/****************************************************************************** * * Copyright (C) 2018 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 hme_coarse.c * * @brief * Contains ME algorithm for the coarse layer. * * @author * Ittiam * * * List of Functions * hme_update_mv_bank_coarse() * hme_coarse() ****************************************************************************** */ /*****************************************************************************/ /* File Includes */ /*****************************************************************************/ /* System include files */ #include #include #include #include #include #include #include /* User include files */ #include "ihevc_typedefs.h" #include "itt_video_api.h" #include "ihevce_api.h" #include "rc_cntrl_param.h" #include "rc_frame_info_collector.h" #include "rc_look_ahead_params.h" #include "ihevc_defs.h" #include "ihevc_structs.h" #include "ihevc_platform_macros.h" #include "ihevc_deblk.h" #include "ihevc_itrans_recon.h" #include "ihevc_chroma_itrans_recon.h" #include "ihevc_chroma_intra_pred.h" #include "ihevc_intra_pred.h" #include "ihevc_inter_pred.h" #include "ihevc_mem_fns.h" #include "ihevc_padding.h" #include "ihevc_weighted_pred.h" #include "ihevc_sao.h" #include "ihevc_resi_trans.h" #include "ihevc_quant_iquant_ssd.h" #include "ihevc_cabac_tables.h" #include "ihevce_defs.h" #include "ihevce_lap_enc_structs.h" #include "ihevce_multi_thrd_structs.h" #include "ihevce_multi_thrd_funcs.h" #include "ihevce_me_common_defs.h" #include "ihevce_had_satd.h" #include "ihevce_error_codes.h" #include "ihevce_bitstream.h" #include "ihevce_cabac.h" #include "ihevce_rdoq_macros.h" #include "ihevce_function_selector.h" #include "ihevce_enc_structs.h" #include "ihevce_entropy_structs.h" #include "ihevce_cmn_utils_instr_set_router.h" #include "ihevce_enc_loop_structs.h" #include "ihevce_bs_compute_ctb.h" #include "ihevce_global_tables.h" #include "ihevce_dep_mngr_interface.h" #include "hme_datatype.h" #include "hme_interface.h" #include "hme_common_defs.h" #include "hme_defs.h" #include "ihevce_me_instr_set_router.h" #include "hme_globals.h" #include "hme_utils.h" #include "hme_coarse.h" #include "hme_refine.h" #include "hme_err_compute.h" #include "hme_common_utils.h" #include "hme_search_algo.h" /******************************************************************************* * MACROS *******************************************************************************/ #define COPY_SEARCH_RESULT(ps_mv, pi1_ref_idx, ps_search_node, shift) \ { \ ps_mv->i2_mv_x = ps_search_node->s_mv.i2_mvx >> (shift); \ ps_mv->i2_mv_y = ps_search_node->s_mv.i2_mvy >> (shift); \ *pi1_ref_idx = ps_search_node->i1_ref_idx; \ } /*****************************************************************************/ /* Function Definitions */ /*****************************************************************************/ /** ******************************************************************************** * @fn void hme_update_mv_bank_coarse(search_results_t *ps_search_results, * layer_mv_t *ps_layer_mv, * S32 i4_blk_x, * S32 i4_blk_y, * search_node_t *ps_search_node_4x8_l, * search_node_t *ps_search_node_8x4_t, * S08 i1_ref_idx, * mvbank_update_prms_t *ps_prms * * @brief Updates the coarse layer MV Bank for a given ref id and blk pos * * @param[in] ps_search_results: Search results data structure * * @param[in, out] ps_layer_mv : MV Bank for this layer * * @param[in] i4_search_blk_x: column number of the 4x4 blk searched * * @param[in] i4_search_blk_y: row number of the 4x4 blk searched * * @param[in] ps_search_node_4x8_t: Best MV of the 4x8T blk * * @param[in] ps_search_node_8x4_l: Best MV of the 8x4L blk * * @param[in] i1_ref_idx : Reference ID that has been searched * * @param[in] ps_prms : Parameters pertaining to the MV Bank update * * @return None ******************************************************************************** */ void hme_update_mv_bank_coarse( search_results_t *ps_search_results, layer_mv_t *ps_layer_mv, S32 i4_search_blk_x, S32 i4_search_blk_y, search_node_t *ps_search_node_4x8_t, search_node_t *ps_search_node_8x4_l, S08 i1_ref_idx, mvbank_update_prms_t *ps_prms) { /* These point to the MV and ref idx posn to be udpated */ hme_mv_t *ps_mv; S08 *pi1_ref_idx; /* Offset within the bank */ S32 i4_offset; S32 i, j, i4_blk_x, i4_blk_y; /* Best results for 8x4R and 4x8B blocks */ search_node_t *ps_search_node_8x4_r, *ps_search_node_4x8_b; /* Number of MVs in a block */ S32 num_mvs = ps_layer_mv->i4_num_mvs_per_ref; search_node_t *aps_search_nodes[4]; /* The search blk may be different in size from the blk used to hold MV */ i4_blk_x = i4_search_blk_x << ps_prms->i4_shift; i4_blk_y = i4_search_blk_y << ps_prms->i4_shift; /* Compute the offset in the MV bank */ i4_offset = i4_blk_x + i4_blk_y * ps_layer_mv->i4_num_blks_per_row; i4_offset *= ps_layer_mv->i4_num_mvs_per_blk; /* Identify the correct offset in the mvbank and the reference id buf */ ps_mv = ps_layer_mv->ps_mv + (i4_offset + (num_mvs * i1_ref_idx)); pi1_ref_idx = ps_layer_mv->pi1_ref_idx + (i4_offset + (num_mvs * i1_ref_idx)); /*************************************************************************/ /* We have atleast 4 distinct results: the 4x8 top (coming from top blk) */ /* 8x4 left (coming from left blk), 8x4 and 4x8 right and bot resp. */ /* If number of results to be stored is 4, then we store all these 4 */ /* results, else we pick best ones */ /*************************************************************************/ ps_search_node_8x4_r = ps_search_results->aps_part_results[i1_ref_idx][PART_ID_2NxN_B]; ps_search_node_4x8_b = ps_search_results->aps_part_results[i1_ref_idx][PART_ID_Nx2N_R]; ASSERT(num_mvs <= 4); /* Doing this to sort best results */ aps_search_nodes[0] = ps_search_node_8x4_r; aps_search_nodes[1] = ps_search_node_4x8_b; aps_search_nodes[2] = ps_search_node_8x4_l; aps_search_nodes[3] = ps_search_node_4x8_t; if(num_mvs == 4) { COPY_SEARCH_RESULT(ps_mv, pi1_ref_idx, aps_search_nodes[0], 0); ps_mv++; pi1_ref_idx++; COPY_SEARCH_RESULT(ps_mv, pi1_ref_idx, aps_search_nodes[1], 0); ps_mv++; pi1_ref_idx++; COPY_SEARCH_RESULT(ps_mv, pi1_ref_idx, aps_search_nodes[2], 0); ps_mv++; pi1_ref_idx++; COPY_SEARCH_RESULT(ps_mv, pi1_ref_idx, aps_search_nodes[3], 0); ps_mv++; pi1_ref_idx++; return; } /* Run through the results, store them in best to worst order */ for(i = 0; i < num_mvs; i++) { for(j = i + 1; j < 4; j++) { if(aps_search_nodes[j]->i4_tot_cost < aps_search_nodes[i]->i4_tot_cost) { SWAP_HME(aps_search_nodes[j], aps_search_nodes[i], search_node_t *); } } COPY_SEARCH_RESULT(ps_mv, pi1_ref_idx, aps_search_nodes[i], 0); ps_mv++; pi1_ref_idx++; } } /** ******************************************************************************** * @fn void hme_coarse_frm_init(me_ctxt_t *ps_ctxt, coarse_prms_t *ps_coarse_prms) * * @brief Frame init entry point Coarse ME. * * @param[in,out] ps_ctxt: ME Handle * * @param[in] ps_coarse_prms : Coarse layer config params * * @return None ******************************************************************************** */ void hme_coarse_frm_init(coarse_me_ctxt_t *ps_ctxt, coarse_prms_t *ps_coarse_prms) { layer_ctxt_t *ps_curr_layer; S32 i4_pic_wd, i4_pic_ht; S32 num_blks_in_pic, num_blks_in_row; BLK_SIZE_T e_search_blk_size = BLK_4x4; S32 blk_size_shift = 2, blk_wd = 4, blk_ht = 4; /* Number of references to search */ S32 i4_num_ref; ps_curr_layer = ps_ctxt->ps_curr_descr->aps_layers[ps_coarse_prms->i4_layer_id]; i4_num_ref = ps_coarse_prms->i4_num_ref; i4_pic_wd = ps_curr_layer->i4_wd; i4_pic_ht = ps_curr_layer->i4_ht; /* Macro updates num_blks_in_pic and num_blks_in_row*/ GET_NUM_BLKS_IN_PIC(i4_pic_wd, i4_pic_ht, blk_size_shift, num_blks_in_row, num_blks_in_pic); /************************************************************************/ /* Initialize the mv bank that holds results of this layer. */ /************************************************************************/ hme_init_mv_bank( ps_curr_layer, BLK_4x4, i4_num_ref, ps_coarse_prms->num_results, ps_ctxt->u1_encode[ps_coarse_prms->i4_layer_id]); return; } /** ******************************************************************************** * @fn void hme_derive_worst_case_search_range(range_prms_t *ps_range, * range_prms_t *ps_pic_limit, * range_prms_t *ps_mv_limit, * S32 i4_x, * S32 i4_y, * S32 blk_wd, * S32 blk_ht) * * @brief given picture limits and blk dimensions and mv search limits, obtains * teh valid search range such that the blk stays within pic boundaries, * where picture boundaries include padded portions of picture * * @param[out] ps_range: updated with actual search range * * @param[in] ps_pic_limit : picture boundaries * * @param[in] ps_mv_limit: Search range limits for the mvs * * @param[in] i4_x : x coordinate of the blk * * @param[in] i4_y : y coordinate of the blk * * @param[in] blk_wd : blk width * * @param[in] blk_ht : blk height * * @return void ******************************************************************************** */ void hme_derive_worst_case_search_range( range_prms_t *ps_range, range_prms_t *ps_pic_limit, range_prms_t *ps_mv_limit, S32 i4_x, S32 i4_y, S32 blk_wd, S32 blk_ht) { /* Taking max x of left block, min x of current block */ ps_range->i2_max_x = MIN((ps_pic_limit->i2_max_x - (S16)blk_wd - (S16)(i4_x - 4)), ps_mv_limit->i2_max_x); ps_range->i2_min_x = MAX((ps_pic_limit->i2_min_x - (S16)i4_x), ps_mv_limit->i2_min_x); /* Taking max y of top block, min y of current block */ ps_range->i2_max_y = MIN((ps_pic_limit->i2_max_y - (S16)blk_ht - (S16)(i4_y - 4)), ps_mv_limit->i2_max_y); ps_range->i2_min_y = MAX((ps_pic_limit->i2_min_y - (S16)i4_y), ps_mv_limit->i2_min_y); } /** ******************************************************************************** * @fn void hme_combine_4x4_sads_and_compute_cost(S08 i1_ref_idx, * range_prms_t *ps_mv_range, * range_prms_t *ps_mv_limit, * hme_mv_t *ps_best_mv_4x8, * hme_mv_t *ps_best_mv_8x4, * pred_ctxt_t *ps_pred_ctxt, * PF_MV_COST_FXN pf_mv_cost_compute, * ME_QUALITY_PRESETS_T e_me_quality_preset, * S16 *pi2_sads_4x4_current, * S16 *pi2_sads_4x4_east, * S16 *pi2_sads_4x4_south, * FILE *fp_dump_sad) * * @brief Does a full search on entire srch window with a given step size in coarse layer * * @param[in] i1_ref_idx : Cur ref idx * * @param[in] ps_layer_ctxt: All info about this layer * * @param[out] ps_best_mv : type hme_mv_t contains best mv x and y * * @param[in] ps_pred_ctxt : Prediction ctxt for cost computation * * @param[in] pf_mv_cost_compute : mv cost computation function * * @return void ******************************************************************************** */ void hme_combine_4x4_sads_and_compute_cost_high_quality( S08 i1_ref_idx, range_prms_t *ps_mv_range, range_prms_t *ps_mv_limit, hme_mv_t *ps_best_mv_4x8, hme_mv_t *ps_best_mv_8x4, pred_ctxt_t *ps_pred_ctxt, PF_MV_COST_FXN pf_mv_cost_compute, S16 *pi2_sads_4x4_current, S16 *pi2_sads_4x4_east, S16 *pi2_sads_4x4_south) { /* These control number of parts and number of pts in grid to search */ S32 stepy, stepx, best_mv_y_4x8, best_mv_x_4x8, best_mv_y_8x4, best_mv_x_8x4; S32 step_shift_x, step_shift_y; S32 mvx, mvy, mv_x_offset, mv_y_offset, mv_x_range, mv_y_range; S32 min_cost_4x8 = MAX_32BIT_VAL; S32 min_cost_8x4 = MAX_32BIT_VAL; search_node_t s_search_node; s_search_node.i1_ref_idx = i1_ref_idx; stepx = stepy = HME_COARSE_STEP_SIZE_HIGH_QUALITY; /*TODO: Calculate Step shift from the #define HME_COARSE_STEP_SIZE_HIGH_QUALITY */ step_shift_x = step_shift_y = 1; mv_x_offset = (-ps_mv_limit->i2_min_x >> step_shift_x); mv_y_offset = (-ps_mv_limit->i2_min_y >> step_shift_y); mv_x_range = (-ps_mv_limit->i2_min_x + ps_mv_limit->i2_max_x) >> step_shift_x; mv_y_range = (-ps_mv_limit->i2_min_y + ps_mv_limit->i2_max_y) >> step_shift_y; /* Run 2loops to sweep over the reference area */ for(mvy = ps_mv_range->i2_min_y; mvy < ps_mv_range->i2_max_y; mvy += stepy) { for(mvx = ps_mv_range->i2_min_x; mvx < ps_mv_range->i2_max_x; mvx += stepx) { S32 sad_4x8, cost_4x8, sad_8x4, cost_8x4; S32 sad_pos = ((mvx >> step_shift_x) + mv_x_offset) + ((mvy >> step_shift_y) + mv_y_offset) * mv_x_range; /* Get SAD by adding SAD for current and neighbour S */ sad_4x8 = pi2_sads_4x4_current[sad_pos] + pi2_sads_4x4_south[sad_pos]; sad_8x4 = pi2_sads_4x4_current[sad_pos] + pi2_sads_4x4_east[sad_pos]; // fprintf(fp_dump_sad,"%d\t",sad); s_search_node.s_mv.i2_mvx = mvx; s_search_node.s_mv.i2_mvy = mvy; cost_4x8 = cost_8x4 = pf_mv_cost_compute(&s_search_node, ps_pred_ctxt, PART_ID_2Nx2N, MV_RES_FPEL); cost_4x8 += sad_4x8; cost_8x4 += sad_8x4; if(cost_4x8 < min_cost_4x8) { best_mv_x_4x8 = mvx; best_mv_y_4x8 = mvy; min_cost_4x8 = cost_4x8; } if(cost_8x4 < min_cost_8x4) { best_mv_x_8x4 = mvx; best_mv_y_8x4 = mvy; min_cost_8x4 = cost_8x4; } } } ps_best_mv_4x8->i2_mv_x = best_mv_x_4x8; ps_best_mv_4x8->i2_mv_y = best_mv_y_4x8; ps_best_mv_8x4->i2_mv_x = best_mv_x_8x4; ps_best_mv_8x4->i2_mv_y = best_mv_y_8x4; } void hme_combine_4x4_sads_and_compute_cost_high_speed( S08 i1_ref_idx, range_prms_t *ps_mv_range, range_prms_t *ps_mv_limit, hme_mv_t *ps_best_mv_4x8, hme_mv_t *ps_best_mv_8x4, pred_ctxt_t *ps_pred_ctxt, PF_MV_COST_FXN pf_mv_cost_compute, S16 *pi2_sads_4x4_current, S16 *pi2_sads_4x4_east, S16 *pi2_sads_4x4_south) { /* These control number of parts and number of pts in grid to search */ S32 stepy, stepx, best_mv_y_4x8, best_mv_x_4x8, best_mv_y_8x4, best_mv_x_8x4; S32 step_shift_x, step_shift_y; S32 mvx, mvy, mv_x_offset, mv_y_offset, mv_x_range, mv_y_range; S32 rnd, lambda, lambda_q_shift; S32 min_cost_4x8 = MAX_32BIT_VAL; S32 min_cost_8x4 = MAX_32BIT_VAL; (void)pf_mv_cost_compute; stepx = stepy = HME_COARSE_STEP_SIZE_HIGH_SPEED; /*TODO: Calculate Step shift from the #define HME_COARSE_STEP_SIZE_HIGH_SPEED */ step_shift_x = step_shift_y = 2; mv_x_offset = (-ps_mv_limit->i2_min_x >> step_shift_x); mv_y_offset = (-ps_mv_limit->i2_min_y >> step_shift_y); mv_x_range = (-ps_mv_limit->i2_min_x + ps_mv_limit->i2_max_x) >> step_shift_x; mv_y_range = (-ps_mv_limit->i2_min_y + ps_mv_limit->i2_max_y) >> step_shift_y; lambda = ps_pred_ctxt->lambda; lambda_q_shift = ps_pred_ctxt->lambda_q_shift; rnd = 1 << (lambda_q_shift - 1); ASSERT(MAX_MVX_SUPPORTED_IN_COARSE_LAYER >= ABS(ps_mv_range->i2_max_x)); ASSERT(MAX_MVY_SUPPORTED_IN_COARSE_LAYER >= ABS(ps_mv_range->i2_max_y)); /* Run 2loops to sweep over the reference area */ for(mvy = ps_mv_range->i2_min_y; mvy < ps_mv_range->i2_max_y; mvy += stepy) { for(mvx = ps_mv_range->i2_min_x; mvx < ps_mv_range->i2_max_x; mvx += stepx) { S32 sad_4x8, cost_4x8, sad_8x4, cost_8x4; S32 sad_pos = ((mvx >> step_shift_x) + mv_x_offset) + ((mvy >> step_shift_y) + mv_y_offset) * mv_x_range; /* Get SAD by adding SAD for current and neighbour S */ sad_4x8 = pi2_sads_4x4_current[sad_pos] + pi2_sads_4x4_south[sad_pos]; sad_8x4 = pi2_sads_4x4_current[sad_pos] + pi2_sads_4x4_east[sad_pos]; // fprintf(fp_dump_sad,"%d\t",sad); cost_4x8 = cost_8x4 = (2 * hme_get_range(ABS(mvx)) - 1) + (2 * hme_get_range(ABS(mvy)) - 1) + i1_ref_idx; cost_4x8 += (mvx != 0) ? 1 : 0; cost_4x8 += (mvy != 0) ? 1 : 0; cost_4x8 = (cost_4x8 * lambda + rnd) >> lambda_q_shift; cost_8x4 += (mvx != 0) ? 1 : 0; cost_8x4 += (mvy != 0) ? 1 : 0; cost_8x4 = (cost_8x4 * lambda + rnd) >> lambda_q_shift; cost_4x8 += sad_4x8; cost_8x4 += sad_8x4; if(cost_4x8 < min_cost_4x8) { best_mv_x_4x8 = mvx; best_mv_y_4x8 = mvy; min_cost_4x8 = cost_4x8; } if(cost_8x4 < min_cost_8x4) { best_mv_x_8x4 = mvx; best_mv_y_8x4 = mvy; min_cost_8x4 = cost_8x4; } } } ps_best_mv_4x8->i2_mv_x = best_mv_x_4x8; ps_best_mv_4x8->i2_mv_y = best_mv_y_4x8; ps_best_mv_8x4->i2_mv_x = best_mv_x_8x4; ps_best_mv_8x4->i2_mv_y = best_mv_y_8x4; } /** ******************************************************************************** * @fn hme_store_4x4_sads(hme_search_prms_t *ps_search_prms, * layer_ctxt_t *ps_layer_ctxt) * * @brief Does a 4x4 sad computation on a given range and stores it in memory * * @param[in] ps_search_prms : Search prms structure containing info like * blk dimensions, search range etc * * @param[in] ps_layer_ctxt: All info about this layer * * @param[in] ps_wt_inp_prms: All info about weighted input * * @param[in] e_me_quality_preset: motion estimation quality preset * * @param[in] pi2_sads_4x4: Memory to store all 4x4 SADs for given range * * @return void ******************************************************************************** */ void hme_store_4x4_sads_high_quality( hme_search_prms_t *ps_search_prms, layer_ctxt_t *ps_layer_ctxt, range_prms_t *ps_mv_limit, wgt_pred_ctxt_t *ps_wt_inp_prms, S16 *pi2_sads_4x4) { S32 sad, i, j; /* Input and reference attributes */ U08 *pu1_inp, *pu1_inp_orig, *pu1_ref; S32 i4_inp_stride, i4_ref_stride, i4_ref_offset; /* The reference is actually an array of ptrs since there are several */ /* reference id. So an array gets passed form calling function */ U08 **ppu1_ref, *pu1_ref_coloc; S32 stepy, stepx, step_shift_x, step_shift_y; S32 mvx, mvy, mv_x_offset, mv_y_offset, mv_x_range, mv_y_range; /* Points to the range limits for mv */ range_prms_t *ps_range_prms; /* Reference index to be searched */ S32 i4_search_idx = ps_search_prms->i1_ref_idx; /* Using the member 0 to store for all ref. idx. */ ps_range_prms = ps_search_prms->aps_mv_range[0]; pu1_inp_orig = ps_wt_inp_prms->apu1_wt_inp[i4_search_idx]; i4_inp_stride = ps_search_prms->i4_inp_stride; /* Move to the location of the search blk in inp buffer */ pu1_inp_orig += ps_search_prms->i4_cu_x_off; pu1_inp_orig += ps_search_prms->i4_cu_y_off * i4_inp_stride; /*************************************************************************/ /* we use either input of previously encoded pictures as reference */ /* in coarse layer */ /*************************************************************************/ i4_ref_stride = ps_layer_ctxt->i4_inp_stride; ppu1_ref = ps_layer_ctxt->ppu1_list_inp; /* colocated position in reference picture */ i4_ref_offset = (i4_ref_stride * ps_search_prms->i4_y_off) + ps_search_prms->i4_x_off; pu1_ref_coloc = ppu1_ref[i4_search_idx] + i4_ref_offset; stepx = stepy = HME_COARSE_STEP_SIZE_HIGH_QUALITY; /*TODO: Calculate Step shift from the #define HME_COARSE_STEP_SIZE_HIGH_QUALITY */ step_shift_x = step_shift_y = 1; mv_x_offset = -(ps_mv_limit->i2_min_x >> step_shift_x); mv_y_offset = -(ps_mv_limit->i2_min_y >> step_shift_y); mv_x_range = (-ps_mv_limit->i2_min_x + ps_mv_limit->i2_max_x) >> step_shift_x; mv_y_range = (-ps_mv_limit->i2_min_y + ps_mv_limit->i2_max_y) >> step_shift_y; /* Run 2loops to sweep over the reference area */ for(mvy = ps_range_prms->i2_min_y; mvy < ps_range_prms->i2_max_y; mvy += stepy) { for(mvx = ps_range_prms->i2_min_x; mvx < ps_range_prms->i2_max_x; mvx += stepx) { /* Set up the reference and inp ptr */ pu1_ref = pu1_ref_coloc + mvx + (mvy * i4_ref_stride); pu1_inp = pu1_inp_orig; /* SAD computation */ { sad = 0; for(i = 0; i < 4; i++) { for(j = 0; j < 4; j++) { sad += (ABS(((S32)pu1_inp[j] - (S32)pu1_ref[j]))); } pu1_inp += i4_inp_stride; pu1_ref += i4_ref_stride; } } pi2_sads_4x4 [((mvx >> step_shift_x) + mv_x_offset) + ((mvy >> step_shift_y) + mv_y_offset) * mv_x_range] = sad; } } } void hme_store_4x4_sads_high_speed( hme_search_prms_t *ps_search_prms, layer_ctxt_t *ps_layer_ctxt, range_prms_t *ps_mv_limit, wgt_pred_ctxt_t *ps_wt_inp_prms, S16 *pi2_sads_4x4) { S32 sad, i, j; /* Input and reference attributes */ U08 *pu1_inp, *pu1_inp_orig, *pu1_ref; S32 i4_inp_stride, i4_ref_stride, i4_ref_offset; /* The reference is actually an array of ptrs since there are several */ /* reference id. So an array gets passed form calling