view x86/dsputil_h264_template_mmx.c @ 11032:01bd040f8607 libavcodec

Unroll main loop so the edge==0 case is seperate. This allows many things to be simplified away. h264 decoder is overall 1% faster with a mbaff sample and 0.1% slower with the cathedral sample, probably because the slow loop filter code must be loaded into the code cache for each first MB of each row but isnt used for the following MBs.

/*
 * Copyright (c) 2005 Zoltan Hidvegi <hzoli -a- hzoli -d- com>,
 *                    Loren Merritt
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * MMX optimized version of (put|avg)_h264_chroma_mc8.
 * H264_CHROMA_MC8_TMPL must be defined to the desired function name
 * H264_CHROMA_OP must be defined to empty for put and pavgb/pavgusb for avg
 * H264_CHROMA_MC8_MV0 must be defined to a (put|avg)_pixels8 function
 */
static void H264_CHROMA_MC8_TMPL(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y, const uint64_t *rnd_reg)
{
    DECLARE_ALIGNED_8(uint64_t, AA);
    DECLARE_ALIGNED_8(uint64_t, DD);
    int i;

    if(y==0 && x==0) {
        /* no filter needed */
        H264_CHROMA_MC8_MV0(dst, src, stride, h);
        return;
    }

    assert(x<8 && y<8 && x>=0 && y>=0);

    if(y==0 || x==0)
    {
        /* 1 dimensional filter only */
        const int dxy = x ? 1 : stride;

        __asm__ volatile(
            "movd %0, %%mm5\n\t"
            "movq %1, %%mm4\n\t"
            "movq %2, %%mm6\n\t"         /* mm6 = rnd >> 3 */
            "punpcklwd %%mm5, %%mm5\n\t"
            "punpckldq %%mm5, %%mm5\n\t" /* mm5 = B = x */
            "pxor %%mm7, %%mm7\n\t"
            "psubw %%mm5, %%mm4\n\t"     /* mm4 = A = 8-x */
            :: "rm"(x+y), "m"(ff_pw_8), "m"(*(rnd_reg+1)));

        for(i=0; i<h; i++) {
            __asm__ volatile(
                /* mm0 = src[0..7], mm1 = src[1..8] */
                "movq %0, %%mm0\n\t"
                "movq %1, %%mm2\n\t"
                :: "m"(src[0]), "m"(src[dxy]));

            __asm__ volatile(
                /* [mm0,mm1] = A * src[0..7] */
                /* [mm2,mm3] = B * src[1..8] */
                "movq %%mm0, %%mm1\n\t"
                "movq %%mm2, %%mm3\n\t"
                "punpcklbw %%mm7, %%mm0\n\t"
                "punpckhbw %%mm7, %%mm1\n\t"
                "punpcklbw %%mm7, %%mm2\n\t"
                "punpckhbw %%mm7, %%mm3\n\t"
                "pmullw %%mm4, %%mm0\n\t"
                "pmullw %%mm4, %%mm1\n\t"
                "pmullw %%mm5, %%mm2\n\t"
                "pmullw %%mm5, %%mm3\n\t"

                /* dst[0..7] = (A * src[0..7] + B * src[1..8] + (rnd >> 3)) >> 3 */
                "paddw %%mm6, %%mm0\n\t"
                "paddw %%mm6, %%mm1\n\t"
                "paddw %%mm2, %%mm0\n\t"
                "paddw %%mm3, %%mm1\n\t"
                "psrlw $3, %%mm0\n\t"
                "psrlw $3, %%mm1\n\t"
                "packuswb %%mm1, %%mm0\n\t"
                H264_CHROMA_OP(%0, %%mm0)
                "movq %%mm0, %0\n\t"
                : "=m" (dst[0]));

