mc.c

图象压缩程序

        src += i_src_stride;
        dst += i_dst_stride;
    }
}
static inline void mc_hv_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    int y;

    src -= 2 * i_src_stride;

    asm volatile(
        "pxor %%mm7,        %%mm7\n"
        "movq x264_w0x10,   %%mm4\n"
        : : );

    for( y = 0; y < i_height; y++ )
    {
        asm volatile(
            "leal   (%0, %1),   %%eax\n"

            "movq       (%0),   %%mm0\n"    /* load pix-2 */
            "movq       %%mm0,  %%mm2\n"
            "punpcklbw  %%mm7,  %%mm0\n"
            "punpckhbw  %%mm7,  %%mm2\n"

            "movq       (%%eax),%%mm1\n"    /* load pix-1 */
            "movq       %%mm1,  %%mm3\n"
            "punpcklbw  %%mm7,  %%mm1\n"
            "punpckhbw  %%mm7,  %%mm3\n"
            "psubw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm1\n"
            "psubw      %%mm1,  %%mm0\n"
            "psubw      %%mm3,  %%mm2\n"
            "psllw      $2,     %%mm3\n"
            "psubw      %%mm3,  %%mm2\n"

            "movq       (%%eax,%1),%%mm1\n"  /* load pix */
            "movq       %%mm1,  %%mm3\n"
            "punpcklbw  %%mm7,  %%mm1\n"
            "punpckhbw  %%mm7,  %%mm3\n"
            "psllw      $2,     %%mm1\n"
            "paddw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm1\n"
            "paddw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm3\n"
            "paddw      %%mm3,  %%mm2\n"
            "psllw      $2,     %%mm3\n"
            "paddw      %%mm3,  %%mm2\n"

            "movq       (%%eax,%1,2),%%mm1\n"  /* load pix+1 */
            "movq       %%mm1,  %%mm3\n"
            "punpcklbw  %%mm7,  %%mm1\n"
            "punpckhbw  %%mm7,  %%mm3\n"
            "psllw      $2,     %%mm1\n"
            "paddw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm1\n"
            "paddw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm3\n"
            "paddw      %%mm3,  %%mm2\n"
            "psllw      $2,     %%mm3\n"
            "paddw      %%mm3,  %%mm2\n"

            "movq       (%0,%1,4),%%mm1\n"  /* load pix+2 */
            "movq       %%mm1,  %%mm3\n"
            "punpcklbw  %%mm7,  %%mm1\n"
            "punpckhbw  %%mm7,  %%mm3\n"
            "psubw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm1\n"
            "psubw      %%mm1,  %%mm0\n"
            "psubw      %%mm3,  %%mm2\n"
            "psllw      $2,     %%mm3\n"
            "psubw      %%mm3,  %%mm2\n"

            "movq       (%%eax,%1,4),%%mm1\n"  /* load pix+3 */
            "movq       %%mm1,  %%mm3\n"
            "punpcklbw  %%mm7,  %%mm1\n"
            "punpckhbw  %%mm7,  %%mm3\n"
            "paddw      %%mm1,  %%mm0\n"
            "paddw      %%mm3,  %%mm2\n"

            "paddw      %%mm4,  %%mm0\n"
            "psraw      $5,     %%mm0\n"
            "paddw      %%mm4,  %%mm2\n"
            "psraw      $5,     %%mm2\n"


            "packuswb   %%mm2,  %%mm0\n"
            "movq       %%mm0,  (%2)\n"
            : : "r"(src), "r"(i_src_stride), "r"(dst) : "%eax" );

        src += i_src_stride;
        dst += i_dst_stride;
    }
}

static inline void mc_hc_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    int x, y;

    asm volatile( "pxor %%mm7,        %%mm7\n" : : );

    for( y = 0; y < i_height; y++ )
    {
        int16_t tap[5+8];

        /* first 8 */
        asm volatile(
            "leal   (%0, %1),   %%eax\n"

            "movq       (%0),   %%mm0\n"    /* load pix-2 */
            "movq       %%mm0,  %%mm2\n"
            "punpcklbw  %%mm7,  %%mm0\n"
            "punpckhbw  %%mm7,  %%mm2\n"

            "movq       (%%eax),%%mm1\n"    /* load pix-1 */
            "movq       %%mm1,  %%mm3\n"
            "punpcklbw  %%mm7,  %%mm1\n"
            "punpckhbw  %%mm7,  %%mm3\n"
            "psubw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm1\n"
            "psubw      %%mm1,  %%mm0\n"
            "psubw      %%mm3,  %%mm2\n"
            "psllw      $2,     %%mm3\n"
            "psubw      %%mm3,  %%mm2\n"

            "movq       (%%eax,%1),%%mm1\n"  /* load pix */
            "movq       %%mm1,  %%mm3\n"
            "punpcklbw  %%mm7,  %%mm1\n"
            "punpckhbw  %%mm7,  %%mm3\n"
            "psllw      $2,     %%mm1\n"
            "paddw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm1\n"
            "paddw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm3\n"
            "paddw      %%mm3,  %%mm2\n"
            "psllw      $2,     %%mm3\n"
            "paddw      %%mm3,  %%mm2\n"

