/src/aom/av1/decoder/decodeframe.c
Line | Count | Source (jump to first uncovered line) |
1 | | /* |
2 | | * Copyright (c) 2016, Alliance for Open Media. All rights reserved. |
3 | | * |
4 | | * This source code is subject to the terms of the BSD 2 Clause License and |
5 | | * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License |
6 | | * was not distributed with this source code in the LICENSE file, you can |
7 | | * obtain it at www.aomedia.org/license/software. If the Alliance for Open |
8 | | * Media Patent License 1.0 was not distributed with this source code in the |
9 | | * PATENTS file, you can obtain it at www.aomedia.org/license/patent. |
10 | | */ |
11 | | |
12 | | #include <assert.h> |
13 | | #include <stdbool.h> |
14 | | #include <stddef.h> |
15 | | |
16 | | #include "config/aom_config.h" |
17 | | #include "config/aom_scale_rtcd.h" |
18 | | |
19 | | #include "aom/aom_codec.h" |
20 | | #include "aom/aom_image.h" |
21 | | #include "aom/internal/aom_codec_internal.h" |
22 | | #include "aom_dsp/aom_dsp_common.h" |
23 | | #include "aom_dsp/binary_codes_reader.h" |
24 | | #include "aom_dsp/bitreader.h" |
25 | | #include "aom_dsp/bitreader_buffer.h" |
26 | | #include "aom_dsp/txfm_common.h" |
27 | | #include "aom_mem/aom_mem.h" |
28 | | #include "aom_ports/aom_timer.h" |
29 | | #include "aom_ports/mem.h" |
30 | | #include "aom_ports/mem_ops.h" |
31 | | #include "aom_scale/aom_scale.h" |
32 | | #include "aom_scale/yv12config.h" |
33 | | #include "aom_util/aom_pthread.h" |
34 | | #include "aom_util/aom_thread.h" |
35 | | |
36 | | #if CONFIG_BITSTREAM_DEBUG || CONFIG_MISMATCH_DEBUG |
37 | | #include "aom_util/debug_util.h" |
38 | | #endif // CONFIG_BITSTREAM_DEBUG || CONFIG_MISMATCH_DEBUG |
39 | | |
40 | | #include "av1/common/alloccommon.h" |
41 | | #include "av1/common/av1_common_int.h" |
42 | | #include "av1/common/blockd.h" |
43 | | #include "av1/common/cdef.h" |
44 | | #include "av1/common/cfl.h" |
45 | | #include "av1/common/common_data.h" |
46 | | #include "av1/common/common.h" |
47 | | #include "av1/common/entropy.h" |
48 | | #include "av1/common/entropymode.h" |
49 | | #include "av1/common/entropymv.h" |
50 | | #include "av1/common/enums.h" |
51 | | #include "av1/common/frame_buffers.h" |
52 | | #include "av1/common/idct.h" |
53 | | #include "av1/common/mv.h" |
54 | | #include "av1/common/mvref_common.h" |
55 | | #include "av1/common/obmc.h" |
56 | | #include "av1/common/pred_common.h" |
57 | | #include "av1/common/quant_common.h" |
58 | | #include "av1/common/reconinter.h" |
59 | | #include "av1/common/reconintra.h" |
60 | | #include "av1/common/resize.h" |
61 | | #include "av1/common/restoration.h" |
62 | | #include "av1/common/scale.h" |
63 | | #include "av1/common/seg_common.h" |
64 | | #include "av1/common/thread_common.h" |
65 | | #include "av1/common/tile_common.h" |
66 | | #include "av1/common/warped_motion.h" |
67 | | |
68 | | #include "av1/decoder/decodeframe.h" |
69 | | #include "av1/decoder/decodemv.h" |
70 | | #include "av1/decoder/decoder.h" |
71 | | #include "av1/decoder/decodetxb.h" |
72 | | #include "av1/decoder/detokenize.h" |
73 | | #if CONFIG_INSPECTION |
74 | | #include "av1/decoder/inspection.h" |
75 | | #endif |
76 | | |
77 | | #define ACCT_STR __func__ |
78 | | |
79 | 94.1k | #define AOM_MIN_THREADS_PER_TILE 1 |
80 | 287k | #define AOM_MAX_THREADS_PER_TILE 2 |
81 | | |
82 | | // This is needed by ext_tile related unit tests. |
83 | | #define EXT_TILE_DEBUG 1 |
84 | | #define MC_TEMP_BUF_PELS \ |
85 | 314k | (((MAX_SB_SIZE)*2 + (AOM_INTERP_EXTEND)*2) * \ |
86 | 314k | ((MAX_SB_SIZE)*2 + (AOM_INTERP_EXTEND)*2)) |
87 | | |
88 | | // Checks that the remaining bits start with a 1 and ends with 0s. |
89 | | // It consumes an additional byte, if already byte aligned before the check. |
90 | 77.7k | int av1_check_trailing_bits(AV1Decoder *pbi, struct aom_read_bit_buffer *rb) { |
91 | | // bit_offset is set to 0 (mod 8) when the reader is already byte aligned |
92 | 77.7k | int bits_before_alignment = 8 - rb->bit_offset % 8; |
93 | 77.7k | int trailing = aom_rb_read_literal(rb, bits_before_alignment); |
94 | 77.7k | if (trailing != (1 << (bits_before_alignment - 1))) { |
95 | 5.20k | pbi->error.error_code = AOM_CODEC_CORRUPT_FRAME; |
96 | 5.20k | return -1; |
97 | 5.20k | } |
98 | 72.5k | return 0; |
99 | 77.7k | } |
100 | | |
101 | | // Use only_chroma = 1 to only set the chroma planes |
102 | | static AOM_INLINE void set_planes_to_neutral_grey( |
103 | | const SequenceHeader *const seq_params, const YV12_BUFFER_CONFIG *const buf, |
104 | 188k | int only_chroma) { |
105 | 188k | if (seq_params->use_highbitdepth) { |
106 | 62.7k | const int val = 1 << (seq_params->bit_depth - 1); |
107 | 246k | for (int plane = only_chroma; plane < MAX_MB_PLANE; plane++) { |
108 | 184k | const int is_uv = plane > 0; |
109 | 184k | uint16_t *const base = CONVERT_TO_SHORTPTR(buf->buffers[plane]); |
110 | | // Set the first row to neutral grey. Then copy the first row to all |
111 | | // subsequent rows. |
112 | 184k | if (buf->crop_heights[is_uv] > 0) { |
113 | 184k | aom_memset16(base, val, buf->crop_widths[is_uv]); |
114 | 82.8M | for (int row_idx = 1; row_idx < buf->crop_heights[is_uv]; row_idx++) { |
115 | 82.7M | memcpy(&base[row_idx * buf->strides[is_uv]], base, |
116 | 82.7M | sizeof(*base) * buf->crop_widths[is_uv]); |
117 | 82.7M | } |
118 | 184k | } |
119 | 184k | } |
120 | 125k | } else { |
121 | 500k | for (int plane = only_chroma; plane < MAX_MB_PLANE; plane++) { |
122 | 374k | const int is_uv = plane > 0; |
123 | 52.4M | for (int row_idx = 0; row_idx < buf->crop_heights[is_uv]; row_idx++) { |
124 | 52.0M | memset(&buf->buffers[plane][row_idx * buf->strides[is_uv]], 1 << 7, |
125 | 52.0M | buf->crop_widths[is_uv]); |
126 | 52.0M | } |
127 | 374k | } |
128 | 125k | } |
129 | 188k | } |
130 | | |
131 | | static AOM_INLINE void loop_restoration_read_sb_coeffs( |
132 | | const AV1_COMMON *const cm, MACROBLOCKD *xd, aom_reader *const r, int plane, |
133 | | int runit_idx); |
134 | | |
135 | 241k | static int read_is_valid(const uint8_t *start, size_t len, const uint8_t *end) { |
136 | 241k | return len != 0 && len <= (size_t)(end - start); |
137 | 241k | } |
138 | | |
139 | | static TX_MODE read_tx_mode(struct aom_read_bit_buffer *rb, |
140 | 203k | int coded_lossless) { |
141 | 203k | if (coded_lossless) return ONLY_4X4; |
142 | 179k | return aom_rb_read_bit(rb) ? TX_MODE_SELECT : TX_MODE_LARGEST; |
143 | 203k | } |
144 | | |
145 | | static REFERENCE_MODE read_frame_reference_mode( |
146 | 203k | const AV1_COMMON *cm, struct aom_read_bit_buffer *rb) { |
147 | 203k | if (frame_is_intra_only(cm)) { |
148 | 112k | return SINGLE_REFERENCE; |
149 | 112k | } else { |
150 | 90.7k | return aom_rb_read_bit(rb) ? REFERENCE_MODE_SELECT : SINGLE_REFERENCE; |
151 | 90.7k | } |
152 | 203k | } |
153 | | |
154 | | static AOM_INLINE void inverse_transform_block(DecoderCodingBlock *dcb, |
155 | | int plane, const TX_TYPE tx_type, |
156 | | const TX_SIZE tx_size, |
157 | | uint8_t *dst, int stride, |
158 | 21.8M | int reduced_tx_set) { |
159 | 21.8M | tran_low_t *const dqcoeff = dcb->dqcoeff_block[plane] + dcb->cb_offset[plane]; |
160 | 21.8M | eob_info *eob_data = dcb->eob_data[plane] + dcb->txb_offset[plane]; |
161 | 21.8M | uint16_t scan_line = eob_data->max_scan_line; |
162 | 21.8M | uint16_t eob = eob_data->eob; |
163 | 21.8M | av1_inverse_transform_block(&dcb->xd, dqcoeff, plane, tx_type, tx_size, dst, |
164 | 21.8M | stride, eob, reduced_tx_set); |
165 | 21.8M | memset(dqcoeff, 0, (scan_line + 1) * sizeof(dqcoeff[0])); |
166 | 21.8M | } |
167 | | |
168 | | static AOM_INLINE void read_coeffs_tx_intra_block( |
169 | | const AV1_COMMON *const cm, DecoderCodingBlock *dcb, aom_reader *const r, |
170 | 106M | const int plane, const int row, const int col, const TX_SIZE tx_size) { |
171 | 106M | MB_MODE_INFO *mbmi = dcb->xd.mi[0]; |
172 | 106M | if (!mbmi->skip_txfm) { |
173 | | #if TXCOEFF_TIMER |
174 | | struct aom_usec_timer timer; |
175 | | aom_usec_timer_start(&timer); |
176 | | #endif |
177 | 30.4M | av1_read_coeffs_txb_facade(cm, dcb, r, plane, row, col, tx_size); |
178 | | #if TXCOEFF_TIMER |
179 | | aom_usec_timer_mark(&timer); |
180 | | const int64_t elapsed_time = aom_usec_timer_elapsed(&timer); |
181 | | cm->txcoeff_timer += elapsed_time; |
182 | | ++cm->txb_count; |
183 | | #endif |
184 | 30.4M | } |
185 | 106M | } |
186 | | |
187 | | static AOM_INLINE void decode_block_void(const AV1_COMMON *const cm, |
188 | | DecoderCodingBlock *dcb, |
189 | | aom_reader *const r, const int plane, |
190 | | const int row, const int col, |
191 | 155M | const TX_SIZE tx_size) { |
192 | 155M | (void)cm; |
193 | 155M | (void)dcb; |
194 | 155M | (void)r; |
195 | 155M | (void)plane; |
196 | 155M | (void)row; |
197 | 155M | (void)col; |
198 | 155M | (void)tx_size; |
199 | 155M | } |
200 | | |
201 | | static AOM_INLINE void predict_inter_block_void(AV1_COMMON *const cm, |
202 | | DecoderCodingBlock *dcb, |
203 | 5.35M | BLOCK_SIZE bsize) { |
204 | 5.35M | (void)cm; |
205 | 5.35M | (void)dcb; |
206 | 5.35M | (void)bsize; |
207 | 5.35M | } |
208 | | |
209 | | static AOM_INLINE void cfl_store_inter_block_void(AV1_COMMON *const cm, |
210 | 5.34M | MACROBLOCKD *const xd) { |
211 | 5.34M | (void)cm; |
212 | 5.34M | (void)xd; |
213 | 5.34M | } |
214 | | |
215 | | static AOM_INLINE void predict_and_reconstruct_intra_block( |
216 | | const AV1_COMMON *const cm, DecoderCodingBlock *dcb, aom_reader *const r, |
217 | 97.5M | const int plane, const int row, const int col, const TX_SIZE tx_size) { |
218 | 97.5M | (void)r; |
219 | 97.5M | MACROBLOCKD *const xd = &dcb->xd; |
220 | 97.5M | MB_MODE_INFO *mbmi = xd->mi[0]; |
221 | 97.5M | PLANE_TYPE plane_type = get_plane_type(plane); |
222 | | |
223 | 97.5M | av1_predict_intra_block_facade(cm, xd, plane, col, row, tx_size); |
224 | | |
225 | 97.5M | if (!mbmi->skip_txfm) { |
226 | 23.0M | eob_info *eob_data = dcb->eob_data[plane] + dcb->txb_offset[plane]; |
227 | 23.0M | if (eob_data->eob) { |
228 | 12.1M | const bool reduced_tx_set_used = cm->features.reduced_tx_set_used; |
229 | | // tx_type was read out in av1_read_coeffs_txb. |
230 | 12.1M | const TX_TYPE tx_type = av1_get_tx_type(xd, plane_type, row, col, tx_size, |
231 | 12.1M | reduced_tx_set_used); |
232 | 12.1M | struct macroblockd_plane *const pd = &xd->plane[plane]; |
233 | 12.1M | uint8_t *dst = &pd->dst.buf[(row * pd->dst.stride + col) << MI_SIZE_LOG2]; |
234 | 12.1M | inverse_transform_block(dcb, plane, tx_type, tx_size, dst, pd->dst.stride, |
235 | 12.1M | reduced_tx_set_used); |
236 | 12.1M | } |
237 | 23.0M | } |
238 | 97.5M | if (plane == AOM_PLANE_Y && store_cfl_required(cm, xd)) { |
239 | 2.41M | cfl_store_tx(xd, row, col, tx_size, mbmi->bsize); |
240 | 2.41M | } |
241 | 97.5M | } |
242 | | |
243 | | static AOM_INLINE void inverse_transform_inter_block( |
244 | | const AV1_COMMON *const cm, DecoderCodingBlock *dcb, aom_reader *const r, |
245 | | const int plane, const int blk_row, const int blk_col, |
246 | 9.73M | const TX_SIZE tx_size) { |
247 | 9.73M | (void)r; |
248 | 9.73M | MACROBLOCKD *const xd = &dcb->xd; |
249 | 9.73M | PLANE_TYPE plane_type = get_plane_type(plane); |
250 | 9.73M | const struct macroblockd_plane *const pd = &xd->plane[plane]; |
251 | 9.73M | const bool reduced_tx_set_used = cm->features.reduced_tx_set_used; |
252 | | // tx_type was read out in av1_read_coeffs_txb. |
253 | 9.73M | const TX_TYPE tx_type = av1_get_tx_type(xd, plane_type, blk_row, blk_col, |
254 | 9.73M | tx_size, reduced_tx_set_used); |
255 | | |
256 | 9.73M | uint8_t *dst = |
257 | 9.73M | &pd->dst.buf[(blk_row * pd->dst.stride + blk_col) << MI_SIZE_LOG2]; |
258 | 9.73M | inverse_transform_block(dcb, plane, tx_type, tx_size, dst, pd->dst.stride, |
259 | 9.73M | reduced_tx_set_used); |
260 | | #if CONFIG_MISMATCH_DEBUG |
261 | | int pixel_c, pixel_r; |
262 | | BLOCK_SIZE bsize = txsize_to_bsize[tx_size]; |
263 | | int blk_w = block_size_wide[bsize]; |
264 | | int blk_h = block_size_high[bsize]; |
265 | | const int mi_row = -xd->mb_to_top_edge >> (3 + MI_SIZE_LOG2); |
266 | | const int mi_col = -xd->mb_to_left_edge >> (3 + MI_SIZE_LOG2); |
267 | | mi_to_pixel_loc(&pixel_c, &pixel_r, mi_col, mi_row, blk_col, blk_row, |
268 | | pd->subsampling_x, pd->subsampling_y); |
269 | | mismatch_check_block_tx(dst, pd->dst.stride, cm->current_frame.order_hint, |
270 | | plane, pixel_c, pixel_r, blk_w, blk_h, |
271 | | xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH); |
272 | | #endif |
273 | 9.73M | } |
274 | | |
275 | | static AOM_INLINE void set_cb_buffer_offsets(DecoderCodingBlock *dcb, |
276 | 183M | TX_SIZE tx_size, int plane) { |
277 | 183M | dcb->cb_offset[plane] += tx_size_wide[tx_size] * tx_size_high[tx_size]; |
278 | 183M | dcb->txb_offset[plane] = |
279 | 183M | dcb->cb_offset[plane] / (TX_SIZE_W_MIN * TX_SIZE_H_MIN); |
280 | 183M | } |
281 | | |
282 | | static AOM_INLINE void decode_reconstruct_tx( |
283 | | AV1_COMMON *cm, ThreadData *const td, aom_reader *r, |
284 | | MB_MODE_INFO *const mbmi, int plane, BLOCK_SIZE plane_bsize, int blk_row, |
285 | 20.5M | int blk_col, int block, TX_SIZE tx_size, int *eob_total) { |
286 | 20.5M | DecoderCodingBlock *const dcb = &td->dcb; |
287 | 20.5M | MACROBLOCKD *const xd = &dcb->xd; |
288 | 20.5M | const struct macroblockd_plane *const pd = &xd->plane[plane]; |
289 | 20.5M | const TX_SIZE plane_tx_size = |
290 | 20.5M | plane ? av1_get_max_uv_txsize(mbmi->bsize, pd->subsampling_x, |
291 | 11.9M | pd->subsampling_y) |
292 | 20.5M | : mbmi->inter_tx_size[av1_get_txb_size_index(plane_bsize, blk_row, |
293 | 8.62M | blk_col)]; |
294 | | // Scale to match transform block unit. |
295 | 20.5M | const int max_blocks_high = max_block_high(xd, plane_bsize, plane); |
296 | 20.5M | const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane); |
297 | | |
298 | 20.6M | if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return; |
299 | | |
300 | 20.5M | if (tx_size == plane_tx_size || plane) { |
301 | 20.1M | td->read_coeffs_tx_inter_block_visit(cm, dcb, r, plane, blk_row, blk_col, |
302 | 20.1M | tx_size); |
303 | | |
304 | 20.1M | td->inverse_tx_inter_block_visit(cm, dcb, r, plane, blk_row, blk_col, |
305 | 20.1M | tx_size); |
306 | 20.1M | eob_info *eob_data = dcb->eob_data[plane] + dcb->txb_offset[plane]; |
307 | 20.1M | *eob_total += eob_data->eob; |
308 | 20.1M | set_cb_buffer_offsets(dcb, tx_size, plane); |
309 | 20.1M | } else { |
310 | 408k | const TX_SIZE sub_txs = sub_tx_size_map[tx_size]; |
311 | 408k | assert(IMPLIES(tx_size <= TX_4X4, sub_txs == tx_size)); |
312 | 479k | assert(IMPLIES(tx_size > TX_4X4, sub_txs < tx_size)); |
313 | 479k | const int bsw = tx_size_wide_unit[sub_txs]; |
314 | 479k | const int bsh = tx_size_high_unit[sub_txs]; |
315 | 479k | const int sub_step = bsw * bsh; |
316 | 479k | const int row_end = |
317 | 479k | AOMMIN(tx_size_high_unit[tx_size], max_blocks_high - blk_row); |
318 | 479k | const int col_end = |
319 | 479k | AOMMIN(tx_size_wide_unit[tx_size], max_blocks_wide - blk_col); |
320 | | |
321 | 479k | assert(bsw > 0 && bsh > 0); |
322 | | |
323 | 1.27M | for (int row = 0; row < row_end; row += bsh) { |
324 | 796k | const int offsetr = blk_row + row; |
325 | 2.16M | for (int col = 0; col < col_end; col += bsw) { |
326 | 1.36M | const int offsetc = blk_col + col; |
327 | | |
328 | 1.36M | decode_reconstruct_tx(cm, td, r, mbmi, plane, plane_bsize, offsetr, |
329 | 1.36M | offsetc, block, sub_txs, eob_total); |
330 | 1.36M | block += sub_step; |
331 | 1.36M | } |
332 | 796k | } |
333 | 479k | } |
334 | 20.5M | } |
335 | | |
336 | | static AOM_INLINE void set_offsets(AV1_COMMON *const cm, MACROBLOCKD *const xd, |
337 | | BLOCK_SIZE bsize, int mi_row, int mi_col, |
338 | 16.6M | int bw, int bh, int x_mis, int y_mis) { |
339 | 16.6M | const int num_planes = av1_num_planes(cm); |
340 | 16.6M | const CommonModeInfoParams *const mi_params = &cm->mi_params; |
341 | 16.6M | const TileInfo *const tile = &xd->tile; |
342 | | |
343 | 16.6M | set_mi_offsets(mi_params, xd, mi_row, mi_col); |
344 | 16.6M | xd->mi[0]->bsize = bsize; |
345 | | #if CONFIG_RD_DEBUG |
346 | | xd->mi[0]->mi_row = mi_row; |
347 | | xd->mi[0]->mi_col = mi_col; |
348 | | #endif |
349 | | |
350 | 16.6M | assert(x_mis && y_mis); |
351 | 87.9M | for (int x = 1; x < x_mis; ++x) xd->mi[x] = xd->mi[0]; |
352 | 16.6M | int idx = mi_params->mi_stride; |
353 | 83.6M | for (int y = 1; y < y_mis; ++y) { |
354 | 66.9M | memcpy(&xd->mi[idx], &xd->mi[0], x_mis * sizeof(xd->mi[0])); |
355 | 66.9M | idx += mi_params->mi_stride; |
356 | 66.9M | } |
357 | | |
358 | 16.6M | set_plane_n4(xd, bw, bh, num_planes); |
359 | 16.6M | set_entropy_context(xd, mi_row, mi_col, num_planes); |
360 | | |
361 | | // Distance of Mb to the various image edges. These are specified to 8th pel |
362 | | // as they are always compared to values that are in 1/8th pel units |
363 | 16.6M | set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, mi_params->mi_rows, |
364 | 16.6M | mi_params->mi_cols); |
365 | | |
366 | 16.6M | av1_setup_dst_planes(xd->plane, bsize, &cm->cur_frame->buf, mi_row, mi_col, 0, |
367 | 16.6M | num_planes); |
368 | 16.6M | } |
369 | | |
370 | | static AOM_INLINE void decode_mbmi_block(AV1Decoder *const pbi, |
371 | | DecoderCodingBlock *dcb, int mi_row, |
372 | | int mi_col, aom_reader *r, |
373 | | PARTITION_TYPE partition, |
374 | 16.6M | BLOCK_SIZE bsize) { |
375 | 16.6M | AV1_COMMON *const cm = &pbi->common; |
376 | 16.6M | const SequenceHeader *const seq_params = cm->seq_params; |
377 | 16.6M | const int bw = mi_size_wide[bsize]; |
378 | 16.6M | const int bh = mi_size_high[bsize]; |
379 | 16.6M | const int x_mis = AOMMIN(bw, cm->mi_params.mi_cols - mi_col); |
380 | 16.6M | const int y_mis = AOMMIN(bh, cm->mi_params.mi_rows - mi_row); |
381 | 16.6M | MACROBLOCKD *const xd = &dcb->xd; |
382 | | |
383 | | #if CONFIG_ACCOUNTING |
384 | | aom_accounting_set_context(&pbi->accounting, mi_col, mi_row); |
385 | | #endif |
386 | 16.6M | set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis); |
387 | 16.6M | xd->mi[0]->partition = partition; |
388 | 16.6M | av1_read_mode_info(pbi, dcb, r, x_mis, y_mis); |
389 | 16.6M | if (bsize >= BLOCK_8X8 && |
390 | 16.6M | (seq_params->subsampling_x || seq_params->subsampling_y)) { |
391 | 9.84M | const BLOCK_SIZE uv_subsize = |
392 | 9.84M | av1_ss_size_lookup[bsize][seq_params->subsampling_x] |
393 | 9.84M | [seq_params->subsampling_y]; |
394 | 9.84M | if (uv_subsize == BLOCK_INVALID) |
395 | 0 | aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME, |
396 | 0 | "Invalid block size."); |
397 | 9.84M | } |
398 | 16.6M | } |
399 | | |
400 | | typedef struct PadBlock { |
401 | | int x0; |
402 | | int x1; |
403 | | int y0; |
404 | | int y1; |
405 | | } PadBlock; |
406 | | |
407 | | #if CONFIG_AV1_HIGHBITDEPTH |
408 | | static AOM_INLINE void highbd_build_mc_border(const uint8_t *src8, |
409 | | int src_stride, uint8_t *dst8, |
410 | | int dst_stride, int x, int y, |
411 | 1.37M | int b_w, int b_h, int w, int h) { |
412 | | // Get a pointer to the start of the real data for this row. |
413 | 1.37M | const uint16_t *src = CONVERT_TO_SHORTPTR(src8); |
414 | 1.37M | uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); |
415 | 1.37M | const uint16_t *ref_row = src - x - y * src_stride; |
416 | | |
417 | 1.37M | if (y >= h) |
418 | 398k | ref_row += (h - 1) * src_stride; |
419 | 975k | else if (y > 0) |
420 | 696k | ref_row += y * src_stride; |
421 | | |
422 | 25.8M | do { |
423 | 25.8M | int right = 0, copy; |
424 | 25.8M | int left = x < 0 ? -x : 0; |
425 | | |
426 | 25.8M | if (left > b_w) left = b_w; |
427 | | |
428 | 25.8M | if (x + b_w > w) right = x + b_w - w; |
429 | | |
430 | 25.8M | if (right > b_w) right = b_w; |
431 | | |
432 | 25.8M | copy = b_w - left - right; |
433 | | |
434 | 25.8M | if (left) aom_memset16(dst, ref_row[0], left); |
435 | | |
436 | 25.8M | if (copy) memcpy(dst + left, ref_row + x + left, copy * sizeof(uint16_t)); |
437 | | |
438 | 25.8M | if (right) aom_memset16(dst + left + copy, ref_row[w - 1], right); |
439 | | |
440 | 25.8M | dst += dst_stride; |
441 | 25.8M | ++y; |
442 | | |
443 | 25.8M | if (y > 0 && y < h) ref_row += src_stride; |
444 | 25.8M | } while (--b_h); |
445 | 1.37M | } |
446 | | #endif // CONFIG_AV1_HIGHBITDEPTH |
447 | | |
448 | | static AOM_INLINE void build_mc_border(const uint8_t *src, int src_stride, |
449 | | uint8_t *dst, int dst_stride, int x, |
450 | 1.72M | int y, int b_w, int b_h, int w, int h) { |
451 | | // Get a pointer to the start of the real data for this row. |
452 | 1.72M | const uint8_t *ref_row = src - x - y * src_stride; |
453 | | |
454 | 1.72M | if (y >= h) |
455 | 116k | ref_row += (h - 1) * src_stride; |
456 | 1.61M | else if (y > 0) |
457 | 1.10M | ref_row += y * src_stride; |
458 | | |
459 | 34.2M | do { |
460 | 34.2M | int right = 0, copy; |
461 | 34.2M | int left = x < 0 ? -x : 0; |
462 | | |
463 | 34.2M | if (left > b_w) left = b_w; |
464 | | |
465 | 34.2M | if (x + b_w > w) right = x + b_w - w; |
466 | | |
467 | 34.2M | if (right > b_w) right = b_w; |
468 | | |
469 | 34.2M | copy = b_w - left - right; |
470 | | |
471 | 34.2M | if (left) memset(dst, ref_row[0], left); |
472 | | |
473 | 34.2M | if (copy) memcpy(dst + left, ref_row + x + left, copy); |
474 | | |
475 | 34.2M | if (right) memset(dst + left + copy, ref_row[w - 1], right); |
476 | | |
477 | 34.2M | dst += dst_stride; |
478 | 34.2M | ++y; |
479 | | |
480 | 34.2M | if (y > 0 && y < h) ref_row += src_stride; |
481 | 34.2M | } while (--b_h); |
482 | 1.72M | } |
483 | | |
484 | | static INLINE int update_extend_mc_border_params( |
485 | | const struct scale_factors *const sf, struct buf_2d *const pre_buf, |
486 | | MV32 scaled_mv, PadBlock *block, int subpel_x_mv, int subpel_y_mv, |
487 | 24.2M | int do_warp, int is_intrabc, int *x_pad, int *y_pad) { |
488 | 24.2M | const int is_scaled = av1_is_scaled(sf); |
489 | | // Get reference width and height. |
490 | 24.2M | int frame_width = pre_buf->width; |
491 | 24.2M | int frame_height = pre_buf->height; |
492 | | |
493 | | // Do border extension if there is motion or |
494 | | // width/height is not a multiple of 8 pixels. |
495 | 24.2M | if ((!is_intrabc) && (!do_warp) && |
496 | 24.2M | (is_scaled || scaled_mv.col || scaled_mv.row || (frame_width & 0x7) || |
497 | 23.3M | (frame_height & 0x7))) { |
498 | 18.3M | if (subpel_x_mv || (sf->x_step_q4 != SUBPEL_SHIFTS)) { |
499 | 18.3M | block->x0 -= AOM_INTERP_EXTEND - 1; |
500 | 18.3M | block->x1 += AOM_INTERP_EXTEND; |
501 | 18.3M | *x_pad = 1; |
502 | 18.3M | } |
503 | | |
504 | 18.3M | if (subpel_y_mv || (sf->y_step_q4 != SUBPEL_SHIFTS)) { |
505 | 18.3M | block->y0 -= AOM_INTERP_EXTEND - 1; |
506 | 18.3M | block->y1 += AOM_INTERP_EXTEND; |
507 | 18.3M | *y_pad = 1; |
508 | 18.3M | } |
509 | | |
510 | | // Skip border extension if block is inside the frame. |
511 | 18.3M | if (block->x0 < 0 || block->x1 > frame_width - 1 || block->y0 < 0 || |
512 | 18.3M | block->y1 > frame_height - 1) { |
513 | 3.10M | return 1; |
514 | 3.10M | } |
515 | 18.3M | } |
516 | 21.1M | return 0; |
517 | 24.2M | } |
518 | | |
519 | | static INLINE void extend_mc_border(const struct scale_factors *const sf, |
520 | | struct buf_2d *const pre_buf, |
521 | | MV32 scaled_mv, PadBlock block, |
522 | | int subpel_x_mv, int subpel_y_mv, |
523 | | int do_warp, int is_intrabc, int highbd, |
524 | | uint8_t *mc_buf, uint8_t **pre, |
525 | 24.2M | int *src_stride) { |
526 | 24.2M | int x_pad = 0, y_pad = 0; |
527 | 24.2M | if (update_extend_mc_border_params(sf, pre_buf, scaled_mv, &block, |
528 | 24.2M | subpel_x_mv, subpel_y_mv, do_warp, |
529 | 24.2M | is_intrabc, &x_pad, &y_pad)) { |
530 | | // Get reference block pointer. |
531 | 3.10M | const uint8_t *const buf_ptr = |
532 | 3.10M | pre_buf->buf0 + block.y0 * pre_buf->stride + block.x0; |
533 | 3.10M | int buf_stride = pre_buf->stride; |
534 | 3.10M | const int b_w = block.x1 - block.x0; |
535 | 3.10M | const int b_h = block.y1 - block.y0; |
536 | | |
537 | 3.10M | #if CONFIG_AV1_HIGHBITDEPTH |
538 | | // Extend the border. |
539 | 3.10M | if (highbd) { |
540 | 1.37M | highbd_build_mc_border(buf_ptr, buf_stride, mc_buf, b_w, block.x0, |
541 | 1.37M | block.y0, b_w, b_h, pre_buf->width, |
542 | 1.37M | pre_buf->height); |
543 | 1.72M | } else { |
544 | 1.72M | build_mc_border(buf_ptr, buf_stride, mc_buf, b_w, block.x0, block.y0, b_w, |
545 | 1.72M | b_h, pre_buf->width, pre_buf->height); |
546 | 1.72M | } |
547 | | #else |
548 | | (void)highbd; |
549 | | build_mc_border(buf_ptr, buf_stride, mc_buf, b_w, block.x0, block.y0, b_w, |
550 | | b_h, pre_buf->width, pre_buf->height); |
551 | | #endif |
552 | 3.10M | *src_stride = b_w; |
553 | 3.10M | *pre = mc_buf + y_pad * (AOM_INTERP_EXTEND - 1) * b_w + |
554 | 3.10M | x_pad * (AOM_INTERP_EXTEND - 1); |
555 | 3.10M | } |
556 | 24.2M | } |
557 | | |
558 | | static AOM_INLINE void dec_calc_subpel_params( |
559 | | const MV *const src_mv, InterPredParams *const inter_pred_params, |
560 | | const MACROBLOCKD *const xd, int mi_x, int mi_y, uint8_t **pre, |
561 | | SubpelParams *subpel_params, int *src_stride, PadBlock *block, |
562 | 24.2M | MV32 *scaled_mv, int *subpel_x_mv, int *subpel_y_mv) { |
563 | 24.2M | const struct scale_factors *sf = inter_pred_params->scale_factors; |
564 | 24.2M | struct buf_2d *pre_buf = &inter_pred_params->ref_frame_buf; |
565 | 24.2M | const int bw = inter_pred_params->block_width; |
566 | 24.2M | const int bh = inter_pred_params->block_height; |
567 | 24.2M | const int is_scaled = av1_is_scaled(sf); |
568 | 24.2M | if (is_scaled) { |
569 | 3.40M | int ssx = inter_pred_params->subsampling_x; |
570 | 3.40M | int ssy = inter_pred_params->subsampling_y; |
571 | 3.40M | int orig_pos_y = inter_pred_params->pix_row << SUBPEL_BITS; |
572 | 3.40M | orig_pos_y += src_mv->row * (1 << (1 - ssy)); |
573 | 3.40M | int orig_pos_x = inter_pred_params->pix_col << SUBPEL_BITS; |
574 | 3.40M | orig_pos_x += src_mv->col * (1 << (1 - ssx)); |
575 | 3.40M | int pos_y = av1_scaled_y(orig_pos_y, sf); |
576 | 3.40M | int pos_x = av1_scaled_x(orig_pos_x, sf); |
577 | 3.40M | pos_x += SCALE_EXTRA_OFF; |
578 | 3.40M | pos_y += SCALE_EXTRA_OFF; |
579 | | |
580 | 3.40M | const int top = -AOM_LEFT_TOP_MARGIN_SCALED(ssy); |
581 | 3.40M | const int left = -AOM_LEFT_TOP_MARGIN_SCALED(ssx); |
582 | 3.40M | const int bottom = (pre_buf->height + AOM_INTERP_EXTEND) |
583 | 3.40M | << SCALE_SUBPEL_BITS; |
584 | 3.40M | const int right = (pre_buf->width + AOM_INTERP_EXTEND) << SCALE_SUBPEL_BITS; |
585 | 3.40M | pos_y = clamp(pos_y, top, bottom); |
586 | 3.40M | pos_x = clamp(pos_x, left, right); |
587 | | |
588 | 3.40M | subpel_params->subpel_x = pos_x & SCALE_SUBPEL_MASK; |
589 | 3.40M | subpel_params->subpel_y = pos_y & SCALE_SUBPEL_MASK; |
590 | 3.40M | subpel_params->xs = sf->x_step_q4; |
591 | 3.40M | subpel_params->ys = sf->y_step_q4; |
592 | | |
593 | | // Get reference block top left coordinate. |
594 | 3.40M | block->x0 = pos_x >> SCALE_SUBPEL_BITS; |
595 | 3.40M | block->y0 = pos_y >> SCALE_SUBPEL_BITS; |
596 | | |
597 | | // Get reference block bottom right coordinate. |
598 | 3.40M | block->x1 = |
599 | 3.40M | ((pos_x + (bw - 1) * subpel_params->xs) >> SCALE_SUBPEL_BITS) + 1; |
600 | 3.40M | block->y1 = |
601 | 3.40M | ((pos_y + (bh - 1) * subpel_params->ys) >> SCALE_SUBPEL_BITS) + 1; |
602 | | |
603 | 3.40M | MV temp_mv; |
604 | 3.40M | temp_mv = clamp_mv_to_umv_border_sb(xd, src_mv, bw, bh, |
