| OLD | NEW |
| 1 /* | 1 /* |
| 2 * Copyright (c) 2014 The WebM project authors. All Rights Reserved. | 2 * Copyright (c) 2014 The WebM project authors. All Rights Reserved. |
| 3 * | 3 * |
| 4 * Use of this source code is governed by a BSD-style license | 4 * Use of this source code is governed by a BSD-style license |
| 5 * that can be found in the LICENSE file in the root of the source | 5 * that can be found in the LICENSE file in the root of the source |
| 6 * tree. An additional intellectual property rights grant can be found | 6 * tree. An additional intellectual property rights grant can be found |
| 7 * in the file PATENTS. All contributing project authors may | 7 * in the file PATENTS. All contributing project authors may |
| 8 * be found in the AUTHORS file in the root of the source tree. | 8 * be found in the AUTHORS file in the root of the source tree. |
| 9 */ | 9 */ |
| 10 | 10 |
| (...skipping 106 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 117 | 117 |
| 118 sync_write(lf_sync, r, c, sb_cols); | 118 sync_write(lf_sync, r, c, sb_cols); |
| 119 } | 119 } |
| 120 } | 120 } |
| 121 } | 121 } |
| 122 | 122 |
| 123 // Row-based multi-threaded loopfilter hook | 123 // Row-based multi-threaded loopfilter hook |
| 124 static int loop_filter_row_worker(void *arg1, void *arg2) { | 124 static int loop_filter_row_worker(void *arg1, void *arg2) { |
| 125 TileWorkerData *const tile_data = (TileWorkerData*)arg1; | 125 TileWorkerData *const tile_data = (TileWorkerData*)arg1; |
| 126 LFWorkerData *const lf_data = &tile_data->lfdata; | 126 LFWorkerData *const lf_data = &tile_data->lfdata; |
| 127 | 127 (void) arg2; |
| 128 loop_filter_rows_mt(lf_data->frame_buffer, lf_data->cm, lf_data->planes, | 128 loop_filter_rows_mt(lf_data->frame_buffer, lf_data->cm, lf_data->planes, |
| 129 lf_data->start, lf_data->stop, lf_data->y_only, | 129 lf_data->start, lf_data->stop, lf_data->y_only, |
| 130 lf_data->lf_sync, lf_data->num_lf_workers); | 130 lf_data->lf_sync, lf_data->num_lf_workers); |
| 131 return 1; | 131 return 1; |
| 132 } | 132 } |
| 133 | 133 |
| 134 // VP9 decoder: Implement multi-threaded loopfilter that uses the tile | 134 // VP9 decoder: Implement multi-threaded loopfilter that uses the tile |
| 135 // threads. | 135 // threads. |
| 136 void vp9_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame, | 136 void vp9_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame, |
| 137 VP9Decoder *pbi, VP9_COMMON *cm, | 137 VP9Decoder *pbi, VP9_COMMON *cm, |
| 138 int frame_filter_level, | 138 int frame_filter_level, |
| 139 int y_only) { | 139 int y_only) { |
| 140 VP9LfSync *const lf_sync = &pbi->lf_row_sync; | 140 VP9LfSync *const lf_sync = &pbi->lf_row_sync; |
| 141 const VP9WorkerInterface *const winterface = vp9_get_worker_interface(); |
| 141 // Number of superblock rows and cols | 142 // Number of superblock rows and cols |
| 142 const int sb_rows = mi_cols_aligned_to_sb(cm->mi_rows) >> MI_BLOCK_SIZE_LOG2; | 143 const int sb_rows = mi_cols_aligned_to_sb(cm->mi_rows) >> MI_BLOCK_SIZE_LOG2; |
| 143 const int tile_cols = 1 << cm->log2_tile_cols; | 144 const int tile_cols = 1 << cm->log2_tile_cols; |
| 144 const int num_workers = MIN(pbi->max_threads & ~1, tile_cols); | 145 const int num_workers = MIN(pbi->max_threads & ~1, tile_cols); |
| 145 int i; | 146 int i; |
| 146 | 147 |
| 147 // Allocate memory used in thread synchronization. | 148 // Allocate memory used in thread synchronization. |
| 148 // This always needs to be done even if frame_filter_level is 0. | 149 // This always needs to be done even if frame_filter_level is 0. |
| 149 if (!cm->current_video_frame || cm->last_height != cm->height) { | 150 if (!cm->current_video_frame || cm->last_height != cm->height) { |
| 150 if (cm->last_height != cm->height) { | 151 if (cm->last_height != cm->height) { |
| (...skipping 39 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 190 vp9_copy(lf_data->planes, pbi->mb.plane); | 191 vp9_copy(lf_data->planes, pbi->mb.plane); |
| 191 lf_data->start = i; | 192 lf_data->start = i; |
| 192 lf_data->stop = sb_rows; | 193 lf_data->stop = sb_rows; |
| 193 lf_data->y_only = y_only; // always do all planes in decoder | 194 lf_data->y_only = y_only; // always do all planes in decoder |
| 194 | 195 |
| 195 lf_data->lf_sync = lf_sync; | 196 lf_data->lf_sync = lf_sync; |
| 196 lf_data->num_lf_workers = num_workers; | 197 lf_data->num_lf_workers = num_workers; |
| 197 | 198 |
| 198 // Start loopfiltering | 199 // Start loopfiltering |
| 199 if (i == num_workers - 1) { | 200 if (i == num_workers - 1) { |
| 200 vp9_worker_execute(worker); | 201 winterface->execute(worker); |
| 201 } else { | 202 } else { |
| 202 vp9_worker_launch(worker); | 203 winterface->launch(worker); |
| 203 } | 204 } |
| 204 } | 205 } |
| 205 | 206 |
| 206 // Wait till all rows are finished | 207 // Wait till all rows are finished |
| 207 for (i = 0; i < num_workers; ++i) { | 208 for (i = 0; i < num_workers; ++i) { |
| 208 vp9_worker_sync(&pbi->tile_workers[i]); | 209 winterface->sync(&pbi->tile_workers[i]); |
| 209 } | 210 } |
| 210 } | 211 } |
| 211 | 212 |
| 212 // Set up nsync by width. | 213 // Set up nsync by width. |
| 213 static int get_sync_range(int width) { | 214 static int get_sync_range(int width) { |
| 214 // nsync numbers are picked by testing. For example, for 4k | 215 // nsync numbers are picked by testing. For example, for 4k |
| 215 // video, using 4 gives best performance. | 216 // video, using 4 gives best performance. |
| 216 if (width < 640) | 217 if (width < 640) |
| 217 return 1; | 218 return 1; |
| 218 else if (width <= 1280) | 219 else if (width <= 1280) |
| (...skipping 52 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 271 } | 272 } |
| 272 vpx_free(lf_sync->cond_); | 273 vpx_free(lf_sync->cond_); |
| 273 } | 274 } |
| 274 #endif // CONFIG_MULTITHREAD | 275 #endif // CONFIG_MULTITHREAD |
| 275 vpx_free(lf_sync->cur_sb_col); | 276 vpx_free(lf_sync->cur_sb_col); |
| 276 // clear the structure as the source of this call may be a resize in which | 277 // clear the structure as the source of this call may be a resize in which |
| 277 // case this call will be followed by an _alloc() which may fail. | 278 // case this call will be followed by an _alloc() which may fail. |
| 278 vp9_zero(*lf_sync); | 279 vp9_zero(*lf_sync); |
| 279 } | 280 } |
| 280 } | 281 } |
| OLD | NEW |