function */ U08 **ppu1_ref, *pu1_ref_coloc; S32 stepy, stepx, step_shift_x, step_shift_y; S32 mvx, mvy, mv_x_offset, mv_y_offset, mv_x_range, mv_y_range; /* Points to the range limits for mv */ range_prms_t *ps_range_prms; /* Reference index to be searched */ S32 i4_search_idx = ps_search_prms->i1_ref_idx; /* Using the member 0 for all ref. idx */ ps_range_prms = ps_search_prms->aps_mv_range[0]; pu1_inp_orig = ps_wt_inp_prms->apu1_wt_inp[i4_search_idx]; i4_inp_stride = ps_search_prms->i4_inp_stride; /* Move to the location of the search blk in inp buffer */ pu1_inp_orig += ps_search_prms->i4_cu_x_off; pu1_inp_orig += ps_search_prms->i4_cu_y_off * i4_inp_stride; /*************************************************************************/ /* we use either input of previously encoded pictures as reference */ /* in coarse layer */ /*************************************************************************/ i4_ref_stride = ps_layer_ctxt->i4_inp_stride; ppu1_ref = ps_layer_ctxt->ppu1_list_inp; /* colocated position in reference picture */ i4_ref_offset = (i4_ref_stride * ps_search_prms->i4_y_off) + ps_search_prms->i4_x_off; pu1_ref_coloc = ppu1_ref[i4_search_idx] + i4_ref_offset; stepx = stepy = HME_COARSE_STEP_SIZE_HIGH_SPEED; /*TODO: Calculate Step shift from the #define HME_COARSE_STEP_SIZE_HIGH_SPEED */ step_shift_x = step_shift_y = 2; mv_x_offset = -(ps_mv_limit->i2_min_x >> step_shift_x); mv_y_offset = -(ps_mv_limit->i2_min_y >> step_shift_y); mv_x_range = (-ps_mv_limit->i2_min_x + ps_mv_limit->i2_max_x) >> step_shift_x; mv_y_range = (-ps_mv_limit->i2_min_y + ps_mv_limit->i2_max_y) >> step_shift_y; /* Run 2loops to sweep over the reference area */ for(mvy = ps_range_prms->i2_min_y; mvy < ps_range_prms->i2_max_y; mvy += stepy) { for(mvx = ps_range_prms->i2_min_x; mvx < ps_range_prms->i2_max_x; mvx += stepx) { /* Set up the reference and inp ptr */ pu1_ref = pu1_ref_coloc + mvx + (mvy * i4_ref_stride); pu1_inp = pu1_inp_orig; /* SAD computation */ { sad = 0; for(i = 0; i < 4; i++) { for(j = 0; j < 4; j++) { sad += (ABS(((S32)pu1_inp[j] - (S32)pu1_ref[j]))); } pu1_inp += i4_inp_stride; pu1_ref += i4_ref_stride; } } pi2_sads_4x4 [((mvx >> step_shift_x) + mv_x_offset) + ((mvy >> step_shift_y) + mv_y_offset) * mv_x_range] = sad; } } } /** ******************************************************************************** * @fn void hme_coarsest(me_ctxt_t *ps_ctxt, coarse_prms_t *ps_coarse_prms) * * @brief Top level entry point for Coarse ME. Runs across blks and searches * at a 4x4 blk granularity by using 4x8 and 8x4 patterns. * * @param[in,out] ps_ctxt: ME Handle * * @param[in] ps_coarse_prms : Coarse layer config params * * @param[in] ps_multi_thrd_ctxt : Multi thread context * * @return None ******************************************************************************** */ void hme_coarsest( coarse_me_ctxt_t *ps_ctxt, coarse_prms_t *ps_coarse_prms, multi_thrd_ctxt_t *ps_multi_thrd_ctxt, WORD32 i4_ping_pong, void **ppv_dep_mngr_hme_sync) { S16 *pi2_cur_ref_sads_4x4; S32 ai4_sad_4x4_block_size[MAX_NUM_REF], ai4_sad_4x4_block_stride[MAX_NUM_REF]; S32 num_rows_coarse; S32 sad_top_offset, sad_current_offset; S32 search_node_top_offset, search_node_left_offset; ME_QUALITY_PRESETS_T e_me_quality_preset = ps_ctxt->s_init_prms.s_me_coding_tools.e_me_quality_presets; search_results_t *ps_search_results; mvbank_update_prms_t s_mv_update_prms; BLK_SIZE_T e_search_blk_size = BLK_4x4; hme_search_prms_t s_search_prms_4x8, s_search_prms_8x4, s_search_prms_4x4; S32 global_id_8x4, global_id_4x8; /*************************************************************************/ /* These directly point to the best search result nodes that will be */ /* updated by the search algorithm, rather than have to go through an */ /* elaborate structure */ /*************************************************************************/ search_node_t *aps_best_search_node_8x4[MAX_NUM_REF]; search_node_t *aps_best_search_node_4x8[MAX_NUM_REF]; /* These point to various spatial candts */ search_node_t *ps_candt_8x4_l, *ps_candt_8x4_t, *ps_candt_8x4_tl; search_node_t *ps_candt_4x8_l, *ps_candt_4x8_t, *ps_candt_4x8_tl; search_node_t *ps_candt_zeromv_8x4, *ps_candt_zeromv_4x8; search_node_t *ps_candt_fs_8x4, *ps_candt_fs_4x8; search_node_t as_top_neighbours[4], as_left_neighbours[3]; /* Holds the global mv for a given ref index */ search_node_t s_candt_global[MAX_NUM_REF]; /* All the search candidates */ search_candt_t as_search_candts_8x4[MAX_INIT_CANDTS]; search_candt_t as_search_candts_4x8[MAX_INIT_CANDTS]; search_candt_t *ps_search_candts_8x4, *ps_search_candts_4x8; /* Actual range per blk and the pic level boundaries */ range_prms_t s_range_prms, s_pic_limit, as_mv_limit[MAX_NUM_REF]; /* Current and prev pic layer ctxt at the coarsest layer */ layer_ctxt_t *ps_curr_layer, *ps_prev_layer; /* best mv of full search */ hme_mv_t best_mv_4x8, best_mv_8x4; /* Book keeping at blk level */ S32 blk_x, num_blks_in_pic, num_blks_in_row, num_4x4_blks_in_row; S32 blk_y; /* Block dimensions */ S32 blk_size_shift = 2, blk_wd = 4, blk_ht = 4; S32 lambda = ps_coarse_prms->lambda; /* Number of references to search */ S32 i4_num_ref; S32 i4_i, id, i; S08 i1_ref_idx; S32 i4_pic_wd, i4_pic_ht; S32 i4_layer_id; S32 end_of_frame; pf_get_wt_inp fp_get_wt_inp; /* Maximum search iterations around any candidate */ S32 i4_max_iters = ps_coarse_prms->i4_max_iters; ps_curr_layer = ps_ctxt->ps_curr_descr->aps_layers[ps_coarse_prms->i4_layer_id]; ps_prev_layer = hme_coarse_get_past_layer_ctxt(ps_ctxt, ps_coarse_prms->i4_layer_id); /* We need only one instance of search results structure */ ps_search_results = &ps_ctxt->s_search_results_8x8; ps_search_candts_8x4 = &as_search_candts_8x4[0]; ps_search_candts_4x8 = &as_search_candts_4x8[0]; end_of_frame = 0; i4_pic_wd = ps_curr_layer->i4_wd; i4_pic_ht = ps_curr_layer->i4_ht; fp_get_wt_inp = ((ihevce_me_optimised_function_list_t *)ps_ctxt->pv_me_optimised_function_list) ->pf_get_wt_inp_8x8; num_rows_coarse = ps_ctxt->i4_num_row_bufs; /*************************************************************************/ /* Coarse Layer always does explicit search. Number of reference frames */ /* to search is a configurable parameter supplied by the application */ /*************************************************************************/ i4_num_ref = ps_coarse_prms->i4_num_ref; i4_layer_id = ps_coarse_prms->i4_layer_id; /*************************************************************************/ /* The search algorithm goes as follows: */ /* */ /* ___ */ /* | e | */ /* ___|___|___ */ /* | c | a | b | */ /* |___|___|___| */ /* | d | */ /* |___| */ /* */ /* For the target block a, we collect best results from 2 8x4 blks */ /* These are c-a and a-b. The 4x8 blks are e-a and a-d */ /* c-a result is already available from results of blk c. a-b is */ /* evaluated in this blk. Likewise e-a result is stored in a row buffer */ /* a-d is evaluated this blk */ /* So we store a row buffer which stores best 4x8 results of all top blk */ /*************************************************************************/ /************************************************************************/ /* Initialize the pointers to the best node. */ /************************************************************************/ for(i4_i = 0; i4_i < i4_num_ref; i4_i++) { aps_best_search_node_8x4[i4_i] = ps_search_results->aps_part_results[i4_i][PART_ID_2NxN_B]; aps_best_search_node_4x8[i4_i] = ps_search_results->aps_part_results[i4_i][PART_ID_Nx2N_R]; } /************************************************************************/ /* Initialize the "searchresults" structure. This will set up the number*/ /* of search types, result updates etc */ /************************************************************************/ { S32 num_results_per_part; /* We evaluate 4 types of results per 4x4 blk. 8x4L and 8x4R and */ /* 4x8 T and 4x8B. So if we are to give 4 results, then we need to */ /* only evaluate 1 result per part. In the coarse layer, we are */ /* limited to 2 results max per part, and max of 8 results. */ num_results_per_part = (ps_coarse_prms->num_results + 3) >> 2; hme_init_search_results( ps_search_results, i4_num_ref, ps_coarse_prms->num_results, num_results_per_part, BLK_8x8, 0, 0, ps_ctxt->au1_is_past); } /* Macro updates num_blks_in_pic and num_blks_in_row*/ GET_NUM_BLKS_IN_PIC(i4_pic_wd, i4_pic_ht, blk_size_shift, num_blks_in_row, num_blks_in_pic); num_4x4_blks_in_row = num_blks_in_row + 1; s_mv_update_prms.e_search_blk_size = e_search_blk_size; s_mv_update_prms.i4_num_ref = i4_num_ref; s_mv_update_prms.i4_shift = 0; /* For full search, support 2 or 4 step size */ if(ps_coarse_prms->do_full_search) { ASSERT((ps_coarse_prms->full_search_step == 2) || (ps_coarse_prms->full_search_step == 4)); } for(i4_i = 0; i4_i < i4_num_ref; i4_i++) { S32 blk, delta_poc; S32 mv_x_clip, mv_y_clip; /* Initialize only the first row */ for(blk = 0; blk < num_blks_in_row; blk++) { INIT_SEARCH_NODE(&ps_ctxt->aps_best_search_nodes_4x8_n_rows[i4_i][blk], i4_i); } delta_poc = ABS(ps_curr_layer->i4_poc - ps_curr_layer->ai4_ref_id_to_poc_lc[i4_i]); /* Setting search range for different references based on the delta poc */ /*************************************************************************/ /* set the MV limit per ref. pic. */ /* - P pic. : Based on the config params. */ /* - B/b pic: Based on the Max/Min MV from prev. P and config. param. */ /*************************************************************************/ { /* TO DO : Remove hard coding of P-P dist. of 4 */ mv_x_clip = (ps_curr_layer->i2_max_mv_x * delta_poc) / 4; /* Only for B/b pic. */ if(1 == ps_ctxt->s_frm_prms.bidir_enabled) { WORD16 i2_mv_y_per_poc; /* Get abs MAX for symmetric search */ i2_mv_y_per_poc = MAX(ps_ctxt->s_coarse_dyn_range_prms.i2_dyn_max_y_per_poc[i4_layer_id], (ABS(ps_ctxt->s_coarse_dyn_range_prms.i2_dyn_min_y_per_poc[i4_layer_id]))); mv_y_clip = i2_mv_y_per_poc * delta_poc; } /* Set the Config. File Params for P pic. */ else { /* TO DO : Remove hard coding of P-P dist. of 4 */ mv_y_clip = (ps_curr_layer->i2_max_mv_y * delta_poc) / 4; } /* Making mv_x and mv_y range multiple of 4 */ mv_x_clip = (((mv_x_clip + 3) >> 2) << 2); mv_y_clip = (((mv_y_clip + 3) >> 2) << 2); /* Clipping the range of mv_x and mv_y */ mv_x_clip = CLIP3(mv_x_clip, 4, MAX_MVX_SUPPORTED_IN_COARSE_LAYER); mv_y_clip = CLIP3(mv_y_clip, 4, MAX_MVY_SUPPORTED_IN_COARSE_LAYER); as_mv_limit[i4_i].i2_min_x = -mv_x_clip; as_mv_limit[i4_i].i2_min_y = -mv_y_clip; as_mv_limit[i4_i].i2_max_x = mv_x_clip; as_mv_limit[i4_i].i2_max_y = mv_y_clip; } /*Populating SAD block size based on search range */ ai4_sad_4x4_block_size[i4_i] = ((2 * mv_x_clip) / ps_coarse_prms->full_search_step) * ((2 * mv_y_clip) / ps_coarse_prms->full_search_step); ai4_sad_4x4_block_stride[i4_i] = (num_blks_in_row + 1) * ai4_sad_4x4_block_size[i4_i]; } for(i = 0; i < 2 * MAX_INIT_CANDTS; i++) { search_node_t *ps_search_node; ps_search_node = &ps_ctxt->s_init_search_node[i]; INIT_SEARCH_NODE(ps_search_node, 0); } for(i = 0; i < 3; i++) { search_node_t *ps_search_node; ps_search_node = &as_left_neighbours[i]; INIT_SEARCH_NODE(ps_search_node, 0); ps_search_node = &as_top_neighbours[i]; INIT_SEARCH_NODE(ps_search_node, 0); } INIT_SEARCH_NODE(&as_top_neighbours[3], 0); /* Set up place holders to hold the search nodes of each initial candt */ for(i = 0; i < MAX_INIT_CANDTS; i++) { ps_search_candts_8x4[i].ps_search_node = &ps_ctxt->s_init_search_node[i]; ps_search_candts_4x8[i].ps_search_node = &ps_ctxt->s_init_search_node[MAX_INIT_CANDTS + i]; ps_search_candts_8x4[i].u1_num_steps_refine = (U08)i4_max_iters; ps_search_candts_4x8[i].u1_num_steps_refine = (U08)i4_max_iters; } /* For Top,TopLeft and Left cand., no need for refinement */ id = 0; if((ps_coarse_prms->do_full_search) && (ME_XTREME_SPEED_25 == e_me_quality_preset)) { /* This search candt has the full search result */ ps_candt_fs_8x4 = ps_search_candts_8x4[id].ps_search_node; id++; } ps_candt_8x4_l = ps_search_candts_8x4[id].ps_search_node; ps_search_candts_8x4[id].u1_num_steps_refine = 0; id++; ps_candt_8x4_t = ps_search_candts_8x4[id].ps_search_node; ps_search_candts_8x4[id].u1_num_steps_refine = 0; id++; ps_candt_8x4_tl = ps_search_candts_8x4[id].ps_search_node; ps_search_candts_8x4[id].u1_num_steps_refine = 0; id++; /* This search candt stores the global candt */ global_id_8x4 = id; id++; if((ps_coarse_prms->do_full_search) && (ME_XTREME_SPEED_25 != e_me_quality_preset)) { /* This search candt has the full search result */ ps_candt_fs_8x4 = ps_search_candts_8x4[id].ps_search_node; id++; } /* Don't increment id as (0,0) is removed from cand. list. Initializing */ /* the pointer for hme_init_pred_ctxt_no_encode() */ ps_candt_zeromv_8x4 = ps_search_candts_8x4[id].ps_search_node; /* For Top,TopLeft and Left cand., no need for refinement */ id = 0; if((ps_coarse_prms->do_full_search) && (ME_XTREME_SPEED_25 == e_me_quality_preset)) { /* This search candt has the full search result */ ps_candt_fs_4x8 = ps_search_candts_4x8[id].ps_search_node; id++; } ps_candt_4x8_l = ps_search_candts_4x8[id].ps_search_node; ps_search_candts_4x8[id].u1_num_steps_refine = 0; id++; ps_candt_4x8_t = ps_search_candts_4x8[id].ps_search_node; ps_search_candts_4x8[id].u1_num_steps_refine = 0; id++; ps_candt_4x8_tl = ps_search_candts_4x8[id].ps_search_node; ps_search_candts_4x8[id].u1_num_steps_refine = 0; id++; /* This search candt stores the global candt */ global_id_4x8 = id; id++; if((ps_coarse_prms->do_full_search) && (ME_XTREME_SPEED_25 != e_me_quality_preset)) { /* This search candt has the full search result */ ps_candt_fs_4x8 = ps_search_candts_4x8[id].ps_search_node; id++; } /* Don't increment id4as (0,0) is removed from cand. list. Initializing */ /* the pointer for hme_init_pred_ctxt_no_encode() */ ps_candt_zeromv_4x8 = ps_search_candts_4x8[id].ps_search_node; /* Zero mv always has 0 mvx and y componnent, ref idx initialized inside */ ps_candt_zeromv_8x4->s_mv.i2_mvx = 0; ps_candt_zeromv_8x4->s_mv.i2_mvy = 0; ps_candt_zeromv_4x8->s_mv.i2_mvx = 0; ps_candt_zeromv_4x8->s_mv.i2_mvy = 0; /* SET UP THE PRED CTXT FOR L0 AND L1 */ { S32 pred_lx; /* Bottom left always not available */ as_left_neighbours[2].u1_is_avail = 0; for(pred_lx = 0; pred_lx < 2; pred_lx++) { pred_ctxt_t *ps_pred_ctxt; ps_pred_ctxt = &ps_search_results->as_pred_ctxt[pred_lx]; hme_init_pred_ctxt_no_encode( ps_pred_ctxt, ps_search_results, as_top_neighbours, as_left_neighbours, NULL, ps_candt_zeromv_8x4, ps_candt_zeromv_8x4, pred_lx, lambda, ps_coarse_prms->lambda_q_shift, ps_ctxt->apu1_ref_bits_tlu_lc, ps_ctxt->ai2_ref_scf); } } /*************************************************************************/ /* Initialize the search parameters for search algo with the following */ /* parameters: No SATD, calculated number of initial candidates, */ /* No post refinement, initial step size and number of iterations as */ /* passed by the calling function. */ /* Also, we use input for this layer search, and not recon. */ /*************************************************************************/ if(e_me_quality_preset == ME_XTREME_SPEED_25) s_search_prms_8x4.i4_num_init_candts = 1; else s_search_prms_8x4.i4_num_init_candts = id; s_search_prms_8x4.i4_use_satd = 0; s_search_prms_8x4.i4_start_step = ps_coarse_prms->i4_start_step; s_search_prms_8x4.i4_num_steps_post_refine = 0; s_search_prms_8x4.i4_use_rec = 0; s_search_prms_8x4.ps_search_candts = ps_search_candts_8x4; s_search_prms_8x4.e_blk_size = BLK_8x4; s_search_prms_8x4.i4_max_iters = ps_coarse_prms->i4_max_iters; /* Coarse layer is always explicit */ if(ME_MEDIUM_SPEED > e_me_quality_preset) { s_search_prms_8x4.pf_mv_cost_compute = compute_mv_cost_coarse; } else { s_search_prms_8x4.pf_mv_cost_compute = compute_mv_cost_coarse_high_speed; } s_search_prms_8x4.i4_inp_stride = 8; s_search_prms_8x4.i4_cu_x_off = s_search_prms_8x4.i4_cu_y_off = 0; if(ps_coarse_prms->do_full_search) s_search_prms_8x4.i4_max_iters = 1; s_search_prms_8x4.i4_part_mask = (1 << PART_ID_2NxN_B); /* Using the member 0 to store for all ref. idx. */ s_search_prms_8x4.aps_mv_range[0] = &s_range_prms; s_search_prms_8x4.ps_search_results = ps_search_results; s_search_prms_8x4.full_search_step = ps_coarse_prms->full_search_step; s_search_prms_4x8 = s_search_prms_8x4; s_search_prms_4x8.ps_search_candts = ps_search_candts_4x8; s_search_prms_4x8.e_blk_size = BLK_4x8; s_search_prms_4x8.i4_part_mask = (1 << PART_ID_Nx2N_R); s_search_prms_4x4 = s_search_prms_8x4; /* Since s_search_prms_4x4 is used only to computer sad at 4x4 level, search candidate is not used */ s_search_prms_4x4.ps_search_candts = ps_search_candts_4x8; s_search_prms_4x4.e_blk_size = BLK_4x4; s_search_prms_4x4.i4_part_mask = (1 << PART_ID_2Nx2N); /*************************************************************************/ /* Picture limit on all 4 sides. This will be used to set mv limits for */ /* every block given its coordinate. */ /*************************************************************************/ SET_PIC_LIMIT( s_pic_limit, ps_curr_layer->i4_pad_x_inp, ps_curr_layer->i4_pad_y_inp, ps_curr_layer->i4_wd, ps_curr_layer->i4_ht, s_search_prms_4x4.i4_num_steps_post_refine); /* Pick the global mv from previous reference */ for(i1_ref_idx = 0; i1_ref_idx < i4_num_ref; i1_ref_idx++) { if(ME_XTREME_SPEED_25 != e_me_quality_preset) { /* Distance of current pic from reference */ S32 i4_delta_poc; hme_mv_t s_mv; i4_delta_poc = ps_curr_layer->i4_poc - ps_curr_layer->ai4_ref_id_to_poc_lc[i1_ref_idx]; hme_get_global_mv(ps_prev_layer, &s_mv, i4_delta_poc); s_candt_global[i1_ref_idx].s_mv.i2_mvx = s_mv.i2_mv_x; s_candt_global[i1_ref_idx].s_mv.i2_mvy = s_mv.i2_mv_y; s_candt_global[i1_ref_idx].i1_ref_idx = i1_ref_idx; /*********************************************************************/ /* Initialize the histogram for each reference index in current */ /* layer ctxt */ /*********************************************************************/ hme_init_histogram( ps_ctxt->aps_mv_hist[i1_ref_idx], (S32)as_mv_limit[i1_ref_idx].i2_max_x, (S32)as_mv_limit[i1_ref_idx].i2_max_y); } /*********************************************************************/ /* Initialize the dyn. search range params. for each reference index */ /* in current layer ctxt */ /*********************************************************************/ /* Only for P pic. For P, both are 0, I&B has them mut. exclusive */ if(ps_ctxt->s_frm_prms.is_i_pic == ps_ctxt->s_frm_prms.bidir_enabled) { INIT_DYN_SEARCH_PRMS( &ps_ctxt->s_coarse_dyn_range_prms.as_dyn_range_prms[i4_layer_id][i1_ref_idx], ps_curr_layer->ai4_ref_id_to_poc_lc[i1_ref_idx]); } } /*************************************************************************/ /* if exhaustive algorithmm then we use only 1 candt 0, 0 */ /* else we use a lot of causal and non causal candts */ /* finally set number to the configured number of candts */ /*************************************************************************/ /* Loop in raster order over each 4x4 blk in a given row till end of frame */ while(0 == end_of_frame) { job_queue_t *ps_job; void *pv_hme_dep_mngr; WORD32 offset_val, check_dep_pos, set_dep_pos; /* Get the current layer HME Dep Mngr */ /* Note : Use layer_id - 1 in HME layers */ pv_hme_dep_mngr = ppv_dep_mngr_hme_sync[ps_coarse_prms->i4_layer_id - 1]; /* Get the current row from the job queue */ ps_job = (job_queue_t *)ihevce_pre_enc_grp_get_next_job( ps_multi_thrd_ctxt, ps_multi_thrd_ctxt->i4_me_coarsest_lyr_type, 1, i4_ping_pong); /* If all rows are done, set the end of process flag to 1, */ /* and the current row to -1 */ if(NULL == ps_job) { blk_y = -1; end_of_frame = 1; } else { ASSERT(ps_multi_thrd_ctxt->i4_me_coarsest_lyr_type == ps_job->i4_pre_enc_task_type); /* Obtain the current row's details from the job */ blk_y = ps_job->s_job_info.s_me_job_info.i4_vert_unit_row_no; if(1 == ps_ctxt->s_frm_prms.is_i_pic) { /* set the output dependency of current row */ ihevce_pre_enc_grp_job_set_out_dep(ps_multi_thrd_ctxt, ps_job, i4_ping_pong); continue; } /* Set Variables for Dep. Checking and Setting */ set_dep_pos = blk_y + 1; if(blk_y > 0) { offset_val = 2; check_dep_pos = blk_y - 1; } else { /* First row should run without waiting */ offset_val = -1; check_dep_pos = 0; } /* Loop over all the blocks in current row */ /* One block extra, since the last block in a row needs East block */ for(blk_x = 0; blk_x < (num_blks_in_row + 1); blk_x++) { /* Wait till top row block is processed */ /* Currently checking till top right block*/ if(blk_x < (num_blks_in_row)) { ihevce_dmgr_chk_row_row_sync( pv_hme_dep_mngr, blk_x, offset_val, check_dep_pos, 0, /* Col Tile No. : Not supported in PreEnc*/ ps_ctxt->thrd_id); } /***************************************************************/ /* Get Weighted input for all references */ /***************************************************************/ fp_get_wt_inp( ps_curr_layer, &ps_ctxt->s_wt_pred, 1 << (blk_size_shift + 1), blk_x << blk_size_shift, (blk_y - 1) << blk_size_shift, 1 << (blk_size_shift + 1), i4_num_ref, ps_ctxt->i4_wt_pred_enable_flag); /* RESET ALL SEARCH RESULTS FOR THE NEW BLK */ hme_reset_search_results( ps_search_results, s_search_prms_8x4.i4_part_mask | s_search_prms_4x8.i4_part_mask, MV_RES_FPEL); /* Compute the search node offsets */ /* MAX is used to clip when left and top neighbours are not availbale at coarse boundaries */ search_node_top_offset = blk_x + ps_ctxt->ai4_row_index[MAX((blk_y - 2), 0)] * num_blks_in_row; search_node_left_offset = MAX((blk_x - 1), 0) + ps_ctxt->ai4_row_index[MAX((blk_y - 1), 0)] * num_blks_in_row; /* Input offset: wrt CU start. Offset for South block */ s_search_prms_4x4.i4_cu_x_off = 0; s_search_prms_4x4.i4_cu_y_off = 4; s_search_prms_4x4.i4_inp_stride = 8; s_search_prms_4x4.i4_x_off = blk_x << blk_size_shift; s_search_prms_4x4.i4_y_off = blk_y << blk_size_shift; s_search_prms_4x8.i4_x_off = s_search_prms_8x4.i4_x_off = blk_x << blk_size_shift; s_search_prms_4x8.i4_y_off = s_search_prms_8x4.i4_y_off = (blk_y - 1) << blk_size_shift; /* This layer will always use explicit ME */ /* Loop across different Ref IDx */ for(i1_ref_idx = 0; i1_ref_idx < i4_num_ref; i1_ref_idx++) { sad_top_offset = (blk_x * ai4_sad_4x4_block_size[i1_ref_idx]) + ps_ctxt->ai4_row_index[MAX((blk_y - 1), 0)] * ai4_sad_4x4_block_stride[i1_ref_idx]; sad_current_offset = (blk_x * ai4_sad_4x4_block_size[i1_ref_idx]) + ps_ctxt->ai4_row_index[blk_y] * ai4_sad_4x4_block_stride[i1_ref_idx]; /* Initialize search node if blk_x == 0, as it doesn't have left neighbours */ if(0 == blk_x) INIT_SEARCH_NODE( &ps_ctxt->aps_best_search_nodes_8x4_n_rows[i1_ref_idx][blk_x], i1_ref_idx); pi2_cur_ref_sads_4x4 = ps_ctxt->api2_sads_4x4_n_rows[i1_ref_idx]; /* Initialize changing params here */ s_search_prms_8x4.i1_ref_idx = i1_ref_idx; s_search_prms_4x8.i1_ref_idx = i1_ref_idx; s_search_prms_4x4.i1_ref_idx = i1_ref_idx; if(num_blks_in_row == blk_x) { S16 *pi2_sads_4x4_current; /* Since the current 4x4 block will be a padded region, which may not match with any of the reference */ pi2_sads_4x4_current = pi2_cur_ref_sads_4x4 + sad_current_offset; memset(pi2_sads_4x4_current, 0, ai4_sad_4x4_block_size[i1_ref_idx]); } /* SAD to be computed and stored for the 4x4 block in 1st row and the last block of all rows*/ if((0 == blk_y) || (num_blks_in_row == blk_x)) { S16 *pi2_sads_4x4_current; /* Computer 4x4 SADs for current block */ /* Pointer to store SADs */ pi2_sads_4x4_current = pi2_cur_ref_sads_4x4 + sad_current_offset; hme_derive_worst_case_search_range( &s_range_prms, &s_pic_limit, &as_mv_limit[i1_ref_idx], blk_x << blk_size_shift, blk_y << blk_size_shift, blk_wd, blk_ht); if(ME_PRISTINE_QUALITY >= e_me_quality_preset) { ((ihevce_me_optimised_function_list_t *) ps_ctxt->pv_me_optimised_function_list) ->pf_store_4x4_sads_high_quality( &s_search_prms_4x4, ps_curr_layer, &as_mv_limit[i1_ref_idx], &ps_ctxt->s_wt_pred, pi2_sads_4x4_current); } else { ((ihevce_me_optimised_function_list_t *) ps_ctxt->pv_me_optimised_function_list) ->pf_store_4x4_sads_high_speed( &s_search_prms_4x4, ps_curr_layer, &as_mv_limit[i1_ref_idx], &ps_ctxt->s_wt_pred, pi2_sads_4x4_current); } } else { /* For the zero mv candt, the ref idx to be modified */ ps_candt_zeromv_8x4->i1_ref_idx = i1_ref_idx; ps_candt_zeromv_4x8->i1_ref_idx = i1_ref_idx; if(ME_XTREME_SPEED_25 != e_me_quality_preset) { /* For the global mvs alone, the search node points to a local variable */ ps_search_candts_8x4[global_id_8x4].ps_search_node = &s_candt_global[i1_ref_idx]; ps_search_candts_4x8[global_id_4x8].ps_search_node = &s_candt_global[i1_ref_idx]; } hme_get_spatial_candt( ps_curr_layer, BLK_4x4, blk_x, blk_y - 1, i1_ref_idx, as_top_neighbours, as_left_neighbours, 0, 1, 0, 0); /* set up the various candts */ *ps_candt_4x8_l = as_left_neighbours[0]; *ps_candt_4x8_t = as_top_neighbours[1]; *ps_candt_4x8_tl = as_top_neighbours[0]; *ps_candt_8x4_l = *ps_candt_4x8_l; *ps_candt_8x4_tl = *ps_candt_4x8_tl; *ps_candt_8x4_t = *ps_candt_4x8_t; { S32 pred_lx; S16 *pi2_sads_4x4_current, *pi2_sads_4x4_top; pred_ctxt_t *ps_pred_ctxt; PF_MV_COST_FXN pf_mv_cost_compute; /* Computer 4x4 SADs for current block */ /* Pointer to store SADs */ pi2_sads_4x4_current = pi2_cur_ref_sads_4x4 + sad_current_offset; hme_derive_worst_case_search_range( &s_range_prms, &s_pic_limit, &as_mv_limit[i1_ref_idx], blk_x << blk_size_shift, blk_y << blk_size_shift, blk_wd, blk_ht); if(i4_pic_ht == blk_y) { memset(pi2_sads_4x4_current, 0, ai4_sad_4x4_block_size[i1_ref_idx]); } else { if(ME_PRISTINE_QUALITY >= e_me_quality_preset) { ((ihevce_me_optimised_function_list_t *) ps_ctxt->pv_me_optimised_function_list) ->pf_store_4x4_sads_high_quality( &s_search_prms_4x4, ps_curr_layer, &as_mv_limit[i1_ref_idx], &ps_ctxt->s_wt_pred, pi2_sads_4x4_current); } else { ((ihevce_me_optimised_function_list_t *) ps_ctxt->pv_me_optimised_function_list) ->pf_store_4x4_sads_high_speed( &s_search_prms_4x4, ps_curr_layer, &as_mv_limit[i1_ref_idx], &ps_ctxt->s_wt_pred, pi2_sads_4x4_current); } } /* Set pred direction to L0 or L1 */ pred_lx = 1 - ps_search_results->pu1_is_past[i1_ref_idx]; /* Suitable context (L0 or L1) */ ps_pred_ctxt = &ps_search_results->as_pred_ctxt[pred_lx]; /* Coarse layer is always explicit */ if(ME_PRISTINE_QUALITY > e_me_quality_preset) { pf_mv_cost_compute = compute_mv_cost_coarse; } else { /* Cost function is not called in high speed case. Below one is just a dummy function */ pf_mv_cost_compute = compute_mv_cost_coarse_high_speed; } /*********************************************************************/ /* Now, compute the mv for the top block */ /*********************************************************************/ pi2_sads_4x4_top = pi2_cur_ref_sads_4x4 + sad_top_offset; /*********************************************************************/ /* For every blk in the picture, the search range needs to be derived*/ /* Any blk can have any mv, but practical search constraints are */ /* imposed by the picture boundary and amt of padding. */ /*********************************************************************/ hme_derive_search_range( &s_range_prms, &s_pic_limit, &as_mv_limit[i1_ref_idx], blk_x << blk_size_shift, (blk_y - 1) << blk_size_shift, blk_wd, blk_ht); /* Computer the mv for the top block */ if(ME_PRISTINE_QUALITY >= e_me_quality_preset) { ((ihevce_me_optimised_function_list_t *) ps_ctxt->pv_me_optimised_function_list) ->pf_combine_4x4_sads_and_compute_cost_high_quality( i1_ref_idx, &s_range_prms, /* Both 4x8 and 8x4 has same search range */ &as_mv_limit[i1_ref_idx], &best_mv_4x8, &best_mv_8x4, ps_pred_ctxt, pf_mv_cost_compute, pi2_sads_4x4_top, /* Current SAD block */ (pi2_sads_4x4_top + ai4_sad_4x4_block_size[i1_ref_idx]), /* East SAD block */ pi2_sads_4x4_current); /* South SAD block */ } else { ((ihevce_me_optimised_function_list_t *) ps_ctxt->pv_me_optimised_function_list) ->pf_combine_4x4_sads_and_compute_cost_high_speed( i1_ref_idx, &s_range_prms, /* Both 4x8 and 8x4 has same search range */ &as_mv_limit[i1_ref_idx], &best_mv_4x8, &best_mv_8x4, ps_pred_ctxt, pf_mv_cost_compute, pi2_sads_4x4_top, /* Current SAD block */ (pi2_sads_4x4_top + ai4_sad_4x4_block_size[i1_ref_idx]), /* East SAD block */ pi2_sads_4x4_current); /* South SAD block */ } ps_candt_fs_4x8->s_mv.i2_mvx = best_mv_4x8.i2_mv_x; ps_candt_fs_4x8->s_mv.i2_mvy = best_mv_4x8.i2_mv_y; ps_candt_fs_4x8->i1_ref_idx = i1_ref_idx; ps_candt_fs_8x4->s_mv.i2_mvx = best_mv_8x4.i2_mv_x; ps_candt_fs_8x4->s_mv.i2_mvy = best_mv_8x4.i2_mv_y; ps_candt_fs_8x4->i1_ref_idx = i1_ref_idx; } /* call the appropriate Search Algo for 4x8S. The 4x8N would */ /* have already been called by top block */ hme_pred_search_square_stepn( &s_search_prms_8x4, ps_curr_layer, &ps_ctxt->s_wt_pred, e_me_quality_preset, (ihevce_me_optimised_function_list_t *) ps_ctxt->pv_me_optimised_function_list ); /* Call the appropriate search algo for 8x4E */ hme_pred_search_square_stepn( &s_search_prms_4x8, ps_curr_layer, &ps_ctxt->s_wt_pred, e_me_quality_preset, (ihevce_me_optimised_function_list_t *) ps_ctxt->pv_me_optimised_function_list); if(ME_XTREME_SPEED_25 != e_me_quality_preset) { /* Histogram updates across different Ref ID for global MV */ hme_update_histogram( ps_ctxt->aps_mv_hist[i1_ref_idx], aps_best_search_node_8x4[i1_ref_idx]->s_mv.i2_mvx, aps_best_search_node_8x4[i1_ref_idx]->s_mv.i2_mvy); hme_update_histogram( ps_ctxt->aps_mv_hist[i1_ref_idx], aps_best_search_node_4x8[i1_ref_idx]->s_mv.i2_mvx, aps_best_search_node_4x8[i1_ref_idx]->s_mv.i2_mvy); } /* update the best results to the mv bank */ hme_update_mv_bank_coarse( ps_search_results, ps_curr_layer->ps_layer_mvbank, blk_x, (blk_y - 1), ps_ctxt->aps_best_search_nodes_4x8_n_rows[i1_ref_idx] + search_node_top_offset, /* Top Candidate */ ps_ctxt->aps_best_search_nodes_8x4_n_rows[i1_ref_idx] + search_node_left_offset, /* Left candidate */ i1_ref_idx, &s_mv_update_prms); /* Copy the best search result to 5 row array for future use */ *(ps_ctxt->aps_best_search_nodes_4x8_n_rows[i1_ref_idx] + blk_x + ps_ctxt->ai4_row_index[blk_y - 1] * num_blks_in_row) = *(aps_best_search_node_4x8[i1_ref_idx]); *(ps_ctxt->aps_best_search_nodes_8x4_n_rows[i1_ref_idx] + blk_x + ps_ctxt->ai4_row_index[blk_y - 1] * num_blks_in_row) = *(aps_best_search_node_8x4[i1_ref_idx]); /* UPDATE the MIN and MAX MVs for Dynamical Search Range for each ref. pic. */ /* Only for P pic. For P, both are 0, I&B has them mut. exclusive */ if(ps_ctxt->s_frm_prms.is_i_pic == ps_ctxt->s_frm_prms.bidir_enabled) { WORD32 num_mvs, i, j; search_node_t *aps_search_nodes[4]; /* Best results for 8x4R and 4x8B blocks */ search_node_t *ps_search_node_8x4_r, *ps_search_node_4x8_b; num_mvs = ps_curr_layer->ps_layer_mvbank->i4_num_mvs_per_ref; /*************************************************************************/ /* We have atleast 4 distinct results: the 4x8 top (coming from top blk) */ /* 8x4 left (coming from left blk), 8x4 and 4x8 right and bot resp. */ /* If number of results to be stored is 4, then we store all these 4 */ /* results, else we pick best ones */ /*************************************************************************/ ps_search_node_8x4_r = ps_search_results->aps_part_results[i1_ref_idx][PART_ID_2NxN_B]; ps_search_node_4x8_b = ps_search_results->aps_part_results[i1_ref_idx][PART_ID_Nx2N_R]; ASSERT(num_mvs <= 4); /* Doing this to sort best results */ aps_search_nodes[0] = ps_search_node_8x4_r; aps_search_nodes[1] = ps_search_node_4x8_b; aps_search_nodes[2] = ps_ctxt->aps_best_search_nodes_8x4_n_rows[i1_ref_idx] + search_node_left_offset; /* Left candidate */ aps_search_nodes[3] = ps_ctxt->aps_best_search_nodes_4x8_n_rows[i1_ref_idx] + search_node_top_offset; /* Top Candidate */ /* Note : Need to be resolved!!! */ /* Added this to match with "hme_update_mv_bank_coarse" */ if(num_mvs != 4) { /* Run through the results, store them in best to worst order */ for(i = 0; i < num_mvs; i++) { for(j = i + 1; j < 4; j++) { if(aps_search_nodes[j]->i4_tot_cost < aps_search_nodes[i]->i4_tot_cost) { SWAP_HME( aps_search_nodes[j], aps_search_nodes[i], search_node_t *); } } } } /* UPDATE the MIN and MAX MVs for Dynamical Search Range for each ref. pic. */ for(i = 0; i < num_mvs; i++) { hme_update_dynamic_search_params( &ps_ctxt->s_coarse_dyn_range_prms .as_dyn_range_prms[i4_layer_id][i1_ref_idx], aps_search_nodes[i]->s_mv.i2_mvy); } } } } /* Update the number of blocks processed in the current row */ ihevce_dmgr_set_row_row_sync( pv_hme_dep_mngr, (blk_x + 1), blk_y, 0 /* Col Tile No. : Not supported in PreEnc*/); } /* set the output dependency after completion of row */ ihevce_pre_enc_grp_job_set_out_dep(ps_multi_thrd_ctxt, ps_job, i4_ping_pong); } } return; }