/****************************************************************************** * * Copyright (C) 2015 The Android Open Source Project * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. * ***************************************************************************** * Originally developed and contributed by Ittiam Systems Pvt. Ltd, Bangalore */ /** ******************************************************************************* * @file * ih264e_intra_modes_eval_ssse3.c * * @brief * This file contains definitions of routines that perform rate distortion * analysis on a macroblock if they are to be coded as intra. * * @author * Ittiam * * @par List of Functions: * ih264e_evaluate_intra16x16_modes_ssse3 * ih264e_evaluate_intra_4x4_modes_ssse3 * ih264e_evaluate_intra_chroma_modes_ssse3 * * @remarks * None * ******************************************************************************* */ /*****************************************************************************/ /* File Includes */ /*****************************************************************************/ /* System include files */ #include #include #include #include #include /* User include files */ #include "ih264e_config.h" #include "ih264_typedefs.h" #include "ih264e_defs.h" #include "iv2.h" #include "ive2.h" #include "ih264_debug.h" #include "ih264_defs.h" #include "ih264_macros.h" #include "ih264_intra_pred_filters.h" #include "ih264_structs.h" #include "ih264_common_tables.h" #include "ih264_trans_quant_itrans_iquant.h" #include "ih264_inter_pred_filters.h" #include "ih264_mem_fns.h" #include "ih264_padding.h" #include "ih264_deblk_edge_filters.h" #include "ime_distortion_metrics.h" #include "ih264e_error.h" #include "ih264e_bitstream.h" #include "ime_defs.h" #include "ime_structs.h" #include "ih264_cabac_tables.h" #include "irc_cntrl_param.h" #include "irc_frame_info_collector.h" #include "ih264e_rate_control.h" #include "ih264e_cabac_structs.h" #include "ih264e_structs.h" #include "ih264e_cabac.h" #include "ih264e_intra_modes_eval.h" #include "ih264e_globals.h" #include "ime_platform_macros.h" /*****************************************************************************/ /* Function Definitions */ /*****************************************************************************/ /** ****************************************************************************** * * @brief * evaluate best intra 16x16 mode (among VERT, HORZ and DC) and do the * prediction. * * @par Description * This function evaluates first three 16x16 modes and compute corresponding * SAD and returns the buffer predicted with best mode. * * @param[in] pu1_src * UWORD8 pointer to the source * * @param[in] pu1_ngbr_pels_i16 * UWORD8 pointer to neighbouring pels * * @param[out] pu1_dst * UWORD8 pointer to the destination * * @param[in] src_strd * integer source stride * * @param[in] dst_strd * integer destination stride * * @param[in] u4_n_avblty * availability of neighbouring pixels * * @param[in] u4_intra_mode * pointer to the variable in which best mode is returned * * @param[in] pu4_sadmin * pointer to the variable in which minimum sad is returned * * @param[in] u4_valid_intra_modes * says what all modes are valid * * @return * None * ****************************************************************************** */ void ih264e_evaluate_intra16x16_modes_ssse3(UWORD8 *pu1_src, UWORD8 *pu1_ngbr_pels_i16, UWORD8 *pu1_dst, UWORD32 src_strd, UWORD32 dst_strd, WORD32 n_avblty, UWORD32 *u4_intra_mode, WORD32 *pu4_sadmin, UWORD32 u4_valid_intra_modes) { UWORD8 *pu1_src_temp; WORD32 left, top, horz_flag, vert_flag, dc_flag; WORD32 sad_vert, sad_horz, sad_dc, min_sad; WORD32 cnt, dcval; WORD32 src_strd2, src_strd3, src_strd4; WORD32 dst_strd2, dst_strd3, dst_strd4; __m128i src1_16x8b, src2_16x8b, src3_16x8b, src4_16x8b; __m128i val1_16x8b, val2_16x8b, val3_16x8b, val4_16x8b; __m128i sad1_8x16b, sad2_8x16b, sad3_8x16b, sad4_8x16b; __m128i sad_8x16b, val_16x8b, zero_vector; sad_vert = INT_MAX; sad_horz = INT_MAX; sad_dc = INT_MAX; src_strd2 = src_strd << 1; src_strd4 = src_strd << 2; src_strd3 = src_strd + src_strd2; dst_strd2 = dst_strd << 1; dst_strd4 = dst_strd << 2; dst_strd3 = dst_strd + dst_strd2; left = (n_avblty & LEFT_MB_AVAILABLE_MASK); top = (n_avblty & TOP_MB_AVAILABLE_MASK) >> 2; zero_vector = _mm_setzero_si128(); horz_flag = left && ((u4_valid_intra_modes & 02) != 0); vert_flag = top && ((u4_valid_intra_modes & 01) != 0); dc_flag = (u4_valid_intra_modes & 04) != 0; if(horz_flag) { pu1_src_temp = pu1_src; val1_16x8b = _mm_set1_epi8(pu1_ngbr_pels_i16[15]); val2_16x8b = _mm_set1_epi8(pu1_ngbr_pels_i16[14]); val3_16x8b = _mm_set1_epi8(pu1_ngbr_pels_i16[13]); val4_16x8b = _mm_set1_epi8(pu1_ngbr_pels_i16[12]); src1_16x8b = _mm_loadu_si128((__m128i *)pu1_src_temp); src2_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd)); src3_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd2)); src4_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd3)); sad1_8x16b = _mm_sad_epu8(val1_16x8b, src1_16x8b); sad2_8x16b = _mm_sad_epu8(val2_16x8b, src2_16x8b); sad3_8x16b = _mm_sad_epu8(val3_16x8b, src3_16x8b); sad4_8x16b = _mm_sad_epu8(val4_16x8b, src4_16x8b); sad1_8x16b = _mm_packs_epi32(sad1_8x16b, sad2_8x16b); sad3_8x16b = _mm_packs_epi32(sad3_8x16b, sad4_8x16b); cnt = 11; sad_8x16b = _mm_packs_epi32(sad1_8x16b, sad3_8x16b); do { pu1_src_temp += src_strd4; val1_16x8b = _mm_set1_epi8(pu1_ngbr_pels_i16[cnt]); val2_16x8b = _mm_set1_epi8(pu1_ngbr_pels_i16[cnt - 1]); val3_16x8b = _mm_set1_epi8(pu1_ngbr_pels_i16[cnt - 2]); val4_16x8b = _mm_set1_epi8(pu1_ngbr_pels_i16[cnt - 3]); src1_16x8b = _mm_loadu_si128((__m128i *)pu1_src_temp); src2_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd)); src3_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd2)); src4_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd3)); sad1_8x16b = _mm_sad_epu8(val1_16x8b, src1_16x8b); sad2_8x16b = _mm_sad_epu8(val2_16x8b, src2_16x8b); sad3_8x16b = _mm_sad_epu8(val3_16x8b, src3_16x8b); sad4_8x16b = _mm_sad_epu8(val4_16x8b, src4_16x8b); sad1_8x16b = _mm_packs_epi32(sad1_8x16b, sad2_8x16b); sad3_8x16b = _mm_packs_epi32(sad3_8x16b, sad4_8x16b); sad1_8x16b = _mm_packs_epi32(sad1_8x16b, sad3_8x16b); cnt -= 4; sad_8x16b = _mm_add_epi16(sad_8x16b, sad1_8x16b); } while(cnt >= 0); sad_8x16b = _mm_hadd_epi16(sad_8x16b, sad_8x16b); sad_8x16b = _mm_hadd_epi16(sad_8x16b, sad_8x16b); sad_8x16b = _mm_hadd_epi16(sad_8x16b, sad_8x16b); sad_horz = _mm_extract_epi16(sad_8x16b, 0); } if(vert_flag) { pu1_src_temp = pu1_src; val1_16x8b = _mm_loadu_si128((__m128i *)(pu1_ngbr_pels_i16 + 17)); src1_16x8b = _mm_loadu_si128((__m128i *)pu1_src_temp); src2_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd)); src3_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd2)); src4_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd3)); sad1_8x16b = _mm_sad_epu8(val1_16x8b, src1_16x8b); sad2_8x16b = _mm_sad_epu8(val1_16x8b, src2_16x8b); sad3_8x16b = _mm_sad_epu8(val1_16x8b, src3_16x8b); sad4_8x16b = _mm_sad_epu8(val1_16x8b, src4_16x8b); sad1_8x16b = _mm_packs_epi32(sad1_8x16b, sad2_8x16b); sad3_8x16b = _mm_packs_epi32(sad3_8x16b, sad4_8x16b); cnt = 11; sad_8x16b = _mm_packs_epi32(sad1_8x16b, sad3_8x16b); do { pu1_src_temp += src_strd4; src1_16x8b = _mm_loadu_si128((__m128i *)pu1_src_temp); src2_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd)); src3_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd2)); src4_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd3)); sad1_8x16b = _mm_sad_epu8(val1_16x8b, src1_16x8b); sad2_8x16b = _mm_sad_epu8(val1_16x8b, src2_16x8b); sad3_8x16b = _mm_sad_epu8(val1_16x8b, src3_16x8b); sad4_8x16b = _mm_sad_epu8(val1_16x8b, src4_16x8b); sad1_8x16b = _mm_packs_epi32(sad1_8x16b, sad2_8x16b); sad3_8x16b = _mm_packs_epi32(sad3_8x16b, sad4_8x16b); sad1_8x16b = _mm_packs_epi32(sad1_8x16b, sad3_8x16b); cnt -= 4; sad_8x16b = _mm_add_epi16(sad_8x16b, sad1_8x16b); } while(cnt >= 0); sad_8x16b = _mm_hadd_epi16(sad_8x16b, sad_8x16b); sad_8x16b = _mm_hadd_epi16(sad_8x16b, sad_8x16b); sad_8x16b = _mm_hadd_epi16(sad_8x16b, sad_8x16b); sad_vert = _mm_extract_epi16(sad_8x16b, 0); } dcval = 0; if(left) { val_16x8b = _mm_loadu_si128((__m128i *)pu1_ngbr_pels_i16); dcval += 8; sad1_8x16b = _mm_sad_epu8(val_16x8b, zero_vector); dcval += _mm_extract_epi16(sad1_8x16b, 0); dcval += _mm_extract_epi16(sad1_8x16b, 4); } if(top) { val_16x8b = _mm_loadu_si128((__m128i *)(pu1_ngbr_pels_i16 + 17)); dcval += 8; sad1_8x16b = _mm_sad_epu8(val_16x8b, zero_vector); dcval += _mm_extract_epi16(sad1_8x16b, 0); dcval += _mm_extract_epi16(sad1_8x16b, 4); } dcval = dcval >> (3 + left + top); dcval += ((left == 0) & (top == 0)) << 7; if(dc_flag) { pu1_src_temp = pu1_src; val1_16x8b = _mm_set1_epi8(dcval); src1_16x8b = _mm_loadu_si128((__m128i *)pu1_src_temp); src2_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd)); src3_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd2)); src4_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd3)); sad1_8x16b = _mm_sad_epu8(val1_16x8b, src1_16x8b); sad2_8x16b = _mm_sad_epu8(val1_16x8b, src2_16x8b); sad3_8x16b = _mm_sad_epu8(val1_16x8b, src3_16x8b); sad4_8x16b = _mm_sad_epu8(val1_16x8b, src4_16x8b); sad1_8x16b = _mm_packs_epi32(sad1_8x16b, sad2_8x16b); sad3_8x16b = _mm_packs_epi32(sad3_8x16b, sad4_8x16b); cnt = 12; sad_8x16b = _mm_packs_epi32(sad1_8x16b, sad3_8x16b); do { pu1_src_temp += src_strd4; src1_16x8b = _mm_loadu_si128((__m128i *)pu1_src_temp); src2_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd)); src3_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd2)); src4_16x8b = _mm_loadu_si128((__m128i *)(pu1_src_temp + src_strd3)); sad1_8x16b = _mm_sad_epu8(val1_16x8b, src1_16x8b); sad2_8x16b = _mm_sad_epu8(val1_16x8b, src2_16x8b); sad3_8x16b = _mm_sad_epu8(val1_16x8b, src3_16x8b); sad4_8x16b = _mm_sad_epu8(val1_16x8b, src4_16x8b); sad1_8x16b = _mm_packs_epi32(sad1_8x16b, sad2_8x16b); sad3_8x16b = _mm_packs_epi32(sad3_8x16b, sad4_8x16b); sad1_8x16b = _mm_packs_epi32(sad1_8x16b, sad3_8x16b); cnt -= 4; sad_8x16b = _mm_add_epi16(sad_8x16b, sad1_8x16b); } while(cnt > 0); sad_8x16b = _mm_hadd_epi16(sad_8x16b, sad_8x16b); sad_8x16b = _mm_hadd_epi16(sad_8x16b, sad_8x16b); sad_8x16b = _mm_hadd_epi16(sad_8x16b, sad_8x16b); sad_dc = _mm_extract_epi16(sad_8x16b, 0); } // Doing prediction for minimum SAD min_sad = MIN3(sad_horz, sad_vert, sad_dc); if(min_sad < *pu4_sadmin) { *pu4_sadmin = min_sad; if(min_sad == sad_vert) { *u4_intra_mode = VERT_I16x16; val1_16x8b = _mm_loadu_si128((__m128i *)(pu1_ngbr_pels_i16 + 17)); cnt = 15; do { _mm_storeu_si128((__m128i *)pu1_dst, val1_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd), val1_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd2), val1_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd3), val1_16x8b); cnt -= 4; pu1_dst += dst_strd4; } while(cnt > 0); } else if(min_sad == sad_horz) { *u4_intra_mode = HORZ_I16x16; cnt = 15; do { val1_16x8b = _mm_set1_epi8(pu1_ngbr_pels_i16[cnt]); val2_16x8b = _mm_set1_epi8(pu1_ngbr_pels_i16[cnt - 1]); val3_16x8b = _mm_set1_epi8(pu1_ngbr_pels_i16[cnt - 2]); val4_16x8b = _mm_set1_epi8(pu1_ngbr_pels_i16[cnt - 3]); _mm_storeu_si128((__m128i *)pu1_dst, val1_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd), val2_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd2), val3_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd3), val4_16x8b); cnt -= 4; pu1_dst += dst_strd4; } while(cnt >= 0); } else { *u4_intra_mode = DC_I16x16; val1_16x8b = _mm_set1_epi8(dcval); cnt = 15; do { _mm_storeu_si128((__m128i *)pu1_dst, val1_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd), val1_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd2), val1_16x8b); _mm_storeu_si128((__m128i *)(pu1_dst + dst_strd3), val1_16x8b); cnt -= 4; pu1_dst += dst_strd4; } while(cnt > 0); } } } /** ****************************************************************************** * * @brief :Evaluate best intra 4x4 mode and do the prediction. * * @par Description * This function evaluates intra 4x4 modes, computes corresponding sad * and returns the buffer predicted with best mode. * * @param[in] pu1_src * UWORD8 pointer to the source * ** @param[in] pu1_ngbr_pels * UWORD8 pointer to neighbouring pels * * @param[out] pu1_dst * UWORD8 pointer to the destination * * @param[in] src_strd * integer source stride * * @param[in] dst_strd * integer destination stride * * @param[in] u4_n_avblty * availability of neighbouring pixels * * @param[in] u4_intra_mode * Pointer to the variable in which best mode is returned * * @param[in] pu4_sadmin * Pointer to the variable in which minimum cost is returned * * @param[in] u4_valid_intra_modes * Says what all modes are valid * * * @param[in] u4_lambda * Lamda value for computing cost from SAD * * @param[in] u4_predictd_mode * Predicted mode for cost computation * * @return none * ****************************************************************************** */ void ih264e_evaluate_intra_4x4_modes_ssse3(UWORD8 *pu1_src, UWORD8 *pu1_ngbr_pels, UWORD8 *pu1_dst, UWORD32 src_strd, UWORD32 dst_strd, WORD32 u4_n_avblty, UWORD32 *u4_intra_mode, WORD32 *pu4_sadmin, UWORD32 u4_valid_intra_modes, UWORD32 u4_lambda, UWORD32 u4_predictd_mode) { WORD32 left, top; WORD32 sad[MAX_I4x4] = { INT_MAX, INT_MAX, INT_MAX, INT_MAX, INT_MAX, INT_MAX, INT_MAX, INT_MAX, INT_MAX }; WORD32 cost[MAX_I4x4] = { INT_MAX, INT_MAX, INT_MAX, INT_MAX, INT_MAX, INT_MAX, INT_MAX, INT_MAX, INT_MAX }; WORD32 min_cost; UWORD32 lambda4 = u4_lambda << 2; WORD32 dst_strd2, dst_strd3; __m128i left_top_16x8b, src_16x8b, pred0_16x8b, sad_8x16b; __m128i pred1_16x8b, pred2_16x8b, pred3_16x8b, pred4_16x8b; __m128i pred5_16x8b, pred6_16x8b, pred7_16x8b, pred8_16x8b; __m128i shuffle_16x8b, zero_vector, mask_low_32b; left = (u4_n_avblty & LEFT_MB_AVAILABLE_MASK); top = (u4_n_avblty & TOP_MB_AVAILABLE_MASK) >> 2; dst_strd2 = dst_strd << 1; dst_strd3 = dst_strd + dst_strd2; // loading the 4x4 source block and neighbouring pixels { __m128i row1_16x8b, row2_16x8b; row1_16x8b = _mm_loadl_epi64((__m128i *)pu1_src); row2_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + src_strd)); left_top_16x8b = _mm_loadu_si128((__m128i *)pu1_ngbr_pels); pu1_src += src_strd << 1; src_16x8b = _mm_unpacklo_epi32(row1_16x8b, row2_16x8b); row1_16x8b = _mm_loadl_epi64((__m128i *)pu1_src); row2_16x8b = _mm_loadl_epi64((__m128i *)(pu1_src + src_strd)); zero_vector = _mm_setzero_si128(); row1_16x8b = _mm_unpacklo_epi32(row1_16x8b, row2_16x8b); src_16x8b = _mm_unpacklo_epi64(src_16x8b, row1_16x8b); } /* Computing SADs*/ if(u4_valid_intra_modes & 1)/* VERT mode valid ????*/ { pred0_16x8b = _mm_srli_si128(left_top_16x8b, 5); pred0_16x8b = _mm_shuffle_epi32(pred0_16x8b, 0); sad_8x16b = _mm_sad_epu8(src_16x8b, pred0_16x8b); sad[VERT_I4x4] = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); cost[VERT_I4x4] = sad[VERT_I4x4] + ((u4_predictd_mode == VERT_I4x4) ? u4_lambda: lambda4); } if(u4_valid_intra_modes & 2)/* HORZ mode valid ????*/ { shuffle_16x8b = _mm_setr_epi8(3, 3, 3, 3, 2, 2, 2, 2, 1, 1, 1, 1, 0, 0, 0, 0); pred1_16x8b = _mm_shuffle_epi8(left_top_16x8b, shuffle_16x8b); sad_8x16b = _mm_sad_epu8(src_16x8b, pred1_16x8b); sad[HORZ_I4x4] = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); cost[HORZ_I4x4] = sad[HORZ_I4x4] + ((u4_predictd_mode == HORZ_I4x4) ? u4_lambda: lambda4); } if(u4_valid_intra_modes & 4)/* DC mode valid ????*/ { if(top + left) { WORD32 shft = 1, dcval = 0; __m128i val_16x8b, temp_16x8b, temp_8x16b; val_16x8b = _mm_setzero_si128(); if(top) { temp_16x8b = _mm_srli_si128(left_top_16x8b, 5); val_16x8b = _mm_alignr_epi8(temp_16x8b, val_16x8b, 4); shft ++; dcval += 2; } if(left) { val_16x8b = _mm_alignr_epi8(left_top_16x8b, val_16x8b, 4); shft++; dcval += 2; } temp_8x16b = _mm_sad_epu8(val_16x8b, zero_vector); dcval += _mm_extract_epi16(temp_8x16b, 4); dcval = dcval >> shft; pred2_16x8b = _mm_set1_epi8(dcval); } else pred2_16x8b = _mm_set1_epi8(128); sad_8x16b = _mm_sad_epu8(src_16x8b, pred2_16x8b); sad[DC_I4x4] = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); cost[DC_I4x4] = sad[DC_I4x4] + ((u4_predictd_mode == DC_I4x4) ? u4_lambda: lambda4); } if(u4_valid_intra_modes > 7)/* if modes other than VERT, HORZ and DC are valid ????*/ { __m128i w11_16x8b, w121_16x8b; __m128i temp1_16x8b, temp2_16x8b; /* Performing FILT121 and FILT11 operation for all neighbour values*/ { __m128i temp1_8x16b, temp2_8x16b, temp3_8x16b; __m128i const_2_8x16b; const_2_8x16b = _mm_set1_epi16(2); temp1_8x16b = _mm_unpacklo_epi8(left_top_16x8b, zero_vector); //l3 l2 l1 l0 tl t0 t1 t2 temp2_8x16b = _mm_slli_si128(temp1_8x16b, 2); // 0 l3 l2 l1 l0 tl t0 t1 temp2_8x16b = _mm_shufflelo_epi16(temp2_8x16b, 0xe5); //l3 l3 l2 l1 l0 tl t0 t1 temp1_8x16b = _mm_add_epi16(temp1_8x16b, temp2_8x16b); //l3+l3 l3+l2 l2+l1... t1+t2 temp2_8x16b = _mm_slli_si128(temp1_8x16b, 2); //l3+l3 l3+l3 l3+l2... t0+t1 temp2_8x16b = _mm_shufflelo_epi16(temp2_8x16b, 0xe5); temp1_8x16b = _mm_add_epi16(temp1_8x16b, temp2_8x16b); //4*l3 l3+2*l3+l2 l3+2*l2+l1... t0+2*t1+t2 temp1_8x16b = _mm_add_epi16(const_2_8x16b, temp1_8x16b); //4*l3+2 3*l3+l2+2 l3+2*l2+l1+2.. t0+2*t1+t2+2 temp1_8x16b = _mm_srli_epi16(temp1_8x16b, 2); temp1_16x8b = _mm_srli_si128(left_top_16x8b, 1); w11_16x8b = _mm_avg_epu8(left_top_16x8b, temp1_16x8b); temp2_16x8b = _mm_srli_si128(left_top_16x8b, 6); temp2_8x16b = _mm_unpacklo_epi8(temp2_16x8b, zero_vector); //t1 t2 t3 t4 t5 t6 t7 0 temp3_8x16b = _mm_srli_si128(temp2_8x16b, 2); //t2 t3 t4 t5 t6 t7 0 0 temp3_8x16b = _mm_shufflehi_epi16(temp3_8x16b, 0xd4); //t2 t3 t4 t5 t6 t7 t7 0 temp2_8x16b = _mm_add_epi16(temp2_8x16b, temp3_8x16b); //t1+t2 t2+t3... t6+t7 t7+t7 0 temp3_8x16b = _mm_srli_si128(temp2_8x16b, 2); //t2+t3 t3+t4... t7+t7 0 0 temp2_8x16b = _mm_add_epi16(temp2_8x16b, temp3_8x16b); //t1+2*t2+t3 t2+2*t3+t4.. t6+2*t7+t7 t7+t7 0 temp2_8x16b = _mm_add_epi16(const_2_8x16b, temp2_8x16b); //t1+2*t2+t3+2 t2+2*t3+t4+2 t3+2*t4+t5+2... t6+2*t7+t7+2 t7+t7+2 2 temp2_8x16b = _mm_srli_epi16(temp2_8x16b, 2); w121_16x8b = _mm_packus_epi16(temp1_8x16b, temp2_8x16b); } if(u4_valid_intra_modes & 8)/* DIAG_DL */ { shuffle_16x8b = _mm_setr_epi8( 7, 8, 9, 10, 8, 9, 10, 11, 9, 10, 11, 12, 10, 11, 12, 13); pred3_16x8b = _mm_shuffle_epi8(w121_16x8b, shuffle_16x8b); sad_8x16b = _mm_sad_epu8(src_16x8b, pred3_16x8b); sad[DIAG_DL_I4x4] = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); cost[DIAG_DL_I4x4] = sad[DIAG_DL_I4x4] + ((u4_predictd_mode == DIAG_DL_I4x4) ? u4_lambda: lambda4); } if(u4_valid_intra_modes & 16)/* DIAG_DR */ { shuffle_16x8b = _mm_setr_epi8(5, 6, 7, 8, 4, 5, 6, 7, 3, 4, 5, 6, 2, 3, 4, 5); pred4_16x8b = _mm_shuffle_epi8(w121_16x8b, shuffle_16x8b); sad_8x16b = _mm_sad_epu8(src_16x8b, pred4_16x8b); sad[DIAG_DR_I4x4] = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); cost[DIAG_DR_I4x4] = sad[DIAG_DR_I4x4] + ((u4_predictd_mode == DIAG_DR_I4x4) ? u4_lambda: lambda4); } if(u4_valid_intra_modes & 32)/* VERT_R mode valid ????