comparison i386/idct_mmx.c @ 2754:a49f140179e9 libavcodec

sort H.264 mmx dsp functions into their own file

2753:ba8ecddf5598

	      mmxext_row_head, mmxext_row, mmxext_row_tail, mmxext_row_mid)

declare_idct (ff_mmx_idct, mmx_table,
	      mmx_row_head, mmx_row, mmx_row_tail, mmx_row_mid)



/* in/out: mma=mma+mmb, mmb=mmb-mma */
#define SUMSUB_BA( a, b ) \
    "paddw "#b", "#a" \n\t"\
    "paddw "#b", "#b" \n\t"\
    "psubw "#a", "#b" \n\t"

#define SUMSUB_BADC( a, b, c, d ) \
    "paddw "#b", "#a" \n\t"\
    "paddw "#d", "#c" \n\t"\
    "paddw "#b", "#b" \n\t"\
    "paddw "#d", "#d" \n\t"\
    "psubw "#a", "#b" \n\t"\
    "psubw "#c", "#d" \n\t"

#define SUMSUBD2_AB( a, b, t ) \
    "movq  "#b", "#t" \n\t"\
    "psraw  $1 , "#b" \n\t"\
    "paddw "#a", "#b" \n\t"\
    "psraw  $1 , "#a" \n\t"\
    "psubw "#t", "#a" \n\t"

#define IDCT4_1D( s02, s13, d02, d13, t ) \
    SUMSUB_BA  ( s02, d02 )\
    SUMSUBD2_AB( s13, d13, t )\
    SUMSUB_BADC( d13, s02, s13, d02 )

#define SBUTTERFLY( a, b, t, n )  \
    "movq   "#a", "#t"       \n\t" /* abcd */\
    "punpckl"#n"  "#b", "#a" \n\t" /* aebf */\
    "punpckh"#n"  "#b", "#t" \n\t" /* cgdh */

#define TRANSPOSE4( a, b, c, d, t ) \
    SBUTTERFLY( a, b, t, wd ) /* a=aebf t=cgdh */\
    SBUTTERFLY( c, d, b, wd ) /* c=imjn b=kolp */\
    SBUTTERFLY( a, c, d, dq ) /* a=aeim d=bfjn */\
    SBUTTERFLY( t, b, c, dq ) /* t=cgko c=dhlp */

#define STORE_DIFF_4P( p, t, z ) \
        "psraw      $6,     "#p" \n\t"\
        "movd       (%0),   "#t" \n\t"\
        "punpcklbw "#z",    "#t" \n\t"\
        "paddsw    "#t",    "#p" \n\t"\
        "packuswb  "#z",    "#p" \n\t"\
        "movd      "#p",    (%0) \n\t"

static const uint64_t ff_pw_32 attribute_used __attribute__ ((aligned(8))) = 0x0020002000200020ULL;

void ff_h264_idct_add_mmx2(uint8_t *dst, int16_t *block, int stride)
{
    /* Load dct coeffs */
    asm volatile(
        "movq   (%0), %%mm0 \n\t"
        "movq  8(%0), %%mm1 \n\t"
        "movq 16(%0), %%mm2 \n\t"
        "movq 24(%0), %%mm3 \n\t"
    :: "r"(block) );

    asm volatile(
        /* mm1=s02+s13  mm2=s02-s13  mm4=d02+d13  mm0=d02-d13 */
        IDCT4_1D( %%mm2, %%mm1, %%mm0, %%mm3, %%mm4 )

        "movq      %0,    %%mm6 \n\t"
        /* in: 1,4,0,2  out: 1,2,3,0 */
        TRANSPOSE4( %%mm3, %%mm1, %%mm0, %%mm2, %%mm4 )

        "paddw     %%mm6, %%mm3 \n\t"

        /* mm2=s02+s13  mm3=s02-s13  mm4=d02+d13  mm1=d02-d13 */
        IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 )

        "pxor %%mm7, %%mm7    \n\t"
    :: "m"(ff_pw_32));

    asm volatile(
    STORE_DIFF_4P( %%mm0, %%mm1, %%mm7)
        "add %1, %0             \n\t"
    STORE_DIFF_4P( %%mm2, %%mm1, %%mm7)
        "add %1, %0             \n\t"
    STORE_DIFF_4P( %%mm3, %%mm1, %%mm7)
        "add %1, %0             \n\t"
    STORE_DIFF_4P( %%mm4, %%mm1, %%mm7)
        : "+r"(dst)
        : "r" ((long)stride)
    );
}

2754:a49f140179e9

	      mmxext_row_head, mmxext_row, mmxext_row_tail, mmxext_row_mid)

declare_idct (ff_mmx_idct, mmx_table,
	      mmx_row_head, mmx_row, mmx_row_tail, mmx_row_mid)