void x264_add4x4_idct_altivec( uint8_t *dst, int16_t dct[16] )
{
vec_u16_t onev = vec_splat_u16(1);
dct[0] += 32; // rounding for the >>6 at the end
vec_s16_t s0, s1, s2, s3;
s0 = vec_ld( 0x00, dct );
s1 = vec_sld( s0, s0, 8 );
s2 = vec_ld( 0x10, dct );
s3 = vec_sld( s2, s2, 8 );
vec_s16_t d0, d1, d2, d3;
IDCT_1D_ALTIVEC( s0, s1, s2, s3, d0, d1, d2, d3 );
vec_s16_t tr0, tr1, tr2, tr3;
VEC_TRANSPOSE_4( d0, d1, d2, d3, tr0, tr1, tr2, tr3 );
vec_s16_t idct0, idct1, idct2, idct3;
IDCT_1D_ALTIVEC( tr0, tr1, tr2, tr3, idct0, idct1, idct2, idct3 );
vec_u8_t perm_ldv = vec_lvsl( 0, dst );
vec_u16_t sixv = vec_splat_u16(6);
LOAD_ZERO;
ALTIVEC_STORE4_SUM_CLIP( &dst[0*FDEC_STRIDE], idct0, perm_ldv );
ALTIVEC_STORE4_SUM_CLIP( &dst[1*FDEC_STRIDE], idct1, perm_ldv );
ALTIVEC_STORE4_SUM_CLIP( &dst[2*FDEC_STRIDE], idct2, perm_ldv );
ALTIVEC_STORE4_SUM_CLIP( &dst[3*FDEC_STRIDE], idct3, perm_ldv );
}
void x264_sub4x4_dct_altivec( int16_t dct[4][4],
uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
{
PREP_DIFF;
PREP_STORE8;
vec_s16_t dct0v, dct1v, dct2v, dct3v;
vec_s16_t tmp0v, tmp1v, tmp2v, tmp3v;
VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 4, dct0v );
VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 4, dct1v );
VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 4, dct2v );
VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 4, dct3v );
VEC_DCT( dct0v, dct1v, dct2v, dct3v, tmp0v, tmp1v, tmp2v, tmp3v );
VEC_TRANSPOSE_4( tmp0v, tmp1v, tmp2v, tmp3v,
dct0v, dct1v, dct2v, dct3v );
VEC_DCT( dct0v, dct1v, dct2v, dct3v, tmp0v, tmp1v, tmp2v, tmp3v );
VEC_TRANSPOSE_4( tmp0v, tmp1v, tmp2v, tmp3v,
dct0v, dct1v, dct2v, dct3v );
VEC_STORE8( dct0v, dct[0] );
VEC_STORE8( dct1v, dct[1] );
VEC_STORE8( dct2v, dct[2] );
VEC_STORE8( dct3v, dct[3] );
}
void x264_sub4x4_dct_altivec( int16_t dct[16], uint8_t *pix1, uint8_t *pix2 )
{
PREP_DIFF_8BYTEALIGNED;
vec_s16_t dct0v, dct1v, dct2v, dct3v;
vec_s16_t tmp0v, tmp1v, tmp2v, tmp3v;
vec_u8_t permHighv;
VEC_DIFF_H_8BYTE_ALIGNED( pix1, FENC_STRIDE, pix2, FDEC_STRIDE, 4, dct0v );
VEC_DIFF_H_8BYTE_ALIGNED( pix1, FENC_STRIDE, pix2, FDEC_STRIDE, 4, dct1v );
VEC_DIFF_H_8BYTE_ALIGNED( pix1, FENC_STRIDE, pix2, FDEC_STRIDE, 4, dct2v );
VEC_DIFF_H_8BYTE_ALIGNED( pix1, FENC_STRIDE, pix2, FDEC_STRIDE, 4, dct3v );
VEC_DCT( dct0v, dct1v, dct2v, dct3v, tmp0v, tmp1v, tmp2v, tmp3v );
VEC_TRANSPOSE_4( tmp0v, tmp1v, tmp2v, tmp3v,
dct0v, dct1v, dct2v, dct3v );
permHighv = (vec_u8_t) CV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17);
VEC_DCT( dct0v, dct1v, dct2v, dct3v, tmp0v, tmp1v, tmp2v, tmp3v );
vec_st(vec_perm(tmp0v, tmp1v, permHighv), 0, dct);
vec_st(vec_perm(tmp2v, tmp3v, permHighv), 16, dct);
}
static void h264_idct_add_altivec(uint8_t *dst, int16_t *block, int stride)
{
vec_s16 va0, va1, va2, va3;
vec_s16 vz0, vz1, vz2, vz3;
vec_s16 vtmp0, vtmp1, vtmp2, vtmp3;
vec_u8 va_u8;
vec_u32 va_u32;
vec_s16 vdst_ss;
const vec_u16 v6us = vec_splat_u16(6);
vec_u8 vdst, vdst_orig;
vec_u8 vdst_mask = vec_lvsl(0, dst);
int element = ((unsigned long)dst & 0xf) >> 2;
LOAD_ZERO;
block[0] += 32; /* add 32 as a DC-level for rounding */
vtmp0 = vec_ld(0,block);
vtmp1 = vec_sld(vtmp0, vtmp0, 8);
vtmp2 = vec_ld(16,block);
vtmp3 = vec_sld(vtmp2, vtmp2, 8);
memset(block, 0, 16 * sizeof(int16_t));
VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3);
VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
va0 = vec_sra(va0,v6us);
va1 = vec_sra(va1,v6us);
va2 = vec_sra(va2,v6us);
va3 = vec_sra(va3,v6us);
VEC_LOAD_U8_ADD_S16_STORE_U8(va0);
dst += stride;
VEC_LOAD_U8_ADD_S16_STORE_U8(va1);
dst += stride;
VEC_LOAD_U8_ADD_S16_STORE_U8(va2);
dst += stride;
VEC_LOAD_U8_ADD_S16_STORE_U8(va3);
}
static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
{
vec_s16 va0, va1, va2, va3;//s16
vec_s16 vz0, vz1, vz2, vz3;//s16
vec_s16 vtmp0, vtmp1, vtmp2, vtmp3;//s16
vec_u8 va_u8;//u8
vec_u32 va_u32;//u32
vec_s16 vdst_ss;//s16
const vec_u16 v6us = __vsplth(6);//u16
vec_u8 vdst, vdst_orig;//u8
vec_u8 vdst_mask = __lvsl(dst,0);//u8
int element = ((unsigned long)dst & 0xf) >> 2;
LOAD_ZERO;
block[0] += 32; /* add 32 as a DC-level for rounding */
vtmp0 = __lvx(block,0);
vtmp1 = __vsldoi(vtmp0, vtmp0, 8);
vtmp2 = __lvx(block,16);
vtmp3 = __vsldoi(vtmp2, vtmp2, 8);
VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3);
VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
va0 = __vsrah(va0,v6us);
va1 = __vsrah(va1,v6us);
va2 = __vsrah(va2,v6us);
va3 = __vsrah(va3,v6us);
VEC_LOAD_U8_ADD_S16_STORE_U8(va0);
dst += stride;
VEC_LOAD_U8_ADD_S16_STORE_U8(va1);
dst += stride;
VEC_LOAD_U8_ADD_S16_STORE_U8(va2);
dst += stride;
VEC_LOAD_U8_ADD_S16_STORE_U8(va3);
}