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 |
11 #include "./vpx_config.h" | 11 #include "./vpx_config.h" |
12 #include "vpx_mem/vpx_mem.h" | 12 #include "vpx_mem/vpx_mem.h" |
13 #include "vp9/common/vp9_entropymode.h" | 13 #include "vp9/common/vp9_entropymode.h" |
14 #include "vp9/common/vp9_thread_common.h" | 14 #include "vp9/common/vp9_thread_common.h" |
15 #include "vp9/common/vp9_reconinter.h" | 15 #include "vp9/common/vp9_reconinter.h" |
| 16 #include "vp9/common/vp9_loopfilter.h" |
16 | 17 |
17 #if CONFIG_MULTITHREAD | 18 #if CONFIG_MULTITHREAD |
18 static INLINE void mutex_lock(pthread_mutex_t *const mutex) { | 19 static INLINE void mutex_lock(pthread_mutex_t *const mutex) { |
19 const int kMaxTryLocks = 4000; | 20 const int kMaxTryLocks = 4000; |
20 int locked = 0; | 21 int locked = 0; |
21 int i; | 22 int i; |
22 | 23 |
23 for (i = 0; i < kMaxTryLocks; ++i) { | 24 for (i = 0; i < kMaxTryLocks; ++i) { |
24 if (!pthread_mutex_trylock(mutex)) { | 25 if (!pthread_mutex_trylock(mutex)) { |
25 locked = 1; | 26 locked = 1; |
(...skipping 59 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
85 } | 86 } |
86 | 87 |
87 // Implement row loopfiltering for each thread. | 88 // Implement row loopfiltering for each thread. |
88 static INLINE | 89 static INLINE |
89 void thread_loop_filter_rows(const YV12_BUFFER_CONFIG *const frame_buffer, | 90 void thread_loop_filter_rows(const YV12_BUFFER_CONFIG *const frame_buffer, |
90 VP9_COMMON *const cm, | 91 VP9_COMMON *const cm, |
91 struct macroblockd_plane planes[MAX_MB_PLANE], | 92 struct macroblockd_plane planes[MAX_MB_PLANE], |
92 int start, int stop, int y_only, | 93 int start, int stop, int y_only, |
93 VP9LfSync *const lf_sync) { | 94 VP9LfSync *const lf_sync) { |
94 const int num_planes = y_only ? 1 : MAX_MB_PLANE; | 95 const int num_planes = y_only ? 1 : MAX_MB_PLANE; |
95 const int use_420 = y_only || (planes[1].subsampling_y == 1 && | |
96 planes[1].subsampling_x == 1); | |
97 const int sb_cols = mi_cols_aligned_to_sb(cm->mi_cols) >> MI_BLOCK_SIZE_LOG2; | 96 const int sb_cols = mi_cols_aligned_to_sb(cm->mi_cols) >> MI_BLOCK_SIZE_LOG2; |
98 int mi_row, mi_col; | 97 int mi_row, mi_col; |
| 98 enum lf_path path; |
| 99 if (y_only) |
| 100 path = LF_PATH_444; |
| 101 else if (planes[1].subsampling_y == 1 && planes[1].subsampling_x == 1) |
| 102 path = LF_PATH_420; |
| 103 else if (planes[1].subsampling_y == 0 && planes[1].subsampling_x == 0) |
| 104 path = LF_PATH_444; |
| 105 else |
| 106 path = LF_PATH_SLOW; |
99 | 107 |
100 for (mi_row = start; mi_row < stop; | 108 for (mi_row = start; mi_row < stop; |
101 mi_row += lf_sync->num_workers * MI_BLOCK_SIZE) { | 109 mi_row += lf_sync->num_workers * MI_BLOCK_SIZE) { |
102 MODE_INFO *const mi = cm->mi + mi_row * cm->mi_stride; | 110 MODE_INFO **const mi = cm->mi_grid_visible + mi_row * cm->mi_stride; |
103 | 111 |
104 for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MI_BLOCK_SIZE) { | 112 for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MI_BLOCK_SIZE) { |
105 const int r = mi_row >> MI_BLOCK_SIZE_LOG2; | 113 const int r = mi_row >> MI_BLOCK_SIZE_LOG2; |
106 const int c = mi_col >> MI_BLOCK_SIZE_LOG2; | 114 const int c = mi_col >> MI_BLOCK_SIZE_LOG2; |
107 LOOP_FILTER_MASK lfm; | 115 LOOP_FILTER_MASK lfm; |
108 int plane; | 116 int plane; |
109 | 117 |
110 sync_read(lf_sync, r, c); | 118 sync_read(lf_sync, r, c); |
111 | 119 |
112 vp9_setup_dst_planes(planes, frame_buffer, mi_row, mi_col); | 120 vp9_setup_dst_planes(planes, frame_buffer, mi_row, mi_col); |
113 | 121 |
114 // TODO(JBB): Make setup_mask work for non 420. | 122 // TODO(JBB): Make setup_mask work for non 420. |
115 if (use_420) | 123 vp9_setup_mask(cm, mi_row, mi_col, mi + mi_col, cm->mi_stride, |
116 vp9_setup_mask(cm, mi_row, mi_col, mi + mi_col, cm->mi_stride, | 124 &lfm); |
117 &lfm); | |
118 | 125 |
119 for (plane = 0; plane < num_planes; ++plane) { | 126 vp9_filter_block_plane_ss00(cm, &planes[0], mi_row, &lfm); |
