    dst1 +=   count;
    src0 += 4*count;
    src1 += 4*count;
    count= - count;
#ifdef PAVGB
    if(count <= -8){
        count += 7;
        __asm__ volatile(
            "pcmpeqw        %%mm7, %%mm7        \n\t"
            "psrlw             $8, %%mm7        \n\t"
            "1:                                \n\t"

            : "r"(src0), "r"(src1), "r"(dst0), "r"(dst1)
        );
        count -= 7;
    }
#endif
    while(count<0){
        dst0[count]= (src0[4*count+0]+src1[4*count+0])>>1;
        dst1[count]= (src0[4*count+2]+src1[4*count+2])>>1;
        count++;
    }
}

    dst0+=   count;
    dst1+=   count;
    src += 4*count;
    count= - count;
#if HAVE_MMX
    if(count <= -8){
        count += 7;
        __asm__ volatile(
            "pcmpeqw       %%mm7, %%mm7        \n\t"
            "psrlw            $8, %%mm7        \n\t"
            "1:                                \n\t"

        );
        count -= 7;
    }
#endif
    src++;
    while(count<0){
        dst0[count]= src[4*count+0];
        dst1[count]= src[4*count+2];
        count++;
    }
}

    dst1 +=   count;
    src0 += 4*count;
    src1 += 4*count;
    count= - count;
#ifdef PAVGB
    if(count <= -8){
        count += 7;
        __asm__ volatile(
            "pcmpeqw        %%mm7, %%mm7        \n\t"
            "psrlw             $8, %%mm7        \n\t"
            "1:                                \n\t"

        count -= 7;
    }
#endif
    src0++;
    src1++;
    while(count<0){
        dst0[count]= (src0[4*count+0]+src1[4*count+0])>>1;
        dst1[count]= (src0[4*count+2]+src1[4*count+2])>>1;
        count++;
    }
}

                                      long lumStride, long chromStride, long srcStride)
{
    long y;
    const long chromWidth= -((-width)>>1);

    for (y=0; y<height; y++){
        RENAME(extract_even)(src, ydst, width);
        if(y&1){
            RENAME(extract_odd2avg)(src-srcStride, src, udst, vdst, chromWidth);
            udst+= chromStride;
            vdst+= chromStride;
        }

                                      long lumStride, long chromStride, long srcStride)
{
    long y;
    const long chromWidth= -((-width)>>1);

    for (y=0; y<height; y++){
        RENAME(extract_even)(src, ydst, width);
        RENAME(extract_odd2)(src, udst, vdst, chromWidth);

        src += srcStride;
        ydst+= lumStride;

                                      long lumStride, long chromStride, long srcStride)
{
    long y;
    const long chromWidth= -((-width)>>1);

    for (y=0; y<height; y++){
        RENAME(extract_even)(src+1, ydst, width);
        if(y&1){
            RENAME(extract_even2avg)(src-srcStride, src, udst, vdst, chromWidth);
            udst+= chromStride;
            vdst+= chromStride;
        }

                                      long lumStride, long chromStride, long srcStride)
{
    long y;
    const long chromWidth= -((-width)>>1);

    for (y=0; y<height; y++){
        RENAME(extract_even)(src+1, ydst, width);
        RENAME(extract_even2)(src, udst, vdst, chromWidth);

        src += srcStride;
        ydst+= lumStride;

            ::: "memory"
        );
#endif
}

static inline void RENAME(rgb2rgb_init)(void){
    rgb15to16       = RENAME(rgb15to16);
    rgb15tobgr24    = RENAME(rgb15tobgr24);
    rgb15to32       = RENAME(rgb15to32);
    rgb16tobgr24    = RENAME(rgb16tobgr24);
    rgb16to32       = RENAME(rgb16to32);

29481:c080f1f5c07e

    end = s + src_size;
    #if HAVE_MMX
    __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");
    mm_end = end - 23;
    __asm__ volatile("movq        %0, %%mm7"::"m"(mask32a):"memory");
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "punpckldq    3%1, %%mm0    \n\t"
            "movd         6%1, %%mm1    \n\t"

        s += 24;
    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
    #endif
    while (s < end) {
    #if HAVE_BIGENDIAN
        /* RGB24 (= R,G,B) -> RGB32 (= A,B,G,R) */
        *dest++ = 255;
        *dest++ = s[2];
        *dest++ = s[1];

