pixel.c

从服务器上下的x264编码器源码……希望对大家有帮助……这个是09年4月11日的

        temp_hv = vec_ld(16, pix2);
        pix2v = vec_perm(temp_lv, temp_hv, perm2vA);
        pix2 += i_stride;

        sum0v = (vec_s32_t) vec_sum4s( vec_sub( vec_max( fencv, pix0v ), vec_min( fencv, pix0v ) ), (vec_u32_t) sum0v );

        sum1v = (vec_s32_t) vec_sum4s( vec_sub( vec_max( fencv, pix1v ), vec_min( fencv, pix1v ) ), (vec_u32_t) sum1v );

        sum2v = (vec_s32_t) vec_sum4s( vec_sub( vec_max( fencv, pix2v ), vec_min( fencv, pix2v ) ), (vec_u32_t) sum2v );

        temp_lv = vec_ld(0, pix0);
        temp_hv = vec_ld(16, pix0);
        pix0v = vec_perm(temp_lv, temp_hv, perm0vB);
        pix0 += i_stride;

        temp_lv = vec_ld(0, pix1);
        temp_hv = vec_ld(16, pix1);
        pix1v = vec_perm(temp_lv, temp_hv, perm1vB);
        pix1 += i_stride;

        temp_lv = vec_ld(0, fenc);
        fencv = vec_perm(temp_lv, temp_hv, permEncv);
        fenc += FENC_STRIDE;

        temp_lv = vec_ld(0, pix2);
        temp_hv = vec_ld(16, pix2);
        pix2v = vec_perm(temp_lv, temp_hv, perm2vB);
        pix2 += i_stride;

        sum0v = (vec_s32_t) vec_sum4s( vec_sub( vec_max( fencv, pix0v ), vec_min( fencv, pix0v ) ), (vec_u32_t) sum0v );

        sum1v = (vec_s32_t) vec_sum4s( vec_sub( vec_max( fencv, pix1v ), vec_min( fencv, pix1v ) ), (vec_u32_t) sum1v );

        sum2v = (vec_s32_t) vec_sum4s( vec_sub( vec_max( fencv, pix2v ), vec_min( fencv, pix2v ) ), (vec_u32_t) sum2v );
    }

    sum0v = vec_sum2s( sum0v, zero_s32v );
    sum1v = vec_sum2s( sum1v, zero_s32v );
    sum2v = vec_sum2s( sum2v, zero_s32v );

    sum0v = vec_splat( sum0v, 1 );
    sum1v = vec_splat( sum1v, 1 );
    sum2v = vec_splat( sum2v, 1 );

    vec_ste( sum0v, 0, &sum0);
    vec_ste( sum1v, 0, &sum1);
    vec_ste( sum2v, 0, &sum2);

    scores[0] = sum0;
    scores[1] = sum1;
    scores[2] = sum2;
}

/***********************************************************************
* SSD routines
**********************************************************************/

static int pixel_ssd_16x16_altivec ( uint8_t *pix1, int i_stride_pix1,
                                     uint8_t *pix2, int i_stride_pix2)
{
    DECLARE_ALIGNED_16( int sum );

    int y;
    LOAD_ZERO;
    vec_u8_t  pix1vA, pix2vA, pix1vB, pix2vB;
    vec_u32_t sumv;
    vec_u8_t maxA, minA, diffA, maxB, minB, diffB;
    vec_u8_t temp_lv, temp_hv;
    vec_u8_t permA, permB;

    sumv = vec_splat_u32(0);

    permA = vec_lvsl(0, pix2);
    permB = vec_lvsl(0, pix2 + i_stride_pix2);

    temp_lv = vec_ld(0, pix2);
    temp_hv = vec_ld(16, pix2);
    pix2vA = vec_perm(temp_lv, temp_hv, permA);
    pix1vA = vec_ld(0, pix1);

    for (y=0; y < 7; y++)
    {
        pix1 += i_stride_pix1;
        pix2 += i_stride_pix2;

        maxA = vec_max(pix1vA, pix2vA);
        minA = vec_min(pix1vA, pix2vA);