*/ { temp1_16x8b = _mm_srli_si128(w121_16x8b, 1); temp1_16x8b = _mm_unpacklo_epi64(temp1_16x8b, w11_16x8b); shuffle_16x8b = _mm_setr_epi8(12, 13, 14, 15, 4, 5, 6, 7, 3, 12, 13, 14, 2, 4, 5, 6); pred5_16x8b = _mm_shuffle_epi8(temp1_16x8b, shuffle_16x8b); sad_8x16b = _mm_sad_epu8(src_16x8b, pred5_16x8b); sad[VERT_R_I4x4] = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); cost[VERT_R_I4x4] = sad[VERT_R_I4x4] + ((u4_predictd_mode == VERT_R_I4x4) ? u4_lambda: lambda4); } if(u4_valid_intra_modes & 64)/* HORZ_D mode valid ????*/ { temp1_16x8b = _mm_unpacklo_epi64(w121_16x8b, w11_16x8b); shuffle_16x8b = _mm_setr_epi8(11, 5, 6, 7, 10, 4, 11, 5, 9, 3, 10, 4, 8, 2, 9, 3); pred6_16x8b = _mm_shuffle_epi8(temp1_16x8b, shuffle_16x8b); sad_8x16b = _mm_sad_epu8(src_16x8b, pred6_16x8b); sad[HORZ_D_I4x4] = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); cost[HORZ_D_I4x4] = sad[HORZ_D_I4x4] + ((u4_predictd_mode == HORZ_D_I4x4) ? u4_lambda: lambda4); } if(u4_valid_intra_modes & 128)/* VERT_L mode valid ????*/ { temp1_16x8b = _mm_srli_si128(w121_16x8b, 5); temp2_16x8b = _mm_srli_si128(w11_16x8b, 5); temp1_16x8b = _mm_unpacklo_epi64(temp1_16x8b, temp2_16x8b); shuffle_16x8b = _mm_setr_epi8(8, 9, 10, 11, 2, 3, 4, 5, 9, 10, 11, 12, 3, 4, 5, 6); pred7_16x8b = _mm_shuffle_epi8(temp1_16x8b, shuffle_16x8b); sad_8x16b = _mm_sad_epu8(src_16x8b, pred7_16x8b); sad[VERT_L_I4x4] = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); cost[VERT_L_I4x4] = sad[VERT_L_I4x4] + ((u4_predictd_mode == VERT_L_I4x4) ? u4_lambda: lambda4); } if(u4_valid_intra_modes & 256)/* HORZ_U mode valid ????*/ { temp1_16x8b = _mm_unpacklo_epi64(w121_16x8b, w11_16x8b); shuffle_16x8b = _mm_setr_epi8(10, 3, 9, 2, 9, 2, 8, 1, 8, 1, 0, 0, 0, 0, 0, 0); pred8_16x8b = _mm_shuffle_epi8(temp1_16x8b, shuffle_16x8b); sad_8x16b = _mm_sad_epu8(src_16x8b, pred8_16x8b); sad[HORZ_U_I4x4] = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); cost[HORZ_U_I4x4] = sad[HORZ_U_I4x4] + ((u4_predictd_mode == HORZ_U_I4x4) ? u4_lambda: lambda4); } min_cost = MIN3(MIN3(cost[0], cost[1], cost[2]), MIN3(cost[3], cost[4], cost[5]), MIN3(cost[6], cost[7], cost[8])); } else { /*Only first three modes valid*/ min_cost = MIN3(cost[0], cost[1], cost[2]); } *pu4_sadmin = min_cost; if(min_cost == cost[0]) { *u4_intra_mode = VERT_I4x4; } else if(min_cost == cost[1]) { *u4_intra_mode = HORZ_I4x4; pred0_16x8b = pred1_16x8b; } else if(min_cost == cost[2]) { *u4_intra_mode = DC_I4x4; pred0_16x8b = pred2_16x8b; } else if(min_cost == cost[3]) { *u4_intra_mode = DIAG_DL_I4x4; pred0_16x8b = pred3_16x8b; } else if(min_cost == cost[4]) { *u4_intra_mode = DIAG_DR_I4x4; pred0_16x8b = pred4_16x8b; } else if(min_cost == cost[5]) { *u4_intra_mode = VERT_R_I4x4; pred0_16x8b = pred5_16x8b; } else if(min_cost == cost[6]) { *u4_intra_mode = HORZ_D_I4x4; pred0_16x8b = pred6_16x8b; } else if(min_cost == cost[7]) { *u4_intra_mode = VERT_L_I4x4; pred0_16x8b = pred7_16x8b; } else if(min_cost == cost[8]) { *u4_intra_mode = HORZ_U_I4x4; pred0_16x8b = pred8_16x8b; } mask_low_32b = _mm_set1_epi8(0xff); mask_low_32b = _mm_srli_si128(mask_low_32b, 12); _mm_maskmoveu_si128(pred0_16x8b, mask_low_32b, (char*)pu1_dst); pred0_16x8b = _mm_srli_si128(pred0_16x8b, 4); _mm_maskmoveu_si128(pred0_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd)); pred0_16x8b = _mm_srli_si128(pred0_16x8b, 4); _mm_maskmoveu_si128(pred0_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd2)); pred0_16x8b = _mm_srli_si128(pred0_16x8b, 4); _mm_maskmoveu_si128(pred0_16x8b, mask_low_32b, (char*)(pu1_dst + dst_strd3)); } /** ****************************************************************************** * * @brief * Evaluate best intra chroma mode (among VERT, HORZ and DC) and do the prediction. * * @par Description * This function evaluates first three intra chroma modes and compute corresponding sad * and return the buffer predicted with best mode. * * @param[in] pu1_src * UWORD8 pointer to the source * ** @param[in] pu1_ngbr_pels * UWORD8 pointer to neighbouring pels * * @param[out] pu1_dst * UWORD8 pointer to the destination * * @param[in] src_strd * integer source stride * * @param[in] dst_strd * integer destination stride * * @param[in] u4_n_avblty * availability of neighbouring