#endif
    end = s + src_size;
#if HAVE_MMX
    __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");
    mm_end = end - 31;
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movq          %1, %%mm0    \n\t"
            "movq         8%1, %%mm1    \n\t"
            "movq        16%1, %%mm4    \n\t"

        s += 32;
    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end) {
#if HAVE_BIGENDIAN
        /* RGB32 (= A,B,G,R) -> RGB24 (= R,G,B) */
        s++;
        dest[2] = *s++;
        dest[1] = *s++;

    end = s + src_size;
#if 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%1         \n\t"
            "movq        %1, %%mm0  \n\t"
            "movq       8%1, %%mm2  \n\t"
            "movq     %%mm0, %%mm1  \n\t"

    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    mm_end = end - 3;
    while (s < mm_end) {
        register unsigned x= *((const uint32_t *)s);
        *((uint32_t *)d) = (x&0x7FFF7FFF) + (x&0x7FE07FE0);
        d+=4;
        s+=4;
    }
    if (s < end) {
        register unsigned short x= *((const uint16_t *)s);
        *((uint16_t *)d) = (x&0x7FFF) + (x&0x7FE0);
    }
}

#if 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%1         \n\t"
            "movq        %1, %%mm0  \n\t"
            "movq       8%1, %%mm2  \n\t"
            "movq     %%mm0, %%mm1  \n\t"

    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    mm_end = end - 3;
    while (s < mm_end) {
        register uint32_t x= *((const uint32_t*)s);
        *((uint32_t *)d) = ((x>>1)&0x7FE07FE0) | (x&0x001F001F);
        s+=4;
        d+=4;
    }
    if (s < end) {
        register uint16_t x= *((const uint16_t*)s);
        *((uint16_t *)d) = ((x>>1)&0x7FE0) | (x&0x001F);
    }
}

    __asm__ volatile(PREFETCH"    %0"::"m"(*src):"memory");
    __asm__ volatile(
        "movq    %0, %%mm7    \n\t"
        "movq    %1, %%mm6    \n\t"
        ::"m"(red_16mask),"m"(green_16mask));
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         4%1, %%mm3    \n\t"
            "punpckldq    8%1, %%mm0    \n\t"

    }
#endif
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end) {
        register int rgb = *(const uint32_t*)s; s += 4;
        *d++ = ((rgb&0xFF)>>3) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>8);
    }
}

    __asm__ volatile(
        "movq          %0, %%mm7    \n\t"
        "movq          %1, %%mm6    \n\t"
        ::"m"(red_16mask),"m"(green_16mask));
    mm_end = end - 15;
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         4%1, %%mm3    \n\t"
            "punpckldq    8%1, %%mm0    \n\t"

        s += 16;
    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end) {
        register int rgb = *(const uint32_t*)s; s += 4;
        *d++ = ((rgb&0xF8)<<8) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>19);
    }
}

    __asm__ volatile(PREFETCH"    %0"::"m"(*src):"memory");
    __asm__ volatile(
        "movq          %0, %%mm7    \n\t"
        "movq          %1, %%mm6    \n\t"
        ::"m"(red_15mask),"m"(green_15mask));
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         4%1, %%mm3    \n\t"
            "punpckldq    8%1, %%mm0    \n\t"

    }
#endif
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end) {
        register int rgb = *(const uint32_t*)s; s += 4;
        *d++ = ((rgb&0xFF)>>3) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>9);
    }
}

    __asm__ volatile(
        "movq          %0, %%mm7    \n\t"
        "movq          %1, %%mm6    \n\t"
        ::"m"(red_15mask),"m"(green_15mask));
    mm_end = end - 15;
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         4%1, %%mm3    \n\t"
            "punpckldq    8%1, %%mm0    \n\t"

        s += 16;
    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end) {
        register int rgb = *(const uint32_t*)s; s += 4;
        *d++ = ((rgb&0xF8)<<7) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>19);
    }
}

    __asm__ volatile(
        "movq         %0, %%mm7     \n\t"
        "movq         %1, %%mm6     \n\t"
        ::"m"(red_16mask),"m"(green_16mask));
    mm_end = end - 11;
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         3%1, %%mm3    \n\t"
            "punpckldq    6%1, %%mm0    \n\t"

        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);
    }

    __asm__ volatile(
        "movq         %0, %%mm7     \n\t"
        "movq         %1, %%mm6     \n\t"
        ::"m"(red_16mask),"m"(green_16mask));
    mm_end = end - 15;
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         3%1, %%mm3    \n\t"
            "punpckldq    6%1, %%mm0    \n\t"

        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);
    }

    __asm__ volatile(
        "movq          %0, %%mm7    \n\t"
        "movq          %1, %%mm6    \n\t"
        ::"m"(red_15mask),"m"(green_15mask));
    mm_end = end - 11;
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "movd         3%1, %%mm3    \n\t"
            "punpckldq    6%1, %%mm0    \n\t"

