diff options
Diffstat (limited to 'common/x86/ih264_luma_intra_pred_filters_ssse3.c')
| -rw-r--r-- | common/x86/ih264_luma_intra_pred_filters_ssse3.c | 288 |
1 files changed, 122 insertions, 166 deletions
diff --git a/common/x86/ih264_luma_intra_pred_filters_ssse3.c b/common/x86/ih264_luma_intra_pred_filters_ssse3.c index 5a35372..a1721d5 100644 --- a/common/x86/ih264_luma_intra_pred_filters_ssse3.c +++ b/common/x86/ih264_luma_intra_pred_filters_ssse3.c @@ -122,28 +122,22 @@ void ih264_intra_pred_luma_4x4_mode_vert_ssse3(UWORD8 *pu1_src, { UWORD8 *pu1_top; WORD32 dst_strd2, dst_strd3; - - __m128i top_16x8b; - __m128i mask_full_128b, mask_low_32b; + WORD32 i4_top; UNUSED(src_strd); UNUSED(ngbr_avail); - mask_full_128b = _mm_set1_epi8(0xff); - pu1_top = pu1_src + BLK_SIZE + 1; - mask_low_32b = _mm_srli_si128(mask_full_128b, 12); - - top_16x8b = _mm_loadl_epi64((__m128i *)pu1_top); + i4_top = *((WORD32 *)pu1_top); dst_strd2 = dst_strd << 1; dst_strd3 = dst_strd + dst_strd2; - _mm_maskmoveu_si128(top_16x8b, mask_low_32b, (char*)pu1_dst); - _mm_maskmoveu_si128(top_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd)); - _mm_maskmoveu_si128(top_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd2)); - _mm_maskmoveu_si128(top_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd3)); + *((WORD32 *)(pu1_dst)) = i4_top; + *((WORD32 *)(pu1_dst + dst_strd)) = i4_top; + *((WORD32 *)(pu1_dst + dst_strd2)) = i4_top; + *((WORD32 *)(pu1_dst + dst_strd3)) = i4_top; } /** @@ -185,39 +179,31 @@ void ih264_intra_pred_luma_4x4_mode_horz_ssse3(UWORD8 *pu1_src, WORD32 dst_strd, WORD32 ngbr_avail) { - UWORD8 *pu1_left; + UWORD8 *pu1_left = NULL; /* Pointer to start of left predictors */ + WORD32 row1,row2,row3,row4; + UWORD8 val; WORD32 dst_strd2, dst_strd3; - WORD32 val1, val2; - - __m128i left_16x8b; - __m128i row1_16x8b, row2_16x8b, row3_16x8b, row4_16x8b; - __m128i mask_full_128b, mask_low_32b; UNUSED(src_strd); UNUSED(ngbr_avail); - - mask_full_128b = _mm_set1_epi8(0xff); - pu1_left = pu1_src + BLK_SIZE - 1; - mask_low_32b = _mm_srli_si128(mask_full_128b, 12); - left_16x8b = _mm_loadl_epi64((__m128i *)(pu1_left - 3)); - - val1 = _mm_extract_epi16(left_16x8b, 1); - val2 = _mm_extract_epi16(left_16x8b, 0); - - row1_16x8b = _mm_set1_epi8(val1 >> 8); - row2_16x8b = _mm_set1_epi8(val1 & 0xff); - row3_16x8b = _mm_set1_epi8(val2 >> 8); - row4_16x8b = _mm_set1_epi8(val2 & 0xff); + val = *pu1_left; + row1 = val + (val << 8) + (val << 16) + (val << 24); + val = *(pu1_left - 1); + row2 = val + (val << 8) + (val << 16) + (val << 24); + val = *(pu1_left - 2); + row3 = val + (val << 8) + (val << 16) + (val << 24); + val = *(pu1_left - 3); + row4 = val + (val << 8) + (val << 16) + (val << 24); dst_strd2 = dst_strd << 1; dst_strd3 = dst_strd + dst_strd2; - _mm_maskmoveu_si128(row1_16x8b, mask_low_32b, (char*)pu1_dst); - _mm_maskmoveu_si128(row2_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd)); - _mm_maskmoveu_si128(row3_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd2)); - _mm_maskmoveu_si128(row4_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd3)); + *((WORD32 *)(pu1_dst)) = row1; + *((WORD32 *)(pu1_dst + dst_strd)) = row2; + *((WORD32 *)(pu1_dst + dst_strd2)) = row3; + *((WORD32 *)(pu1_dst + dst_strd3)) = row4; } /** @@ -259,72 +245,43 @@ void ih264_intra_pred_luma_4x4_mode_dc_ssse3(UWORD8 *pu1_src, WORD32 ngbr_avail) { UWORD8 u1_useleft; /* availability of left predictors (only for DC) */ - UWORD8 u1_usetop; /* availability of top predictors (only for DC) */ - UWORD8 *pu1_left, *pu1_top; - WORD32 dc_val, flag; + UWORD8 u1_usetop; /* availability of top predictors (only for DC) */ + UWORD8 *pu1_left = NULL; /* Pointer to start of left predictors */ + UWORD8 *pu1_top = NULL; /* Pointer to start of top predictors */ WORD32 dst_strd2, dst_strd3; - - __m128i mask_full_128b, mask_low_32b; - __m128i dcval_16x8b; - + WORD32 val = 0; UNUSED(src_strd); UNUSED(ngbr_avail); - - mask_full_128b = _mm_set1_epi8(0xff); - u1_useleft = BOOLEAN(ngbr_avail & LEFT_MB_AVAILABLE_MASK); u1_usetop = BOOLEAN(ngbr_avail & TOP_MB_AVAILABLE_MASK); - - pu1_left = pu1_src + BLK_SIZE - 1; pu1_top = pu1_src + BLK_SIZE + 1; + pu1_left = pu1_src + BLK_SIZE - 1; - mask_low_32b = _mm_srli_si128(mask_full_128b, 12); - - flag = u1_useleft + u1_usetop; - - if(flag) + if(u1_useleft) { - WORD32 shft, ofst = 0; - - __m128i left_16x8b, top_16x8b, val_16x8b, tmp_8x16b, zero_vector; - - if(u1_useleft) - { - left_16x8b = _mm_loadl_epi64((__m128i *)(pu1_left - 3)); - ofst += 2; - } - else - left_16x8b = _mm_setzero_si128(); - - zero_vector = _mm_setzero_si128(); - - if(u1_usetop) - { - top_16x8b = _mm_loadl_epi64((__m128i *)pu1_top); - ofst += 2; - } - else - top_16x8b = _mm_setzero_si128(); - - shft = flag + 1; - val_16x8b = _mm_unpacklo_epi32(left_16x8b, top_16x8b); - tmp_8x16b = _mm_sad_epu8(val_16x8b, zero_vector); - - dc_val = _mm_extract_epi16(tmp_8x16b, 0); - dc_val = (dc_val + ofst) >> shft; + val += *pu1_left--; + val += *pu1_left--; + val += *pu1_left--; + val += *pu1_left + 2; } - else - dc_val = 128; + if(u1_usetop) + { + val += *pu1_top + *(pu1_top + 1) + *(pu1_top + 2) + *(pu1_top + 3) + + 2; + } + /* Since 2 is added if either left/top pred is there, + val still being zero implies both preds are not there */ + val = (val) ? (val >> (1 + u1_useleft + u1_usetop)) : 128; + + val = val + (val << 8) + (val << 16) + (val << 24); dst_strd2 = dst_strd << 1; dst_strd3 = dst_strd + dst_strd2; - dcval_16x8b = _mm_set1_epi8(dc_val); - - _mm_maskmoveu_si128(dcval_16x8b, mask_low_32b, (char*)pu1_dst); - _mm_maskmoveu_si128(dcval_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd)); - _mm_maskmoveu_si128(dcval_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd2)); - _mm_maskmoveu_si128(dcval_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd3)); + *((WORD32 *)(pu1_dst)) = val; + *((WORD32 *)(pu1_dst + dst_strd)) = val; + *((WORD32 *)(pu1_dst + dst_strd2)) = val; + *((WORD32 *)(pu1_dst + dst_strd3)) = val; } /** @@ -371,7 +328,7 @@ void ih264_intra_pred_luma_4x4_mode_diag_dl_ssse3(UWORD8 *pu1_src, __m128i top_16x8b, top_8x16b, top_sh_8x16b; __m128i