Merge "vp9: sync workers at the start of decode_tiles_mt()"
diff --git a/test/invalid_file_test.cc b/test/invalid_file_test.cc
index 34d6236..cf86cf5 100644
--- a/test/invalid_file_test.cc
+++ b/test/invalid_file_test.cc
@@ -144,6 +144,10 @@
const DecodeParam kMultiThreadedVP9InvalidFileTests[] = {
{4, "invalid-vp90-2-08-tile_1x4_frame_parallel_all_key.webm"},
+ {4, "invalid-"
+ "vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf"},
+ {2, "invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf"},
+ {4, "invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf"},
};
INSTANTIATE_TEST_CASE_P(
diff --git a/test/test-data.sha1 b/test/test-data.sha1
index ee6289f..d2bc5a8 100644
--- a/test/test-data.sha1
+++ b/test/test-data.sha1
@@ -681,3 +681,9 @@
c77e4a26616add298a05dd5d12397be22c0e40c5 vp90-2-18-resize.ivf
c12918cf0a716417fba2de35c3fc5ab90e52dfce vp90-2-18-resize.ivf.md5
717da707afcaa1f692ff1946f291054eb75a4f06 screendata.y4m
+b7c1296630cdf1a7ef493d15ff4f9eb2999202f6 invalid-vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf
+0a3884edb3fd8f9d9b500223e650f7de257b67d8 invalid-vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf.res
+fac89b5735be8a86b0dc05159f996a5c3208ae32 invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf
+22e0ee8babe574722baf4ef6d7ff5d7cf80d386c invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf.res
+4506dfdcdf8ee4250924b075a0dcf1f070f72e5a invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf
+d3ea592c8d7b05d14c7ed48befc0a3aaf7709b7a invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf.res
diff --git a/test/test.mk b/test/test.mk
index 0814c2b..e592a70 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -801,8 +801,14 @@
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-00-quantizer-11.webm.ivf.s52984_r01-05_b6-.ivf.res
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-00-quantizer-11.webm.ivf.s52984_r01-05_b6-z.ivf
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-00-quantizer-11.webm.ivf.s52984_r01-05_b6-z.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf.res
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-08-tile_1x4_frame_parallel_all_key.webm
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-08-tile_1x4_frame_parallel_all_key.webm.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf.res
ifeq ($(CONFIG_DECODE_PERF_TESTS),yes)
# BBB VP9 streams
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index 2939e5f..aae56c1 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -813,6 +813,8 @@
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);
diff --git a/vp9/decoder/vp9_decoder.c b/vp9/decoder/vp9_decoder.c
index e79dcf3..ae0da79 100644
--- a/vp9/decoder/vp9_decoder.c
+++ b/vp9/decoder/vp9_decoder.c
@@ -96,10 +96,8 @@
}
vpx_free(pbi->tile_workers);
- if (pbi->num_tile_workers) {
- const int sb_rows =
- mi_cols_aligned_to_sb(cm->mi_rows) >> MI_BLOCK_SIZE_LOG2;
- vp9_loop_filter_dealloc(&pbi->lf_row_sync, sb_rows);
+ if (pbi->num_tile_workers > 0) {
+ vp9_loop_filter_dealloc(&pbi->lf_row_sync);
}
vp9_remove_common(cm);
diff --git a/vp9/decoder/vp9_dthread.c b/vp9/decoder/vp9_dthread.c
index 5dda49a..b82ea6a 100644
--- a/vp9/decoder/vp9_dthread.c
+++ b/vp9/decoder/vp9_dthread.c
@@ -147,17 +147,8 @@
// Allocate memory used in thread synchronization.
// This always needs to be done even if frame_filter_level is 0.
