michael@0: /*
michael@0:  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
michael@0:  *
michael@0:  *  Use of this source code is governed by a BSD-style license
michael@0:  *  that can be found in the LICENSE file in the root of the source
michael@0:  *  tree. An additional intellectual property rights grant can be found
michael@0:  *  in the file PATENTS.  All contributing project authors may
michael@0:  *  be found in the AUTHORS file in the root of the source tree.
michael@0:  */
michael@0: 
michael@0: #include "onyx_int.h"
michael@0: #include "vp8/common/threading.h"
michael@0: #include "vp8/common/common.h"
michael@0: #include "vp8/common/extend.h"
michael@0: #include "bitstream.h"
michael@0: #include "encodeframe.h"
michael@0: 
michael@0: #if CONFIG_MULTITHREAD
michael@0: 
michael@0: extern void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x, int ok_to_skip);
michael@0: 
michael@0: extern void vp8_loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm);
michael@0: 
michael@0: static THREAD_FUNCTION thread_loopfilter(void *p_data)
michael@0: {
michael@0:     VP8_COMP *cpi = (VP8_COMP *)(((LPFTHREAD_DATA *)p_data)->ptr1);
michael@0:     VP8_COMMON *cm = &cpi->common;
michael@0: 
michael@0:     while (1)
michael@0:     {
michael@0:         if (cpi->b_multi_threaded == 0)
michael@0:             break;
michael@0: 
michael@0:         if (sem_wait(&cpi->h_event_start_lpf) == 0)
michael@0:         {
michael@0:             if (cpi->b_multi_threaded == 0) /* we're shutting down */
michael@0:                 break;
michael@0: 
michael@0:             vp8_loopfilter_frame(cpi, cm);
michael@0: 
michael@0:             sem_post(&cpi->h_event_end_lpf);
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     return 0;
michael@0: }
michael@0: 
michael@0: static
michael@0: THREAD_FUNCTION thread_encoding_proc(void *p_data)
michael@0: {
michael@0:     int ithread = ((ENCODETHREAD_DATA *)p_data)->ithread;
michael@0:     VP8_COMP *cpi = (VP8_COMP *)(((ENCODETHREAD_DATA *)p_data)->ptr1);
michael@0:     MB_ROW_COMP *mbri = (MB_ROW_COMP *)(((ENCODETHREAD_DATA *)p_data)->ptr2);
michael@0:     ENTROPY_CONTEXT_PLANES mb_row_left_context;
michael@0: 
michael@0:     while (1)
michael@0:     {
michael@0:         if (cpi->b_multi_threaded == 0)
michael@0:             break;
michael@0: 
michael@0:         if (sem_wait(&cpi->h_event_start_encoding[ithread]) == 0)
michael@0:         {
michael@0:             const int nsync = cpi->mt_sync_range;
michael@0:             VP8_COMMON *cm = &cpi->common;
michael@0:             int mb_row;
michael@0:             MACROBLOCK *x = &mbri->mb;
michael@0:             MACROBLOCKD *xd = &x->e_mbd;
michael@0:             TOKENEXTRA *tp ;
michael@0: #if CONFIG_REALTIME_ONLY & CONFIG_ONTHEFLY_BITPACKING
michael@0:             TOKENEXTRA *tp_start = cpi->tok + (1 + ithread) * (16 * 24);
michael@0:             const int num_part = (1 << cm->multi_token_partition);
michael@0: #endif
michael@0: 
michael@0:             int *segment_counts = mbri->segment_counts;
michael@0:             int *totalrate = &mbri->totalrate;
michael@0: 
michael@0:             if (cpi->b_multi_threaded == 0) /* we're shutting down */
michael@0:                 break;
michael@0: 
michael@0:             for (mb_row = ithread + 1; mb_row < cm->mb_rows; mb_row += (cpi->encoding_thread_count + 1))
michael@0:             {
michael@0: 
michael@0:                 int recon_yoffset, recon_uvoffset;
michael@0:                 int mb_col;
michael@0:                 int ref_fb_idx = cm->lst_fb_idx;
michael@0:                 int dst_fb_idx = cm->new_fb_idx;
michael@0:                 int recon_y_stride = cm->yv12_fb[ref_fb_idx].y_stride;
michael@0:                 int recon_uv_stride = cm->yv12_fb[ref_fb_idx].uv_stride;
michael@0:                 int map_index = (mb_row * cm->mb_cols);
michael@0:                 volatile const int *last_row_current_mb_col;
