Update libvpx snapshot to v0.9.7-p1 (Cayuga).

source/libvpx/vp8/encoder/ethreading.c

 /*
  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 73 matching lines @@
             for (mb_row = ithread + 1; mb_row < cm->mb_rows; mb_row += (cpi->encoding_thread_count + 1))
             {
 
                 int i;
                 int recon_yoffset, recon_uvoffset;
                 int mb_col;
                 int ref_fb_idx = cm->lst_fb_idx;
                 int dst_fb_idx = cm->new_fb_idx;
                 int recon_y_stride = cm->yv12_fb[ref_fb_idx].y_stride;
                 int recon_uv_stride = cm->yv12_fb[ref_fb_idx].uv_stride;
+                int map_index = (mb_row * cm->mb_cols);
                 volatile int *last_row_current_mb_col;
-                INT64 activity_sum = 0;
 
                 tp = cpi->tok + (mb_row * (cm->mb_cols * 16 * 24));
 
                 last_row_current_mb_col = &cpi->mt_current_mb_col[mb_row - 1];
 
                 // reset above block coeffs
                 xd->above_context = cm->above_context;
                 xd->left_context = &mb_row_left_context;
 
                 vp8_zero(mb_row_left_context);
 
                 xd->up_available = (mb_row != 0);
                 recon_yoffset = (mb_row * recon_y_stride * 16);
                 recon_uvoffset = (mb_row * recon_uv_stride * 8);
 
                 cpi->tplist[mb_row].start = tp;
 
                 //printf("Thread mb_row = %d\n", mb_row);
 
+                // Set the mb activity pointer to the start of the row.
+                x->mb_activity_ptr = &cpi->mb_activity_map[map_index];
+
                 // for each macroblock col in image
                 for (mb_col = 0; mb_col < cm->mb_cols; mb_col++)
                 {
-                    int seg_map_index = (mb_row * cm->mb_cols);
-
                     if ((mb_col & (nsync - 1)) == 0)
                     {
                         while (mb_col > (*last_row_current_mb_col - nsync) && *last_row_current_mb_col != cm->mb_cols - 1)
                         {
                             x86_pause_hint();
                             thread_sleep(0);
                         }
                     }
 
                     // Distance of Mb to the various image edges.
@@ skipping 10 matching lines @@
                     x->mv_row_max = ((cm->mb_rows - 1 - mb_row) * 16) + (VP8BORDERINPIXELS - 16);
 
                     xd->dst.y_buffer = cm->yv12_fb[dst_fb_idx].y_buffer + recon_yoffset;
                     xd->dst.u_buffer = cm->yv12_fb[dst_fb_idx].u_buffer + recon_uvoffset;
                     xd->dst.v_buffer = cm->yv12_fb[dst_fb_idx].v_buffer + recon_uvoffset;
                     xd->left_available = (mb_col != 0);
 
                     x->rddiv = cpi->RDDIV;
                     x->rdmult = cpi->RDMULT;
 
+                    //Copy current mb to a buffer
+                    RECON_INVOKE(&xd->rtcd->recon, copy16x16)(x->src.y_buffer, x->src.y_stride, x->thismb, 16);
+
                     if (cpi->oxcf.tuning == VP8_TUNE_SSIM)
-                        activity_sum += vp8_activity_masking(cpi, x);
+                        vp8_activity_masking(cpi, x);
 
                     // Is segmentation enabled
                     // MB level adjutment to quantizer
                     if (xd->segmentation_enabled)
                     {
                         // Code to set segment id in xd->mbmi.segment_id for current MB (with range checking)
-                        if (cpi->segmentation_map[seg_map_index + mb_col] <= 3)
-                            xd->mode_info_context->mbmi.segment_id = cpi->segmentation_map[seg_map_index + mb_col];
+                        if (cpi->segmentation_map[map_index + mb_col] <= 3)
+                            xd->mode_info_context->mbmi.segment_id = cpi->segmentation_map[map_index + mb_col];
                         else
                             xd->mode_info_context->mbmi.segment_id = 0;
 
                         vp8cx_mb_init_quantizer(cpi, x);
                     }
                     else
                         xd->mode_info_context->mbmi.segment_id = 0; // Set to Segment 0 by default
 
-                    x->active_ptr = cpi->active_map + seg_map_index + mb_col;
+                    x->active_ptr = cpi->active_map + map_index + mb_col;
 
                     if (cm->frame_type == KEY_FRAME)
                     {
                         *totalrate += vp8cx_encode_intra_macro_block(cpi, x, &tp);
 #ifdef MODE_STATS
                         y_modes[xd->mbmi.mode] ++;
 #endif
                     }
                     else
                     {
@@ skipping 17 matching lines @@
                         // Count of last ref frame 0,0 useage
                         if ((xd->mode_info_context->mbmi.mode == ZEROMV) && (xd->mode_info_context->mbmi.ref_frame == LAST_FRAME))
                             cpi->inter_zz_count++;
 
