pixel.c

在dsp上实现h.264编解码

/***********************************************************************
 * SATD 8x16
 **********************************************************************/
static int pixel_satd_8x16_altivec( uint8_t *pix1, int i_pix1,
                                    uint8_t *pix2, int i_pix2 )
{
    DECLARE_ALIGNED( int, i_satd, 16 );

    PREP_DIFF;
    vec_s16_t diff0v, diff1v, diff2v, diff3v,
              diff4v, diff5v, diff6v, diff7v;
    vec_s16_t temp0v, temp1v, temp2v, temp3v,
              temp4v, temp5v, temp6v, temp7v;
    vec_s32_t satdv;

    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff0v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff1v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff2v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff3v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff4v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff5v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff6v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff7v );
    VEC_HADAMAR( diff0v, diff1v, diff2v, diff3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diff4v, diff5v, diff6v, diff7v,
                 temp4v, temp5v, temp6v, temp7v );
    VEC_TRANSPOSE_8( temp0v, temp1v, temp2v, temp3v,
                     temp4v, temp5v, temp6v, temp7v,
                     diff0v, diff1v, diff2v, diff3v,
                     diff4v, diff5v, diff6v, diff7v );
    VEC_HADAMAR( diff0v, diff1v, diff2v, diff3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diff4v, diff5v, diff6v, diff7v,
                 temp4v, temp5v, temp6v, temp7v );
    VEC_ADD_ABS( temp0v, zero_s32v, satdv );
    VEC_ADD_ABS( temp1v, satdv,     satdv );
    VEC_ADD_ABS( temp2v, satdv,     satdv );
    VEC_ADD_ABS( temp3v, satdv,     satdv );
    VEC_ADD_ABS( temp4v, satdv,     satdv );
    VEC_ADD_ABS( temp5v, satdv,     satdv );
    VEC_ADD_ABS( temp6v, satdv,     satdv );
    VEC_ADD_ABS( temp7v, satdv,     satdv );

    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff0v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff1v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff2v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff3v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff4v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff5v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff6v );
    VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 8, diff7v );
    VEC_HADAMAR( diff0v, diff1v, diff2v, diff3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diff4v, diff5v, diff6v, diff7v,
                 temp4v, temp5v, temp6v, temp7v );
    VEC_TRANSPOSE_8( temp0v, temp1v, temp2v, temp3v,
                     temp4v, temp5v, temp6v, temp7v,
                     diff0v, diff1v, diff2v, diff3v,
                     diff4v, diff5v, diff6v, diff7v );
    VEC_HADAMAR( diff0v, diff1v, diff2v, diff3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diff4v, diff5v, diff6v, diff7v,
                 temp4v, temp5v, temp6v, temp7v );
    VEC_ADD_ABS( temp0v, satdv,     satdv );
    VEC_ADD_ABS( temp1v, satdv,     satdv );
    VEC_ADD_ABS( temp2v, satdv,     satdv );
    VEC_ADD_ABS( temp3v, satdv,     satdv );
    VEC_ADD_ABS( temp4v, satdv,     satdv );
    VEC_ADD_ABS( temp5v, satdv,     satdv );
    VEC_ADD_ABS( temp6v, satdv,     satdv );
    VEC_ADD_ABS( temp7v, satdv,     satdv );

    satdv = vec_sums( satdv, zero_s32v );
    satdv = vec_splat( satdv, 3 );
    vec_ste( satdv, 0, &i_satd );

    return i_satd / 2;
}

/***********************************************************************
 * SATD 16x8
 **********************************************************************/
static int pixel_satd_16x8_altivec( uint8_t *pix1, int i_pix1,
                                    uint8_t *pix2, int i_pix2 )
{
    DECLARE_ALIGNED( int, i_satd, 16 );

    LOAD_ZERO;
    PREP_LOAD;
    vec_s32_t satdv;
    vec_s16_t pix1v, pix2v;
    vec_s16_t diffh0v, diffh1v, diffh2v, diffh3v,
              diffh4v, diffh5v, diffh6v, diffh7v;
    vec_s16_t diffl0v, diffl1v, diffl2v, diffl3v,
              diffl4v, diffl5v, diffl6v, diffl7v;
    vec_s16_t temp0v, temp1v, temp2v, temp3v,
              temp4v, temp5v, temp6v, temp7v;

    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh0v, diffl0v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh1v, diffl1v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh2v, diffl2v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh3v, diffl3v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh4v, diffl4v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh5v, diffl5v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh6v, diffl6v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh7v, diffl7v );

    VEC_HADAMAR( diffh0v, diffh1v, diffh2v, diffh3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diffh4v, diffh5v, diffh6v, diffh7v,
                 temp4v, temp5v, temp6v, temp7v );

    VEC_TRANSPOSE_8( temp0v, temp1v, temp2v, temp3v,
                     temp4v, temp5v, temp6v, temp7v,
                     diffh0v, diffh1v, diffh2v, diffh3v,
                     diffh4v, diffh5v, diffh6v, diffh7v );