michael@0:                 volatile int *current_mb_col = &cpi->mt_current_mb_col[mb_row];
michael@0: 
michael@0: #if  (CONFIG_REALTIME_ONLY & CONFIG_ONTHEFLY_BITPACKING)
michael@0:                 vp8_writer *w = &cpi->bc[1 + (mb_row % num_part)];
michael@0: #else
michael@0:                 tp = cpi->tok + (mb_row * (cm->mb_cols * 16 * 24));
michael@0:                 cpi->tplist[mb_row].start = tp;
michael@0: #endif
michael@0: 
michael@0:                 last_row_current_mb_col = &cpi->mt_current_mb_col[mb_row - 1];
michael@0: 
michael@0:                 /* reset above block coeffs */
michael@0:                 xd->above_context = cm->above_context;
michael@0:                 xd->left_context = &mb_row_left_context;
michael@0: 
michael@0:                 vp8_zero(mb_row_left_context);
michael@0: 
michael@0:                 xd->up_available = (mb_row != 0);
michael@0:                 recon_yoffset = (mb_row * recon_y_stride * 16);
michael@0:                 recon_uvoffset = (mb_row * recon_uv_stride * 8);
michael@0: 
michael@0:                 /* Set the mb activity pointer to the start of the row. */
michael@0:                 x->mb_activity_ptr = &cpi->mb_activity_map[map_index];
michael@0: 
michael@0:                 /* for each macroblock col in image */
michael@0:                 for (mb_col = 0; mb_col < cm->mb_cols; mb_col++)
michael@0:                 {
michael@0:                     *current_mb_col = mb_col - 1;
michael@0: 
michael@0:                     if ((mb_col & (nsync - 1)) == 0)
michael@0:                     {
michael@0:                         while (mb_col > (*last_row_current_mb_col - nsync))
michael@0:                         {
michael@0:                             x86_pause_hint();
michael@0:                             thread_sleep(0);
michael@0:                         }
michael@0:                     }
michael@0: 
michael@0: #if CONFIG_REALTIME_ONLY & CONFIG_ONTHEFLY_BITPACKING
michael@0:                     tp = tp_start;
michael@0: #endif
michael@0: 
michael@0:                     /* Distance of Mb to the various image edges.
michael@0:                      * These specified to 8th pel as they are always compared
michael@0:                      * to values that are in 1/8th pel units
michael@0:                      */
michael@0:                     xd->mb_to_left_edge = -((mb_col * 16) << 3);
michael@0:                     xd->mb_to_right_edge = ((cm->mb_cols - 1 - mb_col) * 16) << 3;
michael@0:                     xd->mb_to_top_edge = -((mb_row * 16) << 3);
michael@0:                     xd->mb_to_bottom_edge = ((cm->mb_rows - 1 - mb_row) * 16) << 3;
michael@0: 
michael@0:                     /* Set up limit values for motion vectors used to prevent
michael@0:                      * them extending outside the UMV borders
michael@0:                      */
michael@0:                     x->mv_col_min = -((mb_col * 16) + (VP8BORDERINPIXELS - 16));
michael@0:                     x->mv_col_max = ((cm->mb_cols - 1 - mb_col) * 16) + (VP8BORDERINPIXELS - 16);
michael@0:                     x->mv_row_min = -((mb_row * 16) + (VP8BORDERINPIXELS - 16));
michael@0:                     x->mv_row_max = ((cm->mb_rows - 1 - mb_row) * 16) + (VP8BORDERINPIXELS - 16);
michael@0: 
michael@0:                     xd->dst.y_buffer = cm->yv12_fb[dst_fb_idx].y_buffer + recon_yoffset;
michael@0:                     xd->dst.u_buffer = cm->yv12_fb[dst_fb_idx].u_buffer + recon_uvoffset;
michael@0:                     xd->dst.v_buffer = cm->yv12_fb[dst_fb_idx].v_buffer + recon_uvoffset;
michael@0:                     xd->left_available = (mb_col != 0);
michael@0: 
michael@0:                     x->rddiv = cpi->RDDIV;
michael@0:                     x->rdmult = cpi->RDMULT;
michael@0: 
michael@0:                     /* Copy current mb to a buffer */
michael@0:                     vp8_copy_mem16x16(x->src.y_buffer, x->src.y_stride, x->thismb, 16);
michael@0: 
michael@0:                     if (cpi->oxcf.tuning == VP8_TUNE_SSIM)