pixels * * @param[in] u4_intra_mode * pointer to the variable in which best mode is returned * * @param[in] pu4_sadmin * pointer to the variable in which minimum sad is returned * * @param[in] u4_valid_intra_modes * says what all modes are valid * * @return * none * ****************************************************************************** */ void ih264e_evaluate_intra_chroma_modes_ssse3(UWORD8 *pu1_src, UWORD8 *pu1_ngbr_pels, UWORD8 *pu1_dst, UWORD32 src_strd, UWORD32 dst_strd, WORD32 u4_n_avblty, UWORD32 *u4_intra_mode, WORD32 *pu4_sadmin, UWORD32 u4_valid_intra_modes) { WORD32 left, top; WORD32 sad_vert = INT_MAX, sad_horz = INT_MAX, sad_dc = INT_MAX, min_sad; __m128i src1_16x8b, src2_16x8b, src3_16x8b, src4_16x8b; __m128i src5_16x8b, src6_16x8b, src7_16x8b, src8_16x8b; __m128i top_16x8b, left_16x8b; __m128i pred1_16x8b, pred2_16x8b; __m128i tmp1_8x16b, tmp2_8x16b, sad_8x16b; left = (u4_n_avblty & LEFT_MB_AVAILABLE_MASK); top = (u4_n_avblty & TOP_MB_AVAILABLE_MASK) >> 2; //Loading source { src1_16x8b = _mm_loadu_si128((__m128i *)pu1_src); pu1_src += src_strd; src2_16x8b = _mm_loadu_si128((__m128i *)pu1_src); pu1_src += src_strd; src3_16x8b = _mm_loadu_si128((__m128i *)pu1_src); pu1_src += src_strd; src4_16x8b = _mm_loadu_si128((__m128i *)pu1_src); pu1_src += src_strd; src5_16x8b = _mm_loadu_si128((__m128i *)pu1_src); pu1_src += src_strd; src6_16x8b = _mm_loadu_si128((__m128i *)pu1_src); pu1_src += src_strd; src7_16x8b = _mm_loadu_si128((__m128i *)pu1_src); pu1_src += src_strd; src8_16x8b = _mm_loadu_si128((__m128i *)pu1_src); } if(left) { left_16x8b = _mm_loadu_si128((__m128i *)pu1_ngbr_pels); if(u4_valid_intra_modes & 02) //If HORZ mode is valid { __m128i left_tmp_16x8b, left_sh_16x8b; __m128i const_14_15_16x8b; const_14_15_16x8b = _mm_set1_epi16(0x0f0e); left_sh_16x8b = _mm_slli_si128(left_16x8b, 2); pred1_16x8b = _mm_shuffle_epi8(left_16x8b, const_14_15_16x8b); //row 1 pred2_16x8b = _mm_shuffle_epi8(left_sh_16x8b, const_14_15_16x8b); //row 2 tmp1_8x16b = _mm_sad_epu8(src1_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src2_16x8b, pred2_16x8b); left_tmp_16x8b = _mm_slli_si128(left_16x8b, 4); left_sh_16x8b = _mm_slli_si128(left_sh_16x8b, 4); sad_8x16b = _mm_add_epi16(tmp1_8x16b, tmp2_8x16b); pred1_16x8b = _mm_shuffle_epi8(left_tmp_16x8b, const_14_15_16x8b); //row 3 pred2_16x8b = _mm_shuffle_epi8(left_sh_16x8b, const_14_15_16x8b); //row 4 tmp1_8x16b = _mm_sad_epu8(src3_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src4_16x8b, pred2_16x8b); left_tmp_16x8b = _mm_slli_si128(left_tmp_16x8b, 4); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); left_sh_16x8b = _mm_slli_si128(left_sh_16x8b, 4); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); pred1_16x8b = _mm_shuffle_epi8(left_tmp_16x8b, const_14_15_16x8b); //row 5 pred2_16x8b = _mm_shuffle_epi8(left_sh_16x8b, const_14_15_16x8b); //row 6 tmp1_8x16b = _mm_sad_epu8(src5_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src6_16x8b, pred2_16x8b); left_tmp_16x8b = _mm_slli_si128(left_tmp_16x8b, 4); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); left_sh_16x8b = _mm_slli_si128(left_sh_16x8b, 4); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); pred1_16x8b = _mm_shuffle_epi8(left_tmp_16x8b, const_14_15_16x8b); //row 7 pred2_16x8b = _mm_shuffle_epi8(left_sh_16x8b, const_14_15_16x8b); //row 8 tmp1_8x16b = _mm_sad_epu8(src7_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src8_16x8b, pred2_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); sad_horz = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); } } if(top) { UWORD8 *pu1_top; pu1_top = pu1_ngbr_pels + 2 * BLK8x8SIZE + 2; top_16x8b = _mm_loadu_si128((__m128i *)pu1_top); if(u4_valid_intra_modes & 04) //If VERT mode is valid { tmp1_8x16b = _mm_sad_epu8(src1_16x8b, top_16x8b); tmp2_8x16b = _mm_sad_epu8(src2_16x8b, top_16x8b); sad_8x16b = _mm_add_epi16(tmp1_8x16b, tmp2_8x16b); tmp1_8x16b = _mm_sad_epu8(src3_16x8b, top_16x8b); tmp2_8x16b = _mm_sad_epu8(src4_16x8b, top_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); tmp1_8x16b = _mm_sad_epu8(src5_16x8b, top_16x8b); tmp2_8x16b = _mm_sad_epu8(src6_16x8b, top_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); tmp1_8x16b = _mm_sad_epu8(src7_16x8b, top_16x8b); tmp2_8x16b = _mm_sad_epu8(src8_16x8b, top_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); sad_vert = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); } } if(u4_valid_intra_modes & 01) //If DC mode is valid { if(left && top) { WORD32 left_up_u, left_down_u, left_up_v, left_down_v; WORD32 top_left_u, top_right_u, top_left_v, top_right_v; WORD32 dc_1u, dc_1v, dc_2u, dc_2v; __m128i