res1_8x16b, res2_8x16b, res_16x8b; __m128i zero_vector, const_2_8x16b; - __m128i mask_full_128b, mask_low_32b; + WORD32 row1,row2,row3,row4; UNUSED(src_strd); UNUSED(ngbr_avail); @@ -382,13 +339,11 @@ void ih264_intra_pred_luma_4x4_mode_diag_dl_ssse3(UWORD8 *pu1_src, zero_vector = _mm_setzero_si128(); top_8x16b = _mm_unpacklo_epi8(top_16x8b, zero_vector); //t0 t1 t2 t3 t4 t5 t6 t7 - mask_full_128b = _mm_set1_epi8(0xff); top_sh_8x16b = _mm_srli_si128(top_8x16b, 2); //t1 t2 t3 t4 t5 t6 t7 0 const_2_8x16b = _mm_set1_epi16(2); top_sh_8x16b = _mm_shufflehi_epi16(top_sh_8x16b, 0xa4); //t1 t2 t3 t4 t5 t6 t7 t7 res1_8x16b = _mm_add_epi16(top_8x16b, top_sh_8x16b); - mask_low_32b = _mm_srli_si128(mask_full_128b, 12); res2_8x16b = _mm_srli_si128(res1_8x16b, 2); res1_8x16b = _mm_add_epi16(res1_8x16b, const_2_8x16b); @@ -399,13 +354,18 @@ void ih264_intra_pred_luma_4x4_mode_diag_dl_ssse3(UWORD8 *pu1_src, dst_strd3 = dst_strd + dst_strd2; res_16x8b = _mm_packus_epi16(res1_8x16b, res1_8x16b); - _mm_maskmoveu_si128(res_16x8b, mask_low_32b, (char*)pu1_dst); + row1 = _mm_cvtsi128_si32(res_16x8b); res_16x8b = _mm_srli_si128(res_16x8b, 1); - _mm_maskmoveu_si128(res_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd)); + row2 = _mm_cvtsi128_si32(res_16x8b); res_16x8b = _mm_srli_si128(res_16x8b, 1); - _mm_maskmoveu_si128(res_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd2)); + row3 = _mm_cvtsi128_si32(res_16x8b); res_16x8b = _mm_srli_si128(res_16x8b, 1); - _mm_maskmoveu_si128(res_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd3)); + row4 = _mm_cvtsi128_si32(res_16x8b); + + *((WORD32 *)(pu1_dst)) = row1; + *((WORD32 *)(pu1_dst + dst_strd)) = row2; + *((WORD32 *)(pu1_dst + dst_strd2)) = row3; + *((WORD32 *)(pu1_dst + dst_strd3)) = row4; } /** @@ -454,7 +414,7 @@ void ih264_intra_pred_luma_4x4_mode_diag_dr_ssse3(UWORD8 *pu1_src, __m128i res1_8x16b, res2_8x16b; __m128i res1_16x8b, res2_16x8b; __m128i zero_vector, const_2_8x16b; - __m128i mask_full_128b, mask_low_32b; + WORD32 row1,row2,row3,row4; UNUSED(src_strd); UNUSED(ngbr_avail); @@ -468,13 +428,11 @@ void ih264_intra_pred_luma_4x4_mode_diag_dr_ssse3(UWORD8 *pu1_src, top_left_8x16b = _mm_unpacklo_epi8(top_left_16x8b, zero_vector); top_left_sh_8x16b = _mm_unpacklo_epi8(top_left_sh_16x8b, zero_vector); - mask_full_128b = _mm_set1_epi8(0xff); res1_8x16b = _mm_add_epi16(top_left_8x16b, top_left_sh_8x16b); //l3+l2 l2+l1 l1+l0 l0+tl tl+t0 t0+t1 t1+t2 t2+t3... const_2_8x16b = _mm_set1_epi16(2); res2_8x16b = _mm_srli_si128(res1_8x16b, 2); //l2+l1 l1+l0 l0+tl tl+t0 t0+t1 t1+t2 t2+t3... res1_8x16b = _mm_add_epi16(res1_8x16b, const_2_8x16b); - mask_low_32b = _mm_srli_si128(mask_full_128b, 12); res1_8x16b = _mm_add_epi16(res2_8x16b, res1_8x16b); //l3+2*l2+l1+2 l2+2*l1+l0+2... res1_8x16b = _mm_srai_epi16(res1_8x16b, 2); res1_16x8b = _mm_packus_epi16(res1_8x16b, res1_8x16b); @@ -483,12 +441,18 @@ void ih264_intra_pred_luma_4x4_mode_diag_dr_ssse3(UWORD8 *pu1_src, dst_strd3 = dst_strd + dst_strd2; res2_16x8b = _mm_srli_si128(res1_16x8b, 