michael@0:                         vp8_activity_masking(cpi, x);
michael@0: 
michael@0:                     /* Is segmentation enabled */
michael@0:                     /* MB level adjustment to quantizer */
michael@0:                     if (xd->segmentation_enabled)
michael@0:                     {
michael@0:                         /* Code to set segment id in xd->mbmi.segment_id for
michael@0:                          * current MB (with range checking)
michael@0:                          */
michael@0:                         if (cpi->segmentation_map[map_index + mb_col] <= 3)
michael@0:                             xd->mode_info_context->mbmi.segment_id = cpi->segmentation_map[map_index + mb_col];
michael@0:                         else
michael@0:                             xd->mode_info_context->mbmi.segment_id = 0;
michael@0: 
michael@0:                         vp8cx_mb_init_quantizer(cpi, x, 1);
michael@0:                     }
michael@0:                     else
michael@0:                         /* Set to Segment 0 by default */
michael@0:                         xd->mode_info_context->mbmi.segment_id = 0;
michael@0: 
michael@0:                     x->active_ptr = cpi->active_map + map_index + mb_col;
michael@0: 
michael@0:                     if (cm->frame_type == KEY_FRAME)
michael@0:                     {
michael@0:                         *totalrate += vp8cx_encode_intra_macroblock(cpi, x, &tp);
michael@0: #ifdef MODE_STATS
michael@0:                         y_modes[xd->mbmi.mode] ++;
michael@0: #endif
michael@0:                     }
michael@0:                     else
michael@0:                     {
michael@0:                         *totalrate += vp8cx_encode_inter_macroblock(cpi, x, &tp, recon_yoffset, recon_uvoffset, mb_row, mb_col);
michael@0: 
michael@0: #ifdef MODE_STATS
michael@0:                         inter_y_modes[xd->mbmi.mode] ++;
michael@0: 
michael@0:                         if (xd->mbmi.mode == SPLITMV)
michael@0:                         {
michael@0:                             int b;
michael@0: 
michael@0:                             for (b = 0; b < xd->mbmi.partition_count; b++)
michael@0:                             {
michael@0:                                 inter_b_modes[x->partition->bmi[b].mode] ++;
michael@0:                             }
michael@0:                         }
michael@0: 
michael@0: #endif
michael@0: 
michael@0:                         /* Special case code for cyclic refresh
michael@0:                          * If cyclic update enabled then copy
michael@0:                          * xd->mbmi.segment_id; (which may have been updated
michael@0:                          * based on mode during
michael@0:                          * vp8cx_encode_inter_macroblock()) back into the
michael@0:                          * global segmentation map
michael@0:                          */
michael@0:                         if ((cpi->current_layer == 0) &&
michael@0:                             (cpi->cyclic_refresh_mode_enabled &&
michael@0:                              xd->segmentation_enabled))
michael@0:                         {
michael@0:                             const MB_MODE_INFO * mbmi = &xd->mode_info_context->mbmi;
michael@0:                             cpi->segmentation_map[map_index + mb_col] = mbmi->segment_id;
michael@0: 
michael@0:                             /* If the block has been refreshed mark it as clean
michael@0:                              * (the magnitude of the -ve influences how long it
michael@0:                              * will be before we consider another refresh):
michael@0:                              * Else if it was coded (last frame 0,0) and has
michael@0:                              * not already been refreshed then mark it as a
michael@0:                              * candidate for cleanup next time (marked 0) else
michael@0:                              * mark it as dirty (1).