            src += stride;
            dst += stride;
        }
        return;
    }

    /* general case, bilinear */
    __asm__ volatile("movd %2, %%mm4\n\t"
                 "movd %3, %%mm6\n\t"
                 "punpcklwd %%mm4, %%mm4\n\t"
                 "punpcklwd %%mm6, %%mm6\n\t"
                 "punpckldq %%mm4, %%mm4\n\t" /* mm4 = x words */
                 "punpckldq %%mm6, %%mm6\n\t" /* mm6 = y words */
                 "movq %%mm4, %%mm5\n\t"
                 "pmullw %%mm6, %%mm4\n\t"    /* mm4 = x * y */
                 "psllw $3, %%mm5\n\t"
                 "psllw $3, %%mm6\n\t"
                 "movq %%mm5, %%mm7\n\t"
                 "paddw %%mm6, %%mm7\n\t"
                 "movq %%mm4, %1\n\t"         /* DD = x * y */
                 "psubw %%mm4, %%mm5\n\t"     /* mm5 = B = 8x - xy */
                 "psubw %%mm4, %%mm6\n\t"     /* mm6 = C = 8y - xy */
                 "paddw %4, %%mm4\n\t"
                 "psubw %%mm7, %%mm4\n\t"     /* mm4 = A = xy - (8x+8y) + 64 */
                 "pxor %%mm7, %%mm7\n\t"
                 "movq %%mm4, %0\n\t"
                 : "=m" (AA), "=m" (DD) : "rm" (x), "rm" (y), "m" (ff_pw_64));

    __asm__ volatile(
        /* mm0 = src[0..7], mm1 = src[1..8] */
        "movq %0, %%mm0\n\t"
        "movq %1, %%mm1\n\t"
        : : "m" (src[0]), "m" (src[1]));

    for(i=0; i<h; i++) {
        src += stride;

        __asm__ volatile(
            /* mm2 = A * src[0..3] + B * src[1..4] */
            /* mm3 = A * src[4..7] + B * src[5..8] */
            "movq %%mm0, %%mm2\n\t"
            "movq %%mm1, %%mm3\n\t"
            "punpckhbw %%mm7, %%mm0\n\t"
            "punpcklbw %%mm7, %%mm1\n\t"
            "punpcklbw %%mm7, %%mm2\n\t"
            "punpckhbw %%mm7, %%mm3\n\t"
            "pmullw %0, %%mm0\n\t"
            "pmullw %0, %%mm2\n\t"
            "pmullw %%mm5, %%mm1\n\t"
            "pmullw %%mm5, %%mm3\n\t"
            "paddw %%mm1, %%mm2\n\t"
            "paddw %%mm0, %%mm3\n\t"
            : : "m" (AA));

        __asm__ volatile(
            /* [mm2,mm3] += C * src[0..7] */
            "movq %0, %%mm0\n\t"
            "movq %%mm0, %%mm1\n\t"
            "punpcklbw %%mm7, %%mm0\n\t"
            "punpckhbw %%mm7, %%mm1\n\t"
            "pmullw %%mm6, %%mm0\n\t"
            "pmullw %%mm6, %%mm1\n\t"
            "paddw %%mm0, %%mm2\n\t"
            "paddw %%mm1, %%mm3\n\t"
            : : "m" (src[0]));

        __asm__ volatile(
            /* [mm2,mm3] += D * src[1..8] */
            "movq %1, %%mm1\n\t"
            "movq %%mm1, %%mm0\n\t"
            "movq %%mm1, %%mm4\n\t"
            "punpcklbw %%mm7, %%mm0\n\t"
            "punpckhbw %%mm7, %%mm4\n\t"
            "pmullw %2, %%mm0\n\t"
            "pmullw %2, %%mm4\n\t"
            "paddw %%mm0, %%mm2\n\t"
            "paddw %%mm4, %%mm3\n\t"
            "movq %0, %%mm0\n\t"
            : : "m" (src[0]), "m" (src[1]), "m" (DD));

        __asm__ volatile(
            /* dst[0..7] = ([mm2,mm3] + rnd) >> 6 */
            "paddw %1, %%mm2\n\t"
            "paddw %1, %%mm3\n\t"
            "psrlw $6, %%mm2\n\t"
            "psrlw $6, %%mm3\n\t"
            "packuswb %%mm3, %%mm2\n\t"
            H264_CHROMA_OP(%0, %%mm2)
            "movq %%mm2, %0\n\t"
            : "=m" (dst[0]) : "m" (*rnd_reg));
        dst+= stride;
    }
}

static void H264_CHROMA_MC4_TMPL(uint8_t *dst/*align 4*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y, const uint64_t *rnd_reg)
{
    __asm__ volatile(
        "pxor   %%mm7, %%mm7        \n\t"
        "movd %5, %%mm2             \n\t"
        "movd %6, %%mm3             \n\t"
        "movq "MANGLE(ff_pw_8)", %%mm4\n\t"
        "movq "MANGLE(ff_pw_8)", %%mm5\n\t"
        "punpcklwd %%mm2, %%mm2     \n\t"
        "punpcklwd %%mm3, %%mm3     \n\t"
        "punpcklwd %%mm2, %%mm2     \n\t"
        "punpcklwd %%mm3, %%mm3     \n\t"
        "psubw %%mm2, %%mm4         \n\t"
        "psubw %%mm3, %%mm5         \n\t"