                         // Special case code for cyclic refresh
                         // If cyclic update enabled then copy xd->mbmi.segment_id; (which may have been updated based on mode
                         // during vp8cx_encode_inter_macroblock()) back into the global sgmentation map
                         if (cpi->cyclic_refresh_mode_enabled && xd->segmentation_enabled)
                         {
                             const MB_MODE_INFO * mbmi = &xd->mode_info_context->mbmi;
-                            cpi->segmentation_map[seg_map_index + mb_col] = mbmi->segment_id;
+                            cpi->segmentation_map[map_index + mb_col] = mbmi->segment_id;
 
                             // If the block has been refreshed mark it as clean (the magnitude of the -ve influences how long it will be before we consider another refresh):
                             // Else if it was coded (last frame 0,0) and has not already been refreshed then mark it as a candidate for cleanup next time (marked 0)
                             // else mark it as dirty (1).
                             if (mbmi->segment_id)
-                                cpi->cyclic_refresh_map[seg_map_index + mb_col] = -1;
+                                cpi->cyclic_refresh_map[map_index + mb_col] = -1;
                             else if ((mbmi->mode == ZEROMV) && (mbmi->ref_frame == LAST_FRAME))
                             {
-                                if (cpi->cyclic_refresh_map[seg_map_index + mb_col] == 1)
-                                    cpi->cyclic_refresh_map[seg_map_index + mb_col] = 0;
+                                if (cpi->cyclic_refresh_map[map_index + mb_col] == 1)
+                                    cpi->cyclic_refresh_map[map_index + mb_col] = 0;
                             }
                             else
-                                cpi->cyclic_refresh_map[seg_map_index + mb_col] = 1;
+                                cpi->cyclic_refresh_map[map_index + mb_col] = 1;
 
                         }
                     }
                     cpi->tplist[mb_row].stop = tp;
 
-                    x->gf_active_ptr++; // Increment pointer into gf useage flags structure for next mb
+                    // Increment pointer into gf useage flags structure.
+                    x->gf_active_ptr++;
 
+                    // Increment the activity mask pointers.
+                    x->mb_activity_ptr++;
+
+                    /* save the block info */
                     for (i = 0; i < 16; i++)
-                        vpx_memcpy(&xd->mode_info_context->bmi[i], &xd->block[i].bmi, sizeof(xd->block[i].bmi));
+                        xd->mode_info_context->bmi[i] = xd->block[i].bmi;
 
                     // adjust to the next column of macroblocks
                     x->src.y_buffer += 16;
                     x->src.u_buffer += 8;
                     x->src.v_buffer += 8;
 
                     recon_yoffset += 16;
                     recon_uvoffset += 8;
 
                     // Keep track of segment useage
@@ skipping 10 matching lines @@
                 //extend the recon for intra prediction
                 vp8_extend_mb_row(
                     &cm->yv12_fb[dst_fb_idx],
                     xd->dst.y_buffer + 16,
                     xd->dst.u_buffer + 8,
                     xd->dst.v_buffer + 8);
 
                 // this is to account for the border
                 xd->mode_info_context++;
                 x->partition_info++;
-                x->activity_sum += activity_sum;
 
                 x->src.y_buffer += 16 * x->src.y_stride * (cpi->encoding_thread_count + 1) - 16 * cm->mb_cols;
                 x->src.u_buffer += 8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols;
                 x->src.v_buffer += 8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols;
 
                 xd->mode_info_context += xd->mode_info_stride * cpi->encoding_thread_count;
                 x->partition_info += xd->mode_info_stride * cpi->encoding_thread_count;
+                x->gf_active_ptr   += cm->mb_cols * cpi->encoding_thread_count;
 
                 if (mb_row == cm->mb_rows - 1)
                 {
                     //SetEvent(cpi->h_event_main);
                     sem_post(&cpi->h_event_end_encoding); /* signal frame encoding end */
                 }
             }
         }
     }
 
     //printf("exit thread %d\n", ithread);
     return 0;
 }
 
 static void setup_mbby_copy(MACROBLOCK *mbdst, MACROBLOCK *mbsrc)
 {
 
     MACROBLOCK *x = mbsrc;
     MACROBLOCK *z = mbdst;
     int i;
 
     z->ss               = x->ss;
     z->ss_count          = x->ss_count;
     z->searches_per_step  = x->searches_per_step;
     z->errorperbit      = x->errorperbit;
 
     z->sadperbit16      = x->sadperbit16;
     z->sadperbit4       = x->sadperbit4;
-    z->errthresh        = x->errthresh;
 