        temp_lv = vec_ld(0, pix2);
        temp_hv = vec_ld(16, pix2);
        pix2vB = vec_perm(temp_lv, temp_hv, permB);
        pix1vB = vec_ld(0, pix1);

        diffA = vec_sub(maxA, minA);
        sumv = vec_msum(diffA, diffA, sumv);

        pix1 += i_stride_pix1;
        pix2 += i_stride_pix2;

        maxB = vec_max(pix1vB, pix2vB);
        minB = vec_min(pix1vB, pix2vB);

        temp_lv = vec_ld(0, pix2);
        temp_hv = vec_ld(16, pix2);
        pix2vA = vec_perm(temp_lv, temp_hv, permA);
        pix1vA = vec_ld(0, pix1);

        diffB = vec_sub(maxB, minB);
        sumv = vec_msum(diffB, diffB, sumv);
    }

    pix1 += i_stride_pix1;
    pix2 += i_stride_pix2;

    temp_lv = vec_ld(0, pix2);
    temp_hv = vec_ld(16, pix2);
    pix2vB = vec_perm(temp_lv, temp_hv, permB);
    pix1vB = vec_ld(0, pix1);

    maxA = vec_max(pix1vA, pix2vA);
    minA = vec_min(pix1vA, pix2vA);

    maxB = vec_max(pix1vB, pix2vB);
    minB = vec_min(pix1vB, pix2vB);

    diffA = vec_sub(maxA, minA);
    sumv = vec_msum(diffA, diffA, sumv);

    diffB = vec_sub(maxB, minB);
    sumv = vec_msum(diffB, diffB, sumv);

    sumv = (vec_u32_t) vec_sums((vec_s32_t) sumv, zero_s32v);
    sumv = vec_splat(sumv, 3);
    vec_ste((vec_s32_t) sumv, 0, &sum);
    return sum;
}

static int pixel_ssd_8x8_altivec ( uint8_t *pix1, int i_stride_pix1,
                                   uint8_t *pix2, int i_stride_pix2)
{
    DECLARE_ALIGNED_16( int sum );

    int y;
    LOAD_ZERO;
    vec_u8_t  pix1v, pix2v;
    vec_u32_t sumv;
    vec_u8_t maxv, minv, diffv;
    vec_u8_t temp_lv, temp_hv;
    vec_u8_t perm1v, perm2v;

    const vec_u32_t sel = (vec_u32_t)CV(-1,-1,0,0);

    sumv = vec_splat_u32(0);

    perm1v = vec_lvsl(0, pix1);
    perm2v = vec_lvsl(0, pix2);

    for (y=0; y < 8; y++)
    {
        temp_hv = vec_ld(0, pix1);
        temp_lv = vec_ld(7, pix1);
        pix1v = vec_perm(temp_hv, temp_lv, perm1v);

        temp_hv = vec_ld(0, pix2);
        temp_lv = vec_ld(7, pix2);
        pix2v = vec_perm(temp_hv, temp_lv, perm2v);

        maxv = vec_max(pix1v, pix2v);
        minv = vec_min(pix1v, pix2v);

        diffv = vec_sub(maxv, minv);
        sumv = vec_msum(diffv, diffv, sumv);

        pix1 += i_stride_pix1;
        pix2 += i_stride_pix2;
    }

    sumv = vec_sel( zero_u32v, sumv, sel );

    sumv = (vec_u32_t) vec_sums((vec_s32_t) sumv, zero_s32v);
    sumv = vec_splat(sumv, 3);
    vec_ste((vec_s32_t) sumv, 0, &sum);

    return sum;
}


/****************************************************************************
 * variance
 ****************************************************************************/
static int x264_pixel_var_16x16_altivec( uint8_t *pix, int i_stride )
{
    DECLARE_ALIGNED_16(uint32_t sum_tab[4]);
    DECLARE_ALIGNED_16(uint32_t sqr_tab[4]);

