libvpx: Pull from upstream

source/libvpx/vp9/common/vp9_thread_common.c

 /*
  *  Copyright (c) 2014 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.
  */
 
 #include "./vpx_config.h"
 #include "vpx_mem/vpx_mem.h"
 #include "vp9/common/vp9_entropymode.h"
 #include "vp9/common/vp9_thread_common.h"
 #include "vp9/common/vp9_reconinter.h"
+#include "vp9/common/vp9_loopfilter.h"
 
 #if CONFIG_MULTITHREAD
 static INLINE void mutex_lock(pthread_mutex_t *const mutex) {
   const int kMaxTryLocks = 4000;
   int locked = 0;
   int i;
 
   for (i = 0; i < kMaxTryLocks; ++i) {
     if (!pthread_mutex_trylock(mutex)) {
       locked = 1;
@@ skipping 59 matching lines @@
 }
 
 // Implement row loopfiltering for each thread.
 static INLINE
 void thread_loop_filter_rows(const YV12_BUFFER_CONFIG *const frame_buffer,
                              VP9_COMMON *const cm,
                              struct macroblockd_plane planes[MAX_MB_PLANE],
                              int start, int stop, int y_only,
                              VP9LfSync *const lf_sync) {
   const int num_planes = y_only ? 1 : MAX_MB_PLANE;
-  const int use_420 = y_only || (planes[1].subsampling_y == 1 &&
-                                 planes[1].subsampling_x == 1);
   const int sb_cols = mi_cols_aligned_to_sb(cm->mi_cols) >> MI_BLOCK_SIZE_LOG2;
   int mi_row, mi_col;
+  enum lf_path path;
+  if (y_only)
+    path = LF_PATH_444;
+  else if (planes[1].subsampling_y == 1 && planes[1].subsampling_x == 1)
+    path = LF_PATH_420;
+  else if (planes[1].subsampling_y == 0 && planes[1].subsampling_x == 0)
+    path = LF_PATH_444;
+  else
+    path = LF_PATH_SLOW;
 
   for (mi_row = start; mi_row < stop;
        mi_row += lf_sync->num_workers * MI_BLOCK_SIZE) {
-    MODE_INFO *const mi = cm->mi + mi_row * cm->mi_stride;
+    MODE_INFO **const mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
 
     for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MI_BLOCK_SIZE) {
       const int r = mi_row >> MI_BLOCK_SIZE_LOG2;
       const int c = mi_col >> MI_BLOCK_SIZE_LOG2;
       LOOP_FILTER_MASK lfm;
       int plane;
 
       sync_read(lf_sync, r, c);
 
       vp9_setup_dst_planes(planes, frame_buffer, mi_row, mi_col);
 
       // TODO(JBB): Make setup_mask work for non 420.
-      if (use_420)
-        vp9_setup_mask(cm, mi_row, mi_col, mi + mi_col, cm->mi_stride,
-                       &lfm);
+      vp9_setup_mask(cm, mi_row, mi_col, mi + mi_col, cm->mi_stride,
+                     &lfm);
 
-      for (plane = 0; plane < num_planes; ++plane) {
-        if (use_420)
-          vp9_filter_block_plane(cm, &planes[plane], mi_row, &lfm);
-        else
-          vp9_filter_block_plane_non420(cm, &planes[plane], mi + mi_col,
-                                        mi_row, mi_col);
+      vp9_filter_block_plane_ss00(cm, &planes[0], mi_row, &lfm);
+      for (plane = 1; plane < num_planes; ++plane) {
+        switch (path) {
+          case LF_PATH_420:
+            vp9_filter_block_plane_ss11(cm, &planes[plane], mi_row, &lfm);
+            break;
+          case LF_PATH_444:
+            vp9_filter_block_plane_ss00(cm, &planes[plane], mi_row, &lfm);
+            break;
+          case LF_PATH_SLOW:
+            vp9_filter_block_plane_non420(cm, &planes[plane], mi + mi_col,
+                                          mi_row, mi_col);
+            break;
+        }
       }
 
       sync_write(lf_sync, r, c, sb_cols);
     }
   }
 }
 
 // Row-based multi-threaded loopfilter hook
 static int loop_filter_row_worker(VP9LfSync *const lf_sync,
                                   LFWorkerData *const lf_data) {
@@ skipping 18 matching lines @@
   const int num_workers = MIN(nworkers, tile_cols);
   int i;
 
   if (!lf_sync->sync_range || sb_rows != lf_sync->rows ||
       num_workers > lf_sync->num_workers) {
     vp9_loop_filter_dealloc(lf_sync);
     vp9_loop_filter_alloc(lf_sync, cm, sb_rows, cm->width, num_workers);
   }
 
   // Initialize cur_sb_col to -1 for all SB rows.
-  vpx_memset(lf_sync->cur_sb_col, -1, sizeof(*lf_sync->cur_sb_col) * sb_rows);
+  memset(lf_sync->cur_sb_col, -1, sizeof(*lf_sync->cur_sb_col) * sb_rows);
 
   // Set up loopfilter thread data.
   // The decoder is capping num_workers because it has been observed that using
   // more threads on the loopfilter than there are cores will hurt performance
   // on Android. This is because the system will only schedule the tile decode
   // workers on cores equal to the number of tile columns. Then if the decoder
   // tries to use more threads for the loopfilter, it will hurt performance
   // because of contention. If the multithreading code changes in the future
   // then the number of workers used by the loopfilter should be revisited.
   for (i = 0; i < num_workers; ++i) {
@@ skipping 238 matching lines @@
     }
 
     for (i = 0; i < MV_OFFSET_BITS; i++)
       for (j = 0; j < 2; j++)
         comps->bits[i][j] += comps_t->bits[i][j];
 
     for (i = 0; i < MV_FP_SIZE; i++)
       comps->fp[i] += comps_t->fp[i];
   }
 }