comparison i386/dsputil_h264_template_mmx.c @ 4504:6287a2ff4d08 libavcodec

merge asm fragments in H264_CHROMA_MC2_TMPL() 10% faster avg_h264_chroma_mc2_mmx2() 5% faster put_h264_chroma_mc2_mmx2()

4503:f334529b46ec

#ifdef H264_CHROMA_MC2_TMPL
static void H264_CHROMA_MC2_TMPL(uint8_t *dst/*align 2*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
{
    int CD=((1<<16)-1)*x*y + 8*y;
    int AB=((8<<16)-8)*x + 64 - CD;
    int i;

    asm volatile(
        /* mm5 = {A,B,A,B} */
        /* mm6 = {C,D,C,D} */
        "movd %0, %%mm5\n\t"
        "movd %1, %%mm6\n\t"
        "punpckldq %%mm5, %%mm5\n\t"
        "punpckldq %%mm6, %%mm6\n\t"
        "pxor %%mm7, %%mm7\n\t"
        :: "r"(AB), "r"(CD));

    asm volatile(
        /* mm0 = src[0,1,1,2] */
        "movd %0, %%mm0\n\t"
        "punpcklbw %%mm7, %%mm0\n\t"
        "pshufw $0x94, %%mm0, %%mm0\n\t"
        :: "m"(src[0]));

    for(i=0; i<h; i++) {
        asm volatile(
            /* mm1 = A * src[0,1] + B * src[1,2] */
            "movq    %%mm0, %%mm1\n\t"
            "pmaddwd %%mm5, %%mm1\n\t"
            ::);

        src += stride;
        asm volatile(
            /* mm0 = src[0,1,1,2] */
            "movd %0, %%mm0\n\t"
            "punpcklbw %%mm7, %%mm0\n\t"
            "pshufw $0x94, %%mm0, %%mm0\n\t"
            :: "m"(src[0]));

        asm volatile(
            /* mm1 += C * src[0,1] + D * src[1,2] */
            "movq    %%mm0, %%mm2\n\t"
            "pmaddwd %%mm6, %%mm2\n\t"
            "paddw   %%mm2, %%mm1\n\t"
            ::);

        asm volatile(
            /* dst[0,1] = pack((mm1 + 32) >> 6) */
            "paddw %1, %%mm1\n\t"
            "psrlw $6, %%mm1\n\t"
            "packssdw %%mm7, %%mm1\n\t"
            "packuswb %%mm7, %%mm1\n\t"
            /* writes garbage to the right of dst.
             * ok because partitions are processed from left to right. */
            H264_CHROMA_OP4(%0, %%mm1, %%mm3)
            "movd %%mm1, %0\n\t"
            : "=m" (dst[0]) : "m" (ff_pw_32));
        dst += stride;
    }
}
#endif

4504:6287a2ff4d08

#ifdef H264_CHROMA_MC2_TMPL
static void H264_CHROMA_MC2_TMPL(uint8_t *dst/*align 2*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
{
    int CD=((1<<16)-1)*x*y + 8*y;
    int AB=((8<<16)-8)*x + 64 - CD;
    asm volatile(
        /* mm5 = {A,B,A,B} */
        /* mm6 = {C,D,C,D} */
        "movd %0, %%mm5\n\t"
        "movd %1, %%mm6\n\t"
        "punpckldq %%mm5, %%mm5\n\t"
        "punpckldq %%mm6, %%mm6\n\t"
        "pxor %%mm7, %%mm7\n\t"
        /* mm0 = src[0,1,1,2] */
        "movd %2, %%mm0\n\t"
        "punpcklbw %%mm7, %%mm0\n\t"
        "pshufw $0x94, %%mm0, %%mm0\n\t"
        :: "r"(AB), "r"(CD), "m"(src[0]));


    asm volatile(
        "1:\n\t"
        "addl %4, %1\n\t"
        /* mm1 = A * src[0,1] + B * src[1,2] */
        "movq    %%mm0, %%mm1\n\t"
        "pmaddwd %%mm5, %%mm1\n\t"
        /* mm0 = src[0,1,1,2] */
        "movd (%1), %%mm0\n\t"
        "punpcklbw %%mm7, %%mm0\n\t"
        "pshufw $0x94, %%mm0, %%mm0\n\t"
        /* mm1 += C * src[0,1] + D * src[1,2] */
        "movq    %%mm0, %%mm2\n\t"
        "pmaddwd %%mm6, %%mm2\n\t"
        "paddw   %%mm2, %%mm1\n\t"
        /* dst[0,1] = pack((mm1 + 32) >> 6) */
        "paddw %3, %%mm1\n\t"
        "psrlw $6, %%mm1\n\t"
        "packssdw %%mm7, %%mm1\n\t"
        "packuswb %%mm7, %%mm1\n\t"
        /* writes garbage to the right of dst.
            * ok because partitions are processed from left to right. */
        H264_CHROMA_OP4((%0), %%mm1, %%mm3)
        "movd %%mm1, (%0)\n\t"
        "addl %4, %0\n\t"
        "subl $1, %2\n\t"
        "jnz 1b\n\t"
        : "+r" (dst), "+r"(src), "+r"(h) : "m" (ff_pw_32), "r"(stride));

}
#endif