        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);
    }

    __asm__ volatile(
        "movq         %0, %%mm7     \n\t"
        "movq         %1, %%mm6     \n\t"
        ::"m"(red_15mask),"m"(green_15mask));
    mm_end = end - 15;
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"   32%1            \n\t"
            "movd         %1, %%mm0     \n\t"
            "movd        3%1, %%mm3     \n\t"
            "punpckldq   6%1, %%mm0     \n\t"

        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);
    }

    const uint16_t *s = (const uint16_t*)src;
    end = s + src_size/2;
#if HAVE_MMX
    __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");
    mm_end = end - 7;
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movq          %1, %%mm0    \n\t"
            "movq          %1, %%mm1    \n\t"
            "movq          %1, %%mm2    \n\t"

        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;

    const uint16_t *s = (const uint16_t *)src;
    end = s + src_size/2;
#if HAVE_MMX
    __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");
    mm_end = end - 7;
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movq          %1, %%mm0    \n\t"
            "movq          %1, %%mm1    \n\t"
            "movq          %1, %%mm2    \n\t"

        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;

#if HAVE_MMX
    __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");
    __asm__ volatile("pxor    %%mm7,%%mm7    \n\t":::"memory");
    __asm__ volatile("pcmpeqd %%mm6,%%mm6    \n\t":::"memory");
    mm_end = end - 3;
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movq          %1, %%mm0    \n\t"
            "movq          %1, %%mm1    \n\t"
            "movq          %1, %%mm2    \n\t"

        s += 4;
    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end) {
        register uint16_t bgr;
        bgr = *s++;
#if HAVE_BIGENDIAN
        *d++ = 255;
        *d++ = (bgr&0x7C00)>>7;

#if HAVE_MMX
    __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");
    __asm__ volatile("pxor    %%mm7,%%mm7    \n\t":::"memory");
    __asm__ volatile("pcmpeqd %%mm6,%%mm6    \n\t":::"memory");
    mm_end = end - 3;
    while (s < mm_end) {
        __asm__ volatile(
            PREFETCH"    32%1           \n\t"
            "movq          %1, %%mm0    \n\t"
            "movq          %1, %%mm1    \n\t"
            "movq          %1, %%mm2    \n\t"

        s += 4;
    }
    __asm__ volatile(SFENCE:::"memory");
    __asm__ volatile(EMMS:::"memory");
#endif
    while (s < end) {
        register uint16_t bgr;
        bgr = *s++;
#if HAVE_BIGENDIAN
        *d++ = 255;
        *d++ = (bgr&0xF800)>>8;

    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;

                                           long width, long height,
                                           long lumStride, long chromStride, long dstStride, long vertLumPerChroma)
{
    long y;
    const x86_reg chromWidth= width>>1;
    for (y=0; y<height; y++) {
#if HAVE_MMX
//FIXME handle 2 lines at once (fewer prefetches, reuse some chroma, but very likely memory-limited anyway)
        __asm__ volatile(
            "xor                 %%"REG_a", %%"REG_a"   \n\t"
            ASMALIGN(4)

        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));

#elif HAVE_FAST_64BIT
        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);

        }

#else
        int i, *idst = (int32_t *) dst;
        const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;
        for (i = 0; i < chromWidth; i++) {
#if HAVE_BIGENDIAN
            *idst++ = (yc[0] << 24)+ (uc[0] << 16) +
                (yc[1] << 8) + (vc[0] << 0);
#else
            *idst++ = yc[0] + (uc[0] << 8) +

            uc++;
            vc++;
        }
#endif
#endif
        if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) {
            usrc += chromStride;
            vsrc += chromStride;
        }
        ysrc += lumStride;
        dst  += dstStride;

                                           long width, long height,
                                           long lumStride, long chromStride, long dstStride, long vertLumPerChroma)
{
    long y;
    const x86_reg chromWidth= width>>1;
    for (y=0; y<height; y++) {
#if HAVE_MMX
//FIXME handle 2 lines at once (fewer prefetches, reuse some chroma, but very likely memory-limited anyway)
        __asm__ volatile(
            "xor                %%"REG_a", %%"REG_a"    \n\t"
            ASMALIGN(4)

#if HAVE_FAST_64BIT
        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);

        }

#else
        int i, *idst = (int32_t *) dst;
        const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;
        for (i = 0; i < chromWidth; i++) {
#if HAVE_BIGENDIAN
            *idst++ = (uc[0] << 24)+ (yc[0] << 16) +
                (vc[0] << 8) + (yc[1] << 0);
#else
            *idst++ = uc[0] + (yc[0] << 8) +

            uc++;
            vc++;
        }
#endif
#endif
        if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) {
            usrc += chromStride;
            vsrc += chromStride;
        }
        ysrc += lumStride;
        dst += dstStride;