605 | 3.40M | inter_pred_params->subsampling_x, |
606 | 3.40M | inter_pred_params->subsampling_y); |
607 | 3.40M | *scaled_mv = av1_scale_mv(&temp_mv, mi_x, mi_y, sf); |
608 | 3.40M | scaled_mv->row += SCALE_EXTRA_OFF; |
609 | 3.40M | scaled_mv->col += SCALE_EXTRA_OFF; |
610 | | |
611 | 3.40M | *subpel_x_mv = scaled_mv->col & SCALE_SUBPEL_MASK; |
612 | 3.40M | *subpel_y_mv = scaled_mv->row & SCALE_SUBPEL_MASK; |
613 | 20.8M | } else { |
614 | | // Get block position in current frame. |
615 | 20.8M | int pos_x = inter_pred_params->pix_col << SUBPEL_BITS; |
616 | 20.8M | int pos_y = inter_pred_params->pix_row << SUBPEL_BITS; |
617 | | |
618 | 20.8M | const MV mv_q4 = clamp_mv_to_umv_border_sb( |
619 | 20.8M | xd, src_mv, bw, bh, inter_pred_params->subsampling_x, |
620 | 20.8M | inter_pred_params->subsampling_y); |
621 | 20.8M | subpel_params->xs = subpel_params->ys = SCALE_SUBPEL_SHIFTS; |
622 | 20.8M | subpel_params->subpel_x = (mv_q4.col & SUBPEL_MASK) << SCALE_EXTRA_BITS; |
623 | 20.8M | subpel_params->subpel_y = (mv_q4.row & SUBPEL_MASK) << SCALE_EXTRA_BITS; |
624 | | |
625 | | // Get reference block top left coordinate. |
626 | 20.8M | pos_x += mv_q4.col; |
627 | 20.8M | pos_y += mv_q4.row; |
628 | 20.8M | block->x0 = pos_x >> SUBPEL_BITS; |
629 | 20.8M | block->y0 = pos_y >> SUBPEL_BITS; |
630 | | |
631 | | // Get reference block bottom right coordinate. |
632 | 20.8M | block->x1 = (pos_x >> SUBPEL_BITS) + (bw - 1) + 1; |
633 | 20.8M | block->y1 = (pos_y >> SUBPEL_BITS) + (bh - 1) + 1; |
634 | | |
635 | 20.8M | scaled_mv->row = mv_q4.row; |
636 | 20.8M | scaled_mv->col = mv_q4.col; |
637 | 20.8M | *subpel_x_mv = scaled_mv->col & SUBPEL_MASK; |
638 | 20.8M | *subpel_y_mv = scaled_mv->row & SUBPEL_MASK; |
639 | 20.8M | } |
640 | 24.2M | *pre = pre_buf->buf0 + block->y0 * pre_buf->stride + block->x0; |
641 | 24.2M | *src_stride = pre_buf->stride; |
642 | 24.2M | } |
643 | | |
644 | | static AOM_INLINE void dec_calc_subpel_params_and_extend( |
645 | | const MV *const src_mv, InterPredParams *const inter_pred_params, |
646 | | MACROBLOCKD *const xd, int mi_x, int mi_y, int ref, uint8_t **mc_buf, |
647 | 24.2M | uint8_t **pre, SubpelParams *subpel_params, int *src_stride) { |
648 | 24.2M | PadBlock block; |
649 | 24.2M | MV32 scaled_mv; |
650 | 24.2M | int subpel_x_mv, subpel_y_mv; |
651 | 24.2M | dec_calc_subpel_params(src_mv, inter_pred_params, xd, mi_x, mi_y, pre, |
652 | 24.2M | subpel_params, src_stride, &block, &scaled_mv, |
653 | 24.2M | &subpel_x_mv, &subpel_y_mv); |
654 | 24.2M | extend_mc_border( |
655 | 24.2M | inter_pred_params->scale_factors, &inter_pred_params->ref_frame_buf, |
656 | 24.2M | scaled_mv, block, subpel_x_mv, subpel_y_mv, |
657 | 24.2M | inter_pred_params->mode == WARP_PRED, inter_pred_params->is_intrabc, |
658 | 24.2M | inter_pred_params->use_hbd_buf, mc_buf[ref], pre, src_stride); |
659 | 24.2M | } |
660 | | |
661 | | #define IS_DEC 1 |
662 | | #include "av1/common/reconinter_template.inc" |
663 | | #undef IS_DEC |
664 | | |
665 | | static void dec_build_inter_predictors(const AV1_COMMON *cm, |
666 | | DecoderCodingBlock *dcb, int plane, |
667 | | const MB_MODE_INFO *mi, |
668 | | int build_for_obmc, int bw, int bh, |
669 | 19.7M | int mi_x, int mi_y) { |
670 | 19.7M | build_inter_predictors(cm, &dcb->xd, plane, mi, build_for_obmc, bw, bh, mi_x, |
671 | 19.7M | mi_y, dcb->mc_buf); |
672 | 19.7M | } |
673 | | |
674 | | static AOM_INLINE void dec_build_inter_predictor(const AV1_COMMON *cm, |
675 | | DecoderCodingBlock *dcb, |
676 | | int mi_row, int mi_col, |
677 | 5.66M | BLOCK_SIZE bsize) { |
678 | 5.66M | MACROBLOCKD *const xd = &dcb->xd; |
679 | 5.66M | const int num_planes = av1_num_planes(cm); |
680 | 20.7M | for (int plane = 0; plane < num_planes; ++plane) { |
681 | 16.0M | if (plane && !xd->is_chroma_ref) break; |
682 | 15.1M | const int mi_x = mi_col * MI_SIZE; |
683 | 15.1M | const int mi_y = mi_row * MI_SIZE; |
684 | 15.1M | dec_build_inter_predictors(cm, dcb, plane, xd->mi[0], 0, |
685 | 15.1M | xd->plane[plane].width, xd->plane[plane].height, |
686 | 15.1M | mi_x, mi_y); |
687 | 15.1M | if (is_interintra_pred(xd->mi[0])) { |
688 | 1.04M | BUFFER_SET ctx = { { xd->plane[0].dst.buf, xd->plane[1].dst.buf, |
689 | 1.04M | xd->plane[2].dst.buf }, |
690 | 1.04M | { xd->plane[0].dst.stride, xd->plane[1].dst.stride, |
691 | 1.04M | xd->plane[2].dst.stride } }; |
692 | 1.04M | av1_build_interintra_predictor(cm, xd, xd->plane[plane].dst.buf, |
693 | 1.04M | xd->plane[plane].dst.stride, &ctx, plane, |
694 | 1.04M | bsize); |
695 | 1.04M | } |
696 | 15.1M | } |
697 | 5.66M | } |
698 | | |
699 | | static INLINE void dec_build_prediction_by_above_pred( |
700 | | MACROBLOCKD *const xd, int rel_mi_row, int rel_mi_col, uint8_t op_mi_size, |
701 | 962k | int dir, MB_MODE_INFO *above_mbmi, void *fun_ctxt, const int num_planes) { |
702 | 962k | struct build_prediction_ctxt *ctxt = (struct build_prediction_ctxt *)fun_ctxt; |
703 | 962k | const int above_mi_col = xd->mi_col + rel_mi_col; |
704 | 962k | int mi_x, mi_y; |
705 | 962k | MB_MODE_INFO backup_mbmi = *above_mbmi; |
706 | | |
707 | 962k | (void)rel_mi_row; |
708 | 962k | (void)dir; |
709 | | |
710 | 962k | av1_setup_build_prediction_by_above_pred(xd, rel_mi_col, op_mi_size, |
711 | 962k | &backup_mbmi, ctxt, num_planes); |
712 | 962k | mi_x = above_mi_col << MI_SIZE_LOG2; |
713 | 962k | mi_y = xd->mi_row << MI_SIZE_LOG2; |
714 | | |
715 | 962k | const BLOCK_SIZE bsize = xd->mi[0]->bsize; |
716 | | |
717 | 3.84M | for (int j = 0; j < num_planes; ++j) { |
718 | 2.88M | const struct macroblockd_plane *pd = &xd->plane[j]; |
719 | 2.88M | int bw = (op_mi_size * MI_SIZE) >> pd->subsampling_x; |
720 | 2.88M | int bh = clamp(block_size_high[bsize] >> (pd->subsampling_y + 1), 4, |
721 | 2.88M | block_size_high[BLOCK_64X64] >> (pd->subsampling_y + 1)); |
722 | | |
723 | 2.88M | if (av1_skip_u4x4_pred_in_obmc(bsize, pd, 0)) continue; |
724 | 1.82M | dec_build_inter_predictors(ctxt->cm, (DecoderCodingBlock *)ctxt->dcb, j, |
725 | 1.82M | &backup_mbmi, 1, bw, bh, mi_x, mi_y); |
726 | 1.82M | } |
727 | 962k | } |
728 | | |
729 | | static AOM_INLINE void dec_build_prediction_by_above_preds( |
730 | | const AV1_COMMON *cm, DecoderCodingBlock *dcb, |
731 | | uint8_t *tmp_buf[MAX_MB_PLANE], int tmp_width[MAX_MB_PLANE], |
732 | 949k | int tmp_height[MAX_MB_PLANE], int tmp_stride[MAX_MB_PLANE]) { |
733 | 949k | MACROBLOCKD *const xd = &dcb->xd; |
734 | 949k | if (!xd->up_available) return; |
735 | | |
736 | | // Adjust mb_to_bottom_edge to have the correct value for the OBMC |
737 | | // prediction block. This is half the height of the original block, |
738 | | // except for 128-wide blocks, where we only use a height of 32. |
739 | 936k | const int this_height = xd->height * MI_SIZE; |
740 | 936k | const int pred_height = AOMMIN(this_height / 2, 32); |
741 | 936k | xd->mb_to_bottom_edge += GET_MV_SUBPEL(this_height - pred_height); |
742 | 936k | struct build_prediction_ctxt ctxt = { |
743 | 936k | cm, tmp_buf, tmp_width, tmp_height, tmp_stride, xd->mb_to_right_edge, dcb |
744 | 936k | }; |
745 | 936k | const BLOCK_SIZE bsize = xd->mi[0]->bsize; |
746 | 936k | foreach_overlappable_nb_above(cm, xd, |
747 | 936k | max_neighbor_obmc[mi_size_wide_log2[bsize]], |
748 | 936k | dec_build_prediction_by_above_pred, &ctxt); |
749 | | |
750 | 936k | xd->mb_to_left_edge = -GET_MV_SUBPEL(xd->mi_col * MI_SIZE); |
751 | 936k | xd->mb_to_right_edge = ctxt.mb_to_far_edge; |
752 | 936k | xd->mb_to_bottom_edge -= GET_MV_SUBPEL(this_height - pred_height); |
753 | 936k | } |
754 | | |
755 | | static INLINE void dec_build_prediction_by_left_pred( |
756 | | MACROBLOCKD *const xd, int rel_mi_row, int rel_mi_col, uint8_t op_mi_size, |
757 | 959k | int dir, MB_MODE_INFO *left_mbmi, void *fun_ctxt, const int num_planes) { |
758 | 959k | struct build_prediction_ctxt *ctxt = (struct build_prediction_ctxt *)fun_ctxt; |
759 | 959k | const int left_mi_row = xd->mi_row + rel_mi_row; |
760 | 959k | int mi_x, mi_y; |
761 | 959k | MB_MODE_INFO backup_mbmi = *left_mbmi; |
762 | | |
763 | 959k | (void)rel_mi_col; |
764 | 959k | (void)dir; |
765 | | |
766 | 959k | av1_setup_build_prediction_by_left_pred(xd, rel_mi_row, op_mi_size, |
767 | 959k | &backup_mbmi, ctxt, num_planes); |
768 | 959k | mi_x = xd->mi_col << MI_SIZE_LOG2; |
769 | 959k | mi_y = left_mi_row << MI_SIZE_LOG2; |
770 | 959k | const BLOCK_SIZE bsize = xd->mi[0]->bsize; |
771 | | |
772 | 3.83M | for (int j = 0; j < num_planes; ++j) { |
773 | 2.87M | const struct macroblockd_plane *pd = &xd->plane[j]; |
774 | 2.87M | int bw = clamp(block_size_wide[bsize] >> (pd->subsampling_x + 1), 4, |
775 | 2.87M | block_size_wide[BLOCK_64X64] >> (pd->subsampling_x + 1)); |
776 | 2.87M | int bh = (op_mi_size << MI_SIZE_LOG2) >> pd->subsampling_y; |
777 | | |
778 | 2.87M | if (av1_skip_u4x4_pred_in_obmc(bsize, pd, 1)) continue; |
779 | 2.87M | dec_build_inter_predictors(ctxt->cm, (DecoderCodingBlock *)ctxt->dcb, j, |
780 | 2.87M | &backup_mbmi, 1, bw, bh, mi_x, mi_y); |
781 | 2.87M | } |
782 | 959k | } |
783 | | |
784 | | static AOM_INLINE void dec_build_prediction_by_left_preds( |
785 | | const AV1_COMMON *cm, DecoderCodingBlock *dcb, |
786 | | uint8_t *tmp_buf[MAX_MB_PLANE], int tmp_width[MAX_MB_PLANE], |
787 | 949k | int tmp_height[MAX_MB_PLANE], int tmp_stride[MAX_MB_PLANE]) { |
788 | 949k | MACROBLOCKD *const xd = &dcb->xd; |
789 | 949k | if (!xd->left_available) return; |
790 | | |
791 | | // Adjust mb_to_right_edge to have the correct value for the OBMC |
792 | | // prediction block. This is half the width of the original block, |
793 | | // except for 128-wide blocks, where we only use a width of 32. |
794 | 939k | const int this_width = xd->width * MI_SIZE; |
795 | 939k | const int pred_width = AOMMIN(this_width / 2, 32); |
796 | 939k | xd->mb_to_right_edge += GET_MV_SUBPEL(this_width - pred_width); |
797 | | |
798 | 939k | struct build_prediction_ctxt ctxt = { |
799 | 939k | cm, tmp_buf, tmp_width, tmp_height, tmp_stride, xd->mb_to_bottom_edge, dcb |
800 | 939k | }; |
801 | 939k | const BLOCK_SIZE bsize = xd->mi[0]->bsize; |
802 | 939k | foreach_overlappable_nb_left(cm, xd, |
803 | 939k | max_neighbor_obmc[mi_size_high_log2[bsize]], |
804 | 939k | dec_build_prediction_by_left_pred, &ctxt); |
805 | | |
806 | 939k | xd->mb_to_top_edge = -GET_MV_SUBPEL(xd->mi_row * MI_SIZE); |
807 | 939k | xd->mb_to_right_edge -= GET_MV_SUBPEL(this_width - pred_width); |
808 | 939k | xd->mb_to_bottom_edge = ctxt.mb_to_far_edge; |
809 | 939k | } |
810 | | |
811 | | static AOM_INLINE void dec_build_obmc_inter_predictors_sb( |
812 | 949k | const AV1_COMMON *cm, DecoderCodingBlock *dcb) { |
813 | 949k | const int num_planes = av1_num_planes(cm); |
814 | 949k | uint8_t *dst_buf1[MAX_MB_PLANE], *dst_buf2[MAX_MB_PLANE]; |
815 | 949k | int dst_stride1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE }; |
816 | 949k | int dst_stride2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE }; |
817 | 949k | int dst_width1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE }; |
818 | 949k | int dst_width2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE }; |
819 | 949k | int dst_height1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE }; |
820 | 949k | int dst_height2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE }; |
821 | | |
822 | 949k | MACROBLOCKD *const xd = &dcb->xd; |
823 | 949k | av1_setup_obmc_dst_bufs(xd, dst_buf1, dst_buf2); |
824 | | |
825 | 949k | dec_build_prediction_by_above_preds(cm, dcb, dst_buf1, dst_width1, |
826 | 949k | dst_height1, dst_stride1); |
827 | 949k | dec_build_prediction_by_left_preds(cm, dcb, dst_buf2, dst_width2, dst_height2, |
828 | 949k | dst_stride2); |
829 | 949k | const int mi_row = xd->mi_row; |
830 | 949k | const int mi_col = xd->mi_col; |
831 | 949k | av1_setup_dst_planes(xd->plane, xd->mi[0]->bsize, &cm->cur_frame->buf, mi_row, |
832 | 949k | mi_col, 0, num_planes); |
833 | 949k | av1_build_obmc_inter_prediction(cm, xd, dst_buf1, dst_stride1, dst_buf2, |
834 | 949k | dst_stride2); |
835 | 949k | } |
836 | | |
837 | | static AOM_INLINE void cfl_store_inter_block(AV1_COMMON *const cm, |
838 | 5.66M | MACROBLOCKD *const xd) { |
839 | 5.66M | MB_MODE_INFO *mbmi = xd->mi[0]; |
840 | 5.66M | if (store_cfl_required(cm, xd)) { |
841 | 923k | cfl_store_block(xd, mbmi->bsize, mbmi->tx_size); |
842 | 923k | } |
843 | 5.66M | } |
844 | | |
845 | | static AOM_INLINE void predict_inter_block(AV1_COMMON *const cm, |
846 | | DecoderCodingBlock *dcb, |
847 | 5.66M | BLOCK_SIZE bsize) { |
848 | 5.66M | MACROBLOCKD *const xd = &dcb->xd; |
849 | 5.66M | MB_MODE_INFO *mbmi = xd->mi[0]; |
850 | 5.66M | const int num_planes = av1_num_planes(cm); |
851 | 5.66M | const int mi_row = xd->mi_row; |
852 | 5.66M | const int mi_col = xd->mi_col; |
853 | 12.3M | for (int ref = 0; ref < 1 + has_second_ref(mbmi); ++ref) { |
854 | 6.64M | const MV_REFERENCE_FRAME frame = mbmi->ref_frame[ref]; |
855 | 6.64M | if (frame < LAST_FRAME) { |
856 | 58.9k | assert(is_intrabc_block(mbmi)); |
857 | 58.9k | assert(frame == INTRA_FRAME); |
858 | 58.9k | assert(ref == 0); |
859 | 6.58M | } else { |
860 | 6.58M | const RefCntBuffer *ref_buf = get_ref_frame_buf(cm, frame); |
861 | 6.58M | const struct scale_factors *ref_scale_factors = |
862 | 6.58M | get_ref_scale_factors_const(cm, frame); |
863 | | |
864 | 6.58M | xd->block_ref_scale_factors[ref] = ref_scale_factors; |
865 | 6.58M | av1_setup_pre_planes(xd, ref, &ref_buf->buf, mi_row, mi_col, |
866 | 6.58M | ref_scale_factors, num_planes); |
867 | 6.58M | } |
868 | 6.64M | } |
869 | | |
870 | 5.66M | dec_build_inter_predictor(cm, dcb, mi_row, mi_col, bsize); |
871 | 5.66M | if (mbmi->motion_mode == OBMC_CAUSAL) { |
872 | 949k | dec_build_obmc_inter_predictors_sb(cm, dcb); |
873 | 949k | } |
874 | | #if CONFIG_MISMATCH_DEBUG |
875 | | for (int plane = 0; plane < num_planes; ++plane) { |
876 | | const struct macroblockd_plane *pd = &xd->plane[plane]; |
877 | | int pixel_c, pixel_r; |
878 | | mi_to_pixel_loc(&pixel_c, &pixel_r, mi_col, mi_row, 0, 0, pd->subsampling_x, |
879 | | pd->subsampling_y); |
880 | | if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x, |
881 | | pd->subsampling_y)) |
882 | | continue; |
883 | | mismatch_check_block_pre(pd->dst.buf, pd->dst.stride, |
884 | | cm->current_frame.order_hint, plane, pixel_c, |
885 | | pixel_r, pd->width, pd->height, |
886 | | xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH); |
887 | | } |
888 | | #endif |
889 | 5.66M | } |
890 | | |
891 | | static AOM_INLINE void set_color_index_map_offset(MACROBLOCKD *const xd, |
892 | 101k | int plane, aom_reader *r) { |
893 | 101k | (void)r; |
894 | 101k | Av1ColorMapParam params; |
895 | 101k | const MB_MODE_INFO *const mbmi = xd->mi[0]; |
896 | 101k | av1_get_block_dimensions(mbmi->bsize, plane, xd, ¶ms.plane_width, |
897 | 101k | ¶ms.plane_height, NULL, NULL); |
898 | 101k | xd->color_index_map_offset[plane] += params.plane_width * params.plane_height; |
899 | 101k | } |
900 | | |
901 | | static AOM_INLINE void decode_token_recon_block(AV1Decoder *const pbi, |
902 | | ThreadData *const td, |
903 | | aom_reader *r, |
904 | 26.1M | BLOCK_SIZE bsize) { |
905 | 26.1M | AV1_COMMON *const cm = &pbi->common; |
906 | 26.1M | DecoderCodingBlock *const dcb = &td->dcb; |
907 | 26.1M | MACROBLOCKD *const xd = &dcb->xd; |
908 | 26.1M | const int num_planes = av1_num_planes(cm); |
909 | 26.1M | MB_MODE_INFO *mbmi = xd->mi[0]; |
910 | | |
911 | 26.1M | if (!is_inter_block(mbmi)) { |
912 | 15.1M | int row, col; |
913 | 15.1M | assert(bsize == get_plane_block_size(bsize, xd->plane[0].subsampling_x, |
914 | 15.1M | xd->plane[0].subsampling_y)); |
915 | 15.1M | const int max_blocks_wide = max_block_wide(xd, bsize, 0); |
916 | 15.1M | const int max_blocks_high = max_block_high(xd, bsize, 0); |
917 | 15.1M | const BLOCK_SIZE max_unit_bsize = BLOCK_64X64; |
918 | 15.1M | int mu_blocks_wide = mi_size_wide[max_unit_bsize]; |
919 | 15.1M | int mu_blocks_high = mi_size_high[max_unit_bsize]; |
920 | 15.1M | mu_blocks_wide = AOMMIN(max_blocks_wide, mu_blocks_wide); |
921 | 15.1M | mu_blocks_high = AOMMIN(max_blocks_high, mu_blocks_high); |
922 | | |
923 | 30.3M | for (row = 0; row < max_blocks_high; row += mu_blocks_high) { |
924 | 30.9M | for (col = 0; col < max_blocks_wide; col += mu_blocks_wide) { |
925 | 59.6M | for (int plane = 0; plane < num_planes; ++plane) { |
926 | 45.0M | if (plane && !xd->is_chroma_ref) break; |
927 | 43.9M | const struct macroblockd_plane *const pd = &xd->plane[plane]; |
928 | 43.9M | const TX_SIZE tx_size = av1_get_tx_size(plane, xd); |
929 | 43.9M | const int stepr = tx_size_high_unit[tx_size]; |
930 | 43.9M | const int stepc = tx_size_wide_unit[tx_size]; |
931 | | |
932 | 43.9M | const int unit_height = ROUND_POWER_OF_TWO( |
933 | 43.9M | AOMMIN(mu_blocks_high + row, max_blocks_high), pd->subsampling_y); |
934 | 43.9M | const int unit_width = ROUND_POWER_OF_TWO( |
935 | 43.9M | AOMMIN(mu_blocks_wide + col, max_blocks_wide), pd->subsampling_x); |
936 | | |
937 | 99.4M | for (int blk_row = row >> pd->subsampling_y; blk_row < unit_height; |
938 | 55.5M | blk_row += stepr) { |
939 | 219M | for (int blk_col = col >> pd->subsampling_x; blk_col < unit_width; |
940 | 164M | blk_col += stepc) { |
941 | 164M | td->read_coeffs_tx_intra_block_visit(cm, dcb, r, plane, blk_row, |
942 | 164M | blk_col, tx_size); |
943 | 164M | td->predict_and_recon_intra_block_visit( |
944 | 164M | cm, dcb, r, plane, blk_row, blk_col, tx_size); |
945 | 164M | set_cb_buffer_offsets(dcb, tx_size, plane); |
946 | 164M | } |
947 | 55.5M | } |
948 | 43.9M | } |
949 | 15.6M | } |
950 | 15.2M | } |
951 | 15.1M | } else { |
952 | 11.0M | td->predict_inter_block_visit(cm, dcb, bsize); |
953 | | // Reconstruction |
954 | 11.0M | if (!mbmi->skip_txfm) { |
955 | 6.84M | int eobtotal = 0; |
956 | | |
957 | 6.84M | const int max_blocks_wide = max_block_wide(xd, bsize, 0); |
958 | 6.84M | const int max_blocks_high = max_block_high(xd, bsize, 0); |
959 | 6.84M | int row, col; |
960 | | |
961 | 6.84M | const BLOCK_SIZE max_unit_bsize = BLOCK_64X64; |
962 | 6.84M | assert(max_unit_bsize == |
963 | 6.84M | get_plane_block_size(BLOCK_64X64, xd->plane[0].subsampling_x, |
964 | 6.84M | xd->plane[0].subsampling_y)); |
965 | 6.82M | int mu_blocks_wide = mi_size_wide[max_unit_bsize]; |
966 | 6.82M | int mu_blocks_high = mi_size_high[max_unit_bsize]; |
967 | | |
968 | 6.82M | mu_blocks_wide = AOMMIN(max_blocks_wide, mu_blocks_wide); |
969 | 6.82M | mu_blocks_high = AOMMIN(max_blocks_high, mu_blocks_high); |
970 | | |
971 | 13.6M | for (row = 0; row < max_blocks_high; row += mu_blocks_high) { |
972 | 13.7M | for (col = 0; col < max_blocks_wide; col += mu_blocks_wide) { |
973 | 24.9M | for (int plane = 0; plane < num_planes; ++plane) { |
974 | 19.3M | if (plane && !xd->is_chroma_ref) break; |
975 | 18.0M | const struct macroblockd_plane *const pd = &xd->plane[plane]; |
976 | 18.0M | const int ss_x = pd->subsampling_x; |
977 | 18.0M | const int ss_y = pd->subsampling_y; |
978 | 18.0M | const BLOCK_SIZE plane_bsize = |
979 | 18.0M | get_plane_block_size(bsize, ss_x, ss_y); |
980 | 18.0M | const TX_SIZE max_tx_size = |
981 | 18.0M | get_vartx_max_txsize(xd, plane_bsize, plane); |
982 | 18.0M | const int bh_var_tx = tx_size_high_unit[max_tx_size]; |
983 | 18.0M | const int bw_var_tx = tx_size_wide_unit[max_tx_size]; |
984 | 18.0M | int block = 0; |
985 | 18.0M | int step = |
986 | 18.0M | tx_size_wide_unit[max_tx_size] * tx_size_high_unit[max_tx_size]; |
987 | 18.0M | int blk_row, blk_col; |
988 | 18.0M | const int unit_height = ROUND_POWER_OF_TWO( |
989 | 18.0M | AOMMIN(mu_blocks_high + row, max_blocks_high), ss_y); |
990 | 18.0M | const int unit_width = ROUND_POWER_OF_TWO( |
991 | 18.0M | AOMMIN(mu_blocks_wide + col, max_blocks_wide), ss_x); |
992 | | |
993 | 36.0M | for (blk_row = row >> ss_y; blk_row < unit_height; |
994 | 18.0M | blk_row += bh_var_tx) { |
995 | 37.2M | for (blk_col = col >> ss_x; blk_col < unit_width; |
996 | 19.1M | blk_col += bw_var_tx) { |
997 | 19.1M | decode_reconstruct_tx(cm, td, r, mbmi, plane, plane_bsize, |
998 | 19.1M | blk_row, blk_col, block, max_tx_size, |
999 | 19.1M | &eobtotal); |
1000 | 19.1M | block += step; |
1001 | 19.1M | } |
1002 | 18.0M | } |
1003 | 18.0M | } |
1004 | 6.90M | } |
1005 | 6.85M | } |
1006 | 6.82M | } |
1007 | 10.9M | td->cfl_store_inter_block_visit(cm, xd); |
1008 | 10.9M | } |
1009 | | |
1010 | 26.1M | av1_visit_palette(pbi, xd, r, set_color_index_map_offset); |
1011 | 26.1M | } |
1012 | | |
1013 | | static AOM_INLINE void set_inter_tx_size(MB_MODE_INFO *mbmi, int stride_log2, |
1014 | | int tx_w_log2, int tx_h_log2, |
1015 | | int min_txs, int split_size, int txs, |
1016 | 1.17M | int blk_row, int blk_col) { |
1017 | 2.76M | for (int idy = 0; idy < tx_size_high_unit[split_size]; |
1018 | 1.58M | idy += tx_size_high_unit[min_txs]) { |
1019 | 3.94M | for (int idx = 0; idx < tx_size_wide_unit[split_size]; |
1020 | 2.35M | idx += tx_size_wide_unit[min_txs]) { |
1021 | 2.35M | const int index = (((blk_row + idy) >> tx_h_log2) << stride_log2) + |
1022 | 2.35M | ((blk_col + idx) >> tx_w_log2); |
1023 | 2.35M | mbmi->inter_tx_size[index] = txs; |
1024 | 2.35M | } |
1025 | 1.58M | } |
1026 | 1.17M | } |
1027 | | |
1028 | | static AOM_INLINE void read_tx_size_vartx(MACROBLOCKD *xd, MB_MODE_INFO *mbmi, |
1029 | | TX_SIZE tx_size, int depth, |
1030 | | int blk_row, int blk_col, |
1031 | 1.38M | aom_reader *r) { |
1032 | 1.38M | FRAME_CONTEXT *ec_ctx = xd->tile_ctx; |
1033 | 1.38M | int is_split = 0; |
1034 | 1.38M | const BLOCK_SIZE bsize = mbmi->bsize; |
1035 | 1.38M | const int max_blocks_high = max_block_high(xd, bsize, 0); |
1036 | 1.38M | const int max_blocks_wide = max_block_wide(xd, bsize, 0); |
1037 | 1.38M | if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return; |
1038 | 1.38M | assert(tx_size > TX_4X4); |
1039 | 1.38M | TX_SIZE txs = max_txsize_rect_lookup[bsize]; |
1040 | 2.77M | for (int level = 0; level < MAX_VARTX_DEPTH - 1; ++level) |
1041 | 1.38M | txs = sub_tx_size_map[txs]; |
1042 | 1.38M | const int tx_w_log2 = tx_size_wide_log2[txs] - MI_SIZE_LOG2; |
1043 | 1.38M | const int tx_h_log2 = tx_size_high_log2[txs] - MI_SIZE_LOG2; |
1044 | 1.38M | const int bw_log2 = mi_size_wide_log2[bsize]; |
1045 | 1.38M | const int stride_log2 = bw_log2 - tx_w_log2; |
1046 | | |
1047 | 1.38M | if (depth == MAX_VARTX_DEPTH) { |
1048 | 141k | set_inter_tx_size(mbmi, stride_log2, tx_w_log2, tx_h_log2, txs, tx_size, |
1049 | 141k | tx_size, blk_row, blk_col); |
1050 | 141k | mbmi->tx_size = tx_size; |
1051 | 141k | txfm_partition_update(xd->above_txfm_context + blk_col, |
1052 | 141k | xd->left_txfm_context + blk_row, tx_size, tx_size); |
1053 | 141k | return; |
1054 | 141k | } |
1055 | | |
1056 | 1.24M | const int ctx = txfm_partition_context(xd->above_txfm_context + blk_col, |
1057 | 1.24M | xd->left_txfm_context + blk_row, |
1058 | 1.24M | mbmi->bsize, tx_size); |
1059 | 1.24M | is_split = aom_read_symbol(r, ec_ctx->txfm_partition_cdf[ctx], 2, ACCT_STR); |
1060 | | |
1061 | 1.24M | if (is_split) { |
1062 | 289k | const TX_SIZE sub_txs = sub_tx_size_map[tx_size]; |
1063 | 289k | const int bsw = tx_size_wide_unit[sub_txs]; |
1064 | 289k | const int bsh = tx_size_high_unit[sub_txs]; |
1065 | | |
1066 | 289k | if (sub_txs == TX_4X4) { |
1067 | 73.7k | set_inter_tx_size(mbmi, stride_log2, tx_w_log2, tx_h_log2, txs, tx_size, |
1068 | 73.7k | sub_txs, blk_row, blk_col); |
1069 | 73.7k | mbmi->tx_size = sub_txs; |
1070 | 73.7k | txfm_partition_update(xd->above_txfm_context + blk_col, |
1071 | 73.7k | xd->left_txfm_context + blk_row, sub_txs, tx_size); |
1072 | 73.7k | return; |
1073 | 73.7k | } |
1074 | | |
1075 | 215k | assert(bsw > 0 && bsh > 0); |
1076 | 563k | for (int row = 0; row < tx_size_high_unit[tx_size]; row += bsh) { |
1077 | 925k | for (int col = 0; col < tx_size_wide_unit[tx_size]; col += bsw) { |
1078 | 577k | int offsetr = blk_row + row; |
1079 | 577k | int offsetc = blk_col + col; |
1080 | 577k | read_tx_size_vartx(xd, mbmi, sub_txs, depth + 1, offsetr, offsetc, r); |
1081 | 577k | } |
1082 | 347k | } |
1083 | 955k | } else { |
1084 | 955k | set_inter_tx_size(mbmi, stride_log2, tx_w_log2, tx_h_log2, txs, tx_size, |
1085 | 955k | tx_size, blk_row, blk_col); |
1086 | 955k | mbmi->tx_size = tx_size; |
1087 | 955k | txfm_partition_update(xd->above_txfm_context + blk_col, |
1088 | 955k | xd->left_txfm_context + blk_row, tx_size, tx_size); |
1089 | 955k | } |
1090 | 1.24M | } |
1091 | | |
1092 | | static TX_SIZE read_selected_tx_size(const MACROBLOCKD *const xd, |
1093 | 1.48M | aom_reader *r) { |
1094 | | // TODO(debargha): Clean up the logic here. This function should only |
1095 | | // be called for intra. |
1096 | 1.48M | const BLOCK_SIZE bsize = xd->mi[0]->bsize; |
1097 | 1.48M | const int32_t tx_size_cat = bsize_to_tx_size_cat(bsize); |
1098 | 1.48M | const int max_depths = bsize_to_max_depth(bsize); |
1099 | 1.48M | const int ctx = get_tx_size_context(xd); |
1100 | 1.48M | FRAME_CONTEXT *ec_ctx = xd->tile_ctx; |
1101 | 1.48M | const int depth = aom_read_symbol(r, ec_ctx->tx_size_cdf[tx_size_cat][ctx], |
1102 | 1.48M | max_depths + 1, ACCT_STR); |
1103 | 1.48M | assert(depth >= 0 && depth <= max_depths); |
1104 | 1.48M | const TX_SIZE tx_size = depth_to_tx_size(depth, bsize); |
1105 | 1.48M | return tx_size; |
1106 | 1.48M | } |
1107 | | |
1108 | | static TX_SIZE read_tx_size(const MACROBLOCKD *const xd, TX_MODE tx_mode, |
1109 | | int is_inter, int allow_select_inter, |
1110 | 15.8M | aom_reader *r) { |
1111 | 15.8M | const BLOCK_SIZE bsize = xd->mi[0]->bsize; |
1112 | 15.8M | if (xd->lossless[xd->mi[0]->segment_id]) return TX_4X4; |
1113 | | |
1114 | 15.7M | if (block_signals_txsize(bsize)) { |
1115 | 14.8M | if ((!is_inter || allow_select_inter) && tx_mode == TX_MODE_SELECT) { |
1116 | 1.48M | const TX_SIZE coded_tx_size = read_selected_tx_size(xd, r); |
1117 | 1.48M | return coded_tx_size; |
1118 | 13.4M | } else { |
1119 | 13.4M | return tx_size_from_tx_mode(bsize, tx_mode); |
1120 | 13.4M | } |
1121 | 14.8M | } else { |
1122 | 883k | assert(IMPLIES(tx_mode == ONLY_4X4, bsize == BLOCK_4X4)); |
1123 | 884k | return max_txsize_rect_lookup[bsize]; |
1124 | 883k | } |
1125 | 15.7M | } |
1126 | | |
1127 | | static AOM_INLINE void parse_decode_block(AV1Decoder *const pbi, |
1128 | | ThreadData *const td, int mi_row, |
1129 | | int mi_col, aom_reader *r, |
1130 | | PARTITION_TYPE partition, |
1131 | 16.6M | BLOCK_SIZE bsize) { |
1132 | 16.6M | DecoderCodingBlock *const dcb = &td->dcb; |
1133 | 16.6M | MACROBLOCKD *const xd = &dcb->xd; |
1134 | 16.6M | decode_mbmi_block(pbi, dcb, mi_row, mi_col, r, partition, bsize); |
1135 | | |
1136 | 16.6M | av1_visit_palette(pbi, xd, r, av1_decode_palette_tokens); |
1137 | | |
1138 | 16.6M | AV1_COMMON *cm = &pbi->common; |
1139 | 16.6M | const int num_planes = av1_num_planes(cm); |
1140 | 16.6M | MB_MODE_INFO *mbmi = xd->mi[0]; |
1141 | 16.6M | int inter_block_tx = is_inter_block(mbmi) || is_intrabc_block(mbmi); |
1142 | 16.6M | if (cm->features.tx_mode == TX_MODE_SELECT && block_signals_txsize(bsize) && |
1143 | 16.6M | !mbmi->skip_txfm && inter_block_tx && !xd->lossless[mbmi->segment_id]) { |
1144 | 791k | const TX_SIZE max_tx_size = max_txsize_rect_lookup[bsize]; |
1145 | 791k | const int bh = tx_size_high_unit[max_tx_size]; |
1146 | 791k | const int bw = tx_size_wide_unit[max_tx_size]; |
1147 | 791k | const int width = mi_size_wide[bsize]; |
