libvpx: Pull from upstream

source/libvpx/vp9/decoder/vp9_decodeframe.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 109 matching lines @@
   if (is_compound_reference_allowed(cm)) {
     return vp9_read_bit(r) ? (vp9_read_bit(r) ? REFERENCE_MODE_SELECT
                                               : COMPOUND_REFERENCE)
                            : SINGLE_REFERENCE;
   } else {
     return SINGLE_REFERENCE;
   }
 }
 
 static void read_frame_reference_mode_probs(VP9_COMMON *cm, vp9_reader *r) {
-  FRAME_CONTEXT *const fc = &cm->fc;
+  FRAME_CONTEXT *const fc = cm->fc;
   int i;
 
   if (cm->reference_mode == REFERENCE_MODE_SELECT)
     for (i = 0; i < COMP_INTER_CONTEXTS; ++i)
       vp9_diff_update_prob(r, &fc->comp_inter_prob[i]);
 
   if (cm->reference_mode != COMPOUND_REFERENCE)
     for (i = 0; i < REF_CONTEXTS; ++i) {
       vp9_diff_update_prob(r, &fc->single_ref_prob[i][0]);
       vp9_diff_update_prob(r, &fc->single_ref_prob[i][1]);
@@ skipping 519 matching lines @@
   *height = vp9_rb_read_literal(rb, 16) + 1;
 }
 
 static void setup_display_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
   cm->display_width = cm->width;
   cm->display_height = cm->height;
   if (vp9_rb_read_bit(rb))
     vp9_read_frame_size(rb, &cm->display_width, &cm->display_height);
 }
 
+static void resize_mv_buffer(VP9_COMMON *cm) {
+  vpx_free(cm->cur_frame->mvs);
+  cm->cur_frame->mi_rows = cm->mi_rows;
+  cm->cur_frame->mi_cols = cm->mi_cols;
+  cm->cur_frame->mvs = (MV_REF *)vpx_calloc(cm->mi_rows * cm->mi_cols,
+                                            sizeof(*cm->cur_frame->mvs));
+}
+
 static void resize_context_buffers(VP9_COMMON *cm, int width, int height) {
 #if CONFIG_SIZE_LIMIT
   if (width > DECODE_WIDTH_LIMIT || height > DECODE_HEIGHT_LIMIT)
     vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
                        "Width and height beyond allowed size.");
 #endif
   if (cm->width != width || cm->height != height) {
     const int new_mi_rows =
         ALIGN_POWER_OF_TWO(height, MI_SIZE_LOG2) >> MI_SIZE_LOG2;
     const int new_mi_cols =
         ALIGN_POWER_OF_TWO(width,  MI_SIZE_LOG2) >> MI_SIZE_LOG2;
 
     // Allocations in vp9_alloc_context_buffers() depend on individual
     // dimensions as well as the overall size.
     if (new_mi_cols > cm->mi_cols || new_mi_rows > cm->mi_rows) {
       if (vp9_alloc_context_buffers(cm, width, height))
         vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                            "Failed to allocate context buffers");
     } else {
       vp9_set_mb_mi(cm, width, height);
     }
     vp9_init_context_buffers(cm);
     cm->width = width;
     cm->height = height;
   }
+  if (cm->cur_frame->mvs == NULL || cm->mi_rows > cm->cur_frame->mi_rows ||
+      cm->mi_cols > cm->cur_frame->mi_cols) {
+    resize_mv_buffer(cm);
+  }
 }
 
 static void setup_frame_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
   int width, height;
   vp9_read_frame_size(rb, &width, &height);
   resize_context_buffers(cm, width, height);
   setup_display_size(cm, rb);
 
   if (vp9_realloc_frame_buffer(
           get_frame_new_buffer(cm), cm->width, cm->height,
@@ skipping 190 matching lines @@
     if (pbi->max_threads > 1 && !winterface->reset(&pbi->lf_worker)) {
       vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
                          "Loop filter thread creation failed");
     }
   }
 
   if (cm->lf.filter_level) {
     LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1;
     // Be sure to sync as we might be resuming after a failed frame decode.
     winterface->sync(&pbi->lf_worker);
-    lf_data->frame_buffer = get_frame_new_buffer(cm);
-    lf_data->cm = cm;
-    vp9_copy(lf_data->planes, pbi->mb.plane);
-    lf_data->stop = 0;
-    lf_data->y_only = 0;
+    vp9_loop_filter_data_reset(lf_data, get_frame_new_buffer(cm), cm,
+                               pbi->mb.plane);
     vp9_loop_filter_frame_init(cm, cm->lf.filter_level);
   }
 
   assert(tile_rows <= 4);
   assert(tile_cols <= (1 << 6));
 
   // Note: this memset assumes above_context[0], [1] and [2]
   // are allocated as part of the same buffer.
   vpx_memset(cm->above_context, 0,
              sizeof(*cm->above_context) * MAX_MB_PLANE * 2 * aligned_cols);
@@ skipping 138 matching lines @@
 