- if (!cm->current_video_frame || cm->last_height != cm->height) {
- if (cm->last_height != cm->height) {
- const int aligned_last_height =
- ALIGN_POWER_OF_TWO(cm->last_height, MI_SIZE_LOG2);
- const int last_sb_rows =
- mi_cols_aligned_to_sb(aligned_last_height >> MI_SIZE_LOG2) >>
- MI_BLOCK_SIZE_LOG2;
-
- vp9_loop_filter_dealloc(lf_sync, last_sb_rows);
- }
-
+ if (!lf_sync->sync_range || cm->last_height != cm->height) {
+ vp9_loop_filter_dealloc(lf_sync);
vp9_loop_filter_alloc(cm, lf_sync, sb_rows, cm->width);
}
@@ -227,19 +218,22 @@
// Allocate memory for lf row synchronization
void vp9_loop_filter_alloc(VP9_COMMON *cm, VP9LfSync *lf_sync, int rows,
int width) {
+ lf_sync->rows = rows;
#if CONFIG_MULTITHREAD
- int i;
+ {
+ int i;
- CHECK_MEM_ERROR(cm, lf_sync->mutex_,
- vpx_malloc(sizeof(*lf_sync->mutex_) * rows));
- for (i = 0; i < rows; ++i) {
- pthread_mutex_init(&lf_sync->mutex_[i], NULL);
- }
+ CHECK_MEM_ERROR(cm, lf_sync->mutex_,
+ vpx_malloc(sizeof(*lf_sync->mutex_) * rows));
+ for (i = 0; i < rows; ++i) {
+ pthread_mutex_init(&lf_sync->mutex_[i], NULL);
+ }
- CHECK_MEM_ERROR(cm, lf_sync->cond_,
- vpx_malloc(sizeof(*lf_sync->cond_) * rows));
- for (i = 0; i < rows; ++i) {
- pthread_cond_init(&lf_sync->cond_[i], NULL);
+ CHECK_MEM_ERROR(cm, lf_sync->cond_,
+ vpx_malloc(sizeof(*lf_sync->cond_) * rows));
+ for (i = 0; i < rows; ++i) {
+ pthread_cond_init(&lf_sync->cond_[i], NULL);
+ }
}
#endif // CONFIG_MULTITHREAD
@@ -251,23 +245,19 @@
}
// Deallocate lf synchronization related mutex and data
-void vp9_loop_filter_dealloc(VP9LfSync *lf_sync, int rows) {
-#if !CONFIG_MULTITHREAD
- (void)rows;
-#endif // !CONFIG_MULTITHREAD
-
+void vp9_loop_filter_dealloc(VP9LfSync *lf_sync) {
if (lf_sync != NULL) {
#if CONFIG_MULTITHREAD
int i;
if (lf_sync->mutex_ != NULL) {
- for (i = 0; i < rows; ++i) {
+ for (i = 0; i < lf_sync->rows; ++i) {
pthread_mutex_destroy(&lf_sync->mutex_[i]);
}
vpx_free(lf_sync->mutex_);
}
if (lf_sync->cond_ != NULL) {
- for (i = 0; i < rows; ++i) {
+ for (i = 0; i < lf_sync->rows; ++i) {
pthread_cond_destroy(&lf_sync->cond_[i]);
}
vpx_free(lf_sync->cond_);
diff --git a/vp9/decoder/vp9_dthread.h b/vp9/decoder/vp9_dthread.h
index 423bd88..8b02ef7 100644
--- a/vp9/decoder/vp9_dthread.h
+++ b/vp9/decoder/vp9_dthread.h
@@ -38,6 +38,7 @@
// The optimal sync_range for different resolution and platform should be
// determined by testing. Currently, it is chosen to be a power-of-2 number.
int sync_range;
+ int rows;
} VP9LfSync;
// Allocate memory for loopfilter row synchronization.
@@ -45,7 +46,7 @@
int rows, int width);
// Deallocate loopfilter synchronization related mutex and data.
-void vp9_loop_filter_dealloc(VP9LfSync *lf_sync, int rows);
+void vp9_loop_filter_dealloc(VP9LfSync *lf_sync);
// Multi-threaded loopfilter that uses the tile threads.
void vp9_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame,
diff --git a/vp9/encoder/vp9_encoder.h b/vp9/encoder/vp9_encoder.h
index 9f82a5a..b26c8c7 100644
--- a/vp9/encoder/vp9_encoder.h
+++ b/vp9/encoder/vp9_encoder.h
@@ -390,7 +390,7 @@
search_site_config ss_cfg;
int mbmode_cost[INTRA_MODES];
- unsigned inter_mode_cost[INTER_MODE_CONTEXTS][INTER_MODES];
+ unsigned int inter_mode_cost[INTER_MODE_CONTEXTS][INTER_MODES];
int intra_uv_mode_cost[FRAME_TYPES][INTRA_MODES];
int y_mode_costs[INTRA_MODES][INTRA_MODES][INTRA_MODES];
int switchable_interp_costs[SWITCHABLE_FILTER_CONTEXTS][SWITCHABLE_FILTERS];