                                      long width, long height,
                                      long lumStride, long chromStride, long srcStride)
{
    long y;
    const x86_reg chromWidth= width>>1;
    for (y=0; y<height; y+=2) {
#if HAVE_MMX
        __asm__ volatile(
            "xor                 %%"REG_a", %%"REG_a"   \n\t"
            "pcmpeqw                 %%mm7, %%mm7       \n\t"
            "psrlw                      $8, %%mm7       \n\t" // FF,00,FF,00...

            ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
            : "memory", "%"REG_a
        );
#else
        long 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;

    long 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 HAVE_MMX2 || HAVE_AMD3DNOW
        const x86_reg mmxSize= srcWidth&~15;
        __asm__ volatile(
            "mov           %4, %%"REG_a"            \n\t"
            "1:                                     \n\t"

        const x86_reg 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;
        }

    // 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

                                      long width, long height,
                                      long lumStride, long chromStride, long srcStride)
{
    long y;
    const x86_reg chromWidth= width>>1;
    for (y=0; y<height; y+=2) {
#if HAVE_MMX
        __asm__ volatile(
            "xor                 %%"REG_a", %%"REG_a"   \n\t"
            "pcmpeqw             %%mm7, %%mm7   \n\t"
            "psrlw                  $8, %%mm7   \n\t" // FF,00,FF,00...

            ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
            : "memory", "%"REG_a
        );
#else
        long 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;

                                       long lumStride, long chromStride, long srcStride)
{
    long y;
    const x86_reg chromWidth= width>>1;
#if HAVE_MMX
    for (y=0; y<height-2; y+=2) {
        long i;
        for (i=0; i<2; i++) {
            __asm__ volatile(
                "mov                        %2, %%"REG_a"   \n\t"
                "movq  "MANGLE(ff_bgr2YCoeff)", %%mm6       \n\t"
                "movq       "MANGLE(ff_w1111)", %%mm5       \n\t"
                "pxor                    %%mm7, %%mm7       \n\t"

                     SFENCE"     \n\t"
                     :::"memory");
#else
    y=0;
#endif
    for (; y<height; y+=2) {
        long 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;

            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;

    }
}

static void RENAME(interleaveBytes)(uint8_t *src1, uint8_t *src2, uint8_t *dest,
                             long width, long height, long src1Stride,
                             long src2Stride, long dstStride)
{
    long h;

    for (h=0; h < height; h++) {
        long w;

#if HAVE_MMX
#if HAVE_SSE2
        __asm__(

            " jb                     1b             \n\t"
            ::"r"(dest), "r"(src1), "r"(src2), "r" ((x86_reg)width-15)
            : "memory", "%"REG_a
        );
#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;

    __asm__ volatile(
        PREFETCH" %0    \n\t"
        PREFETCH" %1    \n\t"
        ::"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;
#if HAVE_MMX
        for (;x<w-31;x+=32) {
            __asm__ volatile(
                PREFETCH"   32%1        \n\t"
                "movq         %1, %%mm0 \n\t"
                "movq        8%1, %%mm2 \n\t"
                "movq       16%1, %%mm4 \n\t"

                :"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;
#if HAVE_MMX
        for (;x<w-31;x+=32) {
            __asm__ volatile(
                PREFETCH"   32%1        \n\t"
                "movq         %1, %%mm0 \n\t"
                "movq        8%1, %%mm2 \n\t"
                "movq       16%1, %%mm4 \n\t"

                                        long srcStride3, long dstStride)
{
    x86_reg x;
    long y,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;
#if HAVE_MMX
        for (;x<w-7;x+=8) {
            __asm__ volatile(
                PREFETCH"   32(%1, %0)          \n\t"
                PREFETCH"   32(%2, %0)          \n\t"
                PREFETCH"   32(%3, %0)          \n\t"
                "movq      (%1, %0, 4), %%mm0   \n\t" /* Y0Y1Y2Y3Y4Y5Y6Y7 */

                : "+r" (x)
                : "r"(yp), "r" (up), "r"(vp), "r"(d)
                :"memory");
        }
#endif
        for (; x<w; x++) {
            const long 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];

    dst +=   count;
    src += 2*count;
    count= - count;

#if HAVE_MMX
    if(count <= -16) {
        count += 15;
        __asm__ volatile(
            "pcmpeqw       %%mm7, %%mm7        \n\t"
            "psrlw            $8, %%mm7        \n\t"
            "1:                                \n\t"

