h264dsp_mmx.c.svn-base

mediastreamer2是开源的网络传输媒体流的库

        "psubusb    %%mm1, %%mm4 \n\t"
        "psubusb    %%mm1, %%mm5 \n\t"
        "movd       %%mm2, %0    \n\t"
        "movd       %%mm3, %1    \n\t"
        "movd       %%mm4, %2    \n\t"
        "movd       %%mm5, %3    \n\t"
        :"+m"(*(uint32_t*)(dst+0*stride)),
         "+m"(*(uint32_t*)(dst+1*stride)),
         "+m"(*(uint32_t*)(dst+2*stride)),
         "+m"(*(uint32_t*)(dst+3*stride))
    );
}

static void ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
{
    int dc = (block[0] + 32) >> 6;
    int y;
    asm volatile(
        "movd          %0, %%mm0 \n\t"
        "pshufw $0, %%mm0, %%mm0 \n\t"
        "pxor       %%mm1, %%mm1 \n\t"
        "psubw      %%mm0, %%mm1 \n\t"
        "packuswb   %%mm0, %%mm0 \n\t"
        "packuswb   %%mm1, %%mm1 \n\t"
        ::"r"(dc)
    );
    for(y=2; y--; dst += 4*stride){
    asm volatile(
        "movq          %0, %%mm2 \n\t"
        "movq          %1, %%mm3 \n\t"
        "movq          %2, %%mm4 \n\t"
        "movq          %3, %%mm5 \n\t"
        "paddusb    %%mm0, %%mm2 \n\t"
        "paddusb    %%mm0, %%mm3 \n\t"
        "paddusb    %%mm0, %%mm4 \n\t"
        "paddusb    %%mm0, %%mm5 \n\t"
        "psubusb    %%mm1, %%mm2 \n\t"
        "psubusb    %%mm1, %%mm3 \n\t"
        "psubusb    %%mm1, %%mm4 \n\t"
        "psubusb    %%mm1, %%mm5 \n\t"
        "movq       %%mm2, %0    \n\t"
        "movq       %%mm3, %1    \n\t"
        "movq       %%mm4, %2    \n\t"
        "movq       %%mm5, %3    \n\t"
        :"+m"(*(uint64_t*)(dst+0*stride)),
         "+m"(*(uint64_t*)(dst+1*stride)),
         "+m"(*(uint64_t*)(dst+2*stride)),
         "+m"(*(uint64_t*)(dst+3*stride))
    );
    }
}


/***********************************/
/* deblocking */

// out: o = |x-y|>a
// clobbers: t
#define DIFF_GT_MMX(x,y,a,o,t)\
    "movq     "#y", "#t"  \n\t"\
    "movq     "#x", "#o"  \n\t"\
    "psubusb  "#x", "#t"  \n\t"\
    "psubusb  "#y", "#o"  \n\t"\
    "por      "#t", "#o"  \n\t"\
    "psubusb  "#a", "#o"  \n\t"

// out: o = |x-y|>a
// clobbers: t
#define DIFF_GT2_MMX(x,y,a,o,t)\
    "movq     "#y", "#t"  \n\t"\
    "movq     "#x", "#o"  \n\t"\
    "psubusb  "#x", "#t"  \n\t"\
    "psubusb  "#y", "#o"  \n\t"\
    "psubusb  "#a", "#t"  \n\t"\
    "psubusb  "#a", "#o"  \n\t"\
    "pcmpeqb  "#t", "#o"  \n\t"\

// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1
// out: mm5=beta-1, mm7=mask
// clobbers: mm4,mm6
#define H264_DEBLOCK_MASK(alpha1, beta1) \
    "pshufw $0, "#alpha1", %%mm4 \n\t"\
    "pshufw $0, "#beta1 ", %%mm5 \n\t"\
    "packuswb  %%mm4, %%mm4      \n\t"\
    "packuswb  %%mm5, %%mm5      \n\t"\
    DIFF_GT_MMX(%%mm1, %%mm2, %%mm4, %%mm7, %%mm6) /* |p0-q0| > alpha-1 */\
    DIFF_GT_MMX(%%mm0, %%mm1, %%mm5, %%mm4, %%mm6) /* |p1-p0| > beta-1 */\
    "por       %%mm4, %%mm7      \n\t"\
    DIFF_GT_MMX(%%mm3, %%mm2, %%mm5, %%mm4, %%mm6) /* |q1-q0| > beta-1 */\
    "por       %%mm4, %%mm7      \n\t"\
    "pxor      %%mm6, %%mm6      \n\t"\
    "pcmpeqb   %%mm6, %%mm7      \n\t"

// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)
// out: mm1=p0' mm2=q0'
// clobbers: mm0,3-6
#define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\
        "movq    %%mm1              , %%mm5 \n\t"\
        "pxor    %%mm2              , %%mm5 \n\t" /* p0^q0*/\
        "pand    "#pb_01"           , %%mm5 \n\t" /* (p0^q0)&1*/\
        "pcmpeqb %%mm4              , %%mm4 \n\t"\
        "pxor    %%mm4              , %%mm3 \n\t"\
        "pavgb   %%mm0              , %%mm3 \n\t" /* (p1 - q1 + 256)>>1*/\
        "pavgb   "MANGLE(ff_pb_3)"  , %%mm3 \n\t" /*(((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2*/\
        "pxor    %%mm1              , %%mm4 \n\t"\
        "pavgb   %%mm2              , %%mm4 \n\t" /* (q0 - p0 + 256)>>1*/\
        "pavgb   %%mm5              , %%mm3 \n\t"\
        "paddusb %%mm4              , %%mm3 \n\t" /* d+128+33*/\
        "movq    "MANGLE(ff_pb_A1)" , %%mm6 \n\t"\
        "psubusb %%mm3              , %%mm6 \n\t"\
        "psubusb "MANGLE(ff_pb_A1)" , %%mm3 \n\t"\
        "pminub  %%mm7              , %%mm6 \n\t"\
        "pminub  %%mm7              , %%mm3 \n\t"\
        "psubusb %%mm6              , %%mm1 \n\t"\
        "psubusb %%mm3              , %%mm2 \n\t"\
        "paddusb %%mm3              , %%mm1 \n\t"\
        "paddusb %%mm6              , %%mm2 \n\t"

// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=ff_bone
// out: (q1addr) = av_clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
// clobbers: q2, tmp, tc0
#define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
        "movq     %%mm1,  "#tmp"   \n\t"\
        "pavgb    %%mm2,  "#tmp"   \n\t"\
        "pavgb    "#tmp", "#q2"    \n\t" /* avg(p2,avg(p0,q0)) */\
        "pxor   "q2addr", "#tmp"   \n\t"\
        "pand     %8,     "#tmp"   \n\t" /* (p2^avg(p0,q0))&1 */\
        "psubusb  "#tmp", "#q2"    \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\
        "movq     "#p1",  "#tmp"   \n\t"\
        "psubusb  "#tc0", "#tmp"   \n\t"\
        "paddusb  "#p1",  "#tc0"   \n\t"\
        "pmaxub   "#tmp", "#q2"    \n\t"\
        "pminub   "#tc0", "#q2"    \n\t"\
        "movq     "#q2",  "q1addr" \n\t"

static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
{
    DECLARE_ALIGNED_8(uint64_t, tmp0[2]);

    asm volatile(
        "movq    (%1,%3), %%mm0    \n\t" //p1
        "movq    (%1,%3,2), %%mm1  \n\t" //p0
        "movq    (%2),    %%mm2    \n\t" //q0
        "movq    (%2,%3), %%mm3    \n\t" //q1
        H264_DEBLOCK_MASK(%6, %7)

        "movd      %5,    %%mm4    \n\t"
        "punpcklbw %%mm4, %%mm4    \n\t"
        "punpcklwd %%mm4, %%mm4    \n\t"
        "pcmpeqb   %%mm3, %%mm3    \n\t"
        "movq      %%mm4, %%mm6    \n\t"
        "pcmpgtb   %%mm3, %%mm4    \n\t"
        "movq      %%mm6, 8+%0     \n\t"
        "pand      %%mm4, %%mm7    \n\t"
        "movq      %%mm7, %0       \n\t"

        /* filter p1 */
        "movq     (%1),   %%mm3    \n\t" //p2
        DIFF_GT2_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1
        "pand     %%mm7,  %%mm6    \n\t" // mask & |p2-p0|<beta
        "pand     8+%0,   %%mm7    \n\t" // mask & tc0
        "movq     %%mm7,  %%mm4    \n\t"
        "psubb    %%mm6,  %%mm7    \n\t"
        "pand     %%mm4,  %%mm6    \n\t" // mask & |p2-p0|<beta & tc0
        H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%1)", "(%1,%3)", %%mm6, %%mm4)

        /* filter q1 */
        "movq    (%2,%3,2), %%mm4  \n\t" //q2
        DIFF_GT2_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1
        "pand     %0,     %%mm6    \n\t"
        "movq     8+%0,   %%mm5    \n\t" // can be merged with the and below but is slower then
        "pand     %%mm6,  %%mm5    \n\t"
        "psubb    %%mm6,  %%mm7    \n\t"
        "movq    (%2,%3), %%mm3    \n\t"
        H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%2,%3,2)", "(%2,%3)", %%mm5, %%mm6)

        /* filter p0, q0 */
        H264_DEBLOCK_P0_Q0(%8, unused)
        "movq      %%mm1, (%1,%3,2) \n\t"
        "movq      %%mm2, (%2)      \n\t"