val_sh_16x8b; __m128i intrlv_mask_8x16b, zero_vector; intrlv_mask_8x16b = _mm_set1_epi16(0x00ff); zero_vector = _mm_setzero_si128(); val_sh_16x8b = _mm_srli_si128(left_16x8b, 1); tmp1_8x16b = _mm_and_si128(intrlv_mask_8x16b, left_16x8b); tmp2_8x16b = _mm_and_si128(intrlv_mask_8x16b, val_sh_16x8b); tmp1_8x16b = _mm_sad_epu8(zero_vector, tmp1_8x16b); tmp2_8x16b = _mm_sad_epu8(zero_vector, tmp2_8x16b); left_up_u = _mm_extract_epi16(tmp1_8x16b, 4); left_up_v = _mm_extract_epi16(tmp2_8x16b, 4); left_down_u = _mm_extract_epi16(tmp1_8x16b, 0); left_down_v = _mm_extract_epi16(tmp2_8x16b, 0); val_sh_16x8b = _mm_srli_si128(top_16x8b, 1); tmp1_8x16b = _mm_and_si128(intrlv_mask_8x16b, top_16x8b); tmp2_8x16b = _mm_and_si128(intrlv_mask_8x16b, val_sh_16x8b); tmp1_8x16b = _mm_sad_epu8(zero_vector, tmp1_8x16b); tmp2_8x16b = _mm_sad_epu8(zero_vector, tmp2_8x16b); top_left_u = _mm_extract_epi16(tmp1_8x16b, 0); top_left_v = _mm_extract_epi16(tmp2_8x16b, 0); top_right_u = _mm_extract_epi16(tmp1_8x16b, 4); top_right_v = _mm_extract_epi16(tmp2_8x16b, 4); // First four rows dc_1u = (left_up_u + top_left_u + 4) >> 3; dc_1v = (left_up_v + top_left_v + 4) >> 3; dc_2u = (top_right_u + 2) >> 2; dc_2v = (top_right_v + 2) >> 2; pred1_16x8b = _mm_setr_epi8(dc_1u, dc_1v, dc_1u, dc_1v, dc_1u, dc_1v, dc_1u, dc_1v, dc_2u, dc_2v, dc_2u, dc_2v, dc_2u, dc_2v, dc_2u, dc_2v); tmp1_8x16b = _mm_sad_epu8(src1_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src2_16x8b, pred1_16x8b); sad_8x16b = _mm_add_epi16(tmp1_8x16b, tmp2_8x16b); tmp1_8x16b = _mm_sad_epu8(src3_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src4_16x8b, pred1_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); // Second four rows dc_1u = (left_down_u + 2) >> 2; dc_1v = (left_down_v + 2) >> 2; dc_2u = (left_down_u + top_right_u + 4) >> 3; dc_2v = (left_down_v + top_right_v + 4) >> 3; pred2_16x8b = _mm_setr_epi8(dc_1u, dc_1v, dc_1u, dc_1v, dc_1u, dc_1v, dc_1u, dc_1v, dc_2u, dc_2v, dc_2u, dc_2v, dc_2u, dc_2v, dc_2u, dc_2v); tmp1_8x16b = _mm_sad_epu8(src5_16x8b, pred2_16x8b); tmp2_8x16b = _mm_sad_epu8(src6_16x8b, pred2_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); tmp1_8x16b = _mm_sad_epu8(src7_16x8b, pred2_16x8b); tmp2_8x16b = _mm_sad_epu8(src8_16x8b, pred2_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); sad_dc = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); } else if(left) { WORD32 left_up_u, left_down_u, left_up_v, left_down_v; WORD32 dc_u, dc_v; __m128i left_sh_16x8b; __m128i intrlv_mask_8x16b, zero_vector; intrlv_mask_8x16b = _mm_set1_epi16(0x00ff); zero_vector = _mm_setzero_si128(); left_sh_16x8b = _mm_srli_si128(left_16x8b, 1); tmp1_8x16b = _mm_and_si128(intrlv_mask_8x16b, left_16x8b); tmp2_8x16b = _mm_and_si128(intrlv_mask_8x16b, left_sh_16x8b); tmp1_8x16b = _mm_sad_epu8(zero_vector, tmp1_8x16b); tmp2_8x16b = _mm_sad_epu8(zero_vector, tmp2_8x16b); left_up_u = _mm_extract_epi16(tmp1_8x16b, 4); left_up_v = _mm_extract_epi16(tmp2_8x16b, 4); left_down_u = _mm_extract_epi16(tmp1_8x16b, 0); left_down_v = _mm_extract_epi16(tmp2_8x16b, 0); // First four rows dc_u = (left_up_u + 2) >> 2; dc_v = (left_up_v + 2) >> 2; pred1_16x8b = _mm_set1_epi16(dc_u | (dc_v << 8)); tmp1_8x16b = _mm_sad_epu8(src1_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src2_16x8b, pred1_16x8b); sad_8x16b = _mm_add_epi16(tmp1_8x16b, tmp2_8x16b); tmp1_8x16b = _mm_sad_epu8(src3_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src4_16x8b, pred1_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); // Second four rows dc_u = (left_down_u + 2) >> 2; dc_v = (left_down_v + 2) >> 2; pred2_16x8b = _mm_set1_epi16(dc_u | (dc_v << 8)); tmp1_8x16b = _mm_sad_epu8(src5_16x8b, pred2_16x8b); tmp2_8x16b = _mm_sad_epu8(src6_16x8b, pred2_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); tmp1_8x16b = _mm_sad_epu8(src7_16x8b, pred2_16x8b); tmp2_8x16b = _mm_sad_epu8(src8_16x8b, pred2_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); sad_dc = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); } else if(top) { WORD32 top_left_u, top_right_u, top_left_v, top_right_v; WORD32 dc_1u, dc_1v, dc_2u, dc_2v; __m128i top_sh_16x8b; __m128i intrlv_mask_8x16b, zero_vector; intrlv_mask_8x16b = _mm_set1_epi16(0x00ff); zero_vector = _mm_setzero_si128(); top_sh_16x8b = _mm_srli_si128(top_16x8b, 1); tmp1_8x16b = _mm_and_si128(intrlv_mask_8x16b, top_16x8b); tmp2_8x16b = _mm_and_si128(intrlv_mask_8x16b, top_sh_16x8b); tmp1_8x16b = _mm_sad_epu8(zero_vector, tmp1_8x16b); tmp2_8x16b = _mm_sad_epu8(zero_vector, tmp2_8x16b); top_left_u = _mm_extract_epi16(tmp1_8x16b, 0); top_left_v = _mm_extract_epi16(tmp2_8x16b, 0); top_right_u = _mm_extract_epi16(tmp1_8x16b, 4); top_right_v = _mm_extract_epi16(tmp2_8x16b, 4); dc_1u = (top_left_u + 2) >> 2; dc_1v = (top_left_v + 2) >> 2; dc_2u = (top_right_u + 2) >> 2; dc_2v = (top_right_v + 2) >> 2; pred1_16x8b = _mm_setr_epi8(dc_1u, dc_1v, dc_1u, dc_1v, dc_1u, dc_1v, dc_1u, dc_1v, dc_2u, dc_2v, dc_2u, dc_2v, dc_2u, dc_2v, dc_2u, dc_2v); tmp1_8x16b = _mm_sad_epu8(src1_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src2_16x8b, pred1_16x8b); sad_8x16b = _mm_add_epi16(tmp1_8x16b, tmp2_8x16b); tmp1_8x16b = _mm_sad_epu8(src3_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src4_16x8b, pred1_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); tmp1_8x16b = _mm_sad_epu8(src5_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src6_16x8b, pred1_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); tmp1_8x16b = _mm_sad_epu8(src7_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src8_16x8b, pred1_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); sad_dc = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); } else { pred1_16x8b = _mm_set1_epi8(128); tmp1_8x16b = _mm_sad_epu8(src1_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src2_16x8b, pred1_16x8b); sad_8x16b = _mm_add_epi16(tmp1_8x16b, tmp2_8x16b); tmp1_8x16b = _mm_sad_epu8(src3_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src4_16x8b, pred1_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); tmp1_8x16b = _mm_sad_epu8(src5_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src6_16x8b, pred1_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); tmp1_8x16b = _mm_sad_epu8(src7_16x8b, pred1_16x8b); tmp2_8x16b = _mm_sad_epu8(src8_16x8b, pred1_16x8b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp1_8x16b); sad_8x16b = _mm_add_epi16(sad_8x16b, tmp2_8x16b); sad_dc = _mm_extract_epi16(sad_8x16b, 0) + _mm_extract_epi16(sad_8x16b, 4); } } min_sad = MIN3(sad_horz, sad_vert, sad_dc); /* Finding minimum SAD and doing corresponding prediction*/ if(min_sad < *pu4_sadmin) { *pu4_sadmin = min_sad; if(min_sad == sad_dc) { *u4_intra_mode = DC_CH_I8x8; if(!left) pred2_16x8b = pred1_16x8b; _mm_storeu_si128((__m128i *)pu1_dst, pred1_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred1_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred1_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred1_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred2_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred2_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred2_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred2_16x8b); } else if(min_sad == sad_horz) { __m128i left_sh_16x8b, const_14_15_16x8b; *u4_intra_mode = HORZ_CH_I8x8; const_14_15_16x8b = _mm_set1_epi16(0x0f0e); left_sh_16x8b = _mm_slli_si128(left_16x8b, 2); pred1_16x8b = _mm_shuffle_epi8(left_16x8b, const_14_15_16x8b); //row 1 pred2_16x8b = _mm_shuffle_epi8(left_sh_16x8b, const_14_15_16x8b); //row 2 _mm_storeu_si128((__m128i *)pu1_dst, pred1_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred2_16x8b); left_16x8b = _mm_slli_si128(left_16x8b, 4); left_sh_16x8b = _mm_slli_si128(left_sh_16x8b, 4); pred1_16x8b = _mm_shuffle_epi8(left_16x8b, const_14_15_16x8b); //row 3 pred2_16x8b = _mm_shuffle_epi8(left_sh_16x8b, const_14_15_16x8b); //row 4 pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred1_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred2_16x8b); left_16x8b = _mm_slli_si128(left_16x8b, 4); left_sh_16x8b = _mm_slli_si128(left_sh_16x8b, 4); pred1_16x8b = _mm_shuffle_epi8(left_16x8b, const_14_15_16x8b); //row 5 pred2_16x8b = _mm_shuffle_epi8(left_sh_16x8b, const_14_15_16x8b); //row 6 pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred1_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred2_16x8b); left_16x8b = _mm_slli_si128(left_16x8b, 4); left_sh_16x8b = _mm_slli_si128(left_sh_16x8b, 4); pred1_16x8b = _mm_shuffle_epi8(left_16x8b, const_14_15_16x8b); //row 7 pred2_16x8b = _mm_shuffle_epi8(left_sh_16x8b, const_14_15_16x8b); //row 8 pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred1_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, pred2_16x8b); } else { *u4_intra_mode = VERT_CH_I8x8; _mm_storeu_si128((__m128i *)pu1_dst, top_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, top_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, top_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, top_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, top_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, top_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, top_16x8b); pu1_dst += dst_strd; _mm_storeu_si128((__m128i *)pu1_dst, top_16x8b); } } }