michael@0:                              */
michael@0:                             if (mbmi->segment_id)
michael@0:                                 cpi->cyclic_refresh_map[map_index + mb_col] = -1;
michael@0:                             else if ((mbmi->mode == ZEROMV) && (mbmi->ref_frame == LAST_FRAME))
michael@0:                             {
michael@0:                                 if (cpi->cyclic_refresh_map[map_index + mb_col] == 1)
michael@0:                                     cpi->cyclic_refresh_map[map_index + mb_col] = 0;
michael@0:                             }
michael@0:                             else
michael@0:                                 cpi->cyclic_refresh_map[map_index + mb_col] = 1;
michael@0: 
michael@0:                         }
michael@0:                     }
michael@0: 
michael@0: #if CONFIG_REALTIME_ONLY & CONFIG_ONTHEFLY_BITPACKING
michael@0:                     /* pack tokens for this MB */
michael@0:                     {
michael@0:                         int tok_count = tp - tp_start;
michael@0:                         pack_tokens(w, tp_start, tok_count);
michael@0:                     }
michael@0: #else
michael@0:                     cpi->tplist[mb_row].stop = tp;
michael@0: #endif
michael@0:                     /* Increment pointer into gf usage flags structure. */
michael@0:                     x->gf_active_ptr++;
michael@0: 
michael@0:                     /* Increment the activity mask pointers. */
michael@0:                     x->mb_activity_ptr++;
michael@0: 
michael@0:                     /* adjust to the next column of macroblocks */
michael@0:                     x->src.y_buffer += 16;
michael@0:                     x->src.u_buffer += 8;
michael@0:                     x->src.v_buffer += 8;
michael@0: 
michael@0:                     recon_yoffset += 16;
michael@0:                     recon_uvoffset += 8;
michael@0: 
michael@0:                     /* Keep track of segment usage */
michael@0:                     segment_counts[xd->mode_info_context->mbmi.segment_id]++;
michael@0: 
michael@0:                     /* skip to next mb */
michael@0:                     xd->mode_info_context++;
michael@0:                     x->partition_info++;
michael@0:                     xd->above_context++;
michael@0:                 }
michael@0: 
michael@0:                 vp8_extend_mb_row( &cm->yv12_fb[dst_fb_idx],
michael@0:                                     xd->dst.y_buffer + 16,
michael@0:                                     xd->dst.u_buffer + 8,
michael@0:                                     xd->dst.v_buffer + 8);
michael@0: 
michael@0:                 *current_mb_col = mb_col + nsync;
michael@0: 
michael@0:                 /* this is to account for the border */
michael@0:                 xd->mode_info_context++;
michael@0:                 x->partition_info++;
michael@0: 
michael@0:                 x->src.y_buffer += 16 * x->src.y_stride * (cpi->encoding_thread_count + 1) - 16 * cm->mb_cols;
michael@0:                 x->src.u_buffer += 8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols;
michael@0:                 x->src.v_buffer += 8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols;
michael@0: 
michael@0:                 xd->mode_info_context += xd->mode_info_stride * cpi->encoding_thread_count;
michael@0:                 x->partition_info += xd->mode_info_stride * cpi->encoding_thread_count;
michael@0:                 x->gf_active_ptr   += cm->mb_cols * cpi->encoding_thread_count;
michael@0: 
michael@0:                 if (mb_row == cm->mb_rows - 1)
michael@0:                 {
michael@0:                     sem_post(&cpi->h_event_end_encoding); /* signal frame encoding end */
michael@0:                 }
michael@0:             }
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     /* printf("exit thread %d\n", ithread); */
michael@0:     return 0;
michael@0: }
michael@0: 
michael@0: static void setup_mbby_copy(MACROBLOCK *mbdst, MACROBLOCK *mbsrc)
michael@0: {
michael@0: 
michael@0:     MACROBLOCK *x = mbsrc;
michael@0:     MACROBLOCK *z = mbdst;
michael@0:     int i;
michael@0: 
michael@0:     z->ss               = x->ss;
michael@0:     z->ss_count          = x->ss_count;
michael@0:     z->searches_per_step  = x->searches_per_step;
michael@0:     z->errorperbit      = x->errorperbit;
michael@0: 
michael@0:     z->sadperbit16      = x->sadperbit16;
michael@0:     z->sadperbit4       = x->sadperbit4;
michael@0: 
michael@0:     /*
michael@0:     z->mv_col_min    = x->mv_col_min;
michael@0:     z->mv_col_max    = x->mv_col_max;