1148 | 791k | const int height = mi_size_high[bsize]; |
1149 | | |
1150 | 1.58M | for (int idy = 0; idy < height; idy += bh) |
1151 | 1.60M | for (int idx = 0; idx < width; idx += bw) |
1152 | 810k | read_tx_size_vartx(xd, mbmi, max_tx_size, 0, idy, idx, r); |
1153 | 15.9M | } else { |
1154 | 15.9M | mbmi->tx_size = read_tx_size(xd, cm->features.tx_mode, inter_block_tx, |
1155 | 15.9M | !mbmi->skip_txfm, r); |
1156 | 15.9M | if (inter_block_tx) |
1157 | 5.23M | memset(mbmi->inter_tx_size, mbmi->tx_size, sizeof(mbmi->inter_tx_size)); |
1158 | 15.9M | set_txfm_ctxs(mbmi->tx_size, xd->width, xd->height, |
1159 | 15.9M | mbmi->skip_txfm && is_inter_block(mbmi), xd); |
1160 | 15.9M | } |
1161 | | |
1162 | 16.6M | if (cm->delta_q_info.delta_q_present_flag) { |
1163 | 18.2M | for (int i = 0; i < MAX_SEGMENTS; i++) { |
1164 | 16.2M | const int current_qindex = |
1165 | 16.2M | av1_get_qindex(&cm->seg, i, xd->current_base_qindex); |
1166 | 16.2M | const CommonQuantParams *const quant_params = &cm->quant_params; |
1167 | 64.3M | for (int j = 0; j < num_planes; ++j) { |
1168 | 48.1M | const int dc_delta_q = j == 0 ? quant_params->y_dc_delta_q |
1169 | 48.1M | : (j == 1 ? quant_params->u_dc_delta_q |
1170 | 31.8M | : quant_params->v_dc_delta_q); |
1171 | 48.1M | const int ac_delta_q = j == 0 ? 0 |
1172 | 48.1M | : (j == 1 ? quant_params->u_ac_delta_q |
1173 | 31.8M | : quant_params->v_ac_delta_q); |
1174 | 48.1M | xd->plane[j].seg_dequant_QTX[i][0] = av1_dc_quant_QTX( |
1175 | 48.1M | current_qindex, dc_delta_q, cm->seq_params->bit_depth); |
1176 | 48.1M | xd->plane[j].seg_dequant_QTX[i][1] = av1_ac_quant_QTX( |
1177 | 48.1M | current_qindex, ac_delta_q, cm->seq_params->bit_depth); |
1178 | 48.1M | } |
1179 | 16.2M | } |
1180 | 2.03M | } |
1181 | 16.6M | if (mbmi->skip_txfm) av1_reset_entropy_context(xd, bsize, num_planes); |
1182 | | |
1183 | 16.6M | decode_token_recon_block(pbi, td, r, bsize); |
1184 | 16.6M | } |
1185 | | |
1186 | | static AOM_INLINE void set_offsets_for_pred_and_recon(AV1Decoder *const pbi, |
1187 | | ThreadData *const td, |
1188 | | int mi_row, int mi_col, |
1189 | 9.47M | BLOCK_SIZE bsize) { |
1190 | 9.47M | AV1_COMMON *const cm = &pbi->common; |
1191 | 9.47M | const CommonModeInfoParams *const mi_params = &cm->mi_params; |
1192 | 9.47M | DecoderCodingBlock *const dcb = &td->dcb; |
1193 | 9.47M | MACROBLOCKD *const xd = &dcb->xd; |
1194 | 9.47M | const int bw = mi_size_wide[bsize]; |
1195 | 9.47M | const int bh = mi_size_high[bsize]; |
1196 | 9.47M | const int num_planes = av1_num_planes(cm); |
1197 | | |
1198 | 9.47M | const int offset = mi_row * mi_params->mi_stride + mi_col; |
1199 | 9.47M | const TileInfo *const tile = &xd->tile; |
1200 | | |
1201 | 9.47M | xd->mi = mi_params->mi_grid_base + offset; |
1202 | 9.47M | xd->tx_type_map = |
1203 | 9.47M | &mi_params->tx_type_map[mi_row * mi_params->mi_stride + mi_col]; |
1204 | 9.47M | xd->tx_type_map_stride = mi_params->mi_stride; |
1205 | | |
1206 | 9.47M | set_plane_n4(xd, bw, bh, num_planes); |
1207 | | |
1208 | | // Distance of Mb to the various image edges. These are specified to 8th pel |
1209 | | // as they are always compared to values that are in 1/8th pel units |
1210 | 9.47M | set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, mi_params->mi_rows, |
1211 | 9.47M | mi_params->mi_cols); |
1212 | | |
1213 | 9.47M | av1_setup_dst_planes(xd->plane, bsize, &cm->cur_frame->buf, mi_row, mi_col, 0, |
1214 | 9.47M | num_planes); |
1215 | 9.47M | } |
1216 | | |
1217 | | static AOM_INLINE void decode_block(AV1Decoder *const pbi, ThreadData *const td, |
1218 | | int mi_row, int mi_col, aom_reader *r, |
1219 | | PARTITION_TYPE partition, |
1220 | 9.47M | BLOCK_SIZE bsize) { |
1221 | 9.47M | (void)partition; |
1222 | 9.47M | set_offsets_for_pred_and_recon(pbi, td, mi_row, mi_col, bsize); |
1223 | 9.47M | decode_token_recon_block(pbi, td, r, bsize); |
1224 | 9.47M | } |
1225 | | |
1226 | | static PARTITION_TYPE read_partition(MACROBLOCKD *xd, int mi_row, int mi_col, |
1227 | | aom_reader *r, int has_rows, int has_cols, |
1228 | 11.4M | BLOCK_SIZE bsize) { |
1229 | 11.4M | const int ctx = partition_plane_context(xd, mi_row, mi_col, bsize); |
1230 | 11.4M | FRAME_CONTEXT *ec_ctx = xd->tile_ctx; |
1231 | | |
1232 | 11.4M | if (!has_rows && !has_cols) return PARTITION_SPLIT; |
1233 | | |
1234 | 11.3M | assert(ctx >= 0); |
1235 | 11.3M | aom_cdf_prob *partition_cdf = ec_ctx->partition_cdf[ctx]; |
1236 | 11.3M | if (has_rows && has_cols) { |
1237 | 10.9M | return (PARTITION_TYPE)aom_read_symbol( |
1238 | 10.9M | r, partition_cdf, partition_cdf_length(bsize), ACCT_STR); |
1239 | 10.9M | } else if (!has_rows && has_cols) { |
1240 | 149k | assert(bsize > BLOCK_8X8); |
1241 | 149k | aom_cdf_prob cdf[2]; |
1242 | 149k | partition_gather_vert_alike(cdf, partition_cdf, bsize); |
1243 | 149k | assert(cdf[1] == AOM_ICDF(CDF_PROB_TOP)); |
1244 | 149k | return aom_read_cdf(r, cdf, 2, ACCT_STR) ? PARTITION_SPLIT : PARTITION_HORZ; |
1245 | 306k | } else { |
1246 | 306k | assert(has_rows && !has_cols); |
1247 | 306k | assert(bsize > BLOCK_8X8); |
1248 | 306k | aom_cdf_prob cdf[2]; |
1249 | 306k | partition_gather_horz_alike(cdf, partition_cdf, bsize); |
1250 | 306k | assert(cdf[1] == AOM_ICDF(CDF_PROB_TOP)); |
1251 | 306k | return aom_read_cdf(r, cdf, 2, ACCT_STR) ? PARTITION_SPLIT : PARTITION_VERT; |
1252 | 306k | } |
1253 | 11.3M | } |
1254 | | |
1255 | | // TODO(slavarnway): eliminate bsize and subsize in future commits |
1256 | | static AOM_INLINE void decode_partition(AV1Decoder *const pbi, |
1257 | | ThreadData *const td, int mi_row, |
1258 | | int mi_col, aom_reader *reader, |
1259 | | BLOCK_SIZE bsize, |
1260 | 20.3M | int parse_decode_flag) { |
1261 | 20.3M | assert(bsize < BLOCK_SIZES_ALL); |
1262 | 20.3M | AV1_COMMON *const cm = &pbi->common; |
1263 | 20.3M | DecoderCodingBlock *const dcb = &td->dcb; |
1264 | 20.3M | MACROBLOCKD *const xd = &dcb->xd; |
1265 | 20.3M | const int bw = mi_size_wide[bsize]; |
1266 | 20.3M | const int hbs = bw >> 1; |
1267 | 20.3M | PARTITION_TYPE partition; |
1268 | 20.3M | BLOCK_SIZE subsize; |
1269 | 20.3M | const int quarter_step = bw / 4; |
1270 | 20.3M | BLOCK_SIZE bsize2 = get_partition_subsize(bsize, PARTITION_SPLIT); |
1271 | 20.3M | const int has_rows = (mi_row + hbs) < cm->mi_params.mi_rows; |
1272 | 20.3M | const int has_cols = (mi_col + hbs) < cm->mi_params.mi_cols; |
1273 | | |
1274 | 20.3M | if (mi_row >= cm->mi_params.mi_rows || mi_col >= cm->mi_params.mi_cols) |
1275 | 858k | return; |
1276 | | |
1277 | | // parse_decode_flag takes the following values : |
1278 | | // 01 - do parse only |
1279 | | // 10 - do decode only |
1280 | | // 11 - do parse and decode |
1281 | 19.4M | static const block_visitor_fn_t block_visit[4] = { NULL, parse_decode_block, |
1282 | 19.4M | decode_block, |
1283 | 19.4M | parse_decode_block }; |
1284 | | |
1285 | 19.4M | if (parse_decode_flag & 1) { |
1286 | 12.3M | const int num_planes = av1_num_planes(cm); |
1287 | 48.9M | for (int plane = 0; plane < num_planes; ++plane) { |
1288 | 36.6M | int rcol0, rcol1, rrow0, rrow1; |
1289 | | |
1290 | | // Skip some unnecessary work if loop restoration is disabled |
1291 | 36.6M | if (cm->rst_info[plane].frame_restoration_type == RESTORE_NONE) continue; |
1292 | | |
1293 | 20.4M | if (av1_loop_restoration_corners_in_sb(cm, plane, mi_row, mi_col, bsize, |
1294 | 20.4M | &rcol0, &rcol1, &rrow0, &rrow1)) { |
1295 | 265k | const int rstride = cm->rst_info[plane].horz_units; |
1296 | 530k | for (int rrow = rrow0; rrow < rrow1; ++rrow) { |
1297 | 532k | for (int rcol = rcol0; rcol < rcol1; ++rcol) { |
1298 | 267k | const int runit_idx = rcol + rrow * rstride; |
1299 | 267k | loop_restoration_read_sb_coeffs(cm, xd, reader, plane, runit_idx); |
1300 | 267k | } |
1301 | 265k | } |
1302 | 265k | } |
1303 | 20.4M | } |
1304 | | |
1305 | 12.3M | partition = (bsize < BLOCK_8X8) ? PARTITION_NONE |
1306 | 12.3M | : read_partition(xd, mi_row, mi_col, reader, |
1307 | 11.4M | has_rows, has_cols, bsize); |
1308 | 12.3M | } else { |
1309 | 7.12M | partition = get_partition(cm, mi_row, mi_col, bsize); |
1310 | 7.12M | } |
1311 | 19.4M | subsize = get_partition_subsize(bsize, partition); |
1312 | 19.4M | if (subsize == BLOCK_INVALID) { |
1313 | | // When an internal error occurs ensure that xd->mi_row is set appropriately |
1314 | | // w.r.t. current tile, which is used to signal processing of current row is |
1315 | | // done. |
1316 | 0 | xd->mi_row = mi_row; |
1317 | 0 | aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME, |
1318 | 0 | "Partition is invalid for block size %dx%d", |
1319 | 0 | block_size_wide[bsize], block_size_high[bsize]); |
1320 | 0 | } |
1321 | | // Check the bitstream is conformant: if there is subsampling on the |
1322 | | // chroma planes, subsize must subsample to a valid block size. |
1323 | 19.4M | const struct macroblockd_plane *const pd_u = &xd->plane[1]; |
1324 | 19.4M | if (get_plane_block_size(subsize, pd_u->subsampling_x, pd_u->subsampling_y) == |
1325 | 19.4M | BLOCK_INVALID) { |
1326 | | // When an internal error occurs ensure that xd->mi_row is set appropriately |
1327 | | // w.r.t. current tile, which is used to signal processing of current row is |
1328 | | // done. |
1329 | 694 | xd->mi_row = mi_row; |
1330 | 694 | aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME, |
1331 | 694 | "Block size %dx%d invalid with this subsampling mode", |
1332 | 694 | block_size_wide[subsize], block_size_high[subsize]); |
1333 | 694 | } |
1334 | | |
1335 | 19.4M | #define DEC_BLOCK_STX_ARG |
1336 | 26.1M | #define DEC_BLOCK_EPT_ARG partition, |
1337 | 19.4M | #define DEC_BLOCK(db_r, db_c, db_subsize) \ |
1338 | 26.1M | block_visit[parse_decode_flag](pbi, td, DEC_BLOCK_STX_ARG(db_r), (db_c), \ |
1339 | 26.1M | reader, DEC_BLOCK_EPT_ARG(db_subsize)) |
1340 | 19.4M | #define DEC_PARTITION(db_r, db_c, db_subsize) \ |
1341 | 19.4M | decode_partition(pbi, td, DEC_BLOCK_STX_ARG(db_r), (db_c), reader, \ |
1342 | 18.1M | (db_subsize), parse_decode_flag) |
1343 | | |
1344 | 19.4M | switch (partition) { |
1345 | 8.15M | case PARTITION_NONE: DEC_BLOCK(mi_row, mi_col, subsize); break; |
1346 | 1.99M | case PARTITION_HORZ: |
1347 | 1.99M | DEC_BLOCK(mi_row, mi_col, subsize); |
1348 | 1.99M | if (has_rows) DEC_BLOCK(mi_row + hbs, mi_col, subsize); |
1349 | 1.99M | break; |
1350 | 1.70M | case PARTITION_VERT: |
1351 | 1.70M | DEC_BLOCK(mi_row, mi_col, subsize); |
1352 | 1.70M | if (has_cols) DEC_BLOCK(mi_row, mi_col + hbs, subsize); |
1353 | 1.70M | break; |
1354 | 4.53M | case PARTITION_SPLIT: |
1355 | 4.53M | DEC_PARTITION(mi_row, mi_col, subsize); |
1356 | 4.53M | DEC_PARTITION(mi_row, mi_col + hbs, subsize); |
1357 | 4.53M | DEC_PARTITION(mi_row + hbs, mi_col, subsize); |
1358 | 4.53M | DEC_PARTITION(mi_row + hbs, mi_col + hbs, subsize); |
1359 | 4.53M | break; |
1360 | 384k | case PARTITION_HORZ_A: |
1361 | 384k | DEC_BLOCK(mi_row, mi_col, bsize2); |
1362 | 384k | DEC_BLOCK(mi_row, mi_col + hbs, bsize2); |
1363 | 384k | DEC_BLOCK(mi_row + hbs, mi_col, subsize); |
1364 | 384k | break; |
1365 | 398k | case PARTITION_HORZ_B: |
1366 | 398k | DEC_BLOCK(mi_row, mi_col, subsize); |
1367 | 398k | DEC_BLOCK(mi_row + hbs, mi_col, bsize2); |
1368 | 398k | DEC_BLOCK(mi_row + hbs, mi_col + hbs, bsize2); |
1369 | 398k | break; |
1370 | 282k | case PARTITION_VERT_A: |
1371 | 282k | DEC_BLOCK(mi_row, mi_col, bsize2); |
1372 | 282k | DEC_BLOCK(mi_row + hbs, mi_col, bsize2); |
1373 | 282k | DEC_BLOCK(mi_row, mi_col + hbs, subsize); |
1374 | 282k | break; |
1375 | 300k | case PARTITION_VERT_B: |
1376 | 300k | DEC_BLOCK(mi_row, mi_col, subsize); |
1377 | 300k | DEC_BLOCK(mi_row, mi_col + hbs, bsize2); |
1378 | 300k | DEC_BLOCK(mi_row + hbs, mi_col + hbs, bsize2); |
1379 | 300k | break; |
1380 | 1.11M | case PARTITION_HORZ_4: |
1381 | 5.56M | for (int i = 0; i < 4; ++i) { |
1382 | 4.45M | int this_mi_row = mi_row + i * quarter_step; |
1383 | 4.45M | if (i > 0 && this_mi_row >= cm->mi_params.mi_rows) break; |
1384 | 4.45M | DEC_BLOCK(this_mi_row, mi_col, subsize); |
1385 | 4.45M | } |
1386 | 1.11M | break; |
1387 | 634k | case PARTITION_VERT_4: |
1388 | 3.16M | for (int i = 0; i < 4; ++i) { |
1389 | 2.53M | int this_mi_col = mi_col + i * quarter_step; |
1390 | 2.53M | if (i > 0 && this_mi_col >= cm->mi_params.mi_cols) break; |
1391 | 2.53M | DEC_BLOCK(mi_row, this_mi_col, subsize); |
1392 | 2.53M | } |
1393 | 634k | break; |
1394 | 0 | default: assert(0 && "Invalid partition type"); |
1395 | 19.4M | } |
1396 | | |
1397 | 19.4M | #undef DEC_PARTITION |
1398 | 19.4M | #undef DEC_BLOCK |
1399 | 19.4M | #undef DEC_BLOCK_EPT_ARG |
1400 | 19.4M | #undef DEC_BLOCK_STX_ARG |
1401 | | |
1402 | 19.4M | if (parse_decode_flag & 1) |
1403 | 12.3M | update_ext_partition_context(xd, mi_row, mi_col, subsize, bsize, partition); |
1404 | 19.4M | } |
1405 | | |
1406 | | static AOM_INLINE void setup_bool_decoder( |
1407 | | MACROBLOCKD *const xd, const uint8_t *data, const uint8_t *data_end, |
1408 | | const size_t read_size, struct aom_internal_error_info *error_info, |
1409 | 205k | aom_reader *r, uint8_t allow_update_cdf) { |
1410 | | // Validate the calculated partition length. If the buffer |
1411 | | // described by the partition can't be fully read, then restrict |
1412 | | // it to the portion that can be (for EC mode) or throw an error. |
1413 | 205k | if (!read_is_valid(data, read_size, data_end)) { |
1414 | | // When internal error occurs ensure that xd->mi_row is set appropriately |
1415 | | // w.r.t. current tile, which is used to signal processing of current row is |
1416 | | // done in row-mt decoding. |
1417 | 236 | xd->mi_row = xd->tile.mi_row_start; |
1418 | | |
1419 | 236 | aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME, |
1420 | 236 | "Truncated packet or corrupt tile length"); |
1421 | 236 | } |
1422 | 205k | if (aom_reader_init(r, data, read_size)) { |
1423 | | // When internal error occurs ensure that xd->mi_row is set appropriately |
1424 | | // w.r.t. current tile, which is used to signal processing of current row is |
1425 | | // done in row-mt decoding. |
1426 | 0 | xd->mi_row = xd->tile.mi_row_start; |
1427 | |
|
1428 | 0 | aom_internal_error(error_info, AOM_CODEC_MEM_ERROR, |
1429 | 0 | "Failed to allocate bool decoder %d", 1); |
1430 | 0 | } |
1431 | | |
1432 | 205k | r->allow_update_cdf = allow_update_cdf; |
1433 | 205k | } |
1434 | | |
1435 | | static AOM_INLINE void setup_segmentation(AV1_COMMON *const cm, |
1436 | 208k | struct aom_read_bit_buffer *rb) { |
1437 | 208k | struct segmentation *const seg = &cm->seg; |
1438 | | |
1439 | 208k | seg->update_map = 0; |
1440 | 208k | seg->update_data = 0; |
1441 | 208k | seg->temporal_update = 0; |
1442 | | |
1443 | 208k | seg->enabled = aom_rb_read_bit(rb); |
1444 | 208k | if (!seg->enabled) { |
1445 | 184k | if (cm->cur_frame->seg_map) { |
1446 | 184k | memset(cm->cur_frame->seg_map, 0, |
1447 | 184k | (cm->cur_frame->mi_rows * cm->cur_frame->mi_cols)); |
1448 | 184k | } |
1449 | | |
1450 | 184k | memset(seg, 0, sizeof(*seg)); |
1451 | 184k | segfeatures_copy(&cm->cur_frame->seg, seg); |
1452 | 184k | return; |
1453 | 184k | } |
1454 | 23.7k | if (cm->seg.enabled && cm->prev_frame && |
1455 | 23.7k | (cm->mi_params.mi_rows == cm->prev_frame->mi_rows) && |
1456 | 23.7k | (cm->mi_params.mi_cols == cm->prev_frame->mi_cols)) { |
1457 | 8.28k | cm->last_frame_seg_map = cm->prev_frame->seg_map; |
1458 | 15.5k | } else { |
1459 | 15.5k | cm->last_frame_seg_map = NULL; |
1460 | 15.5k | } |
1461 | | // Read update flags |
1462 | 23.7k | if (cm->features.primary_ref_frame == PRIMARY_REF_NONE) { |
1463 | | // These frames can't use previous frames, so must signal map + features |
1464 | 13.1k | seg->update_map = 1; |
1465 | 13.1k | seg->temporal_update = 0; |
1466 | 13.1k | seg->update_data = 1; |
1467 | 13.1k | } else { |
1468 | 10.6k | seg->update_map = aom_rb_read_bit(rb); |
1469 | 10.6k | if (seg->update_map) { |
1470 | 5.53k | seg->temporal_update = aom_rb_read_bit(rb); |
1471 | 5.53k | } else { |
1472 | 5.12k | seg->temporal_update = 0; |
1473 | 5.12k | } |
1474 | 10.6k | seg->update_data = aom_rb_read_bit(rb); |
1475 | 10.6k | } |
1476 | | |
1477 | | // Segmentation data update |
1478 | 23.7k | if (seg->update_data) { |
1479 | 17.8k | av1_clearall_segfeatures(seg); |
1480 | | |
1481 | 157k | for (int i = 0; i < MAX_SEGMENTS; i++) { |
1482 | 1.24M | for (int j = 0; j < SEG_LVL_MAX; j++) { |
1483 | 1.10M | int data = 0; |
1484 | 1.10M | const int feature_enabled = aom_rb_read_bit(rb); |
1485 | 1.10M | if (feature_enabled) { |
1486 | 309k | av1_enable_segfeature(seg, i, j); |
1487 | | |
1488 | 309k | const int data_max = av1_seg_feature_data_max(j); |
1489 | 309k | const int data_min = -data_max; |
1490 | 309k | const int ubits = get_unsigned_bits(data_max); |
1491 | | |
1492 | 309k | if (av1_is_segfeature_signed(j)) { |
1493 | 177k | data = aom_rb_read_inv_signed_literal(rb, ubits); |
1494 | 177k | } else { |
1495 | 131k | data = aom_rb_read_literal(rb, ubits); |
1496 | 131k | } |
1497 | | |
1498 | 309k | data = clamp(data, data_min, data_max); |
1499 | 309k | } |
1500 | 1.10M | av1_set_segdata(seg, i, j, data); |
1501 | 1.10M | } |
1502 | 139k | } |
1503 | 17.8k | av1_calculate_segdata(seg); |
1504 | 17.8k | } else if (cm->prev_frame) { |
1505 | 5.92k | segfeatures_copy(seg, &cm->prev_frame->seg); |
1506 | 5.92k | } |
1507 | 23.7k | segfeatures_copy(&cm->cur_frame->seg, seg); |
1508 | 23.7k | } |
1509 | | |
1510 | | static AOM_INLINE void decode_restoration_mode(AV1_COMMON *cm, |
1511 | 129k | struct aom_read_bit_buffer *rb) { |
1512 | 129k | assert(!cm->features.all_lossless); |
1513 | 129k | const int num_planes = av1_num_planes(cm); |
1514 | 129k | if (cm->features.allow_intrabc) return; |
1515 | 116k | int all_none = 1, chroma_none = 1; |
1516 | 460k | for (int p = 0; p < num_planes; ++p) { |
1517 | 344k | RestorationInfo *rsi = &cm->rst_info[p]; |
1518 | 344k | if (aom_rb_read_bit(rb)) { |
1519 | 46.0k | rsi->frame_restoration_type = |
1520 | 46.0k | aom_rb_read_bit(rb) ? RESTORE_SGRPROJ : RESTORE_WIENER; |
1521 | 298k | } else { |
1522 | 298k | rsi->frame_restoration_type = |
1523 | 298k | aom_rb_read_bit(rb) ? RESTORE_SWITCHABLE : RESTORE_NONE; |
1524 | 298k | } |
1525 | 344k | if (rsi->frame_restoration_type != RESTORE_NONE) { |
1526 | 57.2k | all_none = 0; |
1527 | 57.2k | chroma_none &= p == 0; |
1528 | 57.2k | } |
1529 | 344k | } |
1530 | 116k | if (!all_none) { |
1531 | 27.9k | assert(cm->seq_params->sb_size == BLOCK_64X64 || |
1532 | 27.9k | cm->seq_params->sb_size == BLOCK_128X128); |
1533 | 27.9k | const int sb_size = cm->seq_params->sb_size == BLOCK_128X128 ? 128 : 64; |
1534 | | |
1535 | 106k | for (int p = 0; p < num_planes; ++p) |
1536 | 79.0k | cm->rst_info[p].restoration_unit_size = sb_size; |
1537 | | |
1538 | 27.9k | RestorationInfo *rsi = &cm->rst_info[0]; |
1539 | | |
1540 | 27.9k | if (sb_size == 64) { |
1541 | 6.83k | rsi->restoration_unit_size <<= aom_rb_read_bit(rb); |
1542 | 6.83k | } |
1543 | 27.9k | if (rsi->restoration_unit_size > 64) { |
1544 | 24.3k | rsi->restoration_unit_size <<= aom_rb_read_bit(rb); |
1545 | 24.3k | } |
1546 | 88.9k | } else { |
1547 | 88.9k | const int size = RESTORATION_UNITSIZE_MAX; |
1548 | 353k | for (int p = 0; p < num_planes; ++p) |
1549 | 265k | cm->rst_info[p].restoration_unit_size = size; |
1550 | 88.9k | } |
1551 | | |
1552 | 116k | if (num_planes > 1) { |
1553 | 113k | int s = |
1554 | 113k | AOMMIN(cm->seq_params->subsampling_x, cm->seq_params->subsampling_y); |
1555 | 113k | if (s && !chroma_none) { |
1556 | 18.9k | cm->rst_info[1].restoration_unit_size = |
1557 | 18.9k | cm->rst_info[0].restoration_unit_size >> (aom_rb_read_bit(rb) * s); |
1558 | 94.7k | } else { |
1559 | 94.7k | cm->rst_info[1].restoration_unit_size = |
1560 | 94.7k | cm->rst_info[0].restoration_unit_size; |
1561 | 94.7k | } |
1562 | 113k | cm->rst_info[2].restoration_unit_size = |
1563 | 113k | cm->rst_info[1].restoration_unit_size; |
1564 | 113k | } |
1565 | 116k | } |
1566 | | |
1567 | | static AOM_INLINE void read_wiener_filter(int wiener_win, |
1568 | | WienerInfo *wiener_info, |
1569 | | WienerInfo *ref_wiener_info, |
1570 | 55.3k | aom_reader *rb) { |
1571 | 55.3k | memset(wiener_info->vfilter, 0, sizeof(wiener_info->vfilter)); |
1572 | 55.3k | memset(wiener_info->hfilter, 0, sizeof(wiener_info->hfilter)); |
1573 | | |
1574 | 55.3k | if (wiener_win == WIENER_WIN) |
1575 | 27.8k | wiener_info->vfilter[0] = wiener_info->vfilter[WIENER_WIN - 1] = |
1576 | 27.8k | aom_read_primitive_refsubexpfin( |
1577 | 27.8k | rb, WIENER_FILT_TAP0_MAXV - WIENER_FILT_TAP0_MINV + 1, |
1578 | 27.8k | WIENER_FILT_TAP0_SUBEXP_K, |
1579 | 27.8k | ref_wiener_info->vfilter[0] - WIENER_FILT_TAP0_MINV, ACCT_STR) + |
1580 | 27.8k | WIENER_FILT_TAP0_MINV; |
1581 | 27.4k | else |
1582 | 27.4k | wiener_info->vfilter[0] = wiener_info->vfilter[WIENER_WIN - 1] = 0; |
1583 | 55.3k | wiener_info->vfilter[1] = wiener_info->vfilter[WIENER_WIN - 2] = |
1584 | 55.3k | aom_read_primitive_refsubexpfin( |
1585 | 55.3k | rb, WIENER_FILT_TAP1_MAXV - WIENER_FILT_TAP1_MINV + 1, |
1586 | 55.3k | WIENER_FILT_TAP1_SUBEXP_K, |
1587 | 55.3k | ref_wiener_info->vfilter[1] - WIENER_FILT_TAP1_MINV, ACCT_STR) + |
1588 | 55.3k | WIENER_FILT_TAP1_MINV; |
1589 | 55.3k | wiener_info->vfilter[2] = wiener_info->vfilter[WIENER_WIN - 3] = |
1590 | 55.3k | aom_read_primitive_refsubexpfin( |
1591 | 55.3k | rb, WIENER_FILT_TAP2_MAXV - WIENER_FILT_TAP2_MINV + 1, |
1592 | 55.3k | WIENER_FILT_TAP2_SUBEXP_K, |
1593 | 55.3k | ref_wiener_info->vfilter[2] - WIENER_FILT_TAP2_MINV, ACCT_STR) + |
1594 | 55.3k | WIENER_FILT_TAP2_MINV; |
1595 | | // The central element has an implicit +WIENER_FILT_STEP |
1596 | 55.3k | wiener_info->vfilter[WIENER_HALFWIN] = |
1597 | 55.3k | -2 * (wiener_info->vfilter[0] + wiener_info->vfilter[1] + |
1598 | 55.3k | wiener_info->vfilter[2]); |
1599 | | |
1600 | 55.3k | if (wiener_win == WIENER_WIN) |
1601 | 27.8k | wiener_info->hfilter[0] = wiener_info->hfilter[WIENER_WIN - 1] = |
1602 | 27.8k | aom_read_primitive_refsubexpfin( |
1603 | 27.8k | rb, WIENER_FILT_TAP0_MAXV - WIENER_FILT_TAP0_MINV + 1, |
1604 | 27.8k | WIENER_FILT_TAP0_SUBEXP_K, |
1605 | 27.8k | ref_wiener_info->hfilter[0] - WIENER_FILT_TAP0_MINV, ACCT_STR) + |
1606 | 27.8k | WIENER_FILT_TAP0_MINV; |
1607 | 27.4k | else |
1608 | 27.4k | wiener_info->hfilter[0] = wiener_info->hfilter[WIENER_WIN - 1] = 0; |
1609 | 55.3k | wiener_info->hfilter[1] = wiener_info->hfilter[WIENER_WIN - 2] = |
1610 | 55.3k | aom_read_primitive_refsubexpfin( |
1611 | 55.3k | rb, WIENER_FILT_TAP1_MAXV - WIENER_FILT_TAP1_MINV + 1, |
1612 | 55.3k | WIENER_FILT_TAP1_SUBEXP_K, |
1613 | 55.3k | ref_wiener_info->hfilter[1] - WIENER_FILT_TAP1_MINV, ACCT_STR) + |
1614 | 55.3k | WIENER_FILT_TAP1_MINV; |
1615 | 55.3k | wiener_info->hfilter[2] = wiener_info->hfilter[WIENER_WIN - 3] = |
1616 | 55.3k | aom_read_primitive_refsubexpfin( |
1617 | 55.3k | rb, WIENER_FILT_TAP2_MAXV - WIENER_FILT_TAP2_MINV + 1, |
1618 | 55.3k | WIENER_FILT_TAP2_SUBEXP_K, |
1619 | 55.3k | ref_wiener_info->hfilter[2] - WIENER_FILT_TAP2_MINV, ACCT_STR) + |
1620 | 55.3k | WIENER_FILT_TAP2_MINV; |
1621 | | // The central element has an implicit +WIENER_FILT_STEP |
1622 | 55.3k | wiener_info->hfilter[WIENER_HALFWIN] = |
1623 | 55.3k | -2 * (wiener_info->hfilter[0] + wiener_info->hfilter[1] + |
1624 | 55.3k | wiener_info->hfilter[2]); |
1625 | 55.3k | memcpy(ref_wiener_info, wiener_info, sizeof(*wiener_info)); |
1626 | 55.3k | } |
1627 | | |
1628 | | static AOM_INLINE void read_sgrproj_filter(SgrprojInfo *sgrproj_info, |
1629 | | SgrprojInfo *ref_sgrproj_info, |
1630 | 97.8k | aom_reader *rb) { |
1631 | 97.8k | sgrproj_info->ep = aom_read_literal(rb, SGRPROJ_PARAMS_BITS, ACCT_STR); |
1632 | 97.8k | const sgr_params_type *params = &av1_sgr_params[sgrproj_info->ep]; |
1633 | | |
1634 | 97.8k | if (params->r[0] == 0) { |
1635 | 17.4k | sgrproj_info->xqd[0] = 0; |
1636 | 17.4k | sgrproj_info->xqd[1] = |
1637 | 17.4k | aom_read_primitive_refsubexpfin( |
1638 | 17.4k | rb, SGRPROJ_PRJ_MAX1 - SGRPROJ_PRJ_MIN1 + 1, SGRPROJ_PRJ_SUBEXP_K, |
1639 | 17.4k | ref_sgrproj_info->xqd[1] - SGRPROJ_PRJ_MIN1, ACCT_STR) + |
1640 | 17.4k | SGRPROJ_PRJ_MIN1; |
1641 | 80.4k | } else if (params->r[1] == 0) { |
1642 | 15.7k | sgrproj_info->xqd[0] = |
1643 | 15.7k | aom_read_primitive_refsubexpfin( |
1644 | 15.7k | rb, SGRPROJ_PRJ_MAX0 - SGRPROJ_PRJ_MIN0 + 1, SGRPROJ_PRJ_SUBEXP_K, |
1645 | 15.7k | ref_sgrproj_info->xqd[0] - SGRPROJ_PRJ_MIN0, ACCT_STR) + |
1646 | 15.7k | SGRPROJ_PRJ_MIN0; |
1647 | 15.7k | sgrproj_info->xqd[1] = clamp((1 << SGRPROJ_PRJ_BITS) - sgrproj_info->xqd[0], |
1648 | 15.7k | SGRPROJ_PRJ_MIN1, SGRPROJ_PRJ_MAX1); |
1649 | 64.6k | } else { |
1650 | 64.6k | sgrproj_info->xqd[0] = |
1651 | 64.6k | aom_read_primitive_refsubexpfin( |
1652 | 64.6k | rb, SGRPROJ_PRJ_MAX0 - SGRPROJ_PRJ_MIN0 + 1, SGRPROJ_PRJ_SUBEXP_K, |
1653 | 64.6k | ref_sgrproj_info->xqd[0] - SGRPROJ_PRJ_MIN0, ACCT_STR) + |
1654 | 64.6k | SGRPROJ_PRJ_MIN0; |
1655 | 64.6k | sgrproj_info->xqd[1] = |
1656 | 64.6k | aom_read_primitive_refsubexpfin( |
1657 | 64.6k | rb, SGRPROJ_PRJ_MAX1 - SGRPROJ_PRJ_MIN1 + 1, SGRPROJ_PRJ_SUBEXP_K, |
1658 | 64.6k | ref_sgrproj_info->xqd[1] - SGRPROJ_PRJ_MIN1, ACCT_STR) + |
1659 | 64.6k | SGRPROJ_PRJ_MIN1; |
1660 | 64.6k | } |
1661 | | |
1662 | 97.8k | memcpy(ref_sgrproj_info, sgrproj_info, sizeof(*sgrproj_info)); |
1663 | 97.8k | } |
1664 | | |
1665 | | static AOM_INLINE void loop_restoration_read_sb_coeffs( |
1666 | | const AV1_COMMON *const cm, MACROBLOCKD *xd, aom_reader *const r, int plane, |
1667 | 267k | int runit_idx) { |
1668 | 267k | const RestorationInfo *rsi = &cm->rst_info[plane]; |
1669 | 267k | RestorationUnitInfo *rui = &rsi->unit_info[runit_idx]; |
1670 | 267k | assert(rsi->frame_restoration_type != RESTORE_NONE); |
1671 | | |
1672 | 267k | assert(!cm->features.all_lossless); |
1673 | | |
1674 | 267k | const int wiener_win = (plane > 0) ? WIENER_WIN_CHROMA : WIENER_WIN; |
1675 | 267k | WienerInfo *wiener_info = xd->wiener_info + plane; |
1676 | 267k | SgrprojInfo *sgrproj_info = xd->sgrproj_info + plane; |
1677 | | |
1678 | 267k | if (rsi->frame_restoration_type == RESTORE_SWITCHABLE) { |
1679 | 84.7k | rui->restoration_type = |
1680 | 84.7k | aom_read_symbol(r, xd->tile_ctx->switchable_restore_cdf, |
1681 | 84.7k | RESTORE_SWITCHABLE_TYPES, ACCT_STR); |
1682 | 84.7k | switch (rui->restoration_type) { |
1683 | 29.2k | case RESTORE_WIENER: |
1684 | 29.2k | read_wiener_filter(wiener_win, &rui->wiener_info, wiener_info, r); |
1685 | 29.2k | break; |
1686 | 25.3k | case RESTORE_SGRPROJ: |
1687 | 25.3k | read_sgrproj_filter(&rui->sgrproj_info, sgrproj_info, r); |
1688 | 25.3k | break; |
1689 | 30.0k | default: assert(rui->restoration_type == RESTORE_NONE); break; |
1690 | 84.7k | } |
1691 | 182k | } else if (rsi->frame_restoration_type == RESTORE_WIENER) { |
1692 | 45.7k | if (aom_read_symbol(r, xd->tile_ctx->wiener_restore_cdf, 2, ACCT_STR)) { |
1693 | 26.0k | rui->restoration_type = RESTORE_WIENER; |
1694 | 26.0k | read_wiener_filter(wiener_win, &rui->wiener_info, wiener_info, r); |
1695 | 26.0k | } else { |
1696 | 19.6k | rui->restoration_type = RESTORE_NONE; |
1697 | 19.6k | } |
1698 | 136k | } else if (rsi->frame_restoration_type == RESTORE_SGRPROJ) { |
1699 | 136k | if (aom_read_symbol(r, xd->tile_ctx->sgrproj_restore_cdf, 2, ACCT_STR)) { |
1700 | 72.5k | rui->restoration_type = RESTORE_SGRPROJ; |