3); - _mm_maskmoveu_si128(res2_16x8b, mask_low_32b, (char*)pu1_dst); + + row1 = _mm_cvtsi128_si32(res2_16x8b); res2_16x8b = _mm_srli_si128(res1_16x8b, 2); - _mm_maskmoveu_si128(res2_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd)); + row2 = _mm_cvtsi128_si32(res2_16x8b); res2_16x8b = _mm_srli_si128(res1_16x8b, 1); - _mm_maskmoveu_si128(res2_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd2)); - _mm_maskmoveu_si128(res1_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd3)); + row3 = _mm_cvtsi128_si32(res2_16x8b); + row4 = _mm_cvtsi128_si32(res1_16x8b); + + *((WORD32 *)(pu1_dst)) = row1; + *((WORD32 *)(pu1_dst + dst_strd)) = row2; + *((WORD32 *)(pu1_dst + dst_strd2)) = row3; + *((WORD32 *)(pu1_dst + dst_strd3)) = row4; } /** @@ -537,14 +501,11 @@ void ih264_intra_pred_luma_4x4_mode_vert_r_ssse3(UWORD8 *pu1_src, __m128i w121_a1_8x16b, w121_a2_8x16b, w121_sh_8x16b; __m128i row1_16x8b, row2_16x8b, row3_16x8b, row4_16x8b; __m128i zero_vector, const_2_8x16b; - __m128i mask_full_128b, mask_low_32b; + WORD32 row1,row2,row3,row4; UNUSED(src_strd); UNUSED(ngbr_avail); - mask_full_128b = _mm_set1_epi8(0xff); - mask_low_32b = _mm_srli_si128(mask_full_128b, 12); - pu1_left = pu1_src + BLK_SIZE - 1; val_16x8b = _mm_loadl_epi64((__m128i *)(pu1_left - 2)); @@ -575,10 +536,15 @@ void ih264_intra_pred_luma_4x4_mode_vert_r_ssse3(UWORD8 *pu1_src, dst_strd2 = dst_strd << 1; dst_strd3 = dst_strd + dst_strd2; - _mm_maskmoveu_si128(row1_16x8b, mask_low_32b, (char*)pu1_dst); - _mm_maskmoveu_si128(row2_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd)); - _mm_maskmoveu_si128(row3_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd2)); - _mm_maskmoveu_si128(row4_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd3)); + row1 = _mm_cvtsi128_si32(row1_16x8b); + row2 = _mm_cvtsi128_si32(row2_16x8b); + row3 = _mm_cvtsi128_si32(row3_16x8b); + row4 = _mm_cvtsi128_si32(row4_16x8b); + + *((WORD32 *)(pu1_dst)) = row1; + *((WORD32 *)(pu1_dst + dst_strd)) = row2; + *((WORD32 *)(pu1_dst + dst_strd2)) = row3; + *((WORD32 *)(pu1_dst + dst_strd3)) = row4; } /* @@ -629,14 +595,11 @@ void ih264_intra_pred_luma_4x4_mode_horz_d_ssse3(UWORD8 *pu1_src, __m128i row1_16x8b, row2_16x8b, row3_16x8b, row4_16x8b; __m128i zero_vector, const_2_8x16b; - __m128i mask_full_128b, mask_low_32b; + WORD32 row1,row2,row3,row4; UNUSED(src_strd); UNUSED(ngbr_avail); - mask_full_128b = _mm_set1_epi8(0xff); - mask_low_32b = _mm_srli_si128(mask_full_128b, 12); - pu1_left = pu1_src + BLK_SIZE - 1; val_16x8b = _mm_loadl_epi64((__m128i *)(pu1_left - 3)); @@ -669,10 +632,15 @@ void ih264_intra_pred_luma_4x4_mode_horz_d_ssse3(UWORD8 *pu1_src, row2_16x8b = _mm_srli_si128(row4_16x8b, 4); row3_16x8b = _mm_srli_si128(row4_16x8b, 2); - _mm_maskmoveu_si128(row1_16x8b, mask_low_32b, (char*)pu1_dst); - _mm_maskmoveu_si128(row2_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd)); - _mm_maskmoveu_si128(row3_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd2)); - _mm_maskmoveu_si128(row4_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd3)); + row1 = _mm_cvtsi128_si32(row1_16x8b); + row2 = _mm_cvtsi128_si32(row2_16x8b); + row3 = _mm_cvtsi128_si32(row3_16x8b); + row4 = _mm_cvtsi128_si32(row4_16x8b); + + *((WORD32 *)(pu1_dst)) = row1; + *((WORD32 *)(pu1_dst + dst_strd)) = row2; + *((WORD32 *)(pu1_dst + dst_strd2)) = row3; + *((WORD32 *)(pu1_dst + dst_strd3)) = row4; } /** @@ -721,14 +689,11 @@ void ih264_intra_pred_luma_4x4_mode_vert_l_ssse3(UWORD8 *pu1_src, __m128i row1_16x8b, row2_16x8b, row3_16x8b, row4_16x8b; __m128i zero_vector, const_2_8x16b; - __m128i mask_full_128b, mask_low_32b; + WORD32 row1,row2,row3,row4; UNUSED(src_strd); UNUSED(ngbr_avail); - mask_full_128b = _mm_set1_epi8(0xff); - mask_low_32b = _mm_srli_si128(mask_full_128b, 12); - pu1_top = pu1_src +BLK_SIZE + 1; val_16x8b = _mm_loadl_epi64((__m128i *)pu1_top); @@ -756,10 +721,15 @@ void ih264_intra_pred_luma_4x4_mode_vert_l_ssse3(UWORD8 *pu1_src, row3_16x8b = _mm_srli_si128(row1_16x8b, 1); row4_16x8b = _mm_srli_si128(row2_16x8b, 1); - _mm_maskmoveu_si128(row1_16x8b, mask_low_32b, (char*)pu1_dst); - _mm_maskmoveu_si128(row2_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd)); - _mm_maskmoveu_si128(row3_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd2)); - _mm_maskmoveu_si128(row4_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd3)); + row1 = _mm_cvtsi128_si32(row1_16x8b); + row2 = _mm_cvtsi128_si32(row2_16x8b); + row3 = _mm_cvtsi128_si32(row3_16x8b); + row4 = _mm_cvtsi128_si32(row4_16x8b); + + *((WORD32 *)(pu1_dst)) = row1; + *((WORD32 *)(pu1_dst + dst_strd)) = row2; + *((WORD32 *)(pu1_dst + dst_strd2)) = row3; + *((WORD32 *)(pu1_dst + dst_strd3)) = row4; } /** @@ -809,14 +779,11 @@ void ih264_intra_pred_luma_4x4_mode_horz_u_ssse3(UWORD8 *pu1_src, __m128i row1_16x8b, row2_16x8b, row3_16x8b, row4_16x8b; __m128i zero_vector, const_2_8x16b, rev_16x8b; - __m128i mask_full_128b, mask_low_32b; + WORD32 row1,row2,row3,row4; UNUSED(src_strd); UNUSED(ngbr_avail); - mask_full_128b = _mm_set1_epi8(0xff); - mask_low_32b = _mm_srli_si128(mask_full_128b, 12); - pu1_left = pu1_src + BLK_SIZE - 1; zero_vector = _mm_setzero_si128(); @@ -851,10 +818,15 @@ void ih264_intra_pred_luma_4x4_mode_horz_u_ssse3(UWORD8 *pu1_src, row3_16x8b = _mm_srli_si128(row1_16x8b, 4); row4_16x8b = _mm_srli_si128(row1_16x8b, 6); - _mm_maskmoveu_si128(row1_16x8b, mask_low_32b, (char*)pu1_dst); - _mm_maskmoveu_si128(row2_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd)); - _mm_maskmoveu_si128(row3_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd2)); - _mm_maskmoveu_si128(row4_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd3)); + row1 = _mm_cvtsi128_si32(row1_16x8b); + row2 = _mm_cvtsi128_si32(row2_16x8b); + row3 = _mm_cvtsi128_si32(row3_16x8b); + row4 = _mm_cvtsi128_si32(row4_16x8b); + + *((WORD32 *)(pu1_dst)) = row1; + *((WORD32 *)(pu1_dst + dst_strd)) = row2; + *((WORD32 *)(pu1_dst + dst_strd2)) = row3; + *((WORD32 *)(pu1_dst + dst_strd3)) = row4; } /******************* 8x8 Modes *******************/ @@ -1814,9 +1786,7 @@ void ih264_intra_pred_luma_16x16_mode_horz_ssse3(UWORD8 *pu1_src, { UWORD8 *pu1_left; WORD32 