michael@0:     z->mv_row_min    = x->mv_row_min;
michael@0:     z->mv_row_max    = x->mv_row_max;
michael@0:     */
michael@0: 
michael@0:     z->short_fdct4x4     = x->short_fdct4x4;
michael@0:     z->short_fdct8x4     = x->short_fdct8x4;
michael@0:     z->short_walsh4x4    = x->short_walsh4x4;
michael@0:     z->quantize_b        = x->quantize_b;
michael@0:     z->quantize_b_pair   = x->quantize_b_pair;
michael@0:     z->optimize          = x->optimize;
michael@0: 
michael@0:     /*
michael@0:     z->mvc              = x->mvc;
michael@0:     z->src.y_buffer      = x->src.y_buffer;
michael@0:     z->src.u_buffer      = x->src.u_buffer;
michael@0:     z->src.v_buffer      = x->src.v_buffer;
michael@0:     */
michael@0: 
michael@0:     z->mvcost[0] =  x->mvcost[0];
michael@0:     z->mvcost[1] =  x->mvcost[1];
michael@0:     z->mvsadcost[0] =  x->mvsadcost[0];
michael@0:     z->mvsadcost[1] =  x->mvsadcost[1];
michael@0: 
michael@0:     z->token_costs = x->token_costs;
michael@0:     z->inter_bmode_costs = x->inter_bmode_costs;
michael@0:     z->mbmode_cost = x->mbmode_cost;
michael@0:     z->intra_uv_mode_cost = x->intra_uv_mode_cost;
michael@0:     z->bmode_costs = x->bmode_costs;
michael@0: 
michael@0:     for (i = 0; i < 25; i++)
michael@0:     {
michael@0:         z->block[i].quant           = x->block[i].quant;
michael@0:         z->block[i].quant_fast      = x->block[i].quant_fast;
michael@0:         z->block[i].quant_shift     = x->block[i].quant_shift;
michael@0:         z->block[i].zbin            = x->block[i].zbin;
michael@0:         z->block[i].zrun_zbin_boost = x->block[i].zrun_zbin_boost;
michael@0:         z->block[i].round           = x->block[i].round;
michael@0:         z->block[i].src_stride      = x->block[i].src_stride;
michael@0:     }
michael@0: 
michael@0:     z->q_index           = x->q_index;
michael@0:     z->act_zbin_adj      = x->act_zbin_adj;
michael@0:     z->last_act_zbin_adj = x->last_act_zbin_adj;
michael@0: 
michael@0:     {
michael@0:         MACROBLOCKD *xd = &x->e_mbd;
michael@0:         MACROBLOCKD *zd = &z->e_mbd;
michael@0: 
michael@0:         /*
michael@0:         zd->mode_info_context = xd->mode_info_context;
michael@0:         zd->mode_info        = xd->mode_info;
michael@0: 
michael@0:         zd->mode_info_stride  = xd->mode_info_stride;
michael@0:         zd->frame_type       = xd->frame_type;
michael@0:         zd->up_available     = xd->up_available   ;
michael@0:         zd->left_available   = xd->left_available;
michael@0:         zd->left_context     = xd->left_context;
michael@0:         zd->last_frame_dc     = xd->last_frame_dc;
michael@0:         zd->last_frame_dccons = xd->last_frame_dccons;
michael@0:         zd->gold_frame_dc     = xd->gold_frame_dc;
michael@0:         zd->gold_frame_dccons = xd->gold_frame_dccons;
michael@0:         zd->mb_to_left_edge    = xd->mb_to_left_edge;
michael@0:         zd->mb_to_right_edge   = xd->mb_to_right_edge;
michael@0:         zd->mb_to_top_edge     = xd->mb_to_top_edge   ;
michael@0:         zd->mb_to_bottom_edge  = xd->mb_to_bottom_edge;
michael@0:         zd->gf_active_ptr     = xd->gf_active_ptr;
michael@0:         zd->frames_since_golden       = xd->frames_since_golden;
michael@0:         zd->frames_till_alt_ref_frame   = xd->frames_till_alt_ref_frame;
michael@0:         */
michael@0:         zd->subpixel_predict         = xd->subpixel_predict;
michael@0:         zd->subpixel_predict8x4      = xd->subpixel_predict8x4;
michael@0:         zd->subpixel_predict8x8      = xd->subpixel_predict8x8;
michael@0:         zd->subpixel_predict16x16    = xd->subpixel_predict16x16;
michael@0:         zd->segmentation_enabled     = xd->segmentation_enabled;
michael@0:         zd->mb_segement_abs_delta      = xd->mb_segement_abs_delta;
michael@0:         vpx_memcpy(zd->segment_feature_data, xd->segment_feature_data,
michael@0:                    sizeof(xd->segment_feature_data));
michael@0: 
michael@0:         vpx_memcpy(zd->dequant_y1_dc, xd->dequant_y1_dc,
michael@0:                    sizeof(xd->dequant_y1_dc));
michael@0:         vpx_memcpy(zd->dequant_y1, xd->dequant_y1, sizeof(xd->dequant_y1));
michael@0:         vpx_memcpy(zd->dequant_y2, xd->dequant_y2, sizeof(xd->dequant_y2));
michael@0:         vpx_memcpy(zd->dequant_uv, xd->dequant_uv, sizeof(xd->dequant_uv));
michael@0: 
michael@0: #if 1
michael@0:         /*TODO:  Remove dequant from BLOCKD.  This is a temporary solution until
michael@0:          * the quantizer code uses a passed in pointer to the dequant constants.