        "movd  (%1), %%mm0          \n\t"
        "movd 1(%1), %%mm6          \n\t"
        "add %3, %1                 \n\t"
        "punpcklbw %%mm7, %%mm0     \n\t"
        "punpcklbw %%mm7, %%mm6     \n\t"
        "pmullw %%mm4, %%mm0        \n\t"
        "pmullw %%mm2, %%mm6        \n\t"
        "paddw %%mm0, %%mm6         \n\t"

        "1:                         \n\t"
        "movd  (%1), %%mm0          \n\t"
        "movd 1(%1), %%mm1          \n\t"
        "add %3, %1                 \n\t"
        "punpcklbw %%mm7, %%mm0     \n\t"
        "punpcklbw %%mm7, %%mm1     \n\t"
        "pmullw %%mm4, %%mm0        \n\t"
        "pmullw %%mm2, %%mm1        \n\t"
        "paddw %%mm0, %%mm1         \n\t"
        "movq %%mm1, %%mm0          \n\t"
        "pmullw %%mm5, %%mm6        \n\t"
        "pmullw %%mm3, %%mm1        \n\t"
        "paddw %4, %%mm6            \n\t"
        "paddw %%mm6, %%mm1         \n\t"
        "psrlw $6, %%mm1            \n\t"
        "packuswb %%mm1, %%mm1      \n\t"
        H264_CHROMA_OP4((%0), %%mm1, %%mm6)
        "movd %%mm1, (%0)           \n\t"
        "add %3, %0                 \n\t"
        "movd  (%1), %%mm6          \n\t"
        "movd 1(%1), %%mm1          \n\t"
        "add %3, %1                 \n\t"
        "punpcklbw %%mm7, %%mm6     \n\t"
        "punpcklbw %%mm7, %%mm1     \n\t"
        "pmullw %%mm4, %%mm6        \n\t"
        "pmullw %%mm2, %%mm1        \n\t"
        "paddw %%mm6, %%mm1         \n\t"
        "movq %%mm1, %%mm6          \n\t"
        "pmullw %%mm5, %%mm0        \n\t"
        "pmullw %%mm3, %%mm1        \n\t"
        "paddw %4, %%mm0            \n\t"
        "paddw %%mm0, %%mm1         \n\t"
        "psrlw $6, %%mm1            \n\t"
        "packuswb %%mm1, %%mm1      \n\t"
        H264_CHROMA_OP4((%0), %%mm1, %%mm0)
        "movd %%mm1, (%0)           \n\t"
        "add %3, %0                 \n\t"
        "sub $2, %2                 \n\t"
        "jnz 1b                     \n\t"
        : "+r"(dst), "+r"(src), "+r"(h)
        : "r"((x86_reg)stride), "m"(*rnd_reg), "m"(x), "m"(y)
    );
}

#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 tmp = ((1<<16)-1)*x + 8;
    int CD= tmp*y;
    int AB= (tmp<<3) - 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, %%mm2\n\t"
        "punpcklbw %%mm7, %%mm2\n\t"
        "pshufw $0x94, %%mm2, %%mm2\n\t"
        :: "r"(AB), "r"(CD), "m"(src[0]));


    __asm__ volatile(
        "1:\n\t"
        "add %4, %1\n\t"
        /* mm1 = A * src[0,1] + B * src[1,2] */
        "movq    %%mm2, %%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, %%mm0\n\t"
        "paddw      %3, %%mm1\n\t"
        "paddw   %%mm0, %%mm1\n\t"
        /* dst[0,1] = pack((mm1 + 32) >> 6) */
        "psrlw $6, %%mm1\n\t"
        "packssdw %%mm7, %%mm1\n\t"
        "packuswb %%mm7, %%mm1\n\t"
        H264_CHROMA_OP4((%0), %%mm1, %%mm3)
        "movd %%mm1, %%esi\n\t"
        "movw %%si, (%0)\n\t"
        "add %4, %0\n\t"
        "sub $1, %2\n\t"
        "jnz 1b\n\t"
        : "+r" (dst), "+r"(src), "+r"(h)
        : "m" (ff_pw_32), "r"((x86_reg)stride)
        : "%esi");

}
#endif