        : "=m"(*tmp0)
        : "r"(pix-3*stride), "r"(pix), "r"((long)stride),
          "m"(*tmp0/*unused*/), "m"(*(uint32_t*)tc0), "m"(alpha1), "m"(beta1),
          "m"(ff_bone)
    );
}

static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
{
    if((tc0[0] & tc0[1]) >= 0)
        h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);
    if((tc0[2] & tc0[3]) >= 0)
        h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);
}
static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
{
    //FIXME: could cut some load/stores by merging transpose with filter
    // also, it only needs to transpose 6x8
    DECLARE_ALIGNED_8(uint8_t, trans[8*8]);
    int i;
    for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
        if((tc0[0] & tc0[1]) < 0)
            continue;
        transpose4x4(trans,       pix-4,          8, stride);
        transpose4x4(trans  +4*8, pix,            8, stride);
        transpose4x4(trans+4,     pix-4+4*stride, 8, stride);
        transpose4x4(trans+4+4*8, pix  +4*stride, 8, stride);
        h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);
        transpose4x4(pix-2,          trans  +2*8, stride, 8);
        transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);
    }
}

static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
{
    asm volatile(
        "movq    (%0),    %%mm0     \n\t" //p1
        "movq    (%0,%2), %%mm1     \n\t" //p0
        "movq    (%1),    %%mm2     \n\t" //q0
        "movq    (%1,%2), %%mm3     \n\t" //q1
        H264_DEBLOCK_MASK(%4, %5)
        "movd      %3,    %%mm6     \n\t"
        "punpcklbw %%mm6, %%mm6     \n\t"
        "pand      %%mm6, %%mm7     \n\t" // mm7 = tc&mask
        H264_DEBLOCK_P0_Q0(%6, %7)
        "movq      %%mm1, (%0,%2)   \n\t"
        "movq      %%mm2, (%1)      \n\t"

        :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
           "r"(*(uint32_t*)tc0),
           "m"(alpha1), "m"(beta1), "m"(ff_bone), "m"(ff_pb_3F)
    );
}

static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
{
    h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
}

static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
{
    //FIXME: could cut some load/stores by merging transpose with filter
    DECLARE_ALIGNED_8(uint8_t, trans[8*4]);
    transpose4x4(trans, pix-2, 8, stride);
    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
    h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
    transpose4x4(pix-2, trans, stride, 8);
    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
}

// p0 = (p0 + q1 + 2*p1 + 2) >> 2
#define H264_FILTER_CHROMA4(p0, p1, q1, one) \
    "movq    "#p0", %%mm4  \n\t"\
    "pxor    "#q1", %%mm4  \n\t"\
    "pand   "#one", %%mm4  \n\t" /* mm4 = (p0^q1)&1 */\
    "pavgb   "#q1", "#p0"  \n\t"\
    "psubusb %%mm4, "#p0"  \n\t"\
    "pavgb   "#p1", "#p0"  \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\

static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
{
    asm volatile(
        "movq    (%0),    %%mm0     \n\t"
        "movq    (%0,%2), %%mm1     \n\t"
        "movq    (%1),    %%mm2     \n\t"
        "movq    (%1,%2), %%mm3     \n\t"
        H264_DEBLOCK_MASK(%3, %4)
        "movq    %%mm1,   %%mm5     \n\t"
        "movq    %%mm2,   %%mm6     \n\t"
        H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
        H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
        "psubb   %%mm5,   %%mm1     \n\t"
        "psubb   %%mm6,   %%mm2     \n\t"
        "pand    %%mm7,   %%mm1     \n\t"
        "pand    %%mm7,   %%mm2     \n\t"
        "paddb   %%mm5,   %%mm1     \n\t"
        "paddb   %%mm6,   %%mm2     \n\t"
        "movq    %%mm1,   (%0,%2)   \n\t"
        "movq    %%mm2,   (%1)      \n\t"
        :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
           "m"(alpha1), "m"(beta1), "m"(ff_bone)
    );
}

static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
{
    h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
}

static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
{
    //FIXME: could cut some load/stores by merging transpose with filter
    DECLARE_ALIGNED_8(uint8_t, trans[8*4]);
    transpose4x4(trans, pix-2, 8, stride);
    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
    h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
    transpose4x4(pix-2, trans, stride, 8);
    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
}

static void h264_loop_filter_strength_mmx2( int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
                                            int bidir, int edges, int step, int mask_mv0, int mask_mv1 ) {
    int dir;
    asm volatile(
        "pxor %%mm7, %%mm7 \n\t"
        "movq %0, %%mm6 \n\t"
        "movq %1, %%mm5 \n\t"
        "movq %2, %%mm4 \n\t"
        ::"m"(ff_pb_1), "m"(ff_pb_3), "m"(ff_pb_7)
    );
    // could do a special case for dir==0 && edges==1, but it only reduces the
    // average filter time by 1.2%
    for( dir=1; dir>=0; dir-- ) {
        const int d_idx = dir ? -8 : -1;
        const int mask_mv = dir ? mask_mv1 : mask_mv0;
        DECLARE_ALIGNED_8(const uint64_t, mask_dir) = dir ? 0 : 0xffffffffffffffffULL;
        int b_idx, edge, l;
        for( b_idx=12, edge=0; edge<edges; edge+=step, b_idx+=8*step ) {
            asm volatile(
                "pand %0, %%mm0 \n\t"
                ::"m"(mask_dir)
            );