michael@0:          * This will also require modifications to the x86 and neon assembly.
michael@0:          * */
michael@0:         for (i = 0; i < 16; i++)
michael@0:             zd->block[i].dequant = zd->dequant_y1;
michael@0:         for (i = 16; i < 24; i++)
michael@0:             zd->block[i].dequant = zd->dequant_uv;
michael@0:         zd->block[24].dequant = zd->dequant_y2;
michael@0: #endif
michael@0: 
michael@0: 
michael@0:         vpx_memcpy(z->rd_threshes, x->rd_threshes, sizeof(x->rd_threshes));
michael@0:         vpx_memcpy(z->rd_thresh_mult, x->rd_thresh_mult,
michael@0:                    sizeof(x->rd_thresh_mult));
michael@0: 
michael@0:         z->zbin_over_quant = x->zbin_over_quant;
michael@0:         z->zbin_mode_boost_enabled = x->zbin_mode_boost_enabled;
michael@0:         z->zbin_mode_boost = x->zbin_mode_boost;
michael@0: 
michael@0:         vpx_memset(z->error_bins, 0, sizeof(z->error_bins));
michael@0:     }
michael@0: }
michael@0: 
michael@0: void vp8cx_init_mbrthread_data(VP8_COMP *cpi,
michael@0:                                MACROBLOCK *x,
michael@0:                                MB_ROW_COMP *mbr_ei,
michael@0:                                int count
michael@0:                               )
michael@0: {
michael@0: 
michael@0:     VP8_COMMON *const cm = & cpi->common;
michael@0:     MACROBLOCKD *const xd = & x->e_mbd;
michael@0:     int i;
michael@0: 
michael@0:     for (i = 0; i < count; i++)
michael@0:     {
michael@0:         MACROBLOCK *mb = & mbr_ei[i].mb;
michael@0:         MACROBLOCKD *mbd = &mb->e_mbd;
michael@0: 
michael@0:         mbd->subpixel_predict        = xd->subpixel_predict;
michael@0:         mbd->subpixel_predict8x4     = xd->subpixel_predict8x4;
michael@0:         mbd->subpixel_predict8x8     = xd->subpixel_predict8x8;
michael@0:         mbd->subpixel_predict16x16   = xd->subpixel_predict16x16;
michael@0:         mb->gf_active_ptr            = x->gf_active_ptr;
michael@0: 
michael@0:         vpx_memset(mbr_ei[i].segment_counts, 0, sizeof(mbr_ei[i].segment_counts));
michael@0:         mbr_ei[i].totalrate = 0;
michael@0: 
michael@0:         mb->partition_info = x->pi + x->e_mbd.mode_info_stride * (i + 1);
michael@0: 
michael@0:         mbd->mode_info_context = cm->mi   + x->e_mbd.mode_info_stride * (i + 1);
michael@0:         mbd->mode_info_stride  = cm->mode_info_stride;
michael@0: 
michael@0:         mbd->frame_type = cm->frame_type;
michael@0: 
michael@0:         mb->src = * cpi->Source;
michael@0:         mbd->pre = cm->yv12_fb[cm->lst_fb_idx];
michael@0:         mbd->dst = cm->yv12_fb[cm->new_fb_idx];
michael@0: 
michael@0:         mb->src.y_buffer += 16 * x->src.y_stride * (i + 1);
michael@0:         mb->src.u_buffer +=  8 * x->src.uv_stride * (i + 1);
michael@0:         mb->src.v_buffer +=  8 * x->src.uv_stride * (i + 1);
michael@0: 
michael@0:         vp8_build_block_offsets(mb);
michael@0: 
michael@0:         mbd->left_context = &cm->left_context;
michael@0:         mb->mvc = cm->fc.mvc;
michael@0: 
michael@0:         setup_mbby_copy(&mbr_ei[i].mb, x);
michael@0: 
michael@0:         mbd->fullpixel_mask = 0xffffffff;
michael@0:         if(cm->full_pixel)
michael@0:             mbd->fullpixel_mask = 0xfffffff8;
michael@0: 
michael@0:         vp8_zero(mb->coef_counts);
michael@0:         vp8_zero(x->ymode_count);
michael@0:         mb->skip_true_count = 0;
michael@0:         vp8_zero(mb->MVcount);
michael@0:         mb->prediction_error = 0;
michael@0:         mb->intra_error = 0;
michael@0:         vp8_zero(mb->count_mb_ref_frame_usage);