120 if (use_420) | 127 for (plane = 1; plane < num_planes; ++plane) { |
121 vp9_filter_block_plane(cm, &planes[plane], mi_row, &lfm); | 128 switch (path) { |
122 else | 129 case LF_PATH_420: |
123 vp9_filter_block_plane_non420(cm, &planes[plane], mi + mi_col, | 130 vp9_filter_block_plane_ss11(cm, &planes[plane], mi_row, &lfm); |
124 mi_row, mi_col); | 131 break; |
| 132 case LF_PATH_444: |
| 133 vp9_filter_block_plane_ss00(cm, &planes[plane], mi_row, &lfm); |
| 134 break; |
| 135 case LF_PATH_SLOW: |
| 136 vp9_filter_block_plane_non420(cm, &planes[plane], mi + mi_col, |
| 137 mi_row, mi_col); |
| 138 break; |
| 139 } |
125 } | 140 } |
126 | 141 |
127 sync_write(lf_sync, r, c, sb_cols); | 142 sync_write(lf_sync, r, c, sb_cols); |
128 } | 143 } |
129 } | 144 } |
130 } | 145 } |
131 | 146 |
132 // Row-based multi-threaded loopfilter hook | 147 // Row-based multi-threaded loopfilter hook |
133 static int loop_filter_row_worker(VP9LfSync *const lf_sync, | 148 static int loop_filter_row_worker(VP9LfSync *const lf_sync, |
134 LFWorkerData *const lf_data) { | 149 LFWorkerData *const lf_data) { |
(...skipping 18 matching lines...) Expand all Loading... |
153 const int num_workers = MIN(nworkers, tile_cols); | 168 const int num_workers = MIN(nworkers, tile_cols); |
154 int i; | 169 int i; |
155 | 170 |
156 if (!lf_sync->sync_range || sb_rows != lf_sync->rows || | 171 if (!lf_sync->sync_range || sb_rows != lf_sync->rows || |
157 num_workers > lf_sync->num_workers) { | 172 num_workers > lf_sync->num_workers) { |
158 vp9_loop_filter_dealloc(lf_sync); | 173 vp9_loop_filter_dealloc(lf_sync); |
159 vp9_loop_filter_alloc(lf_sync, cm, sb_rows, cm->width, num_workers); | 174 vp9_loop_filter_alloc(lf_sync, cm, sb_rows, cm->width, num_workers); |
160 } | 175 } |
161 | 176 |
162 // Initialize cur_sb_col to -1 for all SB rows. | 177 // Initialize cur_sb_col to -1 for all SB rows. |
163 vpx_memset(lf_sync->cur_sb_col, -1, sizeof(*lf_sync->cur_sb_col) * sb_rows); | 178 memset(lf_sync->cur_sb_col, -1, sizeof(*lf_sync->cur_sb_col) * sb_rows); |
164 | 179 |
165 // Set up loopfilter thread data. | 180 // Set up loopfilter thread data. |
166 // The decoder is capping num_workers because it has been observed that using | 181 // The decoder is capping num_workers because it has been observed that using |
167 // more threads on the loopfilter than there are cores will hurt performance | 182 // more threads on the loopfilter than there are cores will hurt performance |
168 // on Android. This is because the system will only schedule the tile decode | 183 // on Android. This is because the system will only schedule the tile decode |
169 // workers on cores equal to the number of tile columns. Then if the decoder | 184 // workers on cores equal to the number of tile columns. Then if the decoder |
170 // tries to use more threads for the loopfilter, it will hurt performance | 185 // tries to use more threads for the loopfilter, it will hurt performance |
171 // because of contention. If the multithreading code changes in the future | 186 // because of contention. If the multithreading code changes in the future |
172 // then the number of workers used by the loopfilter should be revisited. | 187 // then the number of workers used by the loopfilter should be revisited. |
173 for (i = 0; i < num_workers; ++i) { | 188 for (i = 0; i < num_workers; ++i) { |
(...skipping 238 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
412 } | 427 } |
413 | 428 |
414 for (i = 0; i < MV_OFFSET_BITS; i++) | 429 for (i = 0; i < MV_OFFSET_BITS; i++) |
415 for (j = 0; j < 2; j++) | 430 for (j = 0; j < 2; j++) |
416 comps->bits[i][j] += comps_t->bits[i][j]; | 431 comps->bits[i][j] += comps_t->bits[i][j]; |
417 | 432 |
418 for (i = 0; i < MV_FP_SIZE; i++) | 433 for (i = 0; i < MV_FP_SIZE; i++) |
419 comps->fp[i] += comps_t->fp[i]; | 434 comps->fp[i] += comps_t->fp[i]; |
420 } | 435 } |
421 } | 436 } |
OLD | NEW |