    LOAD_ZERO;
    vec_u32_t sqr_v = zero_u32v;
    vec_u32_t sum_v = zero_u32v;

    int y;
    for( y = 0; y < 16; ++y )
    {
        vec_u8_t pix0_v = vec_ld(0, pix);
        sum_v = vec_sum4s(pix0_v, sum_v);
        sqr_v = vec_msum(pix0_v, pix0_v, sqr_v);

        pix += i_stride;
    }
    sum_v = (vec_u32_t)vec_sums( (vec_s32_t)sum_v, zero_s32v );
    sqr_v = (vec_u32_t)vec_sums( (vec_s32_t)sqr_v, zero_s32v );
    vec_ste(sum_v, 12, sum_tab);
    vec_ste(sqr_v, 12, sqr_tab);

    uint32_t sum = sum_tab[3];
    uint32_t sqr = sqr_tab[3];
    uint32_t var = sqr - (sum * sum >> 8);
    return var;
}

static int x264_pixel_var_8x8_altivec( uint8_t *pix, int i_stride )
{
    DECLARE_ALIGNED_16(uint32_t sum_tab[4]);
    DECLARE_ALIGNED_16(uint32_t sqr_tab[4]);

    LOAD_ZERO;
    vec_u32_t sqr_v = zero_u32v;
    vec_u32_t sum_v = zero_u32v;

    static const vec_u8_t perm_tab[] = {
        CV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,  /* pix=mod16, i_stride=mod16 */
           0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17),
        CV(0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F,  /* pix=mod8, i_stride=mod16  */
           0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F),
    };
    vec_u8_t perm = perm_tab[ ((uintptr_t)pix & 8) >> 3 ];

    int y;
    for( y = 0; y < 8; y+=2 )
    {
        vec_u8_t pix0_v = vec_ld(0, pix);
        vec_u8_t pix1_v = vec_ld(i_stride, pix);
        vec_u8_t pix_v = vec_perm(pix0_v, pix1_v, perm);
        sum_v = vec_sum4s(pix_v, sum_v);
        sqr_v = vec_msum(pix_v, pix_v, sqr_v);

        pix += i_stride<<1;
    }
    sum_v = (vec_u32_t)vec_sums( (vec_s32_t)sum_v, zero_s32v );
    sqr_v = (vec_u32_t)vec_sums( (vec_s32_t)sqr_v, zero_s32v );
    vec_ste(sum_v, 12, sum_tab);
    vec_ste(sqr_v, 12, sqr_tab);

    uint32_t sum = sum_tab[3];
    uint32_t sqr = sqr_tab[3];
    uint32_t var = sqr - (sum * sum >> 6);
    return var;
}