1701 | 72.5k | read_sgrproj_filter(&rui->sgrproj_info, sgrproj_info, r); |
1702 | 72.5k | } else { |
1703 | 64.1k | rui->restoration_type = RESTORE_NONE; |
1704 | 64.1k | } |
1705 | 136k | } |
1706 | 267k | } |
1707 | | |
1708 | | static AOM_INLINE void setup_loopfilter(AV1_COMMON *cm, |
1709 | 206k | struct aom_read_bit_buffer *rb) { |
1710 | 206k | const int num_planes = av1_num_planes(cm); |
1711 | 206k | struct loopfilter *lf = &cm->lf; |
1712 | | |
1713 | 206k | if (cm->features.allow_intrabc || cm->features.coded_lossless) { |
1714 | | // write default deltas to frame buffer |
1715 | 57.8k | av1_set_default_ref_deltas(cm->cur_frame->ref_deltas); |
1716 | 57.8k | av1_set_default_mode_deltas(cm->cur_frame->mode_deltas); |
1717 | 57.8k | return; |
1718 | 57.8k | } |
1719 | 148k | assert(!cm->features.coded_lossless); |
1720 | 148k | if (cm->prev_frame) { |
1721 | | // write deltas to frame buffer |
1722 | 85.0k | memcpy(lf->ref_deltas, cm->prev_frame->ref_deltas, REF_FRAMES); |
1723 | 85.0k | memcpy(lf->mode_deltas, cm->prev_frame->mode_deltas, MAX_MODE_LF_DELTAS); |
1724 | 85.0k | } else { |
1725 | 63.6k | av1_set_default_ref_deltas(lf->ref_deltas); |
1726 | 63.6k | av1_set_default_mode_deltas(lf->mode_deltas); |
1727 | 63.6k | } |
1728 | 148k | lf->filter_level[0] = aom_rb_read_literal(rb, 6); |
1729 | 148k | lf->filter_level[1] = aom_rb_read_literal(rb, 6); |
1730 | 148k | if (num_planes > 1) { |
1731 | 138k | if (lf->filter_level[0] || lf->filter_level[1]) { |
1732 | 48.5k | lf->filter_level_u = aom_rb_read_literal(rb, 6); |
1733 | 48.5k | lf->filter_level_v = aom_rb_read_literal(rb, 6); |
1734 | 48.5k | } |
1735 | 138k | } |
1736 | 148k | lf->sharpness_level = aom_rb_read_literal(rb, 3); |
1737 | | |
1738 | | // Read in loop filter deltas applied at the MB level based on mode or ref |
1739 | | // frame. |
1740 | 148k | lf->mode_ref_delta_update = 0; |
1741 | | |
1742 | 148k | lf->mode_ref_delta_enabled = aom_rb_read_bit(rb); |
1743 | 148k | if (lf->mode_ref_delta_enabled) { |
1744 | 53.1k | lf->mode_ref_delta_update = aom_rb_read_bit(rb); |
1745 | 53.1k | if (lf->mode_ref_delta_update) { |
1746 | 164k | for (int i = 0; i < REF_FRAMES; i++) |
1747 | 146k | if (aom_rb_read_bit(rb)) |
1748 | 40.7k | lf->ref_deltas[i] = aom_rb_read_inv_signed_literal(rb, 6); |
1749 | | |
1750 | 54.6k | for (int i = 0; i < MAX_MODE_LF_DELTAS; i++) |
1751 | 36.4k | if (aom_rb_read_bit(rb)) |
1752 | 5.80k | lf->mode_deltas[i] = aom_rb_read_inv_signed_literal(rb, 6); |
1753 | 18.2k | } |
1754 | 53.1k | } |
1755 | | |
1756 | | // write deltas to frame buffer |
1757 | 148k | memcpy(cm->cur_frame->ref_deltas, lf->ref_deltas, REF_FRAMES); |
1758 | 148k | memcpy(cm->cur_frame->mode_deltas, lf->mode_deltas, MAX_MODE_LF_DELTAS); |
1759 | 148k | } |
1760 | | |
1761 | | static AOM_INLINE void setup_cdef(AV1_COMMON *cm, |
1762 | 119k | struct aom_read_bit_buffer *rb) { |
1763 | 119k | const int num_planes = av1_num_planes(cm); |
1764 | 119k | CdefInfo *const cdef_info = &cm->cdef_info; |
1765 | | |
1766 | 119k | if (cm->features.allow_intrabc) return; |
1767 | 107k | cdef_info->cdef_damping = aom_rb_read_literal(rb, 2) + 3; |
1768 | 107k | cdef_info->cdef_bits = aom_rb_read_literal(rb, 2); |
1769 | 107k | cdef_info->nb_cdef_strengths = 1 << cdef_info->cdef_bits; |
1770 | 263k | for (int i = 0; i < cdef_info->nb_cdef_strengths; i++) { |
1771 | 156k | cdef_info->cdef_strengths[i] = aom_rb_read_literal(rb, CDEF_STRENGTH_BITS); |
1772 | 156k | cdef_info->cdef_uv_strengths[i] = |
1773 | 156k | num_planes > 1 ? aom_rb_read_literal(rb, CDEF_STRENGTH_BITS) : 0; |
1774 | 156k | } |
1775 | 107k | } |
1776 | | |
1777 | 590k | static INLINE int read_delta_q(struct aom_read_bit_buffer *rb) { |
1778 | 590k | return aom_rb_read_bit(rb) ? aom_rb_read_inv_signed_literal(rb, 6) : 0; |
1779 | 590k | } |
1780 | | |
1781 | | static AOM_INLINE void setup_quantization(CommonQuantParams *quant_params, |
1782 | | int num_planes, |
1783 | | bool separate_uv_delta_q, |
1784 | 208k | struct aom_read_bit_buffer *rb) { |
1785 | 208k | quant_params->base_qindex = aom_rb_read_literal(rb, QINDEX_BITS); |
1786 | 208k | quant_params->y_dc_delta_q = read_delta_q(rb); |
1787 | 208k | if (num_planes > 1) { |
1788 | 188k | int diff_uv_delta = 0; |
1789 | 188k | if (separate_uv_delta_q) diff_uv_delta = aom_rb_read_bit(rb); |
1790 | 188k | quant_params->u_dc_delta_q = read_delta_q(rb); |
1791 | 188k | quant_params->u_ac_delta_q = read_delta_q(rb); |
1792 | 188k | if (diff_uv_delta) { |
1793 | 2.08k | quant_params->v_dc_delta_q = read_delta_q(rb); |
1794 | 2.08k | quant_params->v_ac_delta_q = read_delta_q(rb); |
1795 | 186k | } else { |
1796 | 186k | quant_params->v_dc_delta_q = quant_params->u_dc_delta_q; |
1797 | 186k | quant_params->v_ac_delta_q = quant_params->u_ac_delta_q; |
1798 | 186k | } |
1799 | 188k | } else { |
1800 | 19.9k | quant_params->u_dc_delta_q = 0; |
1801 | 19.9k | quant_params->u_ac_delta_q = 0; |
1802 | 19.9k | quant_params->v_dc_delta_q = 0; |
1803 | 19.9k | quant_params->v_ac_delta_q = 0; |
1804 | 19.9k | } |
1805 | 208k | quant_params->using_qmatrix = aom_rb_read_bit(rb); |
1806 | 208k | if (quant_params->using_qmatrix) { |
1807 | 35.7k | quant_params->qmatrix_level_y = aom_rb_read_literal(rb, QM_LEVEL_BITS); |
1808 | 35.7k | quant_params->qmatrix_level_u = aom_rb_read_literal(rb, QM_LEVEL_BITS); |
1809 | 35.7k | if (!separate_uv_delta_q) |
1810 | 33.6k | quant_params->qmatrix_level_v = quant_params->qmatrix_level_u; |
1811 | 2.11k | else |
1812 | 2.11k | quant_params->qmatrix_level_v = aom_rb_read_literal(rb, QM_LEVEL_BITS); |
1813 | 173k | } else { |
1814 | 173k | quant_params->qmatrix_level_y = 0; |
1815 | 173k | quant_params->qmatrix_level_u = 0; |
1816 | 173k | quant_params->qmatrix_level_v = 0; |
1817 | 173k | } |
1818 | 208k | } |
1819 | | |
1820 | | // Build y/uv dequant values based on segmentation. |
1821 | | static AOM_INLINE void setup_segmentation_dequant(AV1_COMMON *const cm, |
1822 | 206k | MACROBLOCKD *const xd) { |
1823 | 206k | const int bit_depth = cm->seq_params->bit_depth; |
1824 | | // When segmentation is disabled, only the first value is used. The |
1825 | | // remaining are don't cares. |
1826 | 206k | const int max_segments = cm->seg.enabled ? MAX_SEGMENTS : 1; |
1827 | 206k | CommonQuantParams *const quant_params = &cm->quant_params; |
1828 | 567k | for (int i = 0; i < max_segments; ++i) { |
1829 | 360k | const int qindex = xd->qindex[i]; |
1830 | 360k | quant_params->y_dequant_QTX[i][0] = |
1831 | 360k | av1_dc_quant_QTX(qindex, quant_params->y_dc_delta_q, bit_depth); |
1832 | 360k | quant_params->y_dequant_QTX[i][1] = av1_ac_quant_QTX(qindex, 0, bit_depth); |
1833 | 360k | quant_params->u_dequant_QTX[i][0] = |
1834 | 360k | av1_dc_quant_QTX(qindex, quant_params->u_dc_delta_q, bit_depth); |
1835 | 360k | quant_params->u_dequant_QTX[i][1] = |
1836 | 360k | av1_ac_quant_QTX(qindex, quant_params->u_ac_delta_q, bit_depth); |
1837 | 360k | quant_params->v_dequant_QTX[i][0] = |
1838 | 360k | av1_dc_quant_QTX(qindex, quant_params->v_dc_delta_q, bit_depth); |
1839 | 360k | quant_params->v_dequant_QTX[i][1] = |
1840 | 360k | av1_ac_quant_QTX(qindex, quant_params->v_ac_delta_q, bit_depth); |
1841 | 360k | const int use_qmatrix = av1_use_qmatrix(quant_params, xd, i); |
1842 | | // NB: depends on base index so there is only 1 set per frame |
1843 | | // No quant weighting when lossless or signalled not using QM |
1844 | 360k | const int qmlevel_y = |
1845 | 360k | use_qmatrix ? quant_params->qmatrix_level_y : NUM_QM_LEVELS - 1; |
1846 | 7.21M | for (int j = 0; j < TX_SIZES_ALL; ++j) { |
1847 | 6.85M | quant_params->y_iqmatrix[i][j] = |
1848 | 6.85M | av1_iqmatrix(quant_params, qmlevel_y, AOM_PLANE_Y, j); |
1849 | 6.85M | } |
1850 | 360k | const int qmlevel_u = |
1851 | 360k | use_qmatrix ? quant_params->qmatrix_level_u : NUM_QM_LEVELS - 1; |
1852 | 7.21M | for (int j = 0; j < TX_SIZES_ALL; ++j) { |
1853 | 6.85M | quant_params->u_iqmatrix[i][j] = |
1854 | 6.85M | av1_iqmatrix(quant_params, qmlevel_u, AOM_PLANE_U, j); |
1855 | 6.85M | } |
1856 | 360k | const int qmlevel_v = |
1857 | 360k | use_qmatrix ? quant_params->qmatrix_level_v : NUM_QM_LEVELS - 1; |
1858 | 7.21M | for (int j = 0; j < TX_SIZES_ALL; ++j) { |
1859 | 6.85M | quant_params->v_iqmatrix[i][j] = |
1860 | 6.85M | av1_iqmatrix(quant_params, qmlevel_v, AOM_PLANE_V, j); |
1861 | 6.85M | } |
1862 | 360k | } |
1863 | 206k | } |
1864 | | |
1865 | 95.6k | static InterpFilter read_frame_interp_filter(struct aom_read_bit_buffer *rb) { |
1866 | 95.6k | return aom_rb_read_bit(rb) ? SWITCHABLE |
1867 | 95.6k | : aom_rb_read_literal(rb, LOG_SWITCHABLE_FILTERS); |
1868 | 95.6k | } |
1869 | | |
1870 | | static AOM_INLINE void setup_render_size(AV1_COMMON *cm, |
1871 | 137k | struct aom_read_bit_buffer *rb) { |
1872 | 137k | cm->render_width = cm->superres_upscaled_width; |
1873 | 137k | cm->render_height = cm->superres_upscaled_height; |
1874 | 137k | if (aom_rb_read_bit(rb)) |
1875 | 31.1k | av1_read_frame_size(rb, 16, 16, &cm->render_width, &cm->render_height); |
1876 | 137k | } |
1877 | | |
1878 | | // TODO(afergs): make "struct aom_read_bit_buffer *const rb"? |
1879 | | static AOM_INLINE void setup_superres(AV1_COMMON *const cm, |
1880 | | struct aom_read_bit_buffer *rb, |
1881 | 219k | int *width, int *height) { |
1882 | 219k | cm->superres_upscaled_width = *width; |
1883 | 219k | cm->superres_upscaled_height = *height; |
1884 | | |
1885 | 219k | const SequenceHeader *const seq_params = cm->seq_params; |
1886 | 219k | if (!seq_params->enable_superres) return; |
1887 | | |
1888 | 62.4k | if (aom_rb_read_bit(rb)) { |
1889 | 22.0k | cm->superres_scale_denominator = |
1890 | 22.0k | (uint8_t)aom_rb_read_literal(rb, SUPERRES_SCALE_BITS); |
1891 | 22.0k | cm->superres_scale_denominator += SUPERRES_SCALE_DENOMINATOR_MIN; |
1892 | | // Don't edit cm->width or cm->height directly, or the buffers won't get |
1893 | | // resized correctly |
1894 | 22.0k | av1_calculate_scaled_superres_size(width, height, |
1895 | 22.0k | cm->superres_scale_denominator); |
1896 | 40.3k | } else { |
1897 | | // 1:1 scaling - ie. no scaling, scale not provided |
1898 | 40.3k | cm->superres_scale_denominator = SCALE_NUMERATOR; |
1899 | 40.3k | } |
1900 | 62.4k | } |
1901 | | |
1902 | | static AOM_INLINE void resize_context_buffers(AV1_COMMON *cm, int width, |
1903 | 219k | int height) { |
1904 | 219k | #if CONFIG_SIZE_LIMIT |
1905 | 219k | if (width > DECODE_WIDTH_LIMIT || height > DECODE_HEIGHT_LIMIT) |
1906 | 276 | aom_internal_error(cm->error, AOM_CODEC_CORRUPT_FRAME, |
1907 | 276 | "Dimensions of %dx%d beyond allowed size of %dx%d.", |
1908 | 276 | width, height, DECODE_WIDTH_LIMIT, DECODE_HEIGHT_LIMIT); |
1909 | 219k | #endif |
1910 | 219k | if (cm->width != width || cm->height != height) { |
1911 | 66.9k | const int new_mi_rows = CEIL_POWER_OF_TWO(height, MI_SIZE_LOG2); |
1912 | 66.9k | const int new_mi_cols = CEIL_POWER_OF_TWO(width, MI_SIZE_LOG2); |
1913 | | |
1914 | | // Allocations in av1_alloc_context_buffers() depend on individual |
1915 | | // dimensions as well as the overall size. |
1916 | 66.9k | if (new_mi_cols > cm->mi_params.mi_cols || |
1917 | 66.9k | new_mi_rows > cm->mi_params.mi_rows) { |
1918 | 42.4k | if (av1_alloc_context_buffers(cm, width, height, BLOCK_4X4)) { |
1919 | | // The cm->mi_* values have been cleared and any existing context |
1920 | | // buffers have been freed. Clear cm->width and cm->height to be |
1921 | | // consistent and to force a realloc next time. |
1922 | 313 | cm->width = 0; |
1923 | 313 | cm->height = 0; |
1924 | 313 | aom_internal_error(cm->error, AOM_CODEC_MEM_ERROR, |
1925 | 313 | "Failed to allocate context buffers"); |
1926 | 313 | } |
1927 | 42.4k | } else { |
1928 | 24.5k | cm->mi_params.set_mb_mi(&cm->mi_params, width, height, BLOCK_4X4); |
1929 | 24.5k | } |
1930 | 66.9k | av1_init_mi_buffers(&cm->mi_params); |
1931 | 66.9k | cm->width = width; |
1932 | 66.9k | cm->height = height; |
1933 | 66.9k | } |
1934 | | |
1935 | 219k | ensure_mv_buffer(cm->cur_frame, cm); |
1936 | 219k | cm->cur_frame->width = cm->width; |
1937 | 219k | cm->cur_frame->height = cm->height; |
1938 | 219k | } |
1939 | | |
1940 | 214k | static AOM_INLINE void setup_buffer_pool(AV1_COMMON *cm) { |
1941 | 214k | BufferPool *const pool = cm->buffer_pool; |
1942 | 214k | const SequenceHeader *const seq_params = cm->seq_params; |
1943 | | |
1944 | 214k | lock_buffer_pool(pool); |
1945 | 214k | if (aom_realloc_frame_buffer( |
1946 | 214k | &cm->cur_frame->buf, cm->width, cm->height, seq_params->subsampling_x, |
1947 | 214k | seq_params->subsampling_y, seq_params->use_highbitdepth, |
1948 | 214k | AOM_DEC_BORDER_IN_PIXELS, cm->features.byte_alignment, |
1949 | 214k | &cm->cur_frame->raw_frame_buffer, pool->get_fb_cb, pool->cb_priv, |
1950 | 214k | false, 0)) { |
1951 | 87 | unlock_buffer_pool(pool); |
1952 | 87 | aom_internal_error(cm->error, AOM_CODEC_MEM_ERROR, |
1953 | 87 | "Failed to allocate frame buffer"); |
1954 | 87 | } |
1955 | 214k | unlock_buffer_pool(pool); |
1956 | | |
1957 | 214k | cm->cur_frame->buf.bit_depth = (unsigned int)seq_params->bit_depth; |
1958 | 214k | cm->cur_frame->buf.color_primaries = seq_params->color_primaries; |
1959 | 214k | cm->cur_frame->buf.transfer_characteristics = |
1960 | 214k | seq_params->transfer_characteristics; |
1961 | 214k | cm->cur_frame->buf.matrix_coefficients = seq_params->matrix_coefficients; |
1962 | 214k | cm->cur_frame->buf.monochrome = seq_params->monochrome; |
1963 | 214k | cm->cur_frame->buf.chroma_sample_position = |
1964 | 214k | seq_params->chroma_sample_position; |
1965 | 214k | cm->cur_frame->buf.color_range = seq_params->color_range; |
1966 | 214k | cm->cur_frame->buf.render_width = cm->render_width; |
1967 | 214k | cm->cur_frame->buf.render_height = cm->render_height; |
1968 | 214k | } |
1969 | | |
1970 | | static AOM_INLINE void setup_frame_size(AV1_COMMON *cm, |
1971 | | int frame_size_override_flag, |
1972 | 135k | struct aom_read_bit_buffer *rb) { |
1973 | 135k | const SequenceHeader *const seq_params = cm->seq_params; |
1974 | 135k | int width, height; |
1975 | | |
1976 | 135k | if (frame_size_override_flag) { |
1977 | 48.2k | int num_bits_width = seq_params->num_bits_width; |
1978 | 48.2k | int num_bits_height = seq_params->num_bits_height; |
1979 | 48.2k | av1_read_frame_size(rb, num_bits_width, num_bits_height, &width, &height); |
1980 | 48.2k | if (width > seq_params->max_frame_width || |
1981 | 48.2k | height > seq_params->max_frame_height) { |
1982 | 2.76k | aom_internal_error(cm->error, AOM_CODEC_CORRUPT_FRAME, |
1983 | 2.76k | "Frame dimensions are larger than the maximum values"); |
1984 | 2.76k | } |
1985 | 87.3k | } else { |
1986 | 87.3k | width = seq_params->max_frame_width; |
1987 | 87.3k | height = seq_params->max_frame_height; |
1988 | 87.3k | } |
1989 | | |
1990 | 135k | setup_superres(cm, rb, &width, &height); |
1991 | 135k | resize_context_buffers(cm, width, height); |
1992 | 135k | setup_render_size(cm, rb); |
1993 | 135k | setup_buffer_pool(cm); |
1994 | 135k | } |
1995 | | |
1996 | | static AOM_INLINE void setup_sb_size(SequenceHeader *seq_params, |
1997 | 72.4k | struct aom_read_bit_buffer *rb) { |
1998 | 72.4k | set_sb_size(seq_params, aom_rb_read_bit(rb) ? BLOCK_128X128 : BLOCK_64X64); |
1999 | 72.4k | } |
2000 | | |
2001 | | static INLINE int valid_ref_frame_img_fmt(aom_bit_depth_t ref_bit_depth, |
2002 | | int ref_xss, int ref_yss, |
2003 | | aom_bit_depth_t this_bit_depth, |
2004 | 583k | int this_xss, int this_yss) { |
2005 | 583k | return ref_bit_depth == this_bit_depth && ref_xss == this_xss && |
2006 | 583k | ref_yss == this_yss; |
2007 | 583k | } |
2008 | | |
2009 | | static AOM_INLINE void setup_frame_size_with_refs( |
2010 | 86.7k | AV1_COMMON *cm, struct aom_read_bit_buffer *rb) { |
2011 | 86.7k | int width, height; |
2012 | 86.7k | int found = 0; |
2013 | 86.7k | int has_valid_ref_frame = 0; |
2014 | 324k | for (int i = LAST_FRAME; i <= ALTREF_FRAME; ++i) { |
2015 | 319k | if (aom_rb_read_bit(rb)) { |
2016 | 81.5k | const RefCntBuffer *const ref_buf = get_ref_frame_buf(cm, i); |
2017 | | // This will never be NULL in a normal stream, as streams are required to |
2018 | | // have a shown keyframe before any inter frames, which would refresh all |
2019 | | // the reference buffers. However, it might be null if we're starting in |
2020 | | // the middle of a stream, and static analysis will error if we don't do |
2021 | | // a null check here. |
2022 | 81.5k | if (ref_buf == NULL) { |
2023 | 0 | aom_internal_error(cm->error, AOM_CODEC_CORRUPT_FRAME, |
2024 | 0 | "Invalid condition: invalid reference buffer"); |
2025 | 81.5k | } else { |
2026 | 81.5k | const YV12_BUFFER_CONFIG *const buf = &ref_buf->buf; |
2027 | 81.5k | width = buf->y_crop_width; |
2028 | 81.5k | height = buf->y_crop_height; |
2029 | 81.5k | cm->render_width = buf->render_width; |
2030 | 81.5k | cm->render_height = buf->render_height; |
2031 | 81.5k | setup_superres(cm, rb, &width, &height); |
2032 | 81.5k | resize_context_buffers(cm, width, height); |
2033 | 81.5k | found = 1; |
2034 | 81.5k | break; |
2035 | 81.5k | } |
2036 | 81.5k | } |
2037 | 319k | } |
2038 | | |
2039 | 86.7k | const SequenceHeader *const seq_params = cm->seq_params; |
2040 | 86.7k | if (!found) { |
2041 | 5.23k | int num_bits_width = seq_params->num_bits_width; |
2042 | 5.23k | int num_bits_height = seq_params->num_bits_height; |
2043 | | |
2044 | 5.23k | av1_read_frame_size(rb, num_bits_width, num_bits_height, &width, &height); |
2045 | 5.23k | setup_superres(cm, rb, &width, &height); |
2046 | 5.23k | resize_context_buffers(cm, width, height); |
2047 | 5.23k | setup_render_size(cm, rb); |
2048 | 5.23k | } |
2049 | | |
2050 | 86.7k | if (width <= 0 || height <= 0) |
2051 | 0 | aom_internal_error(cm->error, AOM_CODEC_CORRUPT_FRAME, |
2052 | 0 | "Invalid frame size"); |
2053 | | |
2054 | | // Check to make sure at least one of frames that this frame references |
2055 | | // has valid dimensions. |
2056 | 693k | for (int i = LAST_FRAME; i <= ALTREF_FRAME; ++i) { |
2057 | 607k | const RefCntBuffer *const ref_frame = get_ref_frame_buf(cm, i); |
2058 | 607k | has_valid_ref_frame |= |
2059 | 607k | valid_ref_frame_size(ref_frame->buf.y_crop_width, |
2060 | 607k | ref_frame->buf.y_crop_height, width, height); |
2061 | 607k | } |
2062 | 86.7k | if (!has_valid_ref_frame) |
2063 | 2.17k | aom_internal_error(cm->error, AOM_CODEC_CORRUPT_FRAME, |
2064 | 2.17k | "Referenced frame has invalid size"); |
2065 | 670k | for (int i = LAST_FRAME; i <= ALTREF_FRAME; ++i) { |
2066 | 583k | const RefCntBuffer *const ref_frame = get_ref_frame_buf(cm, i); |
2067 | 583k | if (!valid_ref_frame_img_fmt( |
2068 | 583k | ref_frame->buf.bit_depth, ref_frame->buf.subsampling_x, |
2069 | 583k | ref_frame->buf.subsampling_y, seq_params->bit_depth, |
2070 | 583k | seq_params->subsampling_x, seq_params->subsampling_y)) |
2071 | 2.13k | aom_internal_error(cm->error, AOM_CODEC_CORRUPT_FRAME, |
2072 | 2.13k | "Referenced frame has incompatible color format"); |
2073 | 583k | } |
2074 | 86.7k | setup_buffer_pool(cm); |
2075 | 86.7k | } |
2076 | | |
2077 | | // Same function as av1_read_uniform but reading from uncompresses header wb |
2078 | 269k | static int rb_read_uniform(struct aom_read_bit_buffer *const rb, int n) { |
2079 | 269k | const int l = get_unsigned_bits(n); |
2080 | 269k | const int m = (1 << l) - n; |
2081 | 269k | const int v = aom_rb_read_literal(rb, l - 1); |
2082 | 269k | assert(l != 0); |
2083 | 269k | if (v < m) |
2084 | 214k | return v; |
2085 | 54.7k | else |
2086 | 54.7k | return (v << 1) - m + aom_rb_read_bit(rb); |
2087 | 269k | } |
2088 | | |
2089 | | static AOM_INLINE void read_tile_info_max_tile( |
2090 | 210k | AV1_COMMON *const cm, struct aom_read_bit_buffer *const rb) { |
2091 | 210k | const SequenceHeader *const seq_params = cm->seq_params; |
2092 | 210k | CommonTileParams *const tiles = &cm->tiles; |
2093 | 210k | int width_sb = |
2094 | 210k | CEIL_POWER_OF_TWO(cm->mi_params.mi_cols, seq_params->mib_size_log2); |
2095 | 210k | int height_sb = |
2096 | 210k | CEIL_POWER_OF_TWO(cm->mi_params.mi_rows, seq_params->mib_size_log2); |
2097 | | |
2098 | 210k | av1_get_tile_limits(cm); |
2099 | 210k | tiles->uniform_spacing = aom_rb_read_bit(rb); |
2100 | | |
2101 | | // Read tile columns |
2102 | 210k | if (tiles->uniform_spacing) { |
2103 | 100k | tiles->log2_cols = tiles->min_log2_cols; |
2104 | 102k | while (tiles->log2_cols < tiles->max_log2_cols) { |
2105 | 76.2k | if (!aom_rb_read_bit(rb)) { |
2106 | 73.5k | break; |
2107 | 73.5k | } |
2108 | 2.66k | tiles->log2_cols++; |
2109 | 2.66k | } |
2110 | 110k | } else { |
2111 | 110k | int i; |
2112 | 110k | int start_sb; |
2113 | 245k | for (i = 0, start_sb = 0; width_sb > 0 && i < MAX_TILE_COLS; i++) { |
2114 | 135k | const int size_sb = |
2115 | 135k | 1 + rb_read_uniform(rb, AOMMIN(width_sb, tiles->max_width_sb)); |
2116 | 135k | tiles->col_start_sb[i] = start_sb; |
2117 | 135k | start_sb += size_sb; |
2118 | 135k | width_sb -= size_sb; |
2119 | 135k | } |
2120 | 110k | tiles->cols = i; |
2121 | 110k | tiles->col_start_sb[i] = start_sb + width_sb; |
2122 | 110k | } |
2123 | 210k | av1_calculate_tile_cols(seq_params, cm->mi_params.mi_rows, |
2124 | 210k | cm->mi_params.mi_cols, tiles); |
2125 | | |
2126 | | // Read tile rows |
2127 | 210k | if (tiles->uniform_spacing) { |
2128 | 100k | tiles->log2_rows = tiles->min_log2_rows; |
2129 | 106k | while (tiles->log2_rows < tiles->max_log2_rows) { |
2130 | 77.5k | if (!aom_rb_read_bit(rb)) { |
2131 | 70.5k | break; |
2132 | 70.5k | } |
2133 | 6.96k | tiles->log2_rows++; |
2134 | 6.96k | } |
2135 | 110k | } else { |
2136 | 110k | int i; |
2137 | 110k | int start_sb; |
2138 | 244k | for (i = 0, start_sb = 0; height_sb > 0 && i < MAX_TILE_ROWS; i++) { |
2139 | 134k | const int size_sb = |
2140 | 134k | 1 + rb_read_uniform(rb, AOMMIN(height_sb, tiles->max_height_sb)); |
2141 | 134k | tiles->row_start_sb[i] = start_sb; |
2142 | 134k | start_sb += size_sb; |
2143 | 134k | height_sb -= size_sb; |
2144 | 134k | } |
2145 | 110k | tiles->rows = i; |
2146 | 110k | tiles->row_start_sb[i] = start_sb + height_sb; |
2147 | 110k | } |
2148 | 210k | av1_calculate_tile_rows(seq_params, cm->mi_params.mi_rows, tiles); |
2149 | 210k | } |
2150 | | |
2151 | 14.0k | void av1_set_single_tile_decoding_mode(AV1_COMMON *const cm) { |
2152 | 14.0k | cm->tiles.single_tile_decoding = 0; |
2153 | 14.0k | if (cm->tiles.large_scale) { |
2154 | 14.0k | struct loopfilter *lf = &cm->lf; |
2155 | 14.0k | RestorationInfo *const rst_info = cm->rst_info; |
2156 | 14.0k | const CdefInfo *const cdef_info = &cm->cdef_info; |
2157 | | |
2158 | | // Figure out single_tile_decoding by loopfilter_level. |
2159 | 14.0k | const int no_loopfilter = !(lf->filter_level[0] || lf->filter_level[1]); |
2160 | 14.0k | const int no_cdef = cdef_info->cdef_bits == 0 && |
2161 | 14.0k | cdef_info->cdef_strengths[0] == 0 && |
2162 | 14.0k | cdef_info->cdef_uv_strengths[0] == 0; |
2163 | 14.0k | const int no_restoration = |
2164 | 14.0k | rst_info[0].frame_restoration_type == RESTORE_NONE && |
2165 | 14.0k | rst_info[1].frame_restoration_type == RESTORE_NONE && |
2166 | 14.0k | rst_info[2].frame_restoration_type == RESTORE_NONE; |
2167 | 14.0k | assert(IMPLIES(cm->features.coded_lossless, no_loopfilter && no_cdef)); |
2168 | 14.0k | assert(IMPLIES(cm->features.all_lossless, no_restoration)); |
2169 | 14.0k | cm->tiles.single_tile_decoding = no_loopfilter && no_cdef && no_restoration; |
2170 | 14.0k | } |
2171 | 14.0k | } |
2172 | | |
2173 | | static AOM_INLINE void read_tile_info(AV1Decoder *const pbi, |
2174 | 210k | struct aom_read_bit_buffer *const rb) { |
2175 | 210k | AV1_COMMON *const cm = &pbi->common; |
2176 | | |
2177 | 210k | read_tile_info_max_tile(cm, rb); |
2178 | | |
2179 | 210k | pbi->context_update_tile_id = 0; |
2180 | 210k | if (cm->tiles.rows * cm->tiles.cols > 1) { |
2181 | | // tile to use for cdf update |
2182 | 19.0k | pbi->context_update_tile_id = |
2183 | 19.0k | aom_rb_read_literal(rb, cm->tiles.log2_rows + cm->tiles.log2_cols); |
2184 | 19.0k | if (pbi->context_update_tile_id >= cm->tiles.rows * cm->tiles.cols) { |
2185 | 613 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
2186 | 613 | "Invalid context_update_tile_id"); |
2187 | 613 | } |
2188 | | // tile size magnitude |
2189 | 19.0k | pbi->tile_size_bytes = aom_rb_read_literal(rb, 2) + 1; |
2190 | 19.0k | } |
2191 | 210k | } |
2192 | | |
2193 | | #if EXT_TILE_DEBUG |
2194 | | static AOM_INLINE void read_ext_tile_info( |
2195 | 13.5k | AV1Decoder *const pbi, struct aom_read_bit_buffer *const rb) { |
2196 | 13.5k | AV1_COMMON *const cm = &pbi->common; |
2197 | | |
2198 | | // This information is stored as a separate byte. |
2199 | 13.5k | int mod = rb->bit_offset % CHAR_BIT; |
2200 | 13.5k | if (mod > 0) aom_rb_read_literal(rb, CHAR_BIT - mod); |
2201 | 13.5k | assert(rb->bit_offset % CHAR_BIT == 0); |
2202 | | |
2203 | 13.5k | if (cm->tiles.cols * cm->tiles.rows > 1) { |
2204 | | // Read the number of bytes used to store tile size |
2205 | 4.65k | pbi->tile_col_size_bytes = aom_rb_read_literal(rb, 2) + 1; |
2206 | 4.65k | pbi->tile_size_bytes = aom_rb_read_literal(rb, 2) + 1; |
2207 | 4.65k | } |
2208 | 13.5k | } |
2209 | | #endif // EXT_TILE_DEBUG |
2210 | | |
2211 | 36.9k | static size_t mem_get_varsize(const uint8_t *src, int sz) { |
2212 | 36.9k | switch (sz) { |
2213 | 31.4k | case 1: return src[0]; |
2214 | 4.13k | case 2: return mem_get_le16(src); |
2215 | 821 | case 3: return mem_get_le24(src); |
2216 | 510 | case 4: return mem_get_le32(src); |
2217 | 0 | default: assert(0 && "Invalid size"); return -1; |
2218 | 36.9k | } |
2219 | 36.9k | } |
2220 | | |
2221 | | #if EXT_TILE_DEBUG |
2222 | | // Reads the next tile returning its size and adjusting '*data' accordingly |
2223 | | // based on 'is_last'. On return, '*data' is updated to point to the end of the |
2224 | | // raw tile buffer in the bit stream. |
2225 | | static AOM_INLINE void get_ls_tile_buffer( |
2226 | | const uint8_t *const data_end, struct aom_internal_error_info *error_info, |
2227 | | const uint8_t **data, TileBufferDec (*const tile_buffers)[MAX_TILE_COLS], |
2228 | 19.4k | int tile_size_bytes, int col, int row, int tile_copy_mode) { |
2229 | 19.4k | size_t size; |
2230 | | |
2231 | 19.4k | size_t copy_size = 0; |
2232 | 19.4k | const uint8_t *copy_data = NULL; |
2233 | | |
2234 | 19.4k | if (!read_is_valid(*data, tile_size_bytes, data_end)) |
2235 | 234 | aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME, |
2236 | 234 | "Truncated packet or corrupt tile length"); |
2237 | 19.4k | size = mem_get_varsize(*data, tile_size_bytes); |
2238 | | |
2239 | | // If tile_copy_mode = 1, then the top bit of the tile header indicates copy |
2240 | | // mode. |
2241 | 19.4k | if (tile_copy_mode && (size >> (tile_size_bytes * 8 - 1)) == 1) { |
2242 | | // The remaining bits in the top byte signal the row offset |
2243 | 11.8k | int offset = (size >> (tile_size_bytes - 1) * 8) & 0x7f; |
2244 | 11.8k | if (offset > row) { |
2245 | 83 | aom_internal_error( |
2246 | 83 | error_info, AOM_CODEC_CORRUPT_FRAME, |
2247 | 83 | "Invalid row offset in tile copy mode: row=%d offset=%d", row, |
2248 | 83 | offset); |
2249 | 83 | } |
2250 | | |
2251 | | // Currently, only use tiles in same column as reference tiles. |
2252 | 11.8k | copy_data = tile_buffers[row - offset][col].data; |