    VEC_HADAMAR( diffh0v, diffh1v, diffh2v, diffh3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diffh4v, diffh5v, diffh6v, diffh7v,
                 temp4v, temp5v, temp6v, temp7v );

    VEC_ADD_ABS( temp0v, zero_s32v, satdv );
    VEC_ADD_ABS( temp1v, satdv,     satdv );
    VEC_ADD_ABS( temp2v, satdv,     satdv );
    VEC_ADD_ABS( temp3v, satdv,     satdv );
    VEC_ADD_ABS( temp4v, satdv,     satdv );
    VEC_ADD_ABS( temp5v, satdv,     satdv );
    VEC_ADD_ABS( temp6v, satdv,     satdv );
    VEC_ADD_ABS( temp7v, satdv,     satdv );

    VEC_HADAMAR( diffl0v, diffl1v, diffl2v, diffl3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diffl4v, diffl5v, diffl6v, diffl7v,
                 temp4v, temp5v, temp6v, temp7v );

    VEC_TRANSPOSE_8( temp0v, temp1v, temp2v, temp3v,
                     temp4v, temp5v, temp6v, temp7v,
                     diffl0v, diffl1v, diffl2v, diffl3v,
                     diffl4v, diffl5v, diffl6v, diffl7v );

    VEC_HADAMAR( diffl0v, diffl1v, diffl2v, diffl3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diffl4v, diffl5v, diffl6v, diffl7v,
                 temp4v, temp5v, temp6v, temp7v );

    VEC_ADD_ABS( temp0v, satdv,     satdv );
    VEC_ADD_ABS( temp1v, satdv,     satdv );
    VEC_ADD_ABS( temp2v, satdv,     satdv );
    VEC_ADD_ABS( temp3v, satdv,     satdv );
    VEC_ADD_ABS( temp4v, satdv,     satdv );
    VEC_ADD_ABS( temp5v, satdv,     satdv );
    VEC_ADD_ABS( temp6v, satdv,     satdv );
    VEC_ADD_ABS( temp7v, satdv,     satdv );

    satdv = vec_sums( satdv, zero_s32v );
    satdv = vec_splat( satdv, 3 );
    vec_ste( satdv, 0, &i_satd );

    return i_satd / 2;
}

/***********************************************************************
 * SATD 16x16
 **********************************************************************/
static int pixel_satd_16x16_altivec( uint8_t *pix1, int i_pix1,
                                     uint8_t *pix2, int i_pix2 )
{
    DECLARE_ALIGNED( int, i_satd, 16 );

    LOAD_ZERO;
    PREP_LOAD;
    vec_s32_t satdv;
    vec_s16_t pix1v, pix2v;
    vec_s16_t diffh0v, diffh1v, diffh2v, diffh3v,
              diffh4v, diffh5v, diffh6v, diffh7v;
    vec_s16_t diffl0v, diffl1v, diffl2v, diffl3v,
              diffl4v, diffl5v, diffl6v, diffl7v;
    vec_s16_t temp0v, temp1v, temp2v, temp3v,
              temp4v, temp5v, temp6v, temp7v;

    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh0v, diffl0v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh1v, diffl1v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh2v, diffl2v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh3v, diffl3v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh4v, diffl4v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh5v, diffl5v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh6v, diffl6v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh7v, diffl7v );
    VEC_HADAMAR( diffh0v, diffh1v, diffh2v, diffh3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diffh4v, diffh5v, diffh6v, diffh7v,
                 temp4v, temp5v, temp6v, temp7v );
    VEC_TRANSPOSE_8( temp0v, temp1v, temp2v, temp3v,
                     temp4v, temp5v, temp6v, temp7v,
                     diffh0v, diffh1v, diffh2v, diffh3v,
                     diffh4v, diffh5v, diffh6v, diffh7v );
    VEC_HADAMAR( diffh0v, diffh1v, diffh2v, diffh3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diffh4v, diffh5v, diffh6v, diffh7v,
                 temp4v, temp5v, temp6v, temp7v );
    VEC_ADD_ABS( temp0v, zero_s32v, satdv );
    VEC_ADD_ABS( temp1v, satdv,     satdv );
    VEC_ADD_ABS( temp2v, satdv,     satdv );
    VEC_ADD_ABS( temp3v, satdv,     satdv );
    VEC_ADD_ABS( temp4v, satdv,     satdv );
    VEC_ADD_ABS( temp5v, satdv,     satdv );
    VEC_ADD_ABS( temp6v, satdv,     satdv );
    VEC_ADD_ABS( temp7v, satdv,     satdv );
    VEC_HADAMAR( diffl0v, diffl1v, diffl2v, diffl3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diffl4v, diffl5v, diffl6v, diffl7v,
                 temp4v, temp5v, temp6v, temp7v );
    VEC_TRANSPOSE_8( temp0v, temp1v, temp2v, temp3v,
                     temp4v, temp5v, temp6v, temp7v,
                     diffl0v, diffl1v, diffl2v, diffl3v,
                     diffl4v, diffl5v, diffl6v, diffl7v );
    VEC_HADAMAR( diffl0v, diffl1v, diffl2v, diffl3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diffl4v, diffl5v, diffl6v, diffl7v,
                 temp4v, temp5v, temp6v, temp7v );
    VEC_ADD_ABS( temp0v, satdv,     satdv );
    VEC_ADD_ABS( temp1v, satdv,     satdv );
    VEC_ADD_ABS( temp2v, satdv,     satdv );
    VEC_ADD_ABS( temp3v, satdv,     satdv );
    VEC_ADD_ABS( temp4v, satdv,     satdv );
    VEC_ADD_ABS( temp5v, satdv,     satdv );
    VEC_ADD_ABS( temp6v, satdv,     satdv );
    VEC_ADD_ABS( temp7v, satdv,     satdv );