            "movq       (%%eax,%1,2),%%mm1\n"  /* load pix+1 */
            "movq       %%mm1,  %%mm3\n"
            "punpcklbw  %%mm7,  %%mm1\n"
            "punpckhbw  %%mm7,  %%mm3\n"
            "psllw      $2,     %%mm1\n"
            "paddw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm1\n"
            "paddw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm3\n"
            "paddw      %%mm3,  %%mm2\n"
            "psllw      $2,     %%mm3\n"
            "paddw      %%mm3,  %%mm2\n"

            "movq       (%0,%1,4),%%mm1\n"  /* load pix+2 */
            "movq       %%mm1,  %%mm3\n"
            "punpcklbw  %%mm7,  %%mm1\n"
            "punpckhbw  %%mm7,  %%mm3\n"
            "psubw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm1\n"
            "psubw      %%mm1,  %%mm0\n"
            "psubw      %%mm3,  %%mm2\n"
            "psllw      $2,     %%mm3\n"
            "psubw      %%mm3,  %%mm2\n"

            "movq       (%%eax,%1,4),%%mm1\n"  /* load pix+3 */
            "movq       %%mm1,  %%mm3\n"
            "punpcklbw  %%mm7,  %%mm1\n"
            "punpckhbw  %%mm7,  %%mm3\n"
            "paddw      %%mm1,  %%mm0\n"
            "paddw      %%mm3,  %%mm2\n"

            "movq       %%mm0,   (%2)\n"
            "movq       %%mm2,  8(%2)\n"


            "addl   $8,         %%eax\n"
            "addl   $8,         %0\n"


            "movd       (%0),   %%mm0\n"    /* load pix-2 */
            "punpcklbw  %%mm7,  %%mm0\n"

            "movd       (%%eax),%%mm1\n"    /* load pix-1 */
            "punpcklbw  %%mm7,  %%mm1\n"
            "psubw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm1\n"
            "psubw      %%mm1,  %%mm0\n"

            "movd       (%%eax,%1),%%mm1\n"  /* load pix */
            "punpcklbw  %%mm7,  %%mm1\n"
            "psllw      $2,     %%mm1\n"
            "paddw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm1\n"
            "paddw      %%mm1,  %%mm0\n"

            "movd       (%%eax,%1,2),%%mm1\n"  /* load pix+1 */
            "punpcklbw  %%mm7,  %%mm1\n"
            "psllw      $2,     %%mm1\n"
            "paddw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm1\n"
            "paddw      %%mm1,  %%mm0\n"

            "movd       (%0,%1,4),%%mm1\n"  /* load pix+2 */
            "punpcklbw  %%mm7,  %%mm1\n"
            "psubw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm1\n"
            "psubw      %%mm1,  %%mm0\n"

            "movd       (%%eax,%1,4),%%mm1\n"  /* load pix+3 */
            "punpcklbw  %%mm7,  %%mm1\n"
            "paddw      %%mm1,  %%mm0\n"

            "movq       %%mm0,  16(%2)\n"
            : : "r"(src-2*i_src_stride-2), "r"(i_src_stride), "r"(&tap[0]) : "%eax" );

        /* last one */
        tap[8+4] = x264_tapfilter( &src[-2+8+4], i_src_stride );

        for( x = 0; x < 8; x++ )
        {
            dst[x] = x264_mc_clip1( ( tap[0+x] - 5*tap[1+x] + 20 * tap[2+x] + 20 * tap[3+x] -5*tap[4+x] + tap[5+x] + 512 ) >> 10 );
        }

        src += i_src_stride;
        dst += i_dst_stride;
    }
}

/* mc I+H */
static void mc_xy10_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    uint8_t tmp[8*16];
    mc_hh_w8( src, i_src_stride, tmp, 8, i_height );
    pixel_avg_w8( dst, i_dst_stride, src, i_src_stride, tmp, 8, i_height );
}
static void mc_xy30_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    uint8_t tmp[8*16];
    mc_hh_w8( src, i_src_stride, tmp, 8, i_height );
    pixel_avg_w8( dst, i_dst_stride, src+1, i_src_stride, tmp, 8, i_height );
}
/* mc I+V */
static void mc_xy01_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    uint8_t tmp[8*16];
    mc_hv_w8( src, i_src_stride, tmp, 8, i_height );
    pixel_avg_w8( dst, i_dst_stride, src, i_src_stride, tmp, 8, i_height );
}
static void mc_xy03_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    uint8_t tmp[8*16];
    mc_hv_w8( src, i_src_stride, tmp, 8, i_height );
    pixel_avg_w8( dst, i_dst_stride, src+i_src_stride, i_src_stride, tmp, 8, i_height );
}
/* H+V */
static void mc_xy11_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    uint8_t tmp1[8*16];
    uint8_t tmp2[8*16];