2253 | 11.8k | copy_size = tile_buffers[row - offset][col].size; |
2254 | 11.8k | size = 0; |
2255 | 11.8k | } else { |
2256 | 7.62k | size += AV1_MIN_TILE_SIZE_BYTES; |
2257 | 7.62k | } |
2258 | | |
2259 | 19.4k | *data += tile_size_bytes; |
2260 | | |
2261 | 19.4k | if (size > (size_t)(data_end - *data)) |
2262 | 877 | aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME, |
2263 | 877 | "Truncated packet or corrupt tile size"); |
2264 | | |
2265 | 19.4k | if (size > 0) { |
2266 | 6.51k | tile_buffers[row][col].data = *data; |
2267 | 6.51k | tile_buffers[row][col].size = size; |
2268 | 12.9k | } else { |
2269 | 12.9k | tile_buffers[row][col].data = copy_data; |
2270 | 12.9k | tile_buffers[row][col].size = copy_size; |
2271 | 12.9k | } |
2272 | | |
2273 | 19.4k | *data += size; |
2274 | 19.4k | } |
2275 | | |
2276 | | // Returns the end of the last tile buffer |
2277 | | // (tile_buffers[cm->tiles.rows - 1][cm->tiles.cols - 1]). |
2278 | | static const uint8_t *get_ls_tile_buffers( |
2279 | | AV1Decoder *pbi, const uint8_t *data, const uint8_t *data_end, |
2280 | 13.3k | TileBufferDec (*const tile_buffers)[MAX_TILE_COLS]) { |
2281 | 13.3k | AV1_COMMON *const cm = &pbi->common; |
2282 | 13.3k | const int tile_cols = cm->tiles.cols; |
2283 | 13.3k | const int tile_rows = cm->tiles.rows; |
2284 | 13.3k | const int have_tiles = tile_cols * tile_rows > 1; |
2285 | 13.3k | const uint8_t *raw_data_end; // The end of the last tile buffer |
2286 | | |
2287 | 13.3k | if (!have_tiles) { |
2288 | 8.91k | const size_t tile_size = data_end - data; |
2289 | 8.91k | tile_buffers[0][0].data = data; |
2290 | 8.91k | tile_buffers[0][0].size = tile_size; |
2291 | 8.91k | raw_data_end = NULL; |
2292 | 8.91k | } else { |
2293 | | // We locate only the tile buffers that are required, which are the ones |
2294 | | // specified by pbi->dec_tile_col and pbi->dec_tile_row. Also, we always |
2295 | | // need the last (bottom right) tile buffer, as we need to know where the |
2296 | | // end of the compressed frame buffer is for proper superframe decoding. |
2297 | | |
2298 | 4.43k | const uint8_t *tile_col_data_end[MAX_TILE_COLS] = { NULL }; |
2299 | 4.43k | const uint8_t *const data_start = data; |
2300 | | |
2301 | 4.43k | const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows); |
2302 | 4.43k | const int single_row = pbi->dec_tile_row >= 0; |
2303 | 4.43k | const int tile_rows_start = single_row ? dec_tile_row : 0; |
2304 | 4.43k | const int tile_rows_end = single_row ? tile_rows_start + 1 : tile_rows; |
2305 | 4.43k | const int dec_tile_col = AOMMIN(pbi->dec_tile_col, tile_cols); |
2306 | 4.43k | const int single_col = pbi->dec_tile_col >= 0; |
2307 | 4.43k | const int tile_cols_start = single_col ? dec_tile_col : 0; |
2308 | 4.43k | const int tile_cols_end = single_col ? tile_cols_start + 1 : tile_cols; |
2309 | | |
2310 | 4.43k | const int tile_col_size_bytes = pbi->tile_col_size_bytes; |
2311 | 4.43k | const int tile_size_bytes = pbi->tile_size_bytes; |
2312 | 4.43k | int tile_width, tile_height; |
2313 | 4.43k | if (!av1_get_uniform_tile_size(cm, &tile_width, &tile_height)) { |
2314 | 924 | aom_internal_error( |
2315 | 924 | &pbi->error, AOM_CODEC_CORRUPT_FRAME, |
2316 | 924 | "Not all the tiles in the tile list have the same size."); |
2317 | 924 | } |
2318 | 4.43k | const int tile_copy_mode = |
2319 | 4.43k | ((AOMMAX(tile_width, tile_height) << MI_SIZE_LOG2) <= 256) ? 1 : 0; |
2320 | | // Read tile column sizes for all columns (we need the last tile buffer) |
2321 | 8.53k | for (int c = 0; c < tile_cols; ++c) { |
2322 | 4.10k | const int is_last = c == tile_cols - 1; |
2323 | 4.10k | size_t tile_col_size; |
2324 | | |
2325 | 4.10k | if (!is_last) { |
2326 | 1.25k | if (tile_col_size_bytes > data_end - data) { |
2327 | 24 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
2328 | 24 | "Not enough data to read tile_col_size"); |
2329 | 24 | } |
2330 | 1.25k | tile_col_size = mem_get_varsize(data, tile_col_size_bytes); |
2331 | 1.25k | data += tile_col_size_bytes; |
2332 | 1.25k | if (tile_col_size > (size_t)(data_end - data)) { |
2333 | 638 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
2334 | 638 | "tile_col_data_end[%d] is out of bound", c); |
2335 | 638 | } |
2336 | 1.25k | tile_col_data_end[c] = data + tile_col_size; |
2337 | 2.84k | } else { |
2338 | 2.84k | tile_col_size = data_end - data; |
2339 | 2.84k | tile_col_data_end[c] = data_end; |
2340 | 2.84k | } |
2341 | 4.10k | data += tile_col_size; |
2342 | 4.10k | } |
2343 | | |
2344 | 4.43k | data = data_start; |
2345 | | |
2346 | | // Read the required tile sizes. |
2347 | 7.28k | for (int c = tile_cols_start; c < tile_cols_end; ++c) { |
2348 | 2.85k | const int is_last = c == tile_cols - 1; |
2349 | | |
2350 | 2.85k | if (c > 0) data = tile_col_data_end[c - 1]; |
2351 | | |
2352 | 2.85k | if (!is_last) data += tile_col_size_bytes; |
2353 | | |
2354 | | // Get the whole of the last column, otherwise stop at the required tile. |
2355 | 22.2k | for (int r = 0; r < (is_last ? tile_rows : tile_rows_end); ++r) { |
2356 | 19.4k | get_ls_tile_buffer(tile_col_data_end[c], &pbi->error, &data, |
2357 | 19.4k | tile_buffers, tile_size_bytes, c, r, tile_copy_mode); |
2358 | 19.4k | } |
2359 | 2.85k | } |
2360 | | |
2361 | | // If we have not read the last column, then read it to get the last tile. |
2362 | 4.43k | if (tile_cols_end != tile_cols) { |
2363 | 0 | const int c = tile_cols - 1; |
2364 | |
|
2365 | 0 | data = tile_col_data_end[c - 1]; |
2366 | |
|
2367 | 0 | for (int r = 0; r < tile_rows; ++r) { |
2368 | 0 | get_ls_tile_buffer(tile_col_data_end[c], &pbi->error, &data, |
2369 | 0 | tile_buffers, tile_size_bytes, c, r, tile_copy_mode); |
2370 | 0 | } |
2371 | 0 | } |
2372 | 4.43k | raw_data_end = data; |
2373 | 4.43k | } |
2374 | 13.3k | return raw_data_end; |
2375 | 13.3k | } |
2376 | | #endif // EXT_TILE_DEBUG |
2377 | | |
2378 | | static const uint8_t *get_ls_single_tile_buffer( |
2379 | | AV1Decoder *pbi, const uint8_t *data, |
2380 | 0 | TileBufferDec (*const tile_buffers)[MAX_TILE_COLS]) { |
2381 | 0 | assert(pbi->dec_tile_row >= 0 && pbi->dec_tile_col >= 0); |
2382 | 0 | tile_buffers[pbi->dec_tile_row][pbi->dec_tile_col].data = data; |
2383 | 0 | tile_buffers[pbi->dec_tile_row][pbi->dec_tile_col].size = |
2384 | 0 | (size_t)pbi->coded_tile_data_size; |
2385 | 0 | return data + pbi->coded_tile_data_size; |
2386 | 0 | } |
2387 | | |
2388 | | // Reads the next tile returning its size and adjusting '*data' accordingly |
2389 | | // based on 'is_last'. |
2390 | | static AOM_INLINE void get_tile_buffer( |
2391 | | const uint8_t *const data_end, const int tile_size_bytes, int is_last, |
2392 | | struct aom_internal_error_info *error_info, const uint8_t **data, |
2393 | 191k | TileBufferDec *const buf) { |
2394 | 191k | size_t size; |
2395 | | |
2396 | 191k | if (!is_last) { |
2397 | 16.7k | if (!read_is_valid(*data, tile_size_bytes, data_end)) |
2398 | 282 | aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME, |
2399 | 282 | "Not enough data to read tile size"); |
2400 | | |
2401 | 16.7k | size = mem_get_varsize(*data, tile_size_bytes) + AV1_MIN_TILE_SIZE_BYTES; |
2402 | 16.7k | *data += tile_size_bytes; |
2403 | | |
2404 | 16.7k | if (size > (size_t)(data_end - *data)) |
2405 | 1.25k | aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME, |
2406 | 1.25k | "Truncated packet or corrupt tile size"); |
2407 | 174k | } else { |
2408 | 174k | size = data_end - *data; |
2409 | 174k | } |
2410 | | |
2411 | 191k | buf->data = *data; |
2412 | 191k | buf->size = size; |
2413 | | |
2414 | 191k | *data += size; |
2415 | 191k | } |
2416 | | |
2417 | | static AOM_INLINE void get_tile_buffers( |
2418 | | AV1Decoder *pbi, const uint8_t *data, const uint8_t *data_end, |
2419 | | TileBufferDec (*const tile_buffers)[MAX_TILE_COLS], int start_tile, |
2420 | 176k | int end_tile) { |
2421 | 176k | AV1_COMMON *const cm = &pbi->common; |
2422 | 176k | const int tile_cols = cm->tiles.cols; |
2423 | 176k | const int tile_rows = cm->tiles.rows; |
2424 | 176k | int tc = 0; |
2425 | | |
2426 | 357k | for (int r = 0; r < tile_rows; ++r) { |
2427 | 373k | for (int c = 0; c < tile_cols; ++c, ++tc) { |
2428 | 192k | TileBufferDec *const buf = &tile_buffers[r][c]; |
2429 | | |
2430 | 192k | const int is_last = (tc == end_tile); |
2431 | 192k | const size_t hdr_offset = 0; |
2432 | | |
2433 | 192k | if (tc < start_tile || tc > end_tile) continue; |
2434 | | |
2435 | 191k | if (data + hdr_offset >= data_end) |
2436 | 211 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
2437 | 211 | "Data ended before all tiles were read."); |
2438 | 191k | data += hdr_offset; |
2439 | 191k | get_tile_buffer(data_end, pbi->tile_size_bytes, is_last, &pbi->error, |
2440 | 191k | &data, buf); |
2441 | 191k | } |
2442 | 180k | } |
2443 | 176k | } |
2444 | | |
2445 | | static AOM_INLINE void set_cb_buffer(AV1Decoder *pbi, DecoderCodingBlock *dcb, |
2446 | | CB_BUFFER *cb_buffer_base, |
2447 | | const int num_planes, int mi_row, |
2448 | 2.25M | int mi_col) { |
2449 | 2.25M | AV1_COMMON *const cm = &pbi->common; |
2450 | 2.25M | int mib_size_log2 = cm->seq_params->mib_size_log2; |
2451 | 2.25M | int stride = (cm->mi_params.mi_cols >> mib_size_log2) + 1; |
2452 | 2.25M | int offset = (mi_row >> mib_size_log2) * stride + (mi_col >> mib_size_log2); |
2453 | 2.25M | CB_BUFFER *cb_buffer = cb_buffer_base + offset; |
2454 | | |
2455 | 8.86M | for (int plane = 0; plane < num_planes; ++plane) { |
2456 | 6.60M | dcb->dqcoeff_block[plane] = cb_buffer->dqcoeff[plane]; |
2457 | 6.60M | dcb->eob_data[plane] = cb_buffer->eob_data[plane]; |
2458 | 6.60M | dcb->cb_offset[plane] = 0; |
2459 | 6.60M | dcb->txb_offset[plane] = 0; |
2460 | 6.60M | } |
2461 | 2.25M | MACROBLOCKD *const xd = &dcb->xd; |
2462 | 2.25M | xd->plane[0].color_index_map = cb_buffer->color_index_map[0]; |
2463 | 2.25M | xd->plane[1].color_index_map = cb_buffer->color_index_map[1]; |
2464 | 2.25M | xd->color_index_map_offset[0] = 0; |
2465 | 2.25M | xd->color_index_map_offset[1] = 0; |
2466 | 2.25M | } |
2467 | | |
2468 | | static AOM_INLINE void decoder_alloc_tile_data(AV1Decoder *pbi, |
2469 | 16.6k | const int n_tiles) { |
2470 | 16.6k | AV1_COMMON *const cm = &pbi->common; |
2471 | 16.6k | aom_free(pbi->tile_data); |
2472 | 16.6k | pbi->allocated_tiles = 0; |
2473 | 16.6k | CHECK_MEM_ERROR(cm, pbi->tile_data, |
2474 | 16.6k | aom_memalign(32, n_tiles * sizeof(*pbi->tile_data))); |
2475 | 16.6k | pbi->allocated_tiles = n_tiles; |
2476 | 56.6k | for (int i = 0; i < n_tiles; i++) { |
2477 | 40.0k | TileDataDec *const tile_data = pbi->tile_data + i; |
2478 | 40.0k | av1_zero(tile_data->dec_row_mt_sync); |
2479 | 40.0k | } |
2480 | 16.6k | pbi->allocated_row_mt_sync_rows = 0; |
2481 | 16.6k | } |
2482 | | |
2483 | | // Set up nsync by width. |
2484 | 31.2k | static INLINE int get_sync_range(int width) { |
2485 | | // nsync numbers are picked by testing. |
2486 | | #if 0 |
2487 | | if (width < 640) |
2488 | | return 1; |
2489 | | else if (width <= 1280) |
2490 | | return 2; |
2491 | | else if (width <= 4096) |
2492 | | return 4; |
2493 | | else |
2494 | | return 8; |
2495 | | #else |
2496 | 31.2k | (void)width; |
2497 | 31.2k | #endif |
2498 | 31.2k | return 1; |
2499 | 31.2k | } |
2500 | | |
2501 | | // Allocate memory for decoder row synchronization |
2502 | | static AOM_INLINE void dec_row_mt_alloc(AV1DecRowMTSync *dec_row_mt_sync, |
2503 | 31.2k | AV1_COMMON *cm, int rows) { |
2504 | 31.2k | dec_row_mt_sync->allocated_sb_rows = rows; |
2505 | 31.2k | #if CONFIG_MULTITHREAD |
2506 | 31.2k | { |
2507 | 31.2k | int i; |
2508 | | |
2509 | 31.2k | CHECK_MEM_ERROR(cm, dec_row_mt_sync->mutex_, |
2510 | 31.2k | aom_malloc(sizeof(*(dec_row_mt_sync->mutex_)) * rows)); |
2511 | 31.2k | if (dec_row_mt_sync->mutex_) { |
2512 | 95.4k | for (i = 0; i < rows; ++i) { |
2513 | 64.1k | pthread_mutex_init(&dec_row_mt_sync->mutex_[i], NULL); |
2514 | 64.1k | } |
2515 | 31.2k | } |
2516 | | |
2517 | 31.2k | CHECK_MEM_ERROR(cm, dec_row_mt_sync->cond_, |
2518 | 31.2k | aom_malloc(sizeof(*(dec_row_mt_sync->cond_)) * rows)); |
2519 | 31.2k | if (dec_row_mt_sync->cond_) { |
2520 | 95.4k | for (i = 0; i < rows; ++i) { |
2521 | 64.1k | pthread_cond_init(&dec_row_mt_sync->cond_[i], NULL); |
2522 | 64.1k | } |
2523 | 31.2k | } |
2524 | 31.2k | } |
2525 | 31.2k | #endif // CONFIG_MULTITHREAD |
2526 | | |
2527 | 31.2k | CHECK_MEM_ERROR(cm, dec_row_mt_sync->cur_sb_col, |
2528 | 31.2k | aom_malloc(sizeof(*(dec_row_mt_sync->cur_sb_col)) * rows)); |
2529 | | |
2530 | | // Set up nsync. |
2531 | 31.2k | dec_row_mt_sync->sync_range = get_sync_range(cm->width); |
2532 | 31.2k | } |
2533 | | |
2534 | | // Deallocate decoder row synchronization related mutex and data |
2535 | 66.5k | void av1_dec_row_mt_dealloc(AV1DecRowMTSync *dec_row_mt_sync) { |
2536 | 66.5k | if (dec_row_mt_sync != NULL) { |
2537 | 66.5k | #if CONFIG_MULTITHREAD |
2538 | 66.5k | int i; |
2539 | 66.5k | if (dec_row_mt_sync->mutex_ != NULL) { |
2540 | 95.4k | for (i = 0; i < dec_row_mt_sync->allocated_sb_rows; ++i) { |
2541 | 64.1k | pthread_mutex_destroy(&dec_row_mt_sync->mutex_[i]); |
2542 | 64.1k | } |
2543 | 31.2k | aom_free(dec_row_mt_sync->mutex_); |
2544 | 31.2k | } |
2545 | 66.5k | if (dec_row_mt_sync->cond_ != NULL) { |
2546 | 95.4k | for (i = 0; i < dec_row_mt_sync->allocated_sb_rows; ++i) { |
2547 | 64.1k | pthread_cond_destroy(&dec_row_mt_sync->cond_[i]); |
2548 | 64.1k | } |
2549 | 31.2k | aom_free(dec_row_mt_sync->cond_); |
2550 | 31.2k | } |
2551 | 66.5k | #endif // CONFIG_MULTITHREAD |
2552 | 66.5k | aom_free(dec_row_mt_sync->cur_sb_col); |
2553 | | |
2554 | | // clear the structure as the source of this call may be a resize in which |
2555 | | // case this call will be followed by an _alloc() which may fail. |
2556 | 66.5k | av1_zero(*dec_row_mt_sync); |
2557 | 66.5k | } |
2558 | 66.5k | } |
2559 | | |
2560 | | static INLINE void sync_read(AV1DecRowMTSync *const dec_row_mt_sync, int r, |
2561 | 808k | int c) { |
2562 | 808k | #if CONFIG_MULTITHREAD |
2563 | 808k | const int nsync = dec_row_mt_sync->sync_range; |
2564 | | |
2565 | 808k | if (r && !(c & (nsync - 1))) { |
2566 | 491k | pthread_mutex_t *const mutex = &dec_row_mt_sync->mutex_[r - 1]; |
2567 | 491k | pthread_mutex_lock(mutex); |
2568 | | |
2569 | 637k | while (c > dec_row_mt_sync->cur_sb_col[r - 1] - nsync - |
2570 | 637k | dec_row_mt_sync->intrabc_extra_top_right_sb_delay) { |
2571 | 145k | pthread_cond_wait(&dec_row_mt_sync->cond_[r - 1], mutex); |
2572 | 145k | } |
2573 | 491k | pthread_mutex_unlock(mutex); |
2574 | 491k | } |
2575 | | #else |
2576 | | (void)dec_row_mt_sync; |
2577 | | (void)r; |
2578 | | (void)c; |
2579 | | #endif // CONFIG_MULTITHREAD |
2580 | 808k | } |
2581 | | |
2582 | | static INLINE void sync_write(AV1DecRowMTSync *const dec_row_mt_sync, int r, |
2583 | 813k | int c, const int sb_cols) { |
2584 | 813k | #if CONFIG_MULTITHREAD |
2585 | 813k | const int nsync = dec_row_mt_sync->sync_range; |
2586 | 813k | int cur; |
2587 | 813k | int sig = 1; |
2588 | | |
2589 | 813k | if (c < sb_cols - 1) { |
2590 | 580k | cur = c; |
2591 | 580k | if (c % nsync) sig = 0; |
2592 | 580k | } else { |
2593 | 233k | cur = sb_cols + nsync + dec_row_mt_sync->intrabc_extra_top_right_sb_delay; |
2594 | 233k | } |
2595 | | |
2596 | 813k | if (sig) { |
2597 | 813k | pthread_mutex_lock(&dec_row_mt_sync->mutex_[r]); |
2598 | | |
2599 | 813k | dec_row_mt_sync->cur_sb_col[r] = cur; |
2600 | | |
2601 | 813k | pthread_cond_signal(&dec_row_mt_sync->cond_[r]); |
2602 | 813k | pthread_mutex_unlock(&dec_row_mt_sync->mutex_[r]); |
2603 | 813k | } |
2604 | | #else |
2605 | | (void)dec_row_mt_sync; |
2606 | | (void)r; |
2607 | | (void)c; |
2608 | | (void)sb_cols; |
2609 | | #endif // CONFIG_MULTITHREAD |
2610 | 813k | } |
2611 | | |
2612 | | static INLINE void signal_decoding_done_for_erroneous_row( |
2613 | 934 | AV1Decoder *const pbi, const MACROBLOCKD *const xd) { |
2614 | 934 | AV1_COMMON *const cm = &pbi->common; |
2615 | 934 | const TileInfo *const tile = &xd->tile; |
2616 | 934 | const int sb_row_in_tile = |
2617 | 934 | ((xd->mi_row - tile->mi_row_start) >> cm->seq_params->mib_size_log2); |
2618 | 934 | const int sb_cols_in_tile = av1_get_sb_cols_in_tile(cm, tile); |
2619 | 934 | TileDataDec *const tile_data = |
2620 | 934 | pbi->tile_data + tile->tile_row * cm->tiles.cols + tile->tile_col; |
2621 | 934 | AV1DecRowMTSync *dec_row_mt_sync = &tile_data->dec_row_mt_sync; |
2622 | | |
2623 | 934 | sync_write(dec_row_mt_sync, sb_row_in_tile, sb_cols_in_tile - 1, |
2624 | 934 | sb_cols_in_tile); |
2625 | 934 | } |
2626 | | |
2627 | | static AOM_INLINE void decode_tile_sb_row(AV1Decoder *pbi, ThreadData *const td, |
2628 | | const TileInfo *tile_info, |
2629 | 228k | const int mi_row) { |
2630 | 228k | AV1_COMMON *const cm = &pbi->common; |
2631 | 228k | const int num_planes = av1_num_planes(cm); |
2632 | 228k | TileDataDec *const tile_data = pbi->tile_data + |
2633 | 228k | tile_info->tile_row * cm->tiles.cols + |
2634 | 228k | tile_info->tile_col; |
2635 | 228k | const int sb_cols_in_tile = av1_get_sb_cols_in_tile(cm, tile_info); |
2636 | 228k | const int sb_row_in_tile = |
2637 | 228k | (mi_row - tile_info->mi_row_start) >> cm->seq_params->mib_size_log2; |
2638 | 228k | int sb_col_in_tile = 0; |
2639 | 228k | int row_mt_exit = 0; |
2640 | | |
2641 | 1.03M | for (int mi_col = tile_info->mi_col_start; mi_col < tile_info->mi_col_end; |
2642 | 808k | mi_col += cm->seq_params->mib_size, sb_col_in_tile++) { |
2643 | 808k | set_cb_buffer(pbi, &td->dcb, pbi->cb_buffer_base, num_planes, mi_row, |
2644 | 808k | mi_col); |
2645 | | |
2646 | 808k | sync_read(&tile_data->dec_row_mt_sync, sb_row_in_tile, sb_col_in_tile); |
2647 | | |
2648 | 808k | #if CONFIG_MULTITHREAD |
2649 | 808k | pthread_mutex_lock(pbi->row_mt_mutex_); |
2650 | 808k | #endif |
2651 | 808k | row_mt_exit = pbi->frame_row_mt_info.row_mt_exit; |
2652 | 808k | #if CONFIG_MULTITHREAD |
2653 | 808k | pthread_mutex_unlock(pbi->row_mt_mutex_); |
2654 | 808k | #endif |
2655 | | |
2656 | 809k | if (!row_mt_exit) { |
2657 | | // Decoding of the super-block |
2658 | 809k | decode_partition(pbi, td, mi_row, mi_col, td->bit_reader, |
2659 | 809k | cm->seq_params->sb_size, 0x2); |
2660 | 809k | } |
2661 | | |
2662 | 808k | sync_write(&tile_data->dec_row_mt_sync, sb_row_in_tile, sb_col_in_tile, |
2663 | 808k | sb_cols_in_tile); |
2664 | 808k | } |
2665 | 228k | } |
2666 | | |
2667 | 149k | static int check_trailing_bits_after_symbol_coder(aom_reader *r) { |
2668 | 149k | if (aom_reader_has_overflowed(r)) return -1; |
2669 | | |
2670 | 149k | uint32_t nb_bits = aom_reader_tell(r); |
2671 | 149k | uint32_t nb_bytes = (nb_bits + 7) >> 3; |
2672 | 149k | const uint8_t *p = aom_reader_find_begin(r) + nb_bytes; |
2673 | | |
2674 | | // aom_reader_tell() returns 1 for a newly initialized decoder, and the |
2675 | | // return value only increases as values are decoded. So nb_bits > 0, and |
2676 | | // thus p > p_begin. Therefore accessing p[-1] is safe. |
2677 | 149k | uint8_t last_byte = p[-1]; |
2678 | 149k | uint8_t pattern = 128 >> ((nb_bits - 1) & 7); |
2679 | 149k | if ((last_byte & (2 * pattern - 1)) != pattern) return -1; |
2680 | | |
2681 | | // Make sure that all padding bytes are zero as required by the spec. |
2682 | 140k | const uint8_t *p_end = aom_reader_find_end(r); |
2683 | 198k | while (p < p_end) { |
2684 | 58.7k | if (*p != 0) return -1; |
2685 | 57.3k | p++; |
2686 | 57.3k | } |
2687 | 139k | return 0; |
2688 | 140k | } |
2689 | | |
2690 | | static AOM_INLINE void set_decode_func_pointers(ThreadData *td, |
2691 | 499k | int parse_decode_flag) { |
2692 | 499k | td->read_coeffs_tx_intra_block_visit = decode_block_void; |
2693 | 499k | td->predict_and_recon_intra_block_visit = decode_block_void; |
2694 | 499k | td->read_coeffs_tx_inter_block_visit = decode_block_void; |
2695 | 499k | td->inverse_tx_inter_block_visit = decode_block_void; |
2696 | 499k | td->predict_inter_block_visit = predict_inter_block_void; |
2697 | 499k | td->cfl_store_inter_block_visit = cfl_store_inter_block_void; |
2698 | | |
2699 | 499k | if (parse_decode_flag & 0x1) { |
2700 | 292k | td->read_coeffs_tx_intra_block_visit = read_coeffs_tx_intra_block; |
2701 | 292k | td->read_coeffs_tx_inter_block_visit = av1_read_coeffs_txb_facade; |
2702 | 292k | } |
2703 | 499k | if (parse_decode_flag & 0x2) { |
2704 | 269k | td->predict_and_recon_intra_block_visit = |
2705 | 269k | predict_and_reconstruct_intra_block; |
2706 | 269k | td->inverse_tx_inter_block_visit = inverse_transform_inter_block; |
2707 | 269k | td->predict_inter_block_visit = predict_inter_block; |
2708 | 269k | td->cfl_store_inter_block_visit = cfl_store_inter_block; |
2709 | 269k | } |
2710 | 499k | } |
2711 | | |
2712 | | static AOM_INLINE void decode_tile(AV1Decoder *pbi, ThreadData *const td, |
2713 | 62.0k | int tile_row, int tile_col) { |
2714 | 62.0k | TileInfo tile_info; |
2715 | | |
2716 | 62.0k | AV1_COMMON *const cm = &pbi->common; |
2717 | 62.0k | const int num_planes = av1_num_planes(cm); |
2718 | | |
2719 | 62.0k | av1_tile_set_row(&tile_info, cm, tile_row); |
2720 | 62.0k | av1_tile_set_col(&tile_info, cm, tile_col); |
2721 | 62.0k | DecoderCodingBlock *const dcb = &td->dcb; |
2722 | 62.0k | MACROBLOCKD *const xd = &dcb->xd; |
2723 | | |
2724 | 62.0k | av1_zero_above_context(cm, xd, tile_info.mi_col_start, tile_info.mi_col_end, |
2725 | 62.0k | tile_row); |
2726 | 62.0k | av1_reset_loop_filter_delta(xd, num_planes); |
2727 | 62.0k | av1_reset_loop_restoration(xd, num_planes); |
2728 | | |
2729 | 153k | for (int mi_row = tile_info.mi_row_start; mi_row < tile_info.mi_row_end; |
2730 | 125k | mi_row += cm->seq_params->mib_size) { |
2731 | 125k | av1_zero_left_context(xd); |
2732 | | |
2733 | 682k | for (int mi_col = tile_info.mi_col_start; mi_col < tile_info.mi_col_end; |
2734 | 590k | mi_col += cm->seq_params->mib_size) { |
2735 | 590k | set_cb_buffer(pbi, dcb, &td->cb_buffer_base, num_planes, 0, 0); |
2736 | | |
2737 | | // Bit-stream parsing and decoding of the superblock |
2738 | 590k | decode_partition(pbi, td, mi_row, mi_col, td->bit_reader, |
2739 | 590k | cm->seq_params->sb_size, 0x3); |
2740 | | |
2741 | 590k | if (aom_reader_has_overflowed(td->bit_reader)) { |
2742 | 33.8k | aom_merge_corrupted_flag(&dcb->corrupted, 1); |
2743 | 33.8k | return; |
2744 | 33.8k | } |
2745 | 590k | } |
2746 | 125k | } |
2747 | | |
2748 | 28.2k | int corrupted = |
2749 | 28.2k | (check_trailing_bits_after_symbol_coder(td->bit_reader)) ? 1 : 0; |
2750 | 28.2k | aom_merge_corrupted_flag(&dcb->corrupted, corrupted); |
2751 | 28.2k | } |
2752 | | |
2753 | | static const uint8_t *decode_tiles(AV1Decoder *pbi, const uint8_t *data, |
2754 | | const uint8_t *data_end, int start_tile, |
2755 | 65.5k | int end_tile) { |
2756 | 65.5k | AV1_COMMON *const cm = &pbi->common; |
2757 | 65.5k | ThreadData *const td = &pbi->td; |
2758 | 65.5k | CommonTileParams *const tiles = &cm->tiles; |
2759 | 65.5k | const int tile_cols = tiles->cols; |
2760 | 65.5k | const int tile_rows = tiles->rows; |
2761 | 65.5k | const int n_tiles = tile_cols * tile_rows; |
2762 | 65.5k | TileBufferDec(*const tile_buffers)[MAX_TILE_COLS] = pbi->tile_buffers; |
2763 | 65.5k | const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows); |
2764 | 65.5k | const int single_row = pbi->dec_tile_row >= 0; |
2765 | 65.5k | const int dec_tile_col = AOMMIN(pbi->dec_tile_col, tile_cols); |
2766 | 65.5k | const int single_col = pbi->dec_tile_col >= 0; |
2767 | 65.5k | int tile_rows_start; |
2768 | 65.5k | int tile_rows_end; |
2769 | 65.5k | int tile_cols_start; |
2770 | 65.5k | int tile_cols_end; |
2771 | 65.5k | int inv_col_order; |
2772 | 65.5k | int inv_row_order; |
2773 | 65.5k | int tile_row, tile_col; |
2774 | 65.5k | uint8_t allow_update_cdf; |
2775 | 65.5k | const uint8_t *raw_data_end = NULL; |
2776 | | |
2777 | 65.5k | if (tiles->large_scale) { |
2778 | 9.73k | tile_rows_start = single_row ? dec_tile_row : 0; |
2779 | 9.73k | tile_rows_end = single_row ? dec_tile_row + 1 : tile_rows; |
2780 | 9.73k | tile_cols_start = single_col ? dec_tile_col : 0; |
2781 | 9.73k | tile_cols_end = single_col ? tile_cols_start + 1 : tile_cols; |
2782 | 9.73k | inv_col_order = pbi->inv_tile_order && !single_col; |
2783 | 9.73k | inv_row_order = pbi->inv_tile_order && !single_row; |
2784 | 9.73k | allow_update_cdf = 0; |
2785 | 55.8k | } else { |
2786 | 55.8k | tile_rows_start = 0; |
2787 | 55.8k | tile_rows_end = tile_rows; |
2788 | 55.8k | tile_cols_start = 0; |
2789 | 55.8k | tile_cols_end = tile_cols; |
2790 | 55.8k | inv_col_order = pbi->inv_tile_order; |
2791 | 55.8k | inv_row_order = pbi->inv_tile_order; |
2792 | 55.8k | allow_update_cdf = 1; |
2793 | 55.8k | } |
2794 | | |
2795 | | // No tiles to decode. |
2796 | 65.5k | if (tile_rows_end <= tile_rows_start || tile_cols_end <= tile_cols_start || |
2797 | | // First tile is larger than end_tile. |
2798 | 65.5k | tile_rows_start * tiles->cols + tile_cols_start > end_tile || |
2799 | | // Last tile is smaller than start_tile. |
2800 | 65.5k | (tile_rows_end - 1) * tiles->cols + tile_cols_end - 1 < start_tile) |
2801 | 0 | return data; |
2802 | | |
2803 | 65.5k | allow_update_cdf = allow_update_cdf && !cm->features.disable_cdf_update; |
2804 | | |
2805 | 65.5k | assert(tile_rows <= MAX_TILE_ROWS); |
2806 | 65.5k | assert(tile_cols <= MAX_TILE_COLS); |
2807 | | |
2808 | 65.5k | #if EXT_TILE_DEBUG |
2809 | 65.5k | if (tiles->large_scale && !pbi->ext_tile_debug) |
2810 | 0 | raw_data_end = get_ls_single_tile_buffer(pbi, data, tile_buffers); |
2811 | 65.5k | else if (tiles->large_scale && pbi->ext_tile_debug) |
2812 | 9.73k | raw_data_end = get_ls_tile_buffers(pbi, data, data_end, tile_buffers); |
2813 | 55.8k | else |
2814 | 55.8k | #endif // EXT_TILE_DEBUG |
2815 | 55.8k | get_tile_buffers(pbi, data, data_end, tile_buffers, start_tile, end_tile); |
2816 | | |
2817 | 65.5k | if (pbi->tile_data == NULL || n_tiles != pbi->allocated_tiles) { |
2818 | 11.7k | decoder_alloc_tile_data(pbi, n_tiles); |
2819 | 11.7k | } |
2820 | 65.5k | if (pbi->dcb.xd.seg_mask == NULL) |
2821 | 65.5k | CHECK_MEM_ERROR(cm, pbi->dcb.xd.seg_mask, |
2822 | 65.5k | (uint8_t *)aom_memalign( |
2823 | 65.5k | 16, 2 * MAX_SB_SQUARE * sizeof(*pbi->dcb.xd.seg_mask))); |
2824 | | #if CONFIG_ACCOUNTING |
2825 | | if (pbi->acct_enabled) { |
2826 | | aom_accounting_reset(&pbi->accounting); |
2827 | | } |
2828 | | #endif |
2829 | | |
2830 | 65.5k | set_decode_func_pointers(&pbi->td, 0x3); |
2831 | | |
2832 | | // Load all tile information into thread_data. |
2833 | 65.5k | td->dcb = pbi->dcb; |
2834 | | |
2835 | 65.5k | td->dcb.corrupted = 0; |
2836 | 65.5k | td->dcb.mc_buf[0] = td->mc_buf[0]; |
2837 | 65.5k | td->dcb.mc_buf[1] = td->mc_buf[1]; |
2838 | 65.5k | td->dcb.xd.tmp_conv_dst = td->tmp_conv_dst; |
2839 | 189k | for (int j = 0; j < 2; ++j) { |
2840 | 123k | td->dcb.xd.tmp_obmc_bufs[j] = td->tmp_obmc_bufs[j]; |
2841 | 123k | } |
2842 | | |
2843 | 127k | for (tile_row = tile_rows_start; tile_row < tile_rows_end; ++tile_row) { |
2844 | 61.9k | const int row = inv_row_order ? tile_rows - 1 - tile_row : tile_row; |
2845 | | |
2846 | 124k | for (tile_col = tile_cols_start; tile_col < tile_cols_end; ++tile_col) { |
2847 | 62.2k | const int col = inv_col_order ? tile_cols - 1 - tile_col : tile_col; |
2848 | 62.2k | TileDataDec *const tile_data = pbi->tile_data + row * tiles->cols + col; |