/**********************************************************************
 * SA8D routines: sum of 8x8 Hadamard transformed differences
 **********************************************************************/
/* SA8D_1D unrolled by 8 in Altivec */
#define SA8D_1D_ALTIVEC( sa8d0v, sa8d1v, sa8d2v, sa8d3v,  \
                         sa8d4v, sa8d5v, sa8d6v, sa8d7v ) \
{                                                         \
    /* int    a0  =        SRC(0) + SRC(4) */             \
    vec_s16_t a0v = vec_add(sa8d0v, sa8d4v);              \
    /* int    a4  =        SRC(0) - SRC(4) */             \
    vec_s16_t a4v = vec_sub(sa8d0v, sa8d4v);              \
    /* int    a1  =        SRC(1) + SRC(5) */             \
    vec_s16_t a1v = vec_add(sa8d1v, sa8d5v);              \
    /* int    a5  =        SRC(1) - SRC(5) */             \
    vec_s16_t a5v = vec_sub(sa8d1v, sa8d5v);              \
    /* int    a2  =        SRC(2) + SRC(6) */             \
    vec_s16_t a2v = vec_add(sa8d2v, sa8d6v);              \
    /* int    a6  =        SRC(2) - SRC(6) */             \
    vec_s16_t a6v = vec_sub(sa8d2v, sa8d6v);              \
    /* int    a3  =        SRC(3) + SRC(7) */             \
    vec_s16_t a3v = vec_add(sa8d3v, sa8d7v);              \
    /* int    a7  =        SRC(3) - SRC(7) */             \
    vec_s16_t a7v = vec_sub(sa8d3v, sa8d7v);              \
                                                          \
    /* int    b0  =         a0 + a2  */                   \
    vec_s16_t b0v = vec_add(a0v, a2v);                    \
    /* int    b2  =         a0 - a2; */                   \
    vec_s16_t  b2v = vec_sub(a0v, a2v);                   \
    /* int    b1  =         a1 + a3; */                   \
    vec_s16_t b1v = vec_add(a1v, a3v);                    \
    /* int    b3  =         a1 - a3; */                   \
    vec_s16_t b3v = vec_sub(a1v, a3v);                    \
    /* int    b4  =         a4 + a6; */                   \
    vec_s16_t b4v = vec_add(a4v, a6v);                    \
    /* int    b6  =         a4 - a6; */                   \
    vec_s16_t b6v = vec_sub(a4v, a6v);                    \
    /* int    b5  =         a5 + a7; */                   \
    vec_s16_t b5v = vec_add(a5v, a7v);                    \
    /* int    b7  =         a5 - a7; */                   \
    vec_s16_t b7v = vec_sub(a5v, a7v);                    \
                                                          \
    /* DST(0,        b0 + b1) */                          \
    sa8d0v = vec_add(b0v, b1v);                           \
    /* DST(1,        b0 - b1) */                          \
    sa8d1v = vec_sub(b0v, b1v);                           \
    /* DST(2,        b2 + b3) */                          \
    sa8d2v = vec_add(b2v, b3v);                           \
    /* DST(3,        b2 - b3) */                          \
    sa8d3v = vec_sub(b2v, b3v);                           \
    /* DST(4,        b4 + b5) */                          \
    sa8d4v = vec_add(b4v, b5v);                           \
    /* DST(5,        b4 - b5) */                          \
    sa8d5v = vec_sub(b4v, b5v);                           \
    /* DST(6,        b6 + b7) */                          \
    sa8d6v = vec_add(b6v, b7v);                           \
    /* DST(7,        b6 - b7) */                          \
    sa8d7v = vec_sub(b6v, b7v);                           \
}

static int pixel_sa8d_8x8_core_altivec( uint8_t *pix1, int i_pix1,
                                        uint8_t *pix2, int i_pix2 )
{
    int32_t i_satd=0;

    PREP_DIFF;
    PREP_LOAD_SRC( pix1 );
    PREP_LOAD_SRC( pix2 );

    vec_s16_t diff0v, diff1v, diff2v, diff3v, diff4v, diff5v, diff6v, diff7v;

    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff0v, pix2 );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff1v, pix2 );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff2v, pix2 );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff3v, pix2 );

    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff4v, pix2 );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff5v, pix2 );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff6v, pix2 );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff7v, pix2 );

    vec_s16_t sa8d0v, sa8d1v, sa8d2v, sa8d3v, sa8d4v, sa8d5v, sa8d6v, sa8d7v;

    SA8D_1D_ALTIVEC(diff0v, diff1v, diff2v, diff3v,
                    diff4v, diff5v, diff6v, diff7v);

    VEC_TRANSPOSE_8(diff0v, diff1v, diff2v, diff3v,
                    diff4v, diff5v, diff6v, diff7v,
                    sa8d0v, sa8d1v, sa8d2v, sa8d3v,
                    sa8d4v, sa8d5v, sa8d6v, sa8d7v );

    SA8D_1D_ALTIVEC(sa8d0v, sa8d1v, sa8d2v, sa8d3v,
                    sa8d4v, sa8d5v, sa8d6v, sa8d7v );

    /* accumulation of the absolute value of all elements of the resulting bloc */
    vec_s16_t abs0v = VEC_ABS(sa8d0v);
    vec_s16_t abs1v = VEC_ABS(sa8d1v);
    vec_s16_t sum01v = vec_add(abs0v, abs1v);

    vec_s16_t abs2v = VEC_ABS(sa8d2v