    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh0v, diffl0v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh1v, diffl1v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh2v, diffl2v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh3v, diffl3v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh4v, diffl4v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh5v, diffl5v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh6v, diffl6v );
    VEC_DIFF_HL( pix1, i_pix1, pix2, i_pix2, diffh7v, diffl7v );
    VEC_HADAMAR( diffh0v, diffh1v, diffh2v, diffh3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diffh4v, diffh5v, diffh6v, diffh7v,
                 temp4v, temp5v, temp6v, temp7v );
    VEC_TRANSPOSE_8( temp0v, temp1v, temp2v, temp3v,
                     temp4v, temp5v, temp6v, temp7v,
                     diffh0v, diffh1v, diffh2v, diffh3v,
                     diffh4v, diffh5v, diffh6v, diffh7v );
    VEC_HADAMAR( diffh0v, diffh1v, diffh2v, diffh3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diffh4v, diffh5v, diffh6v, diffh7v,
                 temp4v, temp5v, temp6v, temp7v );
    VEC_ADD_ABS( temp0v, satdv,     satdv );
    VEC_ADD_ABS( temp1v, satdv,     satdv );
    VEC_ADD_ABS( temp2v, satdv,     satdv );
    VEC_ADD_ABS( temp3v, satdv,     satdv );
    VEC_ADD_ABS( temp4v, satdv,     satdv );
    VEC_ADD_ABS( temp5v, satdv,     satdv );
    VEC_ADD_ABS( temp6v, satdv,     satdv );
    VEC_ADD_ABS( temp7v, satdv,     satdv );
    VEC_HADAMAR( diffl0v, diffl1v, diffl2v, diffl3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diffl4v, diffl5v, diffl6v, diffl7v,
                 temp4v, temp5v, temp6v, temp7v );
    VEC_TRANSPOSE_8( temp0v, temp1v, temp2v, temp3v,
                     temp4v, temp5v, temp6v, temp7v,
                     diffl0v, diffl1v, diffl2v, diffl3v,
                     diffl4v, diffl5v, diffl6v, diffl7v );
    VEC_HADAMAR( diffl0v, diffl1v, diffl2v, diffl3v,
                 temp0v, temp1v, temp2v, temp3v );
    VEC_HADAMAR( diffl4v, diffl5v, diffl6v, diffl7v,
                 temp4v, temp5v, temp6v, temp7v );
    VEC_ADD_ABS( temp0v, satdv,     satdv );
    VEC_ADD_ABS( temp1v, satdv,     satdv );
    VEC_ADD_ABS( temp2v, satdv,     satdv );
    VEC_ADD_ABS( temp3v, satdv,     satdv );
    VEC_ADD_ABS( temp4v, satdv,     satdv );
    VEC_ADD_ABS( temp5v, satdv,     satdv );
    VEC_ADD_ABS( temp6v, satdv,     satdv );
    VEC_ADD_ABS( temp7v, satdv,     satdv );

    satdv = vec_sums( satdv, zero_s32v );
    satdv = vec_splat( satdv, 3 );
    vec_ste( satdv, 0, &i_satd );

    return i_satd / 2;
}

/****************************************************************************
 * SHU264_pixel_init:
 ****************************************************************************/
void SHU264_pixel_altivec_init( SHU264_pixel_function_t *pixf )
{
    pixf->sad[PIXEL_16x16]  = pixel_sad_16x16_altivec;
    pixf->sad[PIXEL_8x16]   = pixel_sad_8x16_altivec;
    pixf->sad[PIXEL_16x8]   = pixel_sad_16x8_altivec;
    pixf->sad[PIXEL_8x8]    = pixel_sad_8x8_altivec;

    pixf->satd[PIXEL_16x16] = pixel_satd_16x16_altivec;
    pixf->satd[PIXEL_8x16]  = pixel_satd_8x16_altivec;
    pixf->satd[PIXEL_16x8]  = pixel_satd_16x8_altivec;
    pixf->satd[PIXEL_8x8]   = pixel_satd_8x8_altivec;
    pixf->satd[PIXEL_8x4]   = pixel_satd_8x4_altivec;
    pixf->satd[PIXEL_4x8]   = pixel_satd_4x8_altivec;
    pixf->satd[PIXEL_4x4]   = pixel_satd_4x4_altivec;
}