    mc_hv_w8( src, i_src_stride, tmp1, 8, i_height );
    mc_hh_w8( src, i_src_stride, tmp2, 8, i_height );
    pixel_avg_w8( dst, i_dst_stride, tmp1, 8, tmp2, 8, i_height );
}
static void mc_xy31_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    uint8_t tmp1[8*16];
    uint8_t tmp2[8*16];

    mc_hv_w8( src+1, i_src_stride, tmp1, 8, i_height );
    mc_hh_w8( src,   i_src_stride, tmp2, 8, i_height );
    pixel_avg_w8( dst, i_dst_stride, tmp1, 8, tmp2, 8, i_height );
}
static void mc_xy13_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    uint8_t tmp1[8*16];
    uint8_t tmp2[8*16];

    mc_hv_w8( src,              i_src_stride, tmp1, 8, i_height );
    mc_hh_w8( src+i_src_stride, i_src_stride, tmp2, 8, i_height );
    pixel_avg_w8( dst, i_dst_stride, tmp1, 8, tmp2, 8, i_height );
}
static void mc_xy33_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    uint8_t tmp1[8*16];
    uint8_t tmp2[8*16];

    mc_hv_w8( src+1,            i_src_stride, tmp1, 8, i_height );
    mc_hh_w8( src+i_src_stride, i_src_stride, tmp2, 8, i_height );
    pixel_avg_w8( dst, i_dst_stride, tmp1, 8, tmp2, 8, i_height );
}
static void mc_xy21_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    uint8_t tmp1[8*16];
    uint8_t tmp2[8*16];

    mc_hc_w8( src, i_src_stride, tmp1, 8, i_height );
    mc_hh_w8( src, i_src_stride, tmp2, 8, i_height );
    pixel_avg_w8( dst, i_dst_stride, tmp1, 8, tmp2, 8, i_height );
}
static void mc_xy12_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    uint8_t tmp1[8*16];
    uint8_t tmp2[8*16];

    mc_hc_w8( src, i_src_stride, tmp1, 8, i_height );
    mc_hv_w8( src, i_src_stride, tmp2, 8, i_height );
    pixel_avg_w8( dst, i_dst_stride, tmp1, 8, tmp2, 8, i_height );
}
static void mc_xy32_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    uint8_t tmp1[8*16];
    uint8_t tmp2[8*16];

    mc_hc_w8( src,   i_src_stride, tmp1, 8, i_height );
    mc_hv_w8( src+1, i_src_stride, tmp2, 8, i_height );
    pixel_avg_w8( dst, i_dst_stride, tmp1, 8, tmp2, 8, i_height );
}
static void mc_xy23_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    uint8_t tmp1[8*16];
    uint8_t tmp2[8*16];

    mc_hc_w8( src,              i_src_stride, tmp1, 8, i_height );
    mc_hh_w8( src+i_src_stride, i_src_stride, tmp2, 8, i_height );
    pixel_avg_w8( dst, i_dst_stride, tmp1, 8, tmp2, 8, i_height );
}


/*****************************************************************************
 * MC with width == 16 (height <= 16)
 *****************************************************************************/
static void mc_copy_w16( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    int y;

    for( y = 0; y < i_height; y++ )
    {
        memcpy( dst, src, 16 );

        src += i_src_stride;
        dst += i_dst_stride;
    }
}
static inline void mc_hh_w16( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    int x, y;

    for( y = 0; y < i_height; y++ )
    {
        for( x = 0; x < 16; x++ )
        {
            dst[x] = x264_mc_clip1( ( x264_tapfilter1( &src[x] ) + 16 ) >> 5 );
        }
        src += i_src_stride;
        dst += i_dst_stride;
    }
}
static inline void mc_hv_w16( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    mc_hv_w8( src,     i_src_stride, dst,     i_dst_stride, i_height );
    mc_hv_w8( &src[8], i_src_stride, &dst[8], i_dst_stride, i_height );
}

static inline void mc_hc_w16( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
{
    int x, y;

    asm volatile( "pxor %%mm7,        %%mm7\n" : : );

    for( y = 0; y < i_height; y++ )
    {
        int16_t tap[5+16];

        asm volatile(
            "leal   (%0, %1),   %%eax\n"

            "movq       (%0),   %%mm0\n"    /* load pix-2 */
            "movq       %%mm0,  %%mm2\n"
            "punpcklbw  %%mm7,  %%mm0\n"
            "punpckhbw  %%mm7,  %%mm2\n"

            "movq       (%%eax),%%mm1\n"    /* load pix-1 */
            "movq       %%mm1,  %%mm3\n"
            "punpcklbw  %%mm7,  %%mm1\n"
            "punpckhbw  %%mm7,  %%mm3\n"
            "psubw      %%mm1,  %%mm0\n"
            "psllw      $2,     %%mm1\n"
            "psubw      %%mm1,  %%mm0\n"