michael@0:         mb->mbs_tested_so_far = 0;
michael@0:     }
michael@0: }
michael@0: 
michael@0: int vp8cx_create_encoder_threads(VP8_COMP *cpi)
michael@0: {
michael@0:     const VP8_COMMON * cm = &cpi->common;
michael@0: 
michael@0:     cpi->b_multi_threaded = 0;
michael@0:     cpi->encoding_thread_count = 0;
michael@0:     cpi->b_lpf_running = 0;
michael@0: 
michael@0:     if (cm->processor_core_count > 1 && cpi->oxcf.multi_threaded > 1)
michael@0:     {
michael@0:         int ithread;
michael@0:         int th_count = cpi->oxcf.multi_threaded - 1;
michael@0:         int rc = 0;
michael@0: 
michael@0:         /* don't allocate more threads than cores available */
michael@0:         if (cpi->oxcf.multi_threaded > cm->processor_core_count)
michael@0:             th_count = cm->processor_core_count - 1;
michael@0: 
michael@0:         /* we have th_count + 1 (main) threads processing one row each */
michael@0:         /* no point to have more threads than the sync range allows */
michael@0:         if(th_count > ((cm->mb_cols / cpi->mt_sync_range) - 1))
michael@0:         {
michael@0:             th_count = (cm->mb_cols / cpi->mt_sync_range) - 1;
michael@0:         }
michael@0: 
michael@0:         if(th_count == 0)
michael@0:             return 0;
michael@0: 
michael@0:         CHECK_MEM_ERROR(cpi->h_encoding_thread,
michael@0:                         vpx_malloc(sizeof(pthread_t) * th_count));
michael@0:         CHECK_MEM_ERROR(cpi->h_event_start_encoding,
michael@0:                         vpx_malloc(sizeof(sem_t) * th_count));
michael@0:         CHECK_MEM_ERROR(cpi->mb_row_ei,
michael@0:                         vpx_memalign(32, sizeof(MB_ROW_COMP) * th_count));
michael@0:         vpx_memset(cpi->mb_row_ei, 0, sizeof(MB_ROW_COMP) * th_count);
michael@0:         CHECK_MEM_ERROR(cpi->en_thread_data,
michael@0:                         vpx_malloc(sizeof(ENCODETHREAD_DATA) * th_count));
michael@0: 
michael@0:         sem_init(&cpi->h_event_end_encoding, 0, 0);
michael@0: 
michael@0:         cpi->b_multi_threaded = 1;
michael@0:         cpi->encoding_thread_count = th_count;
michael@0: 
michael@0:         /*
michael@0:         printf("[VP8:] multi_threaded encoding is enabled with %d threads\n\n",
michael@0:                (cpi->encoding_thread_count +1));
michael@0:         */
michael@0: 
michael@0:         for (ithread = 0; ithread < th_count; ithread++)
michael@0:         {
michael@0:             ENCODETHREAD_DATA *ethd = &cpi->en_thread_data[ithread];
michael@0: 
michael@0:             /* Setup block ptrs and offsets */
michael@0:             vp8_setup_block_ptrs(&cpi->mb_row_ei[ithread].mb);
michael@0:             vp8_setup_block_dptrs(&cpi->mb_row_ei[ithread].mb.e_mbd);
michael@0: 
michael@0:             sem_init(&cpi->h_event_start_encoding[ithread], 0, 0);
michael@0: 
michael@0:             ethd->ithread = ithread;
michael@0:             ethd->ptr1 = (void *)cpi;
michael@0:             ethd->ptr2 = (void *)&cpi->mb_row_ei[ithread];
michael@0: 
michael@0:             rc = pthread_create(&cpi->h_encoding_thread[ithread], 0,
michael@0:                                 thread_encoding_proc, ethd);
michael@0:             if(rc)
michael@0:                 break;
michael@0:         }
michael@0: 
michael@0:         if(rc)
michael@0:         {
michael@0:             /* shutdown other threads */
michael@0:             cpi->b_multi_threaded = 0;
michael@0:             for(--ithread; ithread >= 0; ithread--)
michael@0:             {