     /*
     z->mv_col_min    = x->mv_col_min;
     z->mv_col_max    = x->mv_col_max;
     z->mv_row_min    = x->mv_row_min;
     z->mv_row_max    = x->mv_row_max;
     z->vector_range = x->vector_range ;
     */
 
     z->vp8_short_fdct4x4     = x->vp8_short_fdct4x4;
     z->vp8_short_fdct8x4     = x->vp8_short_fdct8x4;
     z->short_walsh4x4    = x->short_walsh4x4;
     z->quantize_b        = x->quantize_b;
+    z->quantize_b_pair   = x->quantize_b_pair;
     z->optimize          = x->optimize;
 
     /*
     z->mvc              = x->mvc;
     z->src.y_buffer      = x->src.y_buffer;
     z->src.u_buffer      = x->src.u_buffer;
     z->src.v_buffer      = x->src.v_buffer;
     */
 
 
     vpx_memcpy(z->mvcosts,          x->mvcosts,         sizeof(x->mvcosts));
     z->mvcost[0] = &z->mvcosts[0][mv_max+1];
     z->mvcost[1] = &z->mvcosts[1][mv_max+1];
-    z->mvsadcost[0] = &z->mvsadcosts[0][mv_max+1];
-    z->mvsadcost[1] = &z->mvsadcosts[1][mv_max+1];
+    z->mvsadcost[0] = &z->mvsadcosts[0][mvfp_max+1];
+    z->mvsadcost[1] = &z->mvsadcosts[1][mvfp_max+1];
 
 
     vpx_memcpy(z->token_costs,       x->token_costs,      sizeof(x->token_costs));
     vpx_memcpy(z->inter_bmode_costs,  x->inter_bmode_costs, sizeof(x->inter_bmode_costs));
     //memcpy(z->mvcosts,            x->mvcosts,         sizeof(x->mvcosts));
     //memcpy(z->mvcost,         x->mvcost,          sizeof(x->mvcost));
     vpx_memcpy(z->mbmode_cost,       x->mbmode_cost,      sizeof(x->mbmode_cost));
     vpx_memcpy(z->intra_uv_mode_cost,  x->intra_uv_mode_cost, sizeof(x->intra_uv_mode_cost));
     vpx_memcpy(z->bmode_costs,       x->bmode_costs,      sizeof(x->bmode_costs));
 
@@ skipping 102 matching lines @@
         mb->src.y_buffer += 16 * x->src.y_stride * (i + 1);
         mb->src.u_buffer +=  8 * x->src.uv_stride * (i + 1);
         mb->src.v_buffer +=  8 * x->src.uv_stride * (i + 1);
 
         vp8_build_block_offsets(mb);
 
         vp8_setup_block_dptrs(mbd);
 
         vp8_setup_block_ptrs(mb);
 
-        mb->activity_sum = 0;
-
         mbd->left_context = &cm->left_context;
         mb->mvc = cm->fc.mvc;
 
         setup_mbby_copy(&mbr_ei[i].mb, x);
 
     }
 }
 
 void vp8cx_create_encoder_threads(VP8_COMP *cpi)
 {
     const VP8_COMMON * cm = &cpi->common;
 
     cpi->b_multi_threaded = 0;
     cpi->encoding_thread_count = 0;
-    cpi->processor_core_count = 32; //vp8_get_proc_core_count();
 
-    if (cpi->processor_core_count > 1 && cpi->oxcf.multi_threaded > 1)
+    if (cm->processor_core_count > 1 && cpi->oxcf.multi_threaded > 1)
     {
         int ithread;
         int th_count = cpi->oxcf.multi_threaded - 1;
 
-        if (cpi->oxcf.multi_threaded > cpi->processor_core_count)
-            th_count = cpi->processor_core_count - 1;
+        /* don't allocate more threads than cores available */
+        if (cpi->oxcf.multi_threaded > cm->processor_core_count)
+            th_count = cm->processor_core_count - 1;
 
         /* we have th_count + 1 (main) threads processing one row each */
         /* no point to have more threads than the sync range allows */
         if(th_count > ((cm->mb_cols / cpi->mt_sync_range) - 1))
         {
             th_count = (cm->mb_cols / cpi->mt_sync_range) - 1;
         }
 
         if(th_count == 0)
             return;
@@ skipping 70 matching lines @@
 
         //free thread related resources
         vpx_free(cpi->h_event_start_encoding);
         vpx_free(cpi->h_encoding_thread);
         vpx_free(cpi->mb_row_ei);
         vpx_free(cpi->en_thread_data);
         vpx_free(cpi->mt_current_mb_col);
     }
 }
 #endif