2849 | 62.2k | const TileBufferDec *const tile_bs_buf = &tile_buffers[row][col]; |
2850 | | |
2851 | 62.2k | if (row * tiles->cols + col < start_tile || |
2852 | 62.2k | row * tiles->cols + col > end_tile) |
2853 | 90 | continue; |
2854 | | |
2855 | 62.1k | td->bit_reader = &tile_data->bit_reader; |
2856 | 62.1k | av1_zero(td->cb_buffer_base.dqcoeff); |
2857 | 62.1k | av1_tile_init(&td->dcb.xd.tile, cm, row, col); |
2858 | 62.1k | td->dcb.xd.current_base_qindex = cm->quant_params.base_qindex; |
2859 | 62.1k | setup_bool_decoder(&td->dcb.xd, tile_bs_buf->data, data_end, |
2860 | 62.1k | tile_bs_buf->size, &pbi->error, td->bit_reader, |
2861 | 62.1k | allow_update_cdf); |
2862 | | #if CONFIG_ACCOUNTING |
2863 | | if (pbi->acct_enabled) { |
2864 | | td->bit_reader->accounting = &pbi->accounting; |
2865 | | td->bit_reader->accounting->last_tell_frac = |
2866 | | aom_reader_tell_frac(td->bit_reader); |
2867 | | } else { |
2868 | | td->bit_reader->accounting = NULL; |
2869 | | } |
2870 | | #endif |
2871 | 62.1k | av1_init_macroblockd(cm, &td->dcb.xd); |
2872 | 62.1k | av1_init_above_context(&cm->above_contexts, av1_num_planes(cm), row, |
2873 | 62.1k | &td->dcb.xd); |
2874 | | |
2875 | | // Initialise the tile context from the frame context |
2876 | 62.1k | tile_data->tctx = *cm->fc; |
2877 | 62.1k | td->dcb.xd.tile_ctx = &tile_data->tctx; |
2878 | | |
2879 | | // decode tile |
2880 | 62.1k | decode_tile(pbi, td, row, col); |
2881 | 62.1k | aom_merge_corrupted_flag(&pbi->dcb.corrupted, td->dcb.corrupted); |
2882 | 62.1k | if (pbi->dcb.corrupted) |
2883 | 35.8k | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
2884 | 35.8k | "Failed to decode tile data"); |
2885 | 62.1k | } |
2886 | 61.9k | } |
2887 | | |
2888 | 65.5k | if (tiles->large_scale) { |
2889 | 3.23k | if (n_tiles == 1) { |
2890 | | // Find the end of the single tile buffer |
2891 | 3.23k | return aom_reader_find_end(&pbi->tile_data->bit_reader); |
2892 | 3.23k | } |
2893 | | // Return the end of the last tile buffer |
2894 | 0 | return raw_data_end; |
2895 | 3.23k | } |
2896 | 62.3k | TileDataDec *const tile_data = pbi->tile_data + end_tile; |
2897 | | |
2898 | 62.3k | return aom_reader_find_end(&tile_data->bit_reader); |
2899 | 65.5k | } |
2900 | | |
2901 | 350k | static TileJobsDec *get_dec_job_info(AV1DecTileMT *tile_mt_info) { |
2902 | 350k | TileJobsDec *cur_job_info = NULL; |
2903 | 350k | #if CONFIG_MULTITHREAD |
2904 | 350k | pthread_mutex_lock(tile_mt_info->job_mutex); |
2905 | | |
2906 | 350k | if (tile_mt_info->jobs_dequeued < tile_mt_info->jobs_enqueued) { |
2907 | 143k | cur_job_info = tile_mt_info->job_queue + tile_mt_info->jobs_dequeued; |
2908 | 143k | tile_mt_info->jobs_dequeued++; |
2909 | 143k | } |
2910 | | |
2911 | 350k | pthread_mutex_unlock(tile_mt_info->job_mutex); |
2912 | | #else |
2913 | | (void)tile_mt_info; |
2914 | | #endif |
2915 | 350k | return cur_job_info; |
2916 | 350k | } |
2917 | | |
2918 | | static AOM_INLINE void tile_worker_hook_init( |
2919 | | AV1Decoder *const pbi, DecWorkerData *const thread_data, |
2920 | | const TileBufferDec *const tile_buffer, TileDataDec *const tile_data, |
2921 | 143k | uint8_t allow_update_cdf) { |
2922 | 143k | AV1_COMMON *cm = &pbi->common; |
2923 | 143k | ThreadData *const td = thread_data->td; |
2924 | 143k | int tile_row = tile_data->tile_info.tile_row; |
2925 | 143k | int tile_col = tile_data->tile_info.tile_col; |
2926 | | |
2927 | 143k | td->bit_reader = &tile_data->bit_reader; |
2928 | 143k | av1_zero(td->cb_buffer_base.dqcoeff); |
2929 | | |
2930 | 143k | MACROBLOCKD *const xd = &td->dcb.xd; |
2931 | 143k | av1_tile_init(&xd->tile, cm, tile_row, tile_col); |
2932 | 143k | xd->current_base_qindex = cm->quant_params.base_qindex; |
2933 | | |
2934 | 143k | setup_bool_decoder(xd, tile_buffer->data, thread_data->data_end, |
2935 | 143k | tile_buffer->size, &thread_data->error_info, |
2936 | 143k | td->bit_reader, allow_update_cdf); |
2937 | | #if CONFIG_ACCOUNTING |
2938 | | if (pbi->acct_enabled) { |
2939 | | td->bit_reader->accounting = &pbi->accounting; |
2940 | | td->bit_reader->accounting->last_tell_frac = |
2941 | | aom_reader_tell_frac(td->bit_reader); |
2942 | | } else { |
2943 | | td->bit_reader->accounting = NULL; |
2944 | | } |
2945 | | #endif |
2946 | 143k | av1_init_macroblockd(cm, xd); |
2947 | 143k | xd->error_info = &thread_data->error_info; |
2948 | 143k | av1_init_above_context(&cm->above_contexts, av1_num_planes(cm), tile_row, xd); |
2949 | | |
2950 | | // Initialise the tile context from the frame context |
2951 | 143k | tile_data->tctx = *cm->fc; |
2952 | 143k | xd->tile_ctx = &tile_data->tctx; |
2953 | | #if CONFIG_ACCOUNTING |
2954 | | if (pbi->acct_enabled) { |
2955 | | tile_data->bit_reader.accounting->last_tell_frac = |
2956 | | aom_reader_tell_frac(&tile_data->bit_reader); |
2957 | | } |
2958 | | #endif |
2959 | 143k | } |
2960 | | |
2961 | 0 | static int tile_worker_hook(void *arg1, void *arg2) { |
2962 | 0 | DecWorkerData *const thread_data = (DecWorkerData *)arg1; |
2963 | 0 | AV1Decoder *const pbi = (AV1Decoder *)arg2; |
2964 | 0 | AV1_COMMON *cm = &pbi->common; |
2965 | 0 | ThreadData *const td = thread_data->td; |
2966 | 0 | uint8_t allow_update_cdf; |
2967 | | |
2968 | | // The jmp_buf is valid only for the duration of the function that calls |
2969 | | // setjmp(). Therefore, this function must reset the 'setjmp' field to 0 |
2970 | | // before it returns. |
2971 | 0 | if (setjmp(thread_data->error_info.jmp)) { |
2972 | 0 | thread_data->error_info.setjmp = 0; |
2973 | 0 | thread_data->td->dcb.corrupted = 1; |
2974 | 0 | return 0; |
2975 | 0 | } |
2976 | 0 | thread_data->error_info.setjmp = 1; |
2977 | |
|
2978 | 0 | allow_update_cdf = cm->tiles.large_scale ? 0 : 1; |
2979 | 0 | allow_update_cdf = allow_update_cdf && !cm->features.disable_cdf_update; |
2980 | |
|
2981 | 0 | set_decode_func_pointers(td, 0x3); |
2982 | |
|
2983 | 0 | assert(cm->tiles.cols > 0); |
2984 | 0 | while (!td->dcb.corrupted) { |
2985 | 0 | TileJobsDec *cur_job_info = get_dec_job_info(&pbi->tile_mt_info); |
2986 | |
|
2987 | 0 | if (cur_job_info != NULL) { |
2988 | 0 | const TileBufferDec *const tile_buffer = cur_job_info->tile_buffer; |
2989 | 0 | TileDataDec *const tile_data = cur_job_info->tile_data; |
2990 | 0 | tile_worker_hook_init(pbi, thread_data, tile_buffer, tile_data, |
2991 | 0 | allow_update_cdf); |
2992 | | // decode tile |
2993 | 0 | int tile_row = tile_data->tile_info.tile_row; |
2994 | 0 | int tile_col = tile_data->tile_info.tile_col; |
2995 | 0 | decode_tile(pbi, td, tile_row, tile_col); |
2996 | 0 | } else { |
2997 | 0 | break; |
2998 | 0 | } |
2999 | 0 | } |
3000 | 0 | thread_data->error_info.setjmp = 0; |
3001 | 0 | return !td->dcb.corrupted; |
3002 | 0 | } |
3003 | | |
3004 | | static INLINE int get_max_row_mt_workers_per_tile(AV1_COMMON *cm, |
3005 | 382k | const TileInfo *tile) { |
3006 | | // NOTE: Currently value of max workers is calculated based |
3007 | | // on the parse and decode time. As per the theoretical estimate |
3008 | | // when percentage of parse time is equal to percentage of decode |
3009 | | // time, number of workers needed to parse + decode a tile can not |
3010 | | // exceed more than 2. |
3011 | | // TODO(any): Modify this value if parsing is optimized in future. |
3012 | 382k | int sb_rows = av1_get_sb_rows_in_tile(cm, tile); |
3013 | 382k | int max_workers = |
3014 | 382k | sb_rows == 1 ? AOM_MIN_THREADS_PER_TILE : AOM_MAX_THREADS_PER_TILE; |
3015 | 382k | return max_workers; |
3016 | 382k | } |
3017 | | |
3018 | | // The caller must hold pbi->row_mt_mutex_ when calling this function. |
3019 | | // Returns 1 if either the next job is stored in *next_job_info or 1 is stored |
3020 | | // in *end_of_frame. |
3021 | | // NOTE: The caller waits on pbi->row_mt_cond_ if this function returns 0. |
3022 | | // The return value of this function depends on the following variables: |
3023 | | // - frame_row_mt_info->mi_rows_parse_done |
3024 | | // - frame_row_mt_info->mi_rows_decode_started |
3025 | | // - frame_row_mt_info->row_mt_exit |
3026 | | // Therefore we may need to signal or broadcast pbi->row_mt_cond_ if any of |
3027 | | // these variables is modified. |
3028 | | static int get_next_job_info(AV1Decoder *const pbi, |
3029 | | AV1DecRowMTJobInfo *next_job_info, |
3030 | 523k | int *end_of_frame) { |
3031 | 523k | AV1_COMMON *cm = &pbi->common; |
3032 | 523k | TileDataDec *tile_data; |
3033 | 523k | AV1DecRowMTSync *dec_row_mt_sync; |
3034 | 523k | AV1DecRowMTInfo *frame_row_mt_info = &pbi->frame_row_mt_info; |
3035 | 523k | const int tile_rows_start = frame_row_mt_info->tile_rows_start; |
3036 | 523k | const int tile_rows_end = frame_row_mt_info->tile_rows_end; |
3037 | 523k | const int tile_cols_start = frame_row_mt_info->tile_cols_start; |
3038 | 523k | const int tile_cols_end = frame_row_mt_info->tile_cols_end; |
3039 | 523k | const int start_tile = frame_row_mt_info->start_tile; |
3040 | 523k | const int end_tile = frame_row_mt_info->end_tile; |
3041 | 523k | const int sb_mi_size = mi_size_wide[cm->seq_params->sb_size]; |
3042 | 523k | int num_mis_to_decode, num_threads_working; |
3043 | 523k | int num_mis_waiting_for_decode; |
3044 | 523k | int min_threads_working = INT_MAX; |
3045 | 523k | int max_mis_to_decode = 0; |
3046 | 523k | int tile_row_idx, tile_col_idx; |
3047 | 523k | int tile_row = -1; |
3048 | 523k | int tile_col = -1; |
3049 | | |
3050 | 523k | memset(next_job_info, 0, sizeof(*next_job_info)); |
3051 | | |
3052 | | // Frame decode is completed or error is encountered. |
3053 | 523k | *end_of_frame = (frame_row_mt_info->mi_rows_decode_started == |
3054 | 523k | frame_row_mt_info->mi_rows_to_decode) || |
3055 | 523k | (frame_row_mt_info->row_mt_exit == 1); |
3056 | 523k | if (*end_of_frame) { |
3057 | 207k | return 1; |
3058 | 207k | } |
3059 | | |
3060 | | // Decoding cannot start as bit-stream parsing is not complete. |
3061 | 316k | assert(frame_row_mt_info->mi_rows_parse_done >= |
3062 | 316k | frame_row_mt_info->mi_rows_decode_started); |
3063 | 316k | if (frame_row_mt_info->mi_rows_parse_done == |
3064 | 316k | frame_row_mt_info->mi_rows_decode_started) |
3065 | 82.6k | return 0; |
3066 | | |
3067 | | // Choose the tile to decode. |
3068 | 474k | for (tile_row_idx = tile_rows_start; tile_row_idx < tile_rows_end; |
3069 | 240k | ++tile_row_idx) { |
3070 | 524k | for (tile_col_idx = tile_cols_start; tile_col_idx < tile_cols_end; |
3071 | 283k | ++tile_col_idx) { |
3072 | 283k | if (tile_row_idx * cm->tiles.cols + tile_col_idx < start_tile || |
3073 | 283k | tile_row_idx * cm->tiles.cols + tile_col_idx > end_tile) |
3074 | 384 | continue; |
3075 | | |
3076 | 283k | tile_data = pbi->tile_data + tile_row_idx * cm->tiles.cols + tile_col_idx; |
3077 | 283k | dec_row_mt_sync = &tile_data->dec_row_mt_sync; |
3078 | | |
3079 | 283k | num_threads_working = dec_row_mt_sync->num_threads_working; |
3080 | 283k | num_mis_waiting_for_decode = (dec_row_mt_sync->mi_rows_parse_done - |
3081 | 283k | dec_row_mt_sync->mi_rows_decode_started) * |
3082 | 283k | dec_row_mt_sync->mi_cols; |
3083 | 283k | num_mis_to_decode = |
3084 | 283k | (dec_row_mt_sync->mi_rows - dec_row_mt_sync->mi_rows_decode_started) * |
3085 | 283k | dec_row_mt_sync->mi_cols; |
3086 | | |
3087 | 283k | assert(num_mis_to_decode >= num_mis_waiting_for_decode); |
3088 | | |
3089 | | // Pick the tile which has minimum number of threads working on it. |
3090 | 283k | if (num_mis_waiting_for_decode > 0) { |
3091 | 239k | if (num_threads_working < min_threads_working) { |
3092 | 233k | min_threads_working = num_threads_working; |
3093 | 233k | max_mis_to_decode = 0; |
3094 | 233k | } |
3095 | 239k | if (num_threads_working == min_threads_working && |
3096 | 239k | num_mis_to_decode > max_mis_to_decode && |
3097 | 239k | num_threads_working < |
3098 | 235k | get_max_row_mt_workers_per_tile(cm, &tile_data->tile_info)) { |
3099 | 232k | max_mis_to_decode = num_mis_to_decode; |
3100 | 232k | tile_row = tile_row_idx; |
3101 | 232k | tile_col = tile_col_idx; |
3102 | 232k | } |
3103 | 239k | } |
3104 | 283k | } |
3105 | 240k | } |
3106 | | // No job found to process |
3107 | 233k | if (tile_row == -1 || tile_col == -1) return 0; |
3108 | | |
3109 | 232k | tile_data = pbi->tile_data + tile_row * cm->tiles.cols + tile_col; |
3110 | 232k | dec_row_mt_sync = &tile_data->dec_row_mt_sync; |
3111 | | |
3112 | 232k | next_job_info->tile_row = tile_row; |
3113 | 232k | next_job_info->tile_col = tile_col; |
3114 | 232k | next_job_info->mi_row = dec_row_mt_sync->mi_rows_decode_started + |
3115 | 232k | tile_data->tile_info.mi_row_start; |
3116 | | |
3117 | 232k | dec_row_mt_sync->num_threads_working++; |
3118 | 232k | dec_row_mt_sync->mi_rows_decode_started += sb_mi_size; |
3119 | 232k | frame_row_mt_info->mi_rows_decode_started += sb_mi_size; |
3120 | 232k | assert(frame_row_mt_info->mi_rows_parse_done >= |
3121 | 232k | frame_row_mt_info->mi_rows_decode_started); |
3122 | 232k | #if CONFIG_MULTITHREAD |
3123 | 232k | if (frame_row_mt_info->mi_rows_decode_started == |
3124 | 232k | frame_row_mt_info->mi_rows_to_decode) { |
3125 | 115k | pthread_cond_broadcast(pbi->row_mt_cond_); |
3126 | 115k | } |
3127 | 232k | #endif |
3128 | | |
3129 | 232k | return 1; |
3130 | 232k | } |
3131 | | |
3132 | | static INLINE void signal_parse_sb_row_done(AV1Decoder *const pbi, |
3133 | | TileDataDec *const tile_data, |
3134 | 242k | const int sb_mi_size) { |
3135 | 242k | AV1DecRowMTInfo *frame_row_mt_info = &pbi->frame_row_mt_info; |
3136 | 242k | #if CONFIG_MULTITHREAD |
3137 | 242k | pthread_mutex_lock(pbi->row_mt_mutex_); |
3138 | 242k | #endif |
3139 | 242k | assert(frame_row_mt_info->mi_rows_parse_done >= |
3140 | 242k | frame_row_mt_info->mi_rows_decode_started); |
3141 | 242k | tile_data->dec_row_mt_sync.mi_rows_parse_done += sb_mi_size; |
3142 | 242k | frame_row_mt_info->mi_rows_parse_done += sb_mi_size; |
3143 | 242k | #if CONFIG_MULTITHREAD |
3144 | | // A new decode job is available. Wake up one worker thread to handle the |
3145 | | // new decode job. |
3146 | | // NOTE: This assumes we bump mi_rows_parse_done and mi_rows_decode_started |
3147 | | // by the same increment (sb_mi_size). |
3148 | 242k | pthread_cond_signal(pbi->row_mt_cond_); |
3149 | 242k | pthread_mutex_unlock(pbi->row_mt_mutex_); |
3150 | 242k | #endif |
3151 | 242k | } |
3152 | | |
3153 | | // This function is very similar to decode_tile(). It would be good to figure |
3154 | | // out how to share code. |
3155 | | static AOM_INLINE void parse_tile_row_mt(AV1Decoder *pbi, ThreadData *const td, |
3156 | 143k | TileDataDec *const tile_data) { |
3157 | 143k | AV1_COMMON *const cm = &pbi->common; |
3158 | 143k | const int sb_mi_size = mi_size_wide[cm->seq_params->sb_size]; |
3159 | 143k | const int num_planes = av1_num_planes(cm); |
3160 | 143k | const TileInfo *const tile_info = &tile_data->tile_info; |
3161 | 143k | int tile_row = tile_info->tile_row; |
3162 | 143k | DecoderCodingBlock *const dcb = &td->dcb; |
3163 | 143k | MACROBLOCKD *const xd = &dcb->xd; |
3164 | | |
3165 | 143k | av1_zero_above_context(cm, xd, tile_info->mi_col_start, tile_info->mi_col_end, |
3166 | 143k | tile_row); |
3167 | 143k | av1_reset_loop_filter_delta(xd, num_planes); |
3168 | 143k | av1_reset_loop_restoration(xd, num_planes); |
3169 | | |
3170 | 386k | for (int mi_row = tile_info->mi_row_start; mi_row < tile_info->mi_row_end; |
3171 | 258k | mi_row += cm->seq_params->mib_size) { |
3172 | 258k | av1_zero_left_context(xd); |
3173 | | |
3174 | 1.09M | for (int mi_col = tile_info->mi_col_start; mi_col < tile_info->mi_col_end; |
3175 | 856k | mi_col += cm->seq_params->mib_size) { |
3176 | 856k | set_cb_buffer(pbi, dcb, pbi->cb_buffer_base, num_planes, mi_row, mi_col); |
3177 | | |
3178 | | // Bit-stream parsing of the superblock |
3179 | 856k | decode_partition(pbi, td, mi_row, mi_col, td->bit_reader, |
3180 | 856k | cm->seq_params->sb_size, 0x1); |
3181 | | |
3182 | 856k | if (aom_reader_has_overflowed(td->bit_reader)) { |
3183 | 15.4k | aom_merge_corrupted_flag(&dcb->corrupted, 1); |
3184 | 15.4k | return; |
3185 | 15.4k | } |
3186 | 856k | } |
3187 | 243k | signal_parse_sb_row_done(pbi, tile_data, sb_mi_size); |
3188 | 243k | } |
3189 | | |
3190 | 128k | int corrupted = |
3191 | 128k | (check_trailing_bits_after_symbol_coder(td->bit_reader)) ? 1 : 0; |
3192 | 128k | aom_merge_corrupted_flag(&dcb->corrupted, corrupted); |
3193 | 128k | } |
3194 | | |
3195 | 231k | static int row_mt_worker_hook(void *arg1, void *arg2) { |
3196 | 231k | DecWorkerData *const thread_data = (DecWorkerData *)arg1; |
3197 | 231k | AV1Decoder *const pbi = (AV1Decoder *)arg2; |
3198 | 231k | ThreadData *const td = thread_data->td; |
3199 | 231k | uint8_t allow_update_cdf; |
3200 | 231k | AV1DecRowMTInfo *frame_row_mt_info = &pbi->frame_row_mt_info; |
3201 | 231k | td->dcb.corrupted = 0; |
3202 | | |
3203 | | // The jmp_buf is valid only for the duration of the function that calls |
3204 | | // setjmp(). Therefore, this function must reset the 'setjmp' field to 0 |
3205 | | // before it returns. |
3206 | 231k | if (setjmp(thread_data->error_info.jmp)) { |
3207 | 934 | thread_data->error_info.setjmp = 0; |
3208 | 934 | thread_data->td->dcb.corrupted = 1; |
3209 | 934 | #if CONFIG_MULTITHREAD |
3210 | 934 | pthread_mutex_lock(pbi->row_mt_mutex_); |
3211 | 934 | #endif |
3212 | 934 | frame_row_mt_info->row_mt_exit = 1; |
3213 | 934 | #if CONFIG_MULTITHREAD |
3214 | 934 | pthread_cond_broadcast(pbi->row_mt_cond_); |
3215 | 934 | pthread_mutex_unlock(pbi->row_mt_mutex_); |
3216 | 934 | #endif |
3217 | | // If any SB row (erroneous row) processed by a thread encounters an |
3218 | | // internal error, there is a need to indicate other threads that decoding |
3219 | | // of the erroneous row is complete. This ensures that other threads which |
3220 | | // wait upon the completion of SB's present in erroneous row are not waiting |
3221 | | // indefinitely. |
3222 | 934 | signal_decoding_done_for_erroneous_row(pbi, &thread_data->td->dcb.xd); |
3223 | 934 | return 0; |
3224 | 934 | } |
3225 | 230k | thread_data->error_info.setjmp = 1; |
3226 | | |
3227 | 230k | AV1_COMMON *cm = &pbi->common; |
3228 | 230k | allow_update_cdf = cm->tiles.large_scale ? 0 : 1; |
3229 | 230k | allow_update_cdf = allow_update_cdf && !cm->features.disable_cdf_update; |
3230 | | |
3231 | 230k | set_decode_func_pointers(td, 0x1); |
3232 | | |
3233 | 230k | assert(cm->tiles.cols > 0); |
3234 | 374k | while (!td->dcb.corrupted) { |
3235 | 350k | TileJobsDec *cur_job_info = get_dec_job_info(&pbi->tile_mt_info); |
3236 | | |
3237 | 350k | if (cur_job_info != NULL) { |
3238 | 143k | const TileBufferDec *const tile_buffer = cur_job_info->tile_buffer; |
3239 | 143k | TileDataDec *const tile_data = cur_job_info->tile_data; |
3240 | 143k | tile_worker_hook_init(pbi, thread_data, tile_buffer, tile_data, |
3241 | 143k | allow_update_cdf); |
3242 | 143k | #if CONFIG_MULTITHREAD |
3243 | 143k | pthread_mutex_lock(pbi->row_mt_mutex_); |
3244 | 143k | #endif |
3245 | 143k | tile_data->dec_row_mt_sync.num_threads_working++; |
3246 | 143k | #if CONFIG_MULTITHREAD |
3247 | 143k | pthread_mutex_unlock(pbi->row_mt_mutex_); |
3248 | 143k | #endif |
3249 | | // decode tile |
3250 | 143k | parse_tile_row_mt(pbi, td, tile_data); |
3251 | 143k | #if CONFIG_MULTITHREAD |
3252 | 143k | pthread_mutex_lock(pbi->row_mt_mutex_); |
3253 | 143k | #endif |
3254 | 143k | tile_data->dec_row_mt_sync.num_threads_working--; |
3255 | 143k | #if CONFIG_MULTITHREAD |
3256 | 143k | pthread_mutex_unlock(pbi->row_mt_mutex_); |
3257 | 143k | #endif |
3258 | 206k | } else { |
3259 | 206k | break; |
3260 | 206k | } |
3261 | 350k | } |
3262 | | |
3263 | 230k | if (td->dcb.corrupted) { |
3264 | 22.9k | thread_data->error_info.setjmp = 0; |
3265 | 22.9k | #if CONFIG_MULTITHREAD |
3266 | 22.9k | pthread_mutex_lock(pbi->row_mt_mutex_); |
3267 | 22.9k | #endif |
3268 | 22.9k | frame_row_mt_info->row_mt_exit = 1; |
3269 | 22.9k | #if CONFIG_MULTITHREAD |
3270 | 22.9k | pthread_cond_broadcast(pbi->row_mt_cond_); |
3271 | 22.9k | pthread_mutex_unlock(pbi->row_mt_mutex_); |
3272 | 22.9k | #endif |
3273 | 22.9k | return 0; |
3274 | 22.9k | } |
3275 | | |
3276 | 207k | set_decode_func_pointers(td, 0x2); |
3277 | | |
3278 | 440k | while (1) { |
3279 | 439k | AV1DecRowMTJobInfo next_job_info; |
3280 | 439k | int end_of_frame = 0; |
3281 | | |
3282 | 439k | #if CONFIG_MULTITHREAD |
3283 | 439k | pthread_mutex_lock(pbi->row_mt_mutex_); |
3284 | 439k | #endif |
3285 | 523k | while (!get_next_job_info(pbi, &next_job_info, &end_of_frame)) { |
3286 | 83.4k | #if CONFIG_MULTITHREAD |
3287 | 83.4k | pthread_cond_wait(pbi->row_mt_cond_, pbi->row_mt_mutex_); |
3288 | 83.4k | #endif |
3289 | 83.4k | } |
3290 | 439k | #if CONFIG_MULTITHREAD |
3291 | 439k | pthread_mutex_unlock(pbi->row_mt_mutex_); |
3292 | 439k | #endif |
3293 | | |
3294 | 439k | if (end_of_frame) break; |
3295 | | |
3296 | 232k | int tile_row = next_job_info.tile_row; |
3297 | 232k | int tile_col = next_job_info.tile_col; |
3298 | 232k | int mi_row = next_job_info.mi_row; |
3299 | | |
3300 | 232k | TileDataDec *tile_data = |
3301 | 232k | pbi->tile_data + tile_row * cm->tiles.cols + tile_col; |
3302 | 232k | AV1DecRowMTSync *dec_row_mt_sync = &tile_data->dec_row_mt_sync; |
3303 | | |
3304 | 232k | av1_tile_init(&td->dcb.xd.tile, cm, tile_row, tile_col); |
3305 | 232k | av1_init_macroblockd(cm, &td->dcb.xd); |
3306 | 232k | td->dcb.xd.error_info = &thread_data->error_info; |
3307 | | |
3308 | 232k | decode_tile_sb_row(pbi, td, &tile_data->tile_info, mi_row); |
3309 | | |
3310 | 232k | #if CONFIG_MULTITHREAD |
3311 | 232k | pthread_mutex_lock(pbi->row_mt_mutex_); |
3312 | 232k | #endif |
3313 | 232k | dec_row_mt_sync->num_threads_working--; |
3314 | 232k | #if CONFIG_MULTITHREAD |
3315 | 232k | pthread_mutex_unlock(pbi->row_mt_mutex_); |
3316 | 232k | #endif |
3317 | 232k | } |
3318 | 207k | thread_data->error_info.setjmp = 0; |
3319 | 207k | return !td->dcb.corrupted; |
3320 | 230k | } |
3321 | | |
3322 | | // sorts in descending order |
3323 | 70.5k | static int compare_tile_buffers(const void *a, const void *b) { |
3324 | 70.5k | const TileJobsDec *const buf1 = (const TileJobsDec *)a; |
3325 | 70.5k | const TileJobsDec *const buf2 = (const TileJobsDec *)b; |
3326 | 70.5k | return (((int)buf2->tile_buffer->size) - ((int)buf1->tile_buffer->size)); |
3327 | 70.5k | } |
3328 | | |
3329 | | static AOM_INLINE void enqueue_tile_jobs(AV1Decoder *pbi, AV1_COMMON *cm, |
3330 | | int tile_rows_start, int tile_rows_end, |
3331 | | int tile_cols_start, int tile_cols_end, |
3332 | 123k | int start_tile, int end_tile) { |
3333 | 123k | AV1DecTileMT *tile_mt_info = &pbi->tile_mt_info; |
3334 | 123k | TileJobsDec *tile_job_queue = tile_mt_info->job_queue; |
3335 | 123k | tile_mt_info->jobs_enqueued = 0; |
3336 | 123k | tile_mt_info->jobs_dequeued = 0; |
3337 | | |
3338 | 263k | for (int row = tile_rows_start; row < tile_rows_end; row++) { |
3339 | 286k | for (int col = tile_cols_start; col < tile_cols_end; col++) { |
3340 | 146k | if (row * cm->tiles.cols + col < start_tile || |
3341 | 146k | row * cm->tiles.cols + col > end_tile) |
3342 | 121 | continue; |
3343 | 146k | tile_job_queue->tile_buffer = &pbi->tile_buffers[row][col]; |
3344 | 146k | tile_job_queue->tile_data = pbi->tile_data + row * cm->tiles.cols + col; |
3345 | 146k | tile_job_queue++; |
3346 | 146k | tile_mt_info->jobs_enqueued++; |
3347 | 146k | } |
3348 | 139k | } |
3349 | 123k | } |
3350 | | |
3351 | | static AOM_INLINE void alloc_dec_jobs(AV1DecTileMT *tile_mt_info, |
3352 | | AV1_COMMON *cm, int tile_rows, |
3353 | 4.89k | int tile_cols) { |
3354 | 4.89k | tile_mt_info->alloc_tile_rows = tile_rows; |
3355 | 4.89k | tile_mt_info->alloc_tile_cols = tile_cols; |
3356 | 4.89k | int num_tiles = tile_rows * tile_cols; |
3357 | 4.89k | #if CONFIG_MULTITHREAD |
3358 | 4.89k | { |
3359 | 4.89k | CHECK_MEM_ERROR(cm, tile_mt_info->job_mutex, |
3360 | 4.89k | aom_malloc(sizeof(*tile_mt_info->job_mutex) * num_tiles)); |
3361 | | |
3362 | 29.3k | for (int i = 0; i < num_tiles; i++) { |
3363 | 24.4k | pthread_mutex_init(&tile_mt_info->job_mutex[i], NULL); |
3364 | 24.4k | } |
3365 | 4.89k | } |
3366 | 4.89k | #endif |
3367 | 4.89k | CHECK_MEM_ERROR(cm, tile_mt_info->job_queue, |
3368 | 4.89k | aom_malloc(sizeof(*tile_mt_info->job_queue) * num_tiles)); |
3369 | 4.89k | } |
3370 | | |
3371 | 300k | void av1_free_mc_tmp_buf(ThreadData *thread_data) { |
3372 | 300k | int ref; |
3373 | 902k | for (ref = 0; ref < 2; ref++) { |
3374 | 601k | if (thread_data->mc_buf_use_highbd) |
3375 | 134k | aom_free(CONVERT_TO_SHORTPTR(thread_data->mc_buf[ref])); |
3376 | 467k | else |
3377 | 467k | aom_free(thread_data->mc_buf[ref]); |
3378 | 601k | thread_data->mc_buf[ref] = NULL; |
3379 | 601k | } |
3380 | 300k | thread_data->mc_buf_size = 0; |
3381 | 300k | thread_data->mc_buf_use_highbd = 0; |
3382 | | |
3383 | 300k | aom_free(thread_data->tmp_conv_dst); |
3384 | 300k | thread_data->tmp_conv_dst = NULL; |
3385 | 300k | aom_free(thread_data->seg_mask); |
3386 | 300k | thread_data->seg_mask = NULL; |
3387 | 902k | for (int i = 0; i < 2; ++i) { |
3388 | 601k | aom_free(thread_data->tmp_obmc_bufs[i]); |
3389 | 601k | thread_data->tmp_obmc_bufs[i] = NULL; |
3390 | 601k | } |
3391 | 300k | } |
3392 | | |
3393 | | static AOM_INLINE void allocate_mc_tmp_buf(AV1_COMMON *const cm, |
3394 | | ThreadData *thread_data, |
3395 | 155k | int buf_size, int use_highbd) { |
3396 | 465k | for (int ref = 0; ref < 2; ref++) { |
3397 | | // The mc_buf/hbd_mc_buf must be zeroed to fix a intermittent valgrind error |
3398 | | // 'Conditional jump or move depends on uninitialised value' from the loop |
3399 | | // filter. Uninitialized reads in convolve function (e.g. horiz_4tap path in |
3400 | | // av1_convolve_2d_sr_avx2()) from mc_buf/hbd_mc_buf are seen to be the |
3401 | | // potential reason for this issue. |
3402 | 310k | if (use_highbd) { |
3403 | 134k | uint16_t *hbd_mc_buf; |
3404 | 134k | CHECK_MEM_ERROR(cm, hbd_mc_buf, (uint16_t *)aom_memalign(16, buf_size)); |
3405 | 134k | memset(hbd_mc_buf, 0, buf_size); |
3406 | 134k | thread_data->mc_buf[ref] = CONVERT_TO_BYTEPTR(hbd_mc_buf); |
3407 | 176k | } else { |
3408 | 176k | CHECK_MEM_ERROR(cm, thread_data->mc_buf[ref], |
3409 | 176k | (uint8_t *)aom_memalign(16, buf_size)); |
3410 | 176k | memset(thread_data->mc_buf[ref], 0, buf_size); |
3411 | 176k | } |
3412 | 310k | } |
3413 | 155k | thread_data->mc_buf_size = buf_size; |
3414 | 155k | thread_data->mc_buf_use_highbd = use_highbd; |
3415 | | |
3416 | 155k | CHECK_MEM_ERROR(cm, thread_data->tmp_conv_dst, |
3417 | 155k | aom_memalign(32, MAX_SB_SIZE * MAX_SB_SIZE * |
3418 | 155k | sizeof(*thread_data->tmp_conv_dst))); |
3419 | 155k | CHECK_MEM_ERROR(cm, thread_data->seg_mask, |
3420 | 155k | (uint8_t *)aom_memalign( |
3421 | 155k | 16, 2 * MAX_SB_SQUARE * sizeof(*thread_data->seg_mask))); |
3422 | | |
3423 | 465k | for (int i = 0; i < 2; ++i) { |
3424 | 310k | CHECK_MEM_ERROR( |
3425 | 310k | cm, thread_data->tmp_obmc_bufs[i], |
3426 | 310k | aom_memalign(16, 2 * MAX_MB_PLANE * MAX_SB_SQUARE * |
3427 | 310k | sizeof(*thread_data->tmp_obmc_bufs[i]))); |
3428 | 310k | } |
3429 | 155k | } |
3430 | | |
3431 | | static AOM_INLINE void reset_dec_workers(AV1Decoder *pbi, |
3432 | | AVxWorkerHook worker_hook, |
3433 | 123k | int num_workers) { |
3434 | 123k | const AVxWorkerInterface *const winterface = aom_get_worker_interface(); |