michael@0:                 pthread_join(cpi->h_encoding_thread[ithread], 0);
michael@0:                 sem_destroy(&cpi->h_event_start_encoding[ithread]);
michael@0:             }
michael@0:             sem_destroy(&cpi->h_event_end_encoding);
michael@0: 
michael@0:             /* free thread related resources */
michael@0:             vpx_free(cpi->h_event_start_encoding);
michael@0:             vpx_free(cpi->h_encoding_thread);
michael@0:             vpx_free(cpi->mb_row_ei);
michael@0:             vpx_free(cpi->en_thread_data);
michael@0: 
michael@0:             return -1;
michael@0:         }
michael@0: 
michael@0: 
michael@0:         {
michael@0:             LPFTHREAD_DATA * lpfthd = &cpi->lpf_thread_data;
michael@0: 
michael@0:             sem_init(&cpi->h_event_start_lpf, 0, 0);
michael@0:             sem_init(&cpi->h_event_end_lpf, 0, 0);
michael@0: 
michael@0:             lpfthd->ptr1 = (void *)cpi;
michael@0:             rc = pthread_create(&cpi->h_filter_thread, 0, thread_loopfilter,
michael@0:                                 lpfthd);
michael@0: 
michael@0:             if(rc)
michael@0:             {
michael@0:                 /* shutdown other threads */
michael@0:                 cpi->b_multi_threaded = 0;
michael@0:                 for(--ithread; ithread >= 0; ithread--)
michael@0:                 {
michael@0:                     sem_post(&cpi->h_event_start_encoding[ithread]);
michael@0:                     pthread_join(cpi->h_encoding_thread[ithread], 0);
michael@0:                     sem_destroy(&cpi->h_event_start_encoding[ithread]);
michael@0:                 }
michael@0:                 sem_destroy(&cpi->h_event_end_encoding);
michael@0:                 sem_destroy(&cpi->h_event_end_lpf);
michael@0:                 sem_destroy(&cpi->h_event_start_lpf);
michael@0: 
michael@0:                 /* free thread related resources */
michael@0:                 vpx_free(cpi->h_event_start_encoding);
michael@0:                 vpx_free(cpi->h_encoding_thread);
michael@0:                 vpx_free(cpi->mb_row_ei);
michael@0:                 vpx_free(cpi->en_thread_data);
michael@0: 
michael@0:                 return -2;
michael@0:             }
michael@0:         }
michael@0:     }
michael@0:     return 0;
michael@0: }
michael@0: 
michael@0: void vp8cx_remove_encoder_threads(VP8_COMP *cpi)
michael@0: {
michael@0:     if (cpi->b_multi_threaded)
michael@0:     {
michael@0:         /* shutdown other threads */
michael@0:         cpi->b_multi_threaded = 0;
michael@0:         {
michael@0:             int i;
michael@0: 
michael@0:             for (i = 0; i < cpi->encoding_thread_count; i++)
michael@0:             {
michael@0:                 sem_post(&cpi->h_event_start_encoding[i]);
michael@0:                 pthread_join(cpi->h_encoding_thread[i], 0);
michael@0: 
michael@0:                 sem_destroy(&cpi->h_event_start_encoding[i]);
michael@0:             }
michael@0: 
michael@0:             sem_post(&cpi->h_event_start_lpf);
michael@0:             pthread_join(cpi->h_filter_thread, 0);
michael@0:         }
michael@0: 
michael@0:         sem_destroy(&cpi->h_event_end_encoding);
michael@0:         sem_destroy(&cpi->h_event_end_lpf);
michael@0:         sem_destroy(&cpi->h_event_start_lpf);
michael@0: 
michael@0:         /* free thread related resources */
michael@0:         vpx_free(cpi->h_event_start_encoding);
michael@0:         vpx_free(cpi->h_encoding_thread);
michael@0:         vpx_free(cpi->mb_row_ei);
michael@0:         vpx_free(cpi->en_thread_data);
michael@0:     }
michael@0: }
michael@0: #endif