dst_strd2, dst_strd3, dst_strd4; - WORD32 val1, val2; - __m128i val_16x8b; __m128i row1_16x8b, row2_16x8b, row3_16x8b, row4_16x8b; UNUSED(src_strd); @@ -1826,60 +1796,46 @@ void ih264_intra_pred_luma_16x16_mode_horz_ssse3(UWORD8 *pu1_src, dst_strd4 = dst_strd << 2; - val_16x8b = _mm_loadu_si128((__m128i *)(pu1_left - 15)); - dst_strd2 = dst_strd << 1; dst_strd3 = dst_strd4 - dst_strd; - val1 = _mm_extract_epi16(val_16x8b, 7); - val2 = _mm_extract_epi16(val_16x8b, 6); - - row1_16x8b = _mm_set1_epi8(val1 >> 8); - row2_16x8b = _mm_set1_epi8(val1 & 0xff); - row3_16x8b = _mm_set1_epi8(val2 >> 8); - row4_16x8b = _mm_set1_epi8(val2 & 0xff); + row1_16x8b = _mm_set1_epi8(*(pu1_left)); + row2_16x8b = _mm_set1_epi8(*(pu1_left - 1)); + row3_16x8b = _mm_set1_epi8(*(pu1_left - 2)); + row4_16x8b = _mm_set1_epi8(*(pu1_left - 3)); _mm_storeu_si128((__m128i *)pu1_dst, row1_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd), row2_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd2), row3_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd3), row4_16x8b); - val1 = _mm_extract_epi16(val_16x8b, 5); - val2 = _mm_extract_epi16(val_16x8b, 4); - pu1_dst += dst_strd4; - row1_16x8b = _mm_set1_epi8(val1 >> 8); - row2_16x8b = _mm_set1_epi8(val1 & 0xff); - row3_16x8b = _mm_set1_epi8(val2 >> 8); - row4_16x8b = _mm_set1_epi8(val2 & 0xff); + row1_16x8b = _mm_set1_epi8(*(pu1_left - 4)); + row2_16x8b = _mm_set1_epi8(*(pu1_left - 5)); + row3_16x8b = _mm_set1_epi8(*(pu1_left - 6)); + row4_16x8b = _mm_set1_epi8(*(pu1_left - 7)); _mm_storeu_si128((__m128i *)pu1_dst, row1_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd), row2_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd2), row3_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd3), row4_16x8b); - val1 = _mm_extract_epi16(val_16x8b, 3); - val2 = _mm_extract_epi16(val_16x8b, 2); - pu1_dst += dst_strd4; - row1_16x8b = _mm_set1_epi8(val1 >> 8); - row2_16x8b = _mm_set1_epi8(val1 & 0xff); - row3_16x8b = _mm_set1_epi8(val2 >> 8); - row4_16x8b = _mm_set1_epi8(val2 & 0xff); + row1_16x8b = _mm_set1_epi8(*(pu1_left - 8)); + row2_16x8b = _mm_set1_epi8(*(pu1_left - 9)); + row3_16x8b = _mm_set1_epi8(*(pu1_left - 10)); + row4_16x8b = _mm_set1_epi8(*(pu1_left - 11)); _mm_storeu_si128((__m128i *)pu1_dst, row1_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd), row2_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd2), row3_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd3), row4_16x8b); - val1 = _mm_extract_epi16(val_16x8b, 1); - val2 = _mm_extract_epi16(val_16x8b, 0); - pu1_dst += dst_strd4; - row1_16x8b = _mm_set1_epi8(val1 >> 8); - row2_16x8b = _mm_set1_epi8(val1 & 0xff); - row3_16x8b = _mm_set1_epi8(val2 >> 8); - row4_16x8b = _mm_set1_epi8(val2 & 0xff); + row1_16x8b = _mm_set1_epi8(*(pu1_left - 12)); + row2_16x8b = _mm_set1_epi8(*(pu1_left - 13)); + row3_16x8b = _mm_set1_epi8(*(pu1_left - 14)); + row4_16x8b = _mm_set1_epi8(*(pu1_left - 15)); _mm_storeu_si128((__m128i *)pu1_dst, row1_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd), row2_16x8b); |