3435 | | |
3436 | | // Reset tile decoding hook |
3437 | 354k | for (int worker_idx = 0; worker_idx < num_workers; ++worker_idx) { |
3438 | 231k | AVxWorker *const worker = &pbi->tile_workers[worker_idx]; |
3439 | 231k | DecWorkerData *const thread_data = pbi->thread_data + worker_idx; |
3440 | 231k | thread_data->td->dcb = pbi->dcb; |
3441 | 231k | thread_data->td->dcb.corrupted = 0; |
3442 | 231k | thread_data->td->dcb.mc_buf[0] = thread_data->td->mc_buf[0]; |
3443 | 231k | thread_data->td->dcb.mc_buf[1] = thread_data->td->mc_buf[1]; |
3444 | 231k | thread_data->td->dcb.xd.tmp_conv_dst = thread_data->td->tmp_conv_dst; |
3445 | 231k | if (worker_idx) |
3446 | 107k | thread_data->td->dcb.xd.seg_mask = thread_data->td->seg_mask; |
3447 | 693k | for (int j = 0; j < 2; ++j) { |
3448 | 462k | thread_data->td->dcb.xd.tmp_obmc_bufs[j] = |
3449 | 462k | thread_data->td->tmp_obmc_bufs[j]; |
3450 | 462k | } |
3451 | 231k | winterface->sync(worker); |
3452 | | |
3453 | 231k | worker->hook = worker_hook; |
3454 | 231k | worker->data1 = thread_data; |
3455 | 231k | worker->data2 = pbi; |
3456 | 231k | } |
3457 | | #if CONFIG_ACCOUNTING |
3458 | | if (pbi->acct_enabled) { |
3459 | | aom_accounting_reset(&pbi->accounting); |
3460 | | } |
3461 | | #endif |
3462 | 123k | } |
3463 | | |
3464 | | static AOM_INLINE void launch_dec_workers(AV1Decoder *pbi, |
3465 | | const uint8_t *data_end, |
3466 | 123k | int num_workers) { |
3467 | 123k | const AVxWorkerInterface *const winterface = aom_get_worker_interface(); |
3468 | | |
3469 | 354k | for (int worker_idx = num_workers - 1; worker_idx >= 0; --worker_idx) { |
3470 | 231k | AVxWorker *const worker = &pbi->tile_workers[worker_idx]; |
3471 | 231k | DecWorkerData *const thread_data = (DecWorkerData *)worker->data1; |
3472 | | |
3473 | 231k | thread_data->data_end = data_end; |
3474 | | |
3475 | 231k | worker->had_error = 0; |
3476 | 231k | if (worker_idx == 0) { |
3477 | 123k | winterface->execute(worker); |
3478 | 123k | } else { |
3479 | 107k | winterface->launch(worker); |
3480 | 107k | } |
3481 | 231k | } |
3482 | 123k | } |
3483 | | |
3484 | 123k | static AOM_INLINE void sync_dec_workers(AV1Decoder *pbi, int num_workers) { |
3485 | 123k | const AVxWorkerInterface *const winterface = aom_get_worker_interface(); |
3486 | 123k | int corrupted = 0; |
3487 | | |
3488 | 354k | for (int worker_idx = num_workers; worker_idx > 0; --worker_idx) { |
3489 | 231k | AVxWorker *const worker = &pbi->tile_workers[worker_idx - 1]; |
3490 | 231k | aom_merge_corrupted_flag(&corrupted, !winterface->sync(worker)); |
3491 | 231k | } |
3492 | | |
3493 | 123k | pbi->dcb.corrupted = corrupted; |
3494 | 123k | } |
3495 | | |
3496 | 124k | static AOM_INLINE void decode_mt_init(AV1Decoder *pbi) { |
3497 | 124k | AV1_COMMON *const cm = &pbi->common; |
3498 | 124k | const AVxWorkerInterface *const winterface = aom_get_worker_interface(); |
3499 | 124k | int worker_idx; |
3500 | | |
3501 | | // Create workers and thread_data |
3502 | 124k | if (pbi->num_workers == 0) { |
3503 | 3.82k | const int num_threads = pbi->max_threads; |
3504 | 3.82k | CHECK_MEM_ERROR(cm, pbi->tile_workers, |
3505 | 3.82k | aom_malloc(num_threads * sizeof(*pbi->tile_workers))); |
3506 | 3.82k | CHECK_MEM_ERROR(cm, pbi->thread_data, |
3507 | 3.82k | aom_calloc(num_threads, sizeof(*pbi->thread_data))); |
3508 | | |
3509 | 136k | for (worker_idx = 0; worker_idx < num_threads; ++worker_idx) { |
3510 | 132k | AVxWorker *const worker = &pbi->tile_workers[worker_idx]; |
3511 | 132k | DecWorkerData *const thread_data = pbi->thread_data + worker_idx; |
3512 | | |
3513 | 132k | winterface->init(worker); |
3514 | 132k | worker->thread_name = "aom tile worker"; |
3515 | 132k | if (worker_idx != 0 && !winterface->reset(worker)) { |
3516 | 0 | aom_internal_error(&pbi->error, AOM_CODEC_ERROR, |
3517 | 0 | "Tile decoder thread creation failed"); |
3518 | 0 | } |
3519 | 132k | ++pbi->num_workers; |
3520 | | |
3521 | 132k | if (worker_idx != 0) { |
3522 | | // Allocate thread data. |
3523 | 128k | CHECK_MEM_ERROR(cm, thread_data->td, |
3524 | 128k | aom_memalign(32, sizeof(*thread_data->td))); |
3525 | 128k | av1_zero(*thread_data->td); |
3526 | 128k | } else { |
3527 | | // Main thread acts as a worker and uses the thread data in pbi |
3528 | 3.82k | thread_data->td = &pbi->td; |
3529 | 3.82k | } |
3530 | 132k | thread_data->error_info.error_code = AOM_CODEC_OK; |
3531 | 132k | thread_data->error_info.setjmp = 0; |
3532 | 132k | } |
3533 | 3.82k | } |
3534 | 124k | const int use_highbd = cm->seq_params->use_highbitdepth; |
3535 | 124k | const int buf_size = MC_TEMP_BUF_PELS << use_highbd; |
3536 | 5.62M | for (worker_idx = 1; worker_idx < pbi->max_threads; ++worker_idx) { |
3537 | 5.50M | DecWorkerData *const thread_data = pbi->thread_data + worker_idx; |
3538 | 5.50M | if (thread_data->td->mc_buf_size != buf_size) { |
3539 | 139k | av1_free_mc_tmp_buf(thread_data->td); |
3540 | 139k | allocate_mc_tmp_buf(cm, thread_data->td, buf_size, use_highbd); |
3541 | 139k | } |
3542 | 5.50M | } |
3543 | 124k | } |
3544 | | |
3545 | | static AOM_INLINE void tile_mt_queue(AV1Decoder *pbi, int tile_cols, |
3546 | | int tile_rows, int tile_rows_start, |
3547 | | int tile_rows_end, int tile_cols_start, |
3548 | | int tile_cols_end, int start_tile, |
3549 | 123k | int end_tile) { |
3550 | 123k | AV1_COMMON *const cm = &pbi->common; |
3551 | 123k | if (pbi->tile_mt_info.alloc_tile_cols != tile_cols || |
3552 | 123k | pbi->tile_mt_info.alloc_tile_rows != tile_rows) { |
3553 | 4.89k | av1_dealloc_dec_jobs(&pbi->tile_mt_info); |
3554 | 4.89k | alloc_dec_jobs(&pbi->tile_mt_info, cm, tile_rows, tile_cols); |
3555 | 4.89k | } |
3556 | 123k | enqueue_tile_jobs(pbi, cm, tile_rows_start, tile_rows_end, tile_cols_start, |
3557 | 123k | tile_cols_end, start_tile, end_tile); |
3558 | 123k | qsort(pbi->tile_mt_info.job_queue, pbi->tile_mt_info.jobs_enqueued, |
3559 | 123k | sizeof(pbi->tile_mt_info.job_queue[0]), compare_tile_buffers); |
3560 | 123k | } |
3561 | | |
3562 | | static const uint8_t *decode_tiles_mt(AV1Decoder *pbi, const uint8_t *data, |
3563 | | const uint8_t *data_end, int start_tile, |
3564 | 0 | int end_tile) { |
3565 | 0 | AV1_COMMON *const cm = &pbi->common; |
3566 | 0 | CommonTileParams *const tiles = &cm->tiles; |
3567 | 0 | const int tile_cols = tiles->cols; |
3568 | 0 | const int tile_rows = tiles->rows; |
3569 | 0 | const int n_tiles = tile_cols * tile_rows; |
3570 | 0 | TileBufferDec(*const tile_buffers)[MAX_TILE_COLS] = pbi->tile_buffers; |
3571 | 0 | const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows); |
3572 | 0 | const int single_row = pbi->dec_tile_row >= 0; |
3573 | 0 | const int dec_tile_col = AOMMIN(pbi->dec_tile_col, tile_cols); |
3574 | 0 | const int single_col = pbi->dec_tile_col >= 0; |
3575 | 0 | int tile_rows_start; |
3576 | 0 | int tile_rows_end; |
3577 | 0 | int tile_cols_start; |
3578 | 0 | int tile_cols_end; |
3579 | 0 | int tile_count_tg; |
3580 | 0 | int num_workers; |
3581 | 0 | const uint8_t *raw_data_end = NULL; |
3582 | |
|
3583 | 0 | if (tiles->large_scale) { |
3584 | 0 | tile_rows_start = single_row ? dec_tile_row : 0; |
3585 | 0 | tile_rows_end = single_row ? dec_tile_row + 1 : tile_rows; |
3586 | 0 | tile_cols_start = single_col ? dec_tile_col : 0; |
3587 | 0 | tile_cols_end = single_col ? tile_cols_start + 1 : tile_cols; |
3588 | 0 | } else { |
3589 | 0 | tile_rows_start = 0; |
3590 | 0 | tile_rows_end = tile_rows; |
3591 | 0 | tile_cols_start = 0; |
3592 | 0 | tile_cols_end = tile_cols; |
3593 | 0 | } |
3594 | 0 | tile_count_tg = end_tile - start_tile + 1; |
3595 | 0 | num_workers = AOMMIN(pbi->max_threads, tile_count_tg); |
3596 | | |
3597 | | // No tiles to decode. |
3598 | 0 | if (tile_rows_end <= tile_rows_start || tile_cols_end <= tile_cols_start || |
3599 | | // First tile is larger than end_tile. |
3600 | 0 | tile_rows_start * tile_cols + tile_cols_start > end_tile || |
3601 | | // Last tile is smaller than start_tile. |
3602 | 0 | (tile_rows_end - 1) * tile_cols + tile_cols_end - 1 < start_tile) |
3603 | 0 | return data; |
3604 | | |
3605 | 0 | assert(tile_rows <= MAX_TILE_ROWS); |
3606 | 0 | assert(tile_cols <= MAX_TILE_COLS); |
3607 | 0 | assert(tile_count_tg > 0); |
3608 | 0 | assert(num_workers > 0); |
3609 | 0 | assert(start_tile <= end_tile); |
3610 | 0 | assert(start_tile >= 0 && end_tile < n_tiles); |
3611 | | |
3612 | 0 | decode_mt_init(pbi); |
3613 | | |
3614 | | // get tile size in tile group |
3615 | 0 | #if EXT_TILE_DEBUG |
3616 | 0 | if (tiles->large_scale) assert(pbi->ext_tile_debug == 1); |
3617 | 0 | if (tiles->large_scale) |
3618 | 0 | raw_data_end = get_ls_tile_buffers(pbi, data, data_end, tile_buffers); |
3619 | 0 | else |
3620 | 0 | #endif // EXT_TILE_DEBUG |
3621 | 0 | get_tile_buffers(pbi, data, data_end, tile_buffers, start_tile, end_tile); |
3622 | |
|
3623 | 0 | if (pbi->tile_data == NULL || n_tiles != pbi->allocated_tiles) { |
3624 | 0 | decoder_alloc_tile_data(pbi, n_tiles); |
3625 | 0 | } |
3626 | 0 | if (pbi->dcb.xd.seg_mask == NULL) |
3627 | 0 | CHECK_MEM_ERROR(cm, pbi->dcb.xd.seg_mask, |
3628 | 0 | (uint8_t *)aom_memalign( |
3629 | 0 | 16, 2 * MAX_SB_SQUARE * sizeof(*pbi->dcb.xd.seg_mask))); |
3630 | |
|
3631 | 0 | for (int row = 0; row < tile_rows; row++) { |
3632 | 0 | for (int col = 0; col < tile_cols; col++) { |
3633 | 0 | TileDataDec *tile_data = pbi->tile_data + row * tiles->cols + col; |
3634 | 0 | av1_tile_init(&tile_data->tile_info, cm, row, col); |
3635 | 0 | } |
3636 | 0 | } |
3637 | |
|
3638 | 0 | tile_mt_queue(pbi, tile_cols, tile_rows, tile_rows_start, tile_rows_end, |
3639 | 0 | tile_cols_start, tile_cols_end, start_tile, end_tile); |
3640 | |
|
3641 | 0 | reset_dec_workers(pbi, tile_worker_hook, num_workers); |
3642 | 0 | launch_dec_workers(pbi, data_end, num_workers); |
3643 | 0 | sync_dec_workers(pbi, num_workers); |
3644 | |
|
3645 | 0 | if (pbi->dcb.corrupted) |
3646 | 0 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
3647 | 0 | "Failed to decode tile data"); |
3648 | |
|
3649 | 0 | if (tiles->large_scale) { |
3650 | 0 | if (n_tiles == 1) { |
3651 | | // Find the end of the single tile buffer |
3652 | 0 | return aom_reader_find_end(&pbi->tile_data->bit_reader); |
3653 | 0 | } |
3654 | | // Return the end of the last tile buffer |
3655 | 0 | return raw_data_end; |
3656 | 0 | } |
3657 | 0 | TileDataDec *const tile_data = pbi->tile_data + end_tile; |
3658 | |
|
3659 | 0 | return aom_reader_find_end(&tile_data->bit_reader); |
3660 | 0 | } |
3661 | | |
3662 | 123k | static AOM_INLINE void dec_alloc_cb_buf(AV1Decoder *pbi) { |
3663 | 123k | AV1_COMMON *const cm = &pbi->common; |
3664 | 123k | int size = ((cm->mi_params.mi_rows >> cm->seq_params->mib_size_log2) + 1) * |
3665 | 123k | ((cm->mi_params.mi_cols >> cm->seq_params->mib_size_log2) + 1); |
3666 | | |
3667 | 123k | if (pbi->cb_buffer_alloc_size < size) { |
3668 | 4.40k | av1_dec_free_cb_buf(pbi); |
3669 | 4.40k | CHECK_MEM_ERROR(cm, pbi->cb_buffer_base, |
3670 | 4.40k | aom_memalign(32, sizeof(*pbi->cb_buffer_base) * size)); |
3671 | 4.40k | memset(pbi->cb_buffer_base, 0, sizeof(*pbi->cb_buffer_base) * size); |
3672 | 4.40k | pbi->cb_buffer_alloc_size = size; |
3673 | 4.40k | } |
3674 | 123k | } |
3675 | | |
3676 | | static AOM_INLINE void row_mt_frame_init(AV1Decoder *pbi, int tile_rows_start, |
3677 | | int tile_rows_end, int tile_cols_start, |
3678 | | int tile_cols_end, int start_tile, |
3679 | 123k | int end_tile, int max_sb_rows) { |
3680 | 123k | AV1_COMMON *const cm = &pbi->common; |
3681 | 123k | AV1DecRowMTInfo *frame_row_mt_info = &pbi->frame_row_mt_info; |
3682 | | |
3683 | 123k | frame_row_mt_info->tile_rows_start = tile_rows_start; |
3684 | 123k | frame_row_mt_info->tile_rows_end = tile_rows_end; |
3685 | 123k | frame_row_mt_info->tile_cols_start = tile_cols_start; |
3686 | 123k | frame_row_mt_info->tile_cols_end = tile_cols_end; |
3687 | 123k | frame_row_mt_info->start_tile = start_tile; |
3688 | 123k | frame_row_mt_info->end_tile = end_tile; |
3689 | 123k | frame_row_mt_info->mi_rows_to_decode = 0; |
3690 | 123k | frame_row_mt_info->mi_rows_parse_done = 0; |
3691 | 123k | frame_row_mt_info->mi_rows_decode_started = 0; |
3692 | 123k | frame_row_mt_info->row_mt_exit = 0; |
3693 | | |
3694 | 263k | for (int tile_row = tile_rows_start; tile_row < tile_rows_end; ++tile_row) { |
3695 | 286k | for (int tile_col = tile_cols_start; tile_col < tile_cols_end; ++tile_col) { |
3696 | 146k | if (tile_row * cm->tiles.cols + tile_col < start_tile || |
3697 | 146k | tile_row * cm->tiles.cols + tile_col > end_tile) |
3698 | 121 | continue; |
3699 | | |
3700 | 146k | TileDataDec *const tile_data = |
3701 | 146k | pbi->tile_data + tile_row * cm->tiles.cols + tile_col; |
3702 | 146k | const TileInfo *const tile_info = &tile_data->tile_info; |
3703 | | |
3704 | 146k | tile_data->dec_row_mt_sync.mi_rows_parse_done = 0; |
3705 | 146k | tile_data->dec_row_mt_sync.mi_rows_decode_started = 0; |
3706 | 146k | tile_data->dec_row_mt_sync.num_threads_working = 0; |
3707 | 146k | tile_data->dec_row_mt_sync.mi_rows = |
3708 | 146k | ALIGN_POWER_OF_TWO(tile_info->mi_row_end - tile_info->mi_row_start, |
3709 | 146k | cm->seq_params->mib_size_log2); |
3710 | 146k | tile_data->dec_row_mt_sync.mi_cols = |
3711 | 146k | ALIGN_POWER_OF_TWO(tile_info->mi_col_end - tile_info->mi_col_start, |
3712 | 146k | cm->seq_params->mib_size_log2); |
3713 | 146k | tile_data->dec_row_mt_sync.intrabc_extra_top_right_sb_delay = |
3714 | 146k | av1_get_intrabc_extra_top_right_sb_delay(cm); |
3715 | | |
3716 | 146k | frame_row_mt_info->mi_rows_to_decode += |
3717 | 146k | tile_data->dec_row_mt_sync.mi_rows; |
3718 | | |
3719 | | // Initialize cur_sb_col to -1 for all SB rows. |
3720 | 146k | memset(tile_data->dec_row_mt_sync.cur_sb_col, -1, |
3721 | 146k | sizeof(*tile_data->dec_row_mt_sync.cur_sb_col) * max_sb_rows); |
3722 | 146k | } |
3723 | 139k | } |
3724 | | |
3725 | 123k | #if CONFIG_MULTITHREAD |
3726 | 123k | if (pbi->row_mt_mutex_ == NULL) { |
3727 | 3.63k | CHECK_MEM_ERROR(cm, pbi->row_mt_mutex_, |
3728 | 3.63k | aom_malloc(sizeof(*(pbi->row_mt_mutex_)))); |
3729 | 3.63k | if (pbi->row_mt_mutex_) { |
3730 | 3.63k | pthread_mutex_init(pbi->row_mt_mutex_, NULL); |
3731 | 3.63k | } |
3732 | 3.63k | } |
3733 | | |
3734 | 123k | if (pbi->row_mt_cond_ == NULL) { |
3735 | 3.63k | CHECK_MEM_ERROR(cm, pbi->row_mt_cond_, |
3736 | 3.63k | aom_malloc(sizeof(*(pbi->row_mt_cond_)))); |
3737 | 3.63k | if (pbi->row_mt_cond_) { |
3738 | 3.63k | pthread_cond_init(pbi->row_mt_cond_, NULL); |
3739 | 3.63k | } |
3740 | 3.63k | } |
3741 | 123k | #endif |
3742 | 123k | } |
3743 | | |
3744 | | static const uint8_t *decode_tiles_row_mt(AV1Decoder *pbi, const uint8_t *data, |
3745 | | const uint8_t *data_end, |
3746 | 124k | int start_tile, int end_tile) { |
3747 | 124k | AV1_COMMON *const cm = &pbi->common; |
3748 | 124k | CommonTileParams *const tiles = &cm->tiles; |
3749 | 124k | const int tile_cols = tiles->cols; |
3750 | 124k | const int tile_rows = tiles->rows; |
3751 | 124k | const int n_tiles = tile_cols * tile_rows; |
3752 | 124k | TileBufferDec(*const tile_buffers)[MAX_TILE_COLS] = pbi->tile_buffers; |
3753 | 124k | const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows); |
3754 | 124k | const int single_row = pbi->dec_tile_row >= 0; |
3755 | 124k | const int dec_tile_col = AOMMIN(pbi->dec_tile_col, tile_cols); |
3756 | 124k | const int single_col = pbi->dec_tile_col >= 0; |
3757 | 124k | int tile_rows_start; |
3758 | 124k | int tile_rows_end; |
3759 | 124k | int tile_cols_start; |
3760 | 124k | int tile_cols_end; |
3761 | 124k | int tile_count_tg; |
3762 | 124k | int num_workers = 0; |
3763 | 124k | int max_threads; |
3764 | 124k | const uint8_t *raw_data_end = NULL; |
3765 | 124k | int max_sb_rows = 0; |
3766 | | |
3767 | 124k | if (tiles->large_scale) { |
3768 | 3.61k | tile_rows_start = single_row ? dec_tile_row : 0; |
3769 | 3.61k | tile_rows_end = single_row ? dec_tile_row + 1 : tile_rows; |
3770 | 3.61k | tile_cols_start = single_col ? dec_tile_col : 0; |
3771 | 3.61k | tile_cols_end = single_col ? tile_cols_start + 1 : tile_cols; |
3772 | 120k | } else { |
3773 | 120k | tile_rows_start = 0; |
3774 | 120k | tile_rows_end = tile_rows; |
3775 | 120k | tile_cols_start = 0; |
3776 | 120k | tile_cols_end = tile_cols; |
3777 | 120k | } |
3778 | 124k | tile_count_tg = end_tile - start_tile + 1; |
3779 | 124k | max_threads = pbi->max_threads; |
3780 | | |
3781 | | // No tiles to decode. |
3782 | 124k | if (tile_rows_end <= tile_rows_start || tile_cols_end <= tile_cols_start || |
3783 | | // First tile is larger than end_tile. |
3784 | 124k | tile_rows_start * tile_cols + tile_cols_start > end_tile || |
3785 | | // Last tile is smaller than start_tile. |
3786 | 124k | (tile_rows_end - 1) * tile_cols + tile_cols_end - 1 < start_tile) |
3787 | 0 | return data; |
3788 | | |
3789 | 124k | assert(tile_rows <= MAX_TILE_ROWS); |
3790 | 124k | assert(tile_cols <= MAX_TILE_COLS); |
3791 | 124k | assert(tile_count_tg > 0); |
3792 | 124k | assert(max_threads > 0); |
3793 | 124k | assert(start_tile <= end_tile); |
3794 | 124k | assert(start_tile >= 0 && end_tile < n_tiles); |
3795 | | |
3796 | 124k | (void)tile_count_tg; |
3797 | | |
3798 | 124k | decode_mt_init(pbi); |
3799 | | |
3800 | | // get tile size in tile group |
3801 | 124k | #if EXT_TILE_DEBUG |
3802 | 124k | if (tiles->large_scale) assert(pbi->ext_tile_debug == 1); |
3803 | 124k | if (tiles->large_scale) |
3804 | 3.61k | raw_data_end = get_ls_tile_buffers(pbi, data, data_end, tile_buffers); |
3805 | 120k | else |
3806 | 120k | #endif // EXT_TILE_DEBUG |
3807 | 120k | get_tile_buffers(pbi, data, data_end, tile_buffers, start_tile, end_tile); |
3808 | | |
3809 | 124k | if (pbi->tile_data == NULL || n_tiles != pbi->allocated_tiles) { |
3810 | 4.89k | if (pbi->tile_data != NULL) { |
3811 | 7.00k | for (int i = 0; i < pbi->allocated_tiles; i++) { |
3812 | 5.75k | TileDataDec *const tile_data = pbi->tile_data + i; |
3813 | 5.75k | av1_dec_row_mt_dealloc(&tile_data->dec_row_mt_sync); |
3814 | 5.75k | } |
3815 | 1.25k | } |
3816 | 4.89k | decoder_alloc_tile_data(pbi, n_tiles); |
3817 | 4.89k | } |
3818 | 124k | if (pbi->dcb.xd.seg_mask == NULL) |
3819 | 124k | CHECK_MEM_ERROR(cm, pbi->dcb.xd.seg_mask, |
3820 | 124k | (uint8_t *)aom_memalign( |
3821 | 124k | 16, 2 * MAX_SB_SQUARE * sizeof(*pbi->dcb.xd.seg_mask))); |
3822 | | |
3823 | 263k | for (int row = 0; row < tile_rows; row++) { |
3824 | 286k | for (int col = 0; col < tile_cols; col++) { |
3825 | 146k | TileDataDec *tile_data = pbi->tile_data + row * tiles->cols + col; |
3826 | 146k | av1_tile_init(&tile_data->tile_info, cm, row, col); |
3827 | | |
3828 | 146k | max_sb_rows = AOMMAX(max_sb_rows, |
3829 | 146k | av1_get_sb_rows_in_tile(cm, &tile_data->tile_info)); |
3830 | 146k | num_workers += get_max_row_mt_workers_per_tile(cm, &tile_data->tile_info); |
3831 | 146k | } |
3832 | 139k | } |
3833 | 124k | num_workers = AOMMIN(num_workers, max_threads); |
3834 | | |
3835 | 124k | if (pbi->allocated_row_mt_sync_rows != max_sb_rows) { |
3836 | 42.9k | for (int i = 0; i < n_tiles; ++i) { |
3837 | 31.2k | TileDataDec *const tile_data = pbi->tile_data + i; |
3838 | 31.2k | av1_dec_row_mt_dealloc(&tile_data->dec_row_mt_sync); |
3839 | 31.2k | dec_row_mt_alloc(&tile_data->dec_row_mt_sync, cm, max_sb_rows); |
3840 | 31.2k | } |
3841 | 11.7k | pbi->allocated_row_mt_sync_rows = max_sb_rows; |
3842 | 11.7k | } |
3843 | | |
3844 | 124k | tile_mt_queue(pbi, tile_cols, tile_rows, tile_rows_start, tile_rows_end, |
3845 | 124k | tile_cols_start, tile_cols_end, start_tile, end_tile); |
3846 | | |
3847 | 124k | dec_alloc_cb_buf(pbi); |
3848 | | |
3849 | 124k | row_mt_frame_init(pbi, tile_rows_start, tile_rows_end, tile_cols_start, |
3850 | 124k | tile_cols_end, start_tile, end_tile, max_sb_rows); |
3851 | | |
3852 | 124k | reset_dec_workers(pbi, row_mt_worker_hook, num_workers); |
3853 | 124k | launch_dec_workers(pbi, data_end, num_workers); |
3854 | 124k | sync_dec_workers(pbi, num_workers); |
3855 | | |
3856 | 124k | if (pbi->dcb.corrupted) |
3857 | 7.72k | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
3858 | 7.72k | "Failed to decode tile data"); |
3859 | | |
3860 | 124k | if (tiles->large_scale) { |
3861 | 2.74k | if (n_tiles == 1) { |
3862 | | // Find the end of the single tile buffer |
3863 | 2.74k | return aom_reader_find_end(&pbi->tile_data->bit_reader); |
3864 | 2.74k | } |
3865 | | // Return the end of the last tile buffer |
3866 | 0 | return raw_data_end; |
3867 | 2.74k | } |
3868 | 121k | TileDataDec *const tile_data = pbi->tile_data + end_tile; |
3869 | | |
3870 | 121k | return aom_reader_find_end(&tile_data->bit_reader); |
3871 | 124k | } |
3872 | | |
3873 | 11.6k | static AOM_INLINE void error_handler(void *data) { |
3874 | 11.6k | AV1_COMMON *const cm = (AV1_COMMON *)data; |
3875 | 11.6k | aom_internal_error(cm->error, AOM_CODEC_CORRUPT_FRAME, "Truncated packet"); |
3876 | 11.6k | } |
3877 | | |
3878 | | // Reads the high_bitdepth and twelve_bit fields in color_config() and sets |
3879 | | // seq_params->bit_depth based on the values of those fields and |
3880 | | // seq_params->profile. Reports errors by calling rb->error_handler() or |
3881 | | // aom_internal_error(). |
3882 | | static AOM_INLINE void read_bitdepth( |
3883 | | struct aom_read_bit_buffer *rb, SequenceHeader *seq_params, |
3884 | 72.4k | struct aom_internal_error_info *error_info) { |
3885 | 72.4k | const int high_bitdepth = aom_rb_read_bit(rb); |
3886 | 72.4k | if (seq_params->profile == PROFILE_2 && high_bitdepth) { |
3887 | 21.9k | const int twelve_bit = aom_rb_read_bit(rb); |
3888 | 21.9k | seq_params->bit_depth = twelve_bit ? AOM_BITS_12 : AOM_BITS_10; |
3889 | 50.5k | } else if (seq_params->profile <= PROFILE_2) { |
3890 | 50.3k | seq_params->bit_depth = high_bitdepth ? AOM_BITS_10 : AOM_BITS_8; |
3891 | 50.3k | } else { |
3892 | 185 | aom_internal_error(error_info, AOM_CODEC_UNSUP_BITSTREAM, |
3893 | 185 | "Unsupported profile/bit-depth combination"); |
3894 | 185 | } |
3895 | | #if !CONFIG_AV1_HIGHBITDEPTH |
3896 | | if (seq_params->bit_depth > AOM_BITS_8) { |
3897 | | aom_internal_error(error_info, AOM_CODEC_UNSUP_BITSTREAM, |
3898 | | "Bit-depth %d not supported", seq_params->bit_depth); |
3899 | | } |
3900 | | #endif |
3901 | 72.4k | } |
3902 | | |
3903 | | static void read_film_grain_params(AV1_COMMON *cm, |
3904 | 48.7k | struct aom_read_bit_buffer *rb) { |
3905 | 48.7k | aom_film_grain_t *pars = &cm->film_grain_params; |
3906 | 48.7k | const SequenceHeader *const seq_params = cm->seq_params; |
3907 | | |
3908 | 48.7k | pars->apply_grain = aom_rb_read_bit(rb); |
3909 | 48.7k | if (!pars->apply_grain) { |
3910 | 32.6k | memset(pars, 0, sizeof(*pars)); |
3911 | 32.6k | return; |
3912 | 32.6k | } |
3913 | | |
3914 | 16.1k | pars->random_seed = aom_rb_read_literal(rb, 16); |
3915 | 16.1k | if (cm->current_frame.frame_type == INTER_FRAME) |
3916 | 741 | pars->update_parameters = aom_rb_read_bit(rb); |
3917 | 15.3k | else |
3918 | 15.3k | pars->update_parameters = 1; |
3919 | | |
3920 | 16.1k | pars->bit_depth = seq_params->bit_depth; |
3921 | | |
3922 | 16.1k | if (!pars->update_parameters) { |
3923 | | // inherit parameters from a previous reference frame |
3924 | 500 | int film_grain_params_ref_idx = aom_rb_read_literal(rb, 3); |
3925 | | // Section 6.8.20: It is a requirement of bitstream conformance that |
3926 | | // film_grain_params_ref_idx is equal to ref_frame_idx[ j ] for some value |
3927 | | // of j in the range 0 to REFS_PER_FRAME - 1. |
3928 | 500 | int found = 0; |
3929 | 2.15k | for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) { |
3930 | 2.09k | if (film_grain_params_ref_idx == cm->remapped_ref_idx[i]) { |
3931 | 447 | found = 1; |
3932 | 447 | break; |
3933 | 447 | } |
3934 | 2.09k | } |
3935 | 500 | if (!found) { |
3936 | 52 | aom_internal_error(cm->error, AOM_CODEC_UNSUP_BITSTREAM, |
3937 | 52 | "Invalid film grain reference idx %d. ref_frame_idx = " |
3938 | 52 | "{%d, %d, %d, %d, %d, %d, %d}", |
3939 | 52 | film_grain_params_ref_idx, cm->remapped_ref_idx[0], |
3940 | 52 | cm->remapped_ref_idx[1], cm->remapped_ref_idx[2], |
3941 | 52 | cm->remapped_ref_idx[3], cm->remapped_ref_idx[4], |
3942 | 52 | cm->remapped_ref_idx[5], cm->remapped_ref_idx[6]); |
3943 | 52 | } |
3944 | 500 | RefCntBuffer *const buf = cm->ref_frame_map[film_grain_params_ref_idx]; |
3945 | 500 | if (buf == NULL) { |
3946 | 0 | aom_internal_error(cm->error, AOM_CODEC_UNSUP_BITSTREAM, |
3947 | 0 | "Invalid Film grain reference idx"); |
3948 | 0 | } |
3949 | 500 | if (!buf->film_grain_params_present) { |
3950 | 62 | aom_internal_error(cm->error, AOM_CODEC_UNSUP_BITSTREAM, |
3951 | 62 | "Film grain reference parameters not available"); |
3952 | 62 | } |
3953 | 500 | uint16_t random_seed = pars->random_seed; |
3954 | 500 | *pars = buf->film_grain_params; // inherit paramaters |
3955 | 500 | pars->random_seed = random_seed; // with new random seed |
3956 | 500 | return; |
3957 | 500 | } |
3958 | | |
3959 | | // Scaling functions parameters |
3960 | 15.6k | pars->num_y_points = aom_rb_read_literal(rb, 4); // max 14 |
3961 | 15.6k | if (pars->num_y_points > 14) |
3962 | 51 | aom_internal_error(cm->error, AOM_CODEC_UNSUP_BITSTREAM, |
3963 | 51 | "Number of points for film grain luma scaling function " |
3964 | 51 | "exceeds the maximum value."); |
3965 | 38.0k | for (int i = 0; i < pars->num_y_points; i++) { |
3966 | 22.4k | pars->scaling_points_y[i][0] = aom_rb_read_literal(rb, 8); |
3967 | 22.4k | if (i && pars->scaling_points_y[i - 1][0] >= pars->scaling_points_y[i][0]) |
3968 | 527 | aom_internal_error(cm->error, AOM_CODEC_UNSUP_BITSTREAM, |
3969 | 527 | "First coordinate of the scaling function points " |
3970 | 527 | "shall be increasing."); |
3971 | 22.4k | pars->scaling_points_y[i][1] = aom_rb_read_literal(rb, 8); |
3972 | 22.4k | } |
3973 | | |
3974 | 15.6k | if (!seq_params->monochrome) |
3975 | 14.5k | pars->chroma_scaling_from_luma = aom_rb_read_bit(rb); |
3976 | 1.04k | else |
3977 | 1.04k | pars->chroma_scaling_from_luma = 0; |
3978 | | |
3979 | 15.6k | if (seq_params->monochrome || pars->chroma_scaling_from_luma || |
3980 | 15.6k | ((seq_params->subsampling_x == 1) && (seq_params->subsampling_y == 1) && |
3981 | 8.94k | (pars->num_y_points == 0))) { |
3982 | 8.46k | pars->num_cb_points = 0; |
3983 | 8.46k | pars->num_cr_points = 0; |
3984 | 8.46k | } else { |
3985 | 7.14k | pars->num_cb_points = aom_rb_read_literal(rb, 4); // max 10 |
3986 | 7.14k | if (pars->num_cb_points > 10) |
3987 | 109 | aom_internal_error(cm->error, AOM_CODEC_UNSUP_BITSTREAM, |
3988 | 109 | "Number of points for film grain cb scaling function " |
3989 | 109 | "exceeds the maximum value."); |
3990 | 12.5k | for (int i = 0; i < pars->num_cb_points; i++) { |
3991 | 5.36k | pars->scaling_points_cb[i][0] = aom_rb_read_literal(rb, 8); |
3992 | 5.36k | if (i && |
3993 | 5.36k | pars->scaling_points_cb[i - 1][0] >= pars->scaling_points_cb[i][0]) |
3994 | 172 | aom_internal_error(cm->error, AOM_CODEC_UNSUP_BITSTREAM, |
3995 | 172 | "First coordinate of the scaling function points " |
3996 | 172 | "shall be increasing."); |
3997 | 5.36k | pars->scaling_points_cb[i][1] = aom_rb_read_literal(rb, 8); |
3998 | 5.36k | } |
3999 | | |
4000 | 7.14k | pars->num_cr_points = aom_rb_read_literal(rb, 4); // max 10 |
4001 | 7.14k | if (pars->num_cr_points > 10) |
4002 | 89 | aom_internal_error(cm->error, AOM_CODEC_UNSUP_BITSTREAM, |
