vp9: store TileWorkerData allocations separately

move them from VP9Worker::data[12] to allow the structure to be reused a
bit more naturally by the multi-threaded loopfilter.

Change-Id: I31b49c9e93ca744fd7f6d6ed8696671188fb2c1d

3 files changed, 17 insertions, 7 deletions

diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index dc712f045..08f51827e 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -1065,14 +1065,19 @@ static const uint8_t *decode_tiles_mt(VP9Decoder *pbi,
     // 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");
@@ -1082,8 +1087,11 @@ static const uint8_t *decode_tiles_mt(VP9Decoder *pbi,
 
   // 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]
diff --git a/vp9/decoder/vp9_decoder.c b/vp9/decoder/vp9_decoder.c
index 3c9469c4c..9326f5cde 100644
--- a/vp9/decoder/vp9_decoder.c
+++ b/vp9/decoder/vp9_decoder.c
@@ -94,9 +94,9 @@ void vp9_decoder_remove(VP9Decoder *pbi) {
   for (i = 0; i < pbi->num_tile_workers; ++i) {
     VP9Worker *const worker = &pbi->tile_workers[i];
     vp9_get_worker_interface()->end(worker);
-    vpx_free(worker->data1);
-    vpx_free(worker->data2);
   }
+  vpx_free(pbi->tile_worker_data);
+  vpx_free(pbi->tile_worker_info);
   vpx_free(pbi->tile_workers);
 
   if (pbi->num_tile_workers > 0) {
diff --git a/vp9/decoder/vp9_decoder.h b/vp9/decoder/vp9_decoder.h
index 4f52bb9c4..25b7339ed 100644
--- a/vp9/decoder/vp9_decoder.h
+++ b/vp9/decoder/vp9_decoder.h
@@ -46,6 +46,8 @@ typedef struct VP9Decoder {
 
   VP9Worker lf_worker;
   VP9Worker *tile_workers;
+  TileWorkerData *tile_worker_data;
+  TileInfo *tile_worker_info;
   int num_tile_workers;
 
   TileData *tile_data;