            : "r"(src), "r"(dst)
        );
        count -= 15;
    }
#endif
    while(count<0) {
        dst[count]= src[2*count];
        count++;
    }
}

    dst0+=   count;
    dst1+=   count;
    src += 4*count;
    count= - count;
#if HAVE_MMX
    if(count <= -8) {
        count += 7;
        __asm__ volatile(
            "pcmpeqw       %%mm7, %%mm7        \n\t"
            "psrlw            $8, %%mm7        \n\t"
            "1:                                \n\t"

            : "r"(src), "r"(dst0), "r"(dst1)
        );
        count -= 7;
    }
#endif
    while(count<0) {
        dst0[count]= src[4*count+0];
        dst1[count]= src[4*count+2];
        count++;
    }
}

    dst1 +=   count;
    src0 += 4*count;
    src1 += 4*count;
    count= - count;
#ifdef PAVGB
    if(count <= -8) {
        count += 7;
        __asm__ volatile(
            "pcmpeqw        %%mm7, %%mm7        \n\t"
            "psrlw             $8, %%mm7        \n\t"
            "1:                                \n\t"

            : "r"(src0), "r"(src1), "r"(dst0), "r"(dst1)
        );
        count -= 7;
    }
#endif
    while(count<0) {
        dst0[count]= (src0[4*count+0]+src1[4*count+0])>>1;
        dst1[count]= (src0[4*count+2]+src1[4*count+2])>>1;
        count++;
    }
}

    dst0+=   count;
    dst1+=   count;
    src += 4*count;
    count= - count;
#if HAVE_MMX
    if(count <= -8) {
        count += 7;
        __asm__ volatile(
            "pcmpeqw       %%mm7, %%mm7        \n\t"
            "psrlw            $8, %%mm7        \n\t"
            "1:                                \n\t"

        );
        count -= 7;
    }
#endif
    src++;
    while(count<0) {
        dst0[count]= src[4*count+0];
        dst1[count]= src[4*count+2];
        count++;
    }
}

    dst1 +=   count;
    src0 += 4*count;
    src1 += 4*count;
    count= - count;
#ifdef PAVGB
    if(count <= -8) {
        count += 7;
        __asm__ volatile(
            "pcmpeqw        %%mm7, %%mm7        \n\t"
            "psrlw             $8, %%mm7        \n\t"
            "1:                                \n\t"

        count -= 7;
    }
#endif
    src0++;
    src1++;
    while(count<0) {
        dst0[count]= (src0[4*count+0]+src1[4*count+0])>>1;
        dst1[count]= (src0[4*count+2]+src1[4*count+2])>>1;
        count++;
    }
}

                                      long lumStride, long chromStride, long srcStride)
{
    long y;
    const long chromWidth= -((-width)>>1);

    for (y=0; y<height; y++) {
        RENAME(extract_even)(src, ydst, width);
        if(y&1) {
            RENAME(extract_odd2avg)(src-srcStride, src, udst, vdst, chromWidth);
            udst+= chromStride;
            vdst+= chromStride;
        }

                                      long lumStride, long chromStride, long srcStride)
{
    long y;
    const long chromWidth= -((-width)>>1);

    for (y=0; y<height; y++) {
        RENAME(extract_even)(src, ydst, width);
        RENAME(extract_odd2)(src, udst, vdst, chromWidth);

        src += srcStride;
        ydst+= lumStride;

                                      long lumStride, long chromStride, long srcStride)
{
    long y;
    const long chromWidth= -((-width)>>1);

    for (y=0; y<height; y++) {
        RENAME(extract_even)(src+1, ydst, width);
        if(y&1) {
            RENAME(extract_even2avg)(src-srcStride, src, udst, vdst, chromWidth);
            udst+= chromStride;
            vdst+= chromStride;
        }

                                      long lumStride, long chromStride, long srcStride)
{
    long y;
    const long chromWidth= -((-width)>>1);

    for (y=0; y<height; y++) {
        RENAME(extract_even)(src+1, ydst, width);
        RENAME(extract_even2)(src, udst, vdst, chromWidth);

        src += srcStride;
        ydst+= lumStride;

            ::: "memory"
        );
#endif
}

static inline void RENAME(rgb2rgb_init)(void)
{
    rgb15to16       = RENAME(rgb15to16);
    rgb15tobgr24    = RENAME(rgb15tobgr24);
    rgb15to32       = RENAME(rgb15to32);
    rgb16tobgr24    = RENAME(rgb16tobgr24);
    rgb16to32       = RENAME(rgb16to32);