4003 | 89 | "Number of points for film grain cr scaling function " |
4004 | 89 | "exceeds the maximum value."); |
4005 | 16.7k | for (int i = 0; i < pars->num_cr_points; i++) { |
4006 | 9.59k | pars->scaling_points_cr[i][0] = aom_rb_read_literal(rb, 8); |
4007 | 9.59k | if (i && |
4008 | 9.59k | pars->scaling_points_cr[i - 1][0] >= pars->scaling_points_cr[i][0]) |
4009 | 135 | aom_internal_error(cm->error, AOM_CODEC_UNSUP_BITSTREAM, |
4010 | 135 | "First coordinate of the scaling function points " |
4011 | 135 | "shall be increasing."); |
4012 | 9.59k | pars->scaling_points_cr[i][1] = aom_rb_read_literal(rb, 8); |
4013 | 9.59k | } |
4014 | | |
4015 | 7.14k | if ((seq_params->subsampling_x == 1) && (seq_params->subsampling_y == 1) && |
4016 | 7.14k | (((pars->num_cb_points == 0) && (pars->num_cr_points != 0)) || |
4017 | 704 | ((pars->num_cb_points != 0) && (pars->num_cr_points == 0)))) |
4018 | 22 | aom_internal_error(cm->error, AOM_CODEC_UNSUP_BITSTREAM, |
4019 | 22 | "In YCbCr 4:2:0, film grain shall be applied " |
4020 | 22 | "to both chroma components or neither."); |
4021 | 7.14k | } |
4022 | | |
4023 | 15.6k | pars->scaling_shift = aom_rb_read_literal(rb, 2) + 8; // 8 + value |
4024 | | |
4025 | | // AR coefficients |
4026 | | // Only sent if the corresponsing scaling function has |
4027 | | // more than 0 points |
4028 | | |
4029 | 15.6k | pars->ar_coeff_lag = aom_rb_read_literal(rb, 2); |
4030 | | |
4031 | 15.6k | int num_pos_luma = 2 * pars->ar_coeff_lag * (pars->ar_coeff_lag + 1); |
4032 | 15.6k | int num_pos_chroma = num_pos_luma; |
4033 | 15.6k | if (pars->num_y_points > 0) ++num_pos_chroma; |
4034 | | |
4035 | 15.6k | if (pars->num_y_points) |
4036 | 34.5k | for (int i = 0; i < num_pos_luma; i++) |
4037 | 24.2k | pars->ar_coeffs_y[i] = aom_rb_read_literal(rb, 8) - 128; |
4038 | | |
4039 | 15.6k | if (pars->num_cb_points || pars->chroma_scaling_from_luma) |
4040 | 36.6k | for (int i = 0; i < num_pos_chroma; i++) |
4041 | 29.7k | pars->ar_coeffs_cb[i] = aom_rb_read_literal(rb, 8) - 128; |
4042 | | |
4043 | 15.6k | if (pars->num_cr_points || pars->chroma_scaling_from_luma) |
4044 | 38.2k | for (int i = 0; i < num_pos_chroma; i++) |
4045 | 30.3k | pars->ar_coeffs_cr[i] = aom_rb_read_literal(rb, 8) - 128; |
4046 | | |
4047 | 15.6k | pars->ar_coeff_shift = aom_rb_read_literal(rb, 2) + 6; // 6 + value |
4048 | | |
4049 | 15.6k | pars->grain_scale_shift = aom_rb_read_literal(rb, 2); |
4050 | | |
4051 | 15.6k | if (pars->num_cb_points) { |
4052 | 1.27k | pars->cb_mult = aom_rb_read_literal(rb, 8); |
4053 | 1.27k | pars->cb_luma_mult = aom_rb_read_literal(rb, 8); |
4054 | 1.27k | pars->cb_offset = aom_rb_read_literal(rb, 9); |
4055 | 1.27k | } |
4056 | | |
4057 | 15.6k | if (pars->num_cr_points) { |
4058 | 2.38k | pars->cr_mult = aom_rb_read_literal(rb, 8); |
4059 | 2.38k | pars->cr_luma_mult = aom_rb_read_literal(rb, 8); |
4060 | 2.38k | pars->cr_offset = aom_rb_read_literal(rb, 9); |
4061 | 2.38k | } |
4062 | | |
4063 | 15.6k | pars->overlap_flag = aom_rb_read_bit(rb); |
4064 | | |
4065 | 15.6k | pars->clip_to_restricted_range = aom_rb_read_bit(rb); |
4066 | 15.6k | } |
4067 | | |
4068 | | static AOM_INLINE void read_film_grain(AV1_COMMON *cm, |
4069 | 202k | struct aom_read_bit_buffer *rb) { |
4070 | 202k | if (cm->seq_params->film_grain_params_present && |
4071 | 202k | (cm->show_frame || cm->showable_frame)) { |
4072 | 48.7k | read_film_grain_params(cm, rb); |
4073 | 153k | } else { |
4074 | 153k | memset(&cm->film_grain_params, 0, sizeof(cm->film_grain_params)); |
4075 | 153k | } |
4076 | 202k | cm->film_grain_params.bit_depth = cm->seq_params->bit_depth; |
4077 | 202k | memcpy(&cm->cur_frame->film_grain_params, &cm->film_grain_params, |
4078 | 202k | sizeof(aom_film_grain_t)); |
4079 | 202k | } |
4080 | | |
4081 | | void av1_read_color_config(struct aom_read_bit_buffer *rb, |
4082 | | int allow_lowbitdepth, SequenceHeader *seq_params, |
4083 | 72.4k | struct aom_internal_error_info *error_info) { |
4084 | 72.4k | read_bitdepth(rb, seq_params, error_info); |
4085 | | |
4086 | 72.4k | seq_params->use_highbitdepth = |
4087 | 72.4k | seq_params->bit_depth > AOM_BITS_8 || !allow_lowbitdepth; |
4088 | | // monochrome bit (not needed for PROFILE_1) |
4089 | 72.4k | const int is_monochrome = |
4090 | 72.4k | seq_params->profile != PROFILE_1 ? aom_rb_read_bit(rb) : 0; |
4091 | 72.4k | seq_params->monochrome = is_monochrome; |
4092 | 72.4k | int color_description_present_flag = aom_rb_read_bit(rb); |
4093 | 72.4k | if (color_description_present_flag) { |
4094 | 5.08k | seq_params->color_primaries = aom_rb_read_literal(rb, 8); |
4095 | 5.08k | seq_params->transfer_characteristics = aom_rb_read_literal(rb, 8); |
4096 | 5.08k | seq_params->matrix_coefficients = aom_rb_read_literal(rb, 8); |
4097 | 67.3k | } else { |
4098 | 67.3k | seq_params->color_primaries = AOM_CICP_CP_UNSPECIFIED; |
4099 | 67.3k | seq_params->transfer_characteristics = AOM_CICP_TC_UNSPECIFIED; |
4100 | 67.3k | seq_params->matrix_coefficients = AOM_CICP_MC_UNSPECIFIED; |
4101 | 67.3k | } |
4102 | 72.4k | if (is_monochrome) { |
4103 | | // [16,235] (including xvycc) vs [0,255] range |
4104 | 14.2k | seq_params->color_range = aom_rb_read_bit(rb); |
4105 | 14.2k | seq_params->subsampling_y = seq_params->subsampling_x = 1; |
4106 | 14.2k | seq_params->chroma_sample_position = AOM_CSP_UNKNOWN; |
4107 | 14.2k | seq_params->separate_uv_delta_q = 0; |
4108 | 14.2k | return; |
4109 | 14.2k | } |
4110 | 58.2k | if (seq_params->color_primaries == AOM_CICP_CP_BT_709 && |
4111 | 58.2k | seq_params->transfer_characteristics == AOM_CICP_TC_SRGB && |
4112 | 58.2k | seq_params->matrix_coefficients == AOM_CICP_MC_IDENTITY) { |
4113 | 2 | seq_params->subsampling_y = seq_params->subsampling_x = 0; |
4114 | 2 | seq_params->color_range = 1; // assume full color-range |
4115 | 2 | if (!(seq_params->profile == PROFILE_1 || |
4116 | 2 | (seq_params->profile == PROFILE_2 && |
4117 | 2 | seq_params->bit_depth == AOM_BITS_12))) { |
4118 | 2 | aom_internal_error( |
4119 | 2 | error_info, AOM_CODEC_UNSUP_BITSTREAM, |
4120 | 2 | "sRGB colorspace not compatible with specified profile"); |
4121 | 2 | } |
4122 | 58.2k | } else { |
4123 | | // [16,235] (including xvycc) vs [0,255] range |
4124 | 58.2k | seq_params->color_range = aom_rb_read_bit(rb); |
4125 | 58.2k | if (seq_params->profile == PROFILE_0) { |
4126 | | // 420 only |
4127 | 25.3k | seq_params->subsampling_x = seq_params->subsampling_y = 1; |
4128 | 32.8k | } else if (seq_params->profile == PROFILE_1) { |
4129 | | // 444 only |
4130 | 14.8k | seq_params->subsampling_x = seq_params->subsampling_y = 0; |
4131 | 18.0k | } else { |
4132 | 18.0k | assert(seq_params->profile == PROFILE_2); |
4133 | 17.0k | if (seq_params->bit_depth == AOM_BITS_12) { |
4134 | 14.2k | seq_params->subsampling_x = aom_rb_read_bit(rb); |
4135 | 14.2k | if (seq_params->subsampling_x) |
4136 | 1.10k | seq_params->subsampling_y = aom_rb_read_bit(rb); // 422 or 420 |
4137 | 13.1k | else |
4138 | 13.1k | seq_params->subsampling_y = 0; // 444 |
4139 | 14.2k | } else { |
4140 | | // 422 |
4141 | 2.85k | seq_params->subsampling_x = 1; |
4142 | 2.85k | seq_params->subsampling_y = 0; |
4143 | 2.85k | } |
4144 | 17.0k | } |
4145 | 57.3k | if (seq_params->matrix_coefficients == AOM_CICP_MC_IDENTITY && |
4146 | 57.3k | (seq_params->subsampling_x || seq_params->subsampling_y)) { |
4147 | 11 | aom_internal_error( |
4148 | 11 | error_info, AOM_CODEC_UNSUP_BITSTREAM, |
4149 | 11 | "Identity CICP Matrix incompatible with non 4:4:4 color sampling"); |
4150 | 11 | } |
4151 | 57.3k | if (seq_params->subsampling_x && seq_params->subsampling_y) { |
4152 | 25.4k | seq_params->chroma_sample_position = aom_rb_read_literal(rb, 2); |
4153 | 25.4k | } |
4154 | 57.3k | } |
4155 | 57.3k | seq_params->separate_uv_delta_q = aom_rb_read_bit(rb); |
4156 | 57.3k | } |
4157 | | |
4158 | | void av1_read_timing_info_header(aom_timing_info_t *timing_info, |
4159 | | struct aom_internal_error_info *error, |
4160 | 4.31k | struct aom_read_bit_buffer *rb) { |
4161 | 4.31k | timing_info->num_units_in_display_tick = |
4162 | 4.31k | aom_rb_read_unsigned_literal(rb, |
4163 | 4.31k | 32); // Number of units in a display tick |
4164 | 4.31k | timing_info->time_scale = aom_rb_read_unsigned_literal(rb, 32); // Time scale |
4165 | 4.31k | if (timing_info->num_units_in_display_tick == 0 || |
4166 | 4.31k | timing_info->time_scale == 0) { |
4167 | 329 | aom_internal_error( |
4168 | 329 | error, AOM_CODEC_UNSUP_BITSTREAM, |
4169 | 329 | "num_units_in_display_tick and time_scale must be greater than 0."); |
4170 | 329 | } |
4171 | 4.31k | timing_info->equal_picture_interval = |
4172 | 4.31k | aom_rb_read_bit(rb); // Equal picture interval bit |
4173 | 4.31k | if (timing_info->equal_picture_interval) { |
4174 | 1.86k | const uint32_t num_ticks_per_picture_minus_1 = aom_rb_read_uvlc(rb); |
4175 | 1.86k | if (num_ticks_per_picture_minus_1 == UINT32_MAX) { |
4176 | 195 | aom_internal_error( |
4177 | 195 | error, AOM_CODEC_UNSUP_BITSTREAM, |
4178 | 195 | "num_ticks_per_picture_minus_1 cannot be (1 << 32) - 1."); |
4179 | 195 | } |
4180 | 1.86k | timing_info->num_ticks_per_picture = num_ticks_per_picture_minus_1 + 1; |
4181 | 1.86k | } |
4182 | 4.31k | } |
4183 | | |
4184 | | void av1_read_decoder_model_info(aom_dec_model_info_t *decoder_model_info, |
4185 | 1.09k | struct aom_read_bit_buffer *rb) { |
4186 | 1.09k | decoder_model_info->encoder_decoder_buffer_delay_length = |
4187 | 1.09k | aom_rb_read_literal(rb, 5) + 1; |
4188 | 1.09k | decoder_model_info->num_units_in_decoding_tick = |
4189 | 1.09k | aom_rb_read_unsigned_literal(rb, |
4190 | 1.09k | 32); // Number of units in a decoding tick |
4191 | 1.09k | decoder_model_info->buffer_removal_time_length = |
4192 | 1.09k | aom_rb_read_literal(rb, 5) + 1; |
4193 | 1.09k | decoder_model_info->frame_presentation_time_length = |
4194 | 1.09k | aom_rb_read_literal(rb, 5) + 1; |
4195 | 1.09k | } |
4196 | | |
4197 | | void av1_read_op_parameters_info(aom_dec_model_op_parameters_t *op_params, |
4198 | | int buffer_delay_length, |
4199 | 1.32k | struct aom_read_bit_buffer *rb) { |
4200 | 1.32k | op_params->decoder_buffer_delay = |
4201 | 1.32k | aom_rb_read_unsigned_literal(rb, buffer_delay_length); |
4202 | 1.32k | op_params->encoder_buffer_delay = |
4203 | 1.32k | aom_rb_read_unsigned_literal(rb, buffer_delay_length); |
4204 | 1.32k | op_params->low_delay_mode_flag = aom_rb_read_bit(rb); |
4205 | 1.32k | } |
4206 | | |
4207 | | static AOM_INLINE void read_temporal_point_info( |
4208 | 1.09k | AV1_COMMON *const cm, struct aom_read_bit_buffer *rb) { |
4209 | 1.09k | cm->frame_presentation_time = aom_rb_read_unsigned_literal( |
4210 | 1.09k | rb, cm->seq_params->decoder_model_info.frame_presentation_time_length); |
4211 | 1.09k | } |
4212 | | |
4213 | | void av1_read_sequence_header(AV1_COMMON *cm, struct aom_read_bit_buffer *rb, |
4214 | 73.1k | SequenceHeader *seq_params) { |
4215 | 73.1k | const int num_bits_width = aom_rb_read_literal(rb, 4) + 1; |
4216 | 73.1k | const int num_bits_height = aom_rb_read_literal(rb, 4) + 1; |
4217 | 73.1k | const int max_frame_width = aom_rb_read_literal(rb, num_bits_width) + 1; |
4218 | 73.1k | const int max_frame_height = aom_rb_read_literal(rb, num_bits_height) + 1; |
4219 | | |
4220 | 73.1k | seq_params->num_bits_width = num_bits_width; |
4221 | 73.1k | seq_params->num_bits_height = num_bits_height; |
4222 | 73.1k | seq_params->max_frame_width = max_frame_width; |
4223 | 73.1k | seq_params->max_frame_height = max_frame_height; |
4224 | | |
4225 | 73.1k | if (seq_params->reduced_still_picture_hdr) { |
4226 | 20.4k | seq_params->frame_id_numbers_present_flag = 0; |
4227 | 52.7k | } else { |
4228 | 52.7k | seq_params->frame_id_numbers_present_flag = aom_rb_read_bit(rb); |
4229 | 52.7k | } |
4230 | 73.1k | if (seq_params->frame_id_numbers_present_flag) { |
4231 | | // We must always have delta_frame_id_length < frame_id_length, |
4232 | | // in order for a frame to be referenced with a unique delta. |
4233 | | // Avoid wasting bits by using a coding that enforces this restriction. |
4234 | 14.8k | seq_params->delta_frame_id_length = aom_rb_read_literal(rb, 4) + 2; |
4235 | 14.8k | seq_params->frame_id_length = |
4236 | 14.8k | aom_rb_read_literal(rb, 3) + seq_params->delta_frame_id_length + 1; |
4237 | 14.8k | if (seq_params->frame_id_length > 16) |
4238 | 154 | aom_internal_error(cm->error, AOM_CODEC_CORRUPT_FRAME, |
4239 | 154 | "Invalid frame_id_length"); |
4240 | 14.8k | } |
4241 | | |
4242 | 73.1k | setup_sb_size(seq_params, rb); |
4243 | | |
4244 | 73.1k | seq_params->enable_filter_intra = aom_rb_read_bit(rb); |
4245 | 73.1k | seq_params->enable_intra_edge_filter = aom_rb_read_bit(rb); |
4246 | | |
4247 | 73.1k | if (seq_params->reduced_still_picture_hdr) { |
4248 | 20.4k | seq_params->enable_interintra_compound = 0; |
4249 | 20.4k | seq_params->enable_masked_compound = 0; |
4250 | 20.4k | seq_params->enable_warped_motion = 0; |
4251 | 20.4k | seq_params->enable_dual_filter = 0; |
4252 | 20.4k | seq_params->order_hint_info.enable_order_hint = 0; |
4253 | 20.4k | seq_params->order_hint_info.enable_dist_wtd_comp = 0; |
4254 | 20.4k | seq_params->order_hint_info.enable_ref_frame_mvs = 0; |
4255 | 20.4k | seq_params->force_screen_content_tools = 2; // SELECT_SCREEN_CONTENT_TOOLS |
4256 | 20.4k | seq_params->force_integer_mv = 2; // SELECT_INTEGER_MV |
4257 | 20.4k | seq_params->order_hint_info.order_hint_bits_minus_1 = -1; |
4258 | 52.7k | } else { |
4259 | 52.7k | seq_params->enable_interintra_compound = aom_rb_read_bit(rb); |
4260 | 52.7k | seq_params->enable_masked_compound = aom_rb_read_bit(rb); |
4261 | 52.7k | seq_params->enable_warped_motion = aom_rb_read_bit(rb); |
4262 | 52.7k | seq_params->enable_dual_filter = aom_rb_read_bit(rb); |
4263 | | |
4264 | 52.7k | seq_params->order_hint_info.enable_order_hint = aom_rb_read_bit(rb); |
4265 | 52.7k | seq_params->order_hint_info.enable_dist_wtd_comp = |
4266 | 52.7k | seq_params->order_hint_info.enable_order_hint ? aom_rb_read_bit(rb) : 0; |
4267 | 52.7k | seq_params->order_hint_info.enable_ref_frame_mvs = |
4268 | 52.7k | seq_params->order_hint_info.enable_order_hint ? aom_rb_read_bit(rb) : 0; |
4269 | | |
4270 | 52.7k | if (aom_rb_read_bit(rb)) { |
4271 | 33.2k | seq_params->force_screen_content_tools = |
4272 | 33.2k | 2; // SELECT_SCREEN_CONTENT_TOOLS |
4273 | 33.2k | } else { |
4274 | 19.5k | seq_params->force_screen_content_tools = aom_rb_read_bit(rb); |
4275 | 19.5k | } |
4276 | | |
4277 | 52.7k | if (seq_params->force_screen_content_tools > 0) { |
4278 | 46.6k | if (aom_rb_read_bit(rb)) { |
4279 | 36.9k | seq_params->force_integer_mv = 2; // SELECT_INTEGER_MV |
4280 | 36.9k | } else { |
4281 | 9.75k | seq_params->force_integer_mv = aom_rb_read_bit(rb); |
4282 | 9.75k | } |
4283 | 46.6k | } else { |
4284 | 6.12k | seq_params->force_integer_mv = 2; // SELECT_INTEGER_MV |
4285 | 6.12k | } |
4286 | 52.7k | seq_params->order_hint_info.order_hint_bits_minus_1 = |
4287 | 52.7k | seq_params->order_hint_info.enable_order_hint |
4288 | 52.7k | ? aom_rb_read_literal(rb, 3) |
4289 | 52.7k | : -1; |
4290 | 52.7k | } |
4291 | | |
4292 | 73.1k | seq_params->enable_superres = aom_rb_read_bit(rb); |
4293 | 73.1k | seq_params->enable_cdef = aom_rb_read_bit(rb); |
4294 | 73.1k | seq_params->enable_restoration = aom_rb_read_bit(rb); |
4295 | 73.1k | } |
4296 | | |
4297 | | static int read_global_motion_params(WarpedMotionParams *params, |
4298 | | const WarpedMotionParams *ref_params, |
4299 | | struct aom_read_bit_buffer *rb, |
4300 | 633k | int allow_hp) { |
4301 | 633k | TransformationType type = aom_rb_read_bit(rb); |
4302 | 633k | if (type != IDENTITY) { |
4303 | 31.7k | if (aom_rb_read_bit(rb)) |
4304 | 25.2k | type = ROTZOOM; |
4305 | 6.48k | else |
4306 | 6.48k | type = aom_rb_read_bit(rb) ? TRANSLATION : AFFINE; |
4307 | 31.7k | } |
4308 | | |
4309 | 633k | *params = default_warp_params; |
4310 | 633k | params->wmtype = type; |
4311 | | |
4312 | 633k | if (type >= ROTZOOM) { |
4313 | 29.1k | params->wmmat[2] = aom_rb_read_signed_primitive_refsubexpfin( |
4314 | 29.1k | rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K, |
4315 | 29.1k | (ref_params->wmmat[2] >> GM_ALPHA_PREC_DIFF) - |
4316 | 29.1k | (1 << GM_ALPHA_PREC_BITS)) * |
4317 | 29.1k | GM_ALPHA_DECODE_FACTOR + |
4318 | 29.1k | (1 << WARPEDMODEL_PREC_BITS); |
4319 | 29.1k | params->wmmat[3] = aom_rb_read_signed_primitive_refsubexpfin( |
4320 | 29.1k | rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K, |
4321 | 29.1k | (ref_params->wmmat[3] >> GM_ALPHA_PREC_DIFF)) * |
4322 | 29.1k | GM_ALPHA_DECODE_FACTOR; |
4323 | 29.1k | } |
4324 | | |
4325 | 633k | if (type >= AFFINE) { |
4326 | 3.53k | params->wmmat[4] = aom_rb_read_signed_primitive_refsubexpfin( |
4327 | 3.53k | rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K, |
4328 | 3.53k | (ref_params->wmmat[4] >> GM_ALPHA_PREC_DIFF)) * |
4329 | 3.53k | GM_ALPHA_DECODE_FACTOR; |
4330 | 3.53k | params->wmmat[5] = aom_rb_read_signed_primitive_refsubexpfin( |
4331 | 3.53k | rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K, |
4332 | 3.53k | (ref_params->wmmat[5] >> GM_ALPHA_PREC_DIFF) - |
4333 | 3.53k | (1 << GM_ALPHA_PREC_BITS)) * |
4334 | 3.53k | GM_ALPHA_DECODE_FACTOR + |
4335 | 3.53k | (1 << WARPEDMODEL_PREC_BITS); |
4336 | 629k | } else { |
4337 | 629k | params->wmmat[4] = -params->wmmat[3]; |
4338 | 629k | params->wmmat[5] = params->wmmat[2]; |
4339 | 629k | } |
4340 | | |
4341 | 633k | if (type >= TRANSLATION) { |
4342 | 31.3k | const int trans_bits = (type == TRANSLATION) |
4343 | 31.3k | ? GM_ABS_TRANS_ONLY_BITS - !allow_hp |
4344 | 31.3k | : GM_ABS_TRANS_BITS; |
4345 | 31.3k | const int trans_dec_factor = |
4346 | 31.3k | (type == TRANSLATION) ? GM_TRANS_ONLY_DECODE_FACTOR * (1 << !allow_hp) |
4347 | 31.3k | : GM_TRANS_DECODE_FACTOR; |
4348 | 31.3k | const int trans_prec_diff = (type == TRANSLATION) |
4349 | 31.3k | ? GM_TRANS_ONLY_PREC_DIFF + !allow_hp |
4350 | 31.3k | : GM_TRANS_PREC_DIFF; |
4351 | 31.3k | params->wmmat[0] = aom_rb_read_signed_primitive_refsubexpfin( |
4352 | 31.3k | rb, (1 << trans_bits) + 1, SUBEXPFIN_K, |
4353 | 31.3k | (ref_params->wmmat[0] >> trans_prec_diff)) * |
4354 | 31.3k | trans_dec_factor; |
4355 | 31.3k | params->wmmat[1] = aom_rb_read_signed_primitive_refsubexpfin( |
4356 | 31.3k | rb, (1 << trans_bits) + 1, SUBEXPFIN_K, |
4357 | 31.3k | (ref_params->wmmat[1] >> trans_prec_diff)) * |
4358 | 31.3k | trans_dec_factor; |
4359 | 31.3k | } |
4360 | | |
4361 | 633k | int good_shear_params = av1_get_shear_params(params); |
4362 | 633k | if (!good_shear_params) return 0; |
4363 | | |
4364 | 617k | return 1; |
4365 | 633k | } |
4366 | | |
4367 | | static AOM_INLINE void read_global_motion(AV1_COMMON *cm, |
4368 | 90.7k | struct aom_read_bit_buffer *rb) { |
4369 | 724k | for (int frame = LAST_FRAME; frame <= ALTREF_FRAME; ++frame) { |
4370 | 633k | const WarpedMotionParams *ref_params = |
4371 | 633k | cm->prev_frame ? &cm->prev_frame->global_motion[frame] |
4372 | 633k | : &default_warp_params; |
4373 | 633k | int good_params = |
4374 | 633k | read_global_motion_params(&cm->global_motion[frame], ref_params, rb, |
4375 | 633k | cm->features.allow_high_precision_mv); |
4376 | 633k | if (!good_params) { |
4377 | | #if WARPED_MOTION_DEBUG |
4378 | | printf("Warning: unexpected global motion shear params from aomenc\n"); |
4379 | | #endif |
4380 | 16.2k | cm->global_motion[frame].invalid = 1; |
4381 | 16.2k | } |
4382 | | |
4383 | | // TODO(sarahparker, debargha): The logic in the commented out code below |
4384 | | // does not work currently and causes mismatches when resize is on. Fix it |
4385 | | // before turning the optimization back on. |
4386 | | /* |
4387 | | YV12_BUFFER_CONFIG *ref_buf = get_ref_frame(cm, frame); |
4388 | | if (cm->width == ref_buf->y_crop_width && |
4389 | | cm->height == ref_buf->y_crop_height) { |
4390 | | read_global_motion_params(&cm->global_motion[frame], |
4391 | | &cm->prev_frame->global_motion[frame], rb, |
4392 | | cm->features.allow_high_precision_mv); |
4393 | | } else { |
4394 | | cm->global_motion[frame] = default_warp_params; |
4395 | | } |
4396 | | */ |
4397 | | /* |
4398 | | printf("Dec Ref %d [%d/%d]: %d %d %d %d\n", |
4399 | | frame, cm->current_frame.frame_number, cm->show_frame, |
4400 | | cm->global_motion[frame].wmmat[0], |
4401 | | cm->global_motion[frame].wmmat[1], |
4402 | | cm->global_motion[frame].wmmat[2], |
4403 | | cm->global_motion[frame].wmmat[3]); |
4404 | | */ |
4405 | 633k | } |
4406 | 90.7k | memcpy(cm->cur_frame->global_motion, cm->global_motion, |
4407 | 90.7k | REF_FRAMES * sizeof(WarpedMotionParams)); |
4408 | 90.7k | } |
4409 | | |
4410 | | // Release the references to the frame buffers in cm->ref_frame_map and reset |
4411 | | // all elements of cm->ref_frame_map to NULL. |
4412 | 113k | static AOM_INLINE void reset_ref_frame_map(AV1_COMMON *const cm) { |
4413 | 113k | BufferPool *const pool = cm->buffer_pool; |
4414 | | |
4415 | 1.01M | for (int i = 0; i < REF_FRAMES; i++) { |
4416 | 906k | decrease_ref_count(cm->ref_frame_map[i], pool); |
4417 | 906k | cm->ref_frame_map[i] = NULL; |
4418 | 906k | } |
4419 | 113k | } |
4420 | | |
4421 | | // If the refresh_frame_flags bitmask is set, update reference frame id values |
4422 | | // and mark frames as valid for reference. |
4423 | 218k | static AOM_INLINE void update_ref_frame_id(AV1Decoder *const pbi) { |
4424 | 218k | AV1_COMMON *const cm = &pbi->common; |
4425 | 218k | int refresh_frame_flags = cm->current_frame.refresh_frame_flags; |
4426 | 1.96M | for (int i = 0; i < REF_FRAMES; i++) { |
4427 | 1.74M | if ((refresh_frame_flags >> i) & 1) { |
4428 | 805k | cm->ref_frame_id[i] = cm->current_frame_id; |
4429 | 805k | pbi->valid_for_referencing[i] = 1; |
4430 | 805k | } |
4431 | 1.74M | } |
4432 | 218k | } |
4433 | | |
4434 | | static AOM_INLINE void show_existing_frame_reset(AV1Decoder *const pbi, |
4435 | 167 | int existing_frame_idx) { |
4436 | 167 | AV1_COMMON *const cm = &pbi->common; |
4437 | | |
4438 | 167 | assert(cm->show_existing_frame); |
4439 | | |
4440 | 167 | cm->current_frame.frame_type = KEY_FRAME; |
4441 | | |
4442 | 167 | cm->current_frame.refresh_frame_flags = (1 << REF_FRAMES) - 1; |
4443 | | |
4444 | 1.33k | for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) { |
4445 | 1.16k | cm->remapped_ref_idx[i] = INVALID_IDX; |
4446 | 1.16k | } |
4447 | | |
4448 | 167 | if (pbi->need_resync) { |
4449 | 87 | reset_ref_frame_map(cm); |
4450 | 87 | pbi->need_resync = 0; |
4451 | 87 | } |
4452 | | |
4453 | | // Note that the displayed frame must be valid for referencing in order to |
4454 | | // have been selected. |
4455 | 167 | cm->current_frame_id = cm->ref_frame_id[existing_frame_idx]; |
4456 | 167 | update_ref_frame_id(pbi); |
4457 | | |
4458 | 167 | cm->features.refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED; |
4459 | 167 | } |
4460 | | |
4461 | 21.8k | static INLINE void reset_frame_buffers(AV1_COMMON *cm) { |
4462 | 21.8k | RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs; |
4463 | 21.8k | int i; |
4464 | | |
4465 | 21.8k | lock_buffer_pool(cm->buffer_pool); |
4466 | 21.8k | reset_ref_frame_map(cm); |
4467 | 21.8k | assert(cm->cur_frame->ref_count == 1); |
4468 | 371k | for (i = 0; i < cm->buffer_pool->num_frame_bufs; ++i) { |
4469 | | // Reset all unreferenced frame buffers. We can also reset cm->cur_frame |
4470 | | // because we are the sole owner of cm->cur_frame. |
4471 | 349k | if (frame_bufs[i].ref_count > 0 && &frame_bufs[i] != cm->cur_frame) { |
4472 | 4.14k | continue; |
4473 | 4.14k | } |
4474 | 345k | frame_bufs[i].order_hint = 0; |
4475 | 345k | av1_zero(frame_bufs[i].ref_order_hints); |
4476 | 345k | } |
4477 | 21.8k | av1_zero_unused_internal_frame_buffers(&cm->buffer_pool->int_frame_buffers); |
4478 | 21.8k | unlock_buffer_pool(cm->buffer_pool); |
4479 | 21.8k | } |
4480 | | |
4481 | | // On success, returns 0. On failure, calls aom_internal_error and does not |
4482 | | // return. |
4483 | | static int read_uncompressed_header(AV1Decoder *pbi, |
4484 | 266k | struct aom_read_bit_buffer *rb) { |
4485 | 266k | AV1_COMMON *const cm = &pbi->common; |
4486 | 266k | const SequenceHeader *const seq_params = cm->seq_params; |
4487 | 266k | CurrentFrame *const current_frame = &cm->current_frame; |
4488 | 266k | FeatureFlags *const features = &cm->features; |
4489 | 266k | MACROBLOCKD *const xd = &pbi->dcb.xd; |
4490 | 266k | BufferPool *const pool = cm->buffer_pool; |
4491 | 266k | RefCntBuffer *const frame_bufs = pool->frame_bufs; |
4492 | 266k | aom_s_frame_info *sframe_info = &pbi->sframe_info; |
4493 | 266k | sframe_info->is_s_frame = 0; |
4494 | 266k | sframe_info->is_s_frame_at_altref = 0; |
4495 | | |
4496 | 266k | if (!pbi->sequence_header_ready) { |
4497 | 1.24k | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
4498 | 1.24k | "No sequence header"); |
4499 | 1.24k | } |
4500 | | |
4501 | 266k | if (seq_params->reduced_still_picture_hdr) { |
4502 | 34.4k | cm->show_existing_frame = 0; |
4503 | 34.4k | cm->show_frame = 1; |
4504 | 34.4k | current_frame->frame_type = KEY_FRAME; |
4505 | 34.4k | if (pbi->sequence_header_changed) { |
4506 | | // This is the start of a new coded video sequence. |
4507 | 6.16k | pbi->sequence_header_changed = 0; |
4508 | 6.16k | pbi->decoding_first_frame = 1; |
4509 | 6.16k | reset_frame_buffers(cm); |
4510 | 6.16k | } |
4511 | 34.4k | features->error_resilient_mode = 1; |
4512 | 231k | } else { |
4513 | 231k | cm->show_existing_frame = aom_rb_read_bit(rb); |
4514 | 231k | pbi->reset_decoder_state = 0; |
4515 | | |
4516 | 231k | if (cm->show_existing_frame) { |
4517 | 8.31k | if (pbi->sequence_header_changed) { |
4518 | 152 | aom_internal_error( |
4519 | 152 | &pbi->error, AOM_CODEC_CORRUPT_FRAME, |
4520 | 152 | "New sequence header starts with a show_existing_frame."); |
4521 | 152 | } |
4522 | | // Show an existing frame directly. |
4523 | 8.31k | const int existing_frame_idx = aom_rb_read_literal(rb, 3); |
4524 | 8.31k | RefCntBuffer *const frame_to_show = cm->ref_frame_map[existing_frame_idx]; |
4525 | 8.31k | if (frame_to_show == NULL) { |
4526 | 4.74k | aom_internal_error(&pbi->error, AOM_CODEC_UNSUP_BITSTREAM, |
4527 | 4.74k | "Buffer does not contain a decoded frame"); |
4528 | 4.74k | } |
4529 | 8.31k | if (seq_params->decoder_model_info_present_flag && |
4530 | 8.31k | seq_params->timing_info.equal_picture_interval == 0) { |
4531 | 91 | read_temporal_point_info(cm, rb); |
4532 | 91 | } |
4533 | 8.31k | if (seq_params->frame_id_numbers_present_flag) { |
4534 | 2.20k | int frame_id_length = seq_params->frame_id_length; |
4535 | 2.20k | int display_frame_id = aom_rb_read_literal(rb, frame_id_length); |
4536 | | /* Compare display_frame_id with ref_frame_id and check valid for |
4537 | | * referencing */ |
4538 | 2.20k | if (display_frame_id != cm->ref_frame_id[existing_frame_idx] || |
4539 | 2.20k | pbi->valid_for_referencing[existing_frame_idx] == 0) |
4540 | 1.75k | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
4541 | 1.75k | "Reference buffer frame ID mismatch"); |
4542 | 2.20k | } |
4543 | 8.31k | lock_buffer_pool(pool); |
4544 | 8.31k | assert(frame_to_show->ref_count > 0); |
4545 | | // cm->cur_frame should be the buffer referenced by the return value |
4546 | | // of the get_free_fb() call in assign_cur_frame_new_fb() (called by |
4547 | | // av1_receive_compressed_data()), so the ref