   // TODO(jzern): See if we can remove the restriction of passing in max
   // threads to the decoder.
   if (pbi->num_tile_workers == 0) {
     const int num_threads = pbi->max_threads & ~1;
     int i;
     // TODO(jzern): Allocate one less worker, as in the current code we only
     // use num_threads - 1 workers.
     CHECK_MEM_ERROR(cm, pbi->tile_workers,
                     vpx_malloc(num_threads * sizeof(*pbi->tile_workers)));
+    // Ensure tile data offsets will be properly aligned. This may fail on
+    // platforms without DECLARE_ALIGNED().
+    assert((sizeof(*pbi->tile_worker_data) % 16) == 0);
+    CHECK_MEM_ERROR(cm, pbi->tile_worker_data,
+                    vpx_memalign(32, num_threads *
+                                 sizeof(*pbi->tile_worker_data)));
+    CHECK_MEM_ERROR(cm, pbi->tile_worker_info,
+                    vpx_malloc(num_threads * sizeof(*pbi->tile_worker_info)));
     for (i = 0; i < num_threads; ++i) {
       VP9Worker *const worker = &pbi->tile_workers[i];
       ++pbi->num_tile_workers;
 
       winterface->init(worker);
-      CHECK_MEM_ERROR(cm, worker->data1,
-                      vpx_memalign(32, sizeof(TileWorkerData)));
-      CHECK_MEM_ERROR(cm, worker->data2, vpx_malloc(sizeof(TileInfo)));
       if (i < num_threads - 1 && !winterface->reset(worker)) {
         vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
                            "Tile decoder thread creation failed");
       }
     }
   }
 
   // Reset tile decoding hook
   for (n = 0; n < num_workers; ++n) {
-    winterface->sync(&pbi->tile_workers[n]);
-    pbi->tile_workers[n].hook = (VP9WorkerHook)tile_worker_hook;
+    VP9Worker *const worker = &pbi->tile_workers[n];
+    winterface->sync(worker);
+    worker->hook = (VP9WorkerHook)tile_worker_hook;
+    worker->data1 = &pbi->tile_worker_data[n];
+    worker->data2 = &pbi->tile_worker_info[n];
   }
 
   // Note: this memset assumes above_context[0], [1] and [2]
   // are allocated as part of the same buffer.
   vpx_memset(cm->above_context, 0,
              sizeof(*cm->above_context) * MAX_MB_PLANE * 2 * aligned_mi_cols);
   vpx_memset(cm->above_seg_context, 0,
              sizeof(*cm->above_seg_context) * aligned_mi_cols);
 
   // Load tile data into tile_buffers
@@ skipping 282 matching lines @@
     vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
                        "Invalid header size");
 
   return sz;
 }
 
 static int read_compressed_header(VP9Decoder *pbi, const uint8_t *data,
                                   size_t partition_size) {
   VP9_COMMON *const cm = &pbi->common;
   MACROBLOCKD *const xd = &pbi->mb;
-  FRAME_CONTEXT *const fc = &cm->fc;
+  FRAME_CONTEXT *const fc = cm->fc;
   vp9_reader r;
   int k;
 
   if (vp9_reader_init(&r, data, partition_size, pbi->decrypt_cb,
                       pbi->decrypt_state))
     vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                        "Failed to allocate bool decoder 0");
 
   cm->tx_mode = xd->lossless ? ONLY_4X4 : read_tx_mode(&r);
   if (cm->tx_mode == TX_MODE_SELECT)
@@ skipping 125 matching lines @@
     return;
   }
 
   data += vp9_rb_bytes_read(&rb);
   if (!read_is_valid(data, first_partition_size, data_end))
     vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
                        "Truncated packet or corrupt header length");
 
   init_macroblockd(cm, &pbi->mb);
 
-  if (!cm->error_resilient_mode)
-    set_prev_mi(cm);
-  else
-    cm->prev_mi = NULL;
+  cm->use_prev_frame_mvs = !cm->error_resilient_mode &&
+                           cm->width == cm->last_width &&
+                           cm->height == cm->last_height &&
+                           !cm->intra_only &&
+                           cm->last_show_frame;
 
   setup_plane_dequants(cm, xd, cm->base_qindex);
   vp9_setup_block_planes(xd, cm->subsampling_x, cm->subsampling_y);
 
-  cm->fc = cm->frame_contexts[cm->frame_context_idx];
+  *cm->fc = cm->frame_contexts[cm->frame_context_idx];
   vp9_zero(cm->counts);
   vp9_zero(xd->dqcoeff);
 
   xd->corrupted = 0;
   new_fb->corrupted = read_compressed_header(pbi, data, first_partition_size);
 
   // TODO(jzern): remove frame_parallel_decoding_mode restriction for
   // single-frame tile decoding.
   if (pbi->max_threads > 1 && tile_rows == 1 && tile_cols > 1 &&
       cm->frame_parallel_decoding_mode) {
     *p_data_end = decode_tiles_mt(pbi, data + first_partition_size, data_end);
     if (!xd->corrupted) {
       // If multiple threads are used to decode tiles, then we use those threads
       // to do parallel loopfiltering.
-      vp9_loop_filter_frame_mt(new_fb, pbi, cm, cm->lf.filter_level, 0);
+      vp9_loop_filter_frame_mt(&pbi->lf_row_sync, new_fb, pbi->mb.plane, cm,
+                               pbi->tile_workers, pbi->num_tile_workers,
+                               cm->lf.filter_level, 0);
     }
   } else {
     *p_data_end = decode_tiles(pbi, data + first_partition_size, data_end);
   }
 
   new_fb->corrupted |= xd->corrupted;
 
   if (!new_fb->corrupted) {
     if (!cm->error_resilient_mode && !cm->frame_parallel_decoding_mode) {
       vp9_adapt_coef_probs(cm);
 
       if (!frame_is_intra_only(cm)) {
         vp9_adapt_mode_probs(cm);
         vp9_adapt_mv_probs(cm, cm->allow_high_precision_mv);
       }
     } else {
       debug_check_frame_counts(cm);
     }
   } else {
     vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
                        "Decode failed. Frame data is corrupted.");
   }
 
   if (cm->refresh_frame_context)
-    cm->frame_contexts[cm->frame_context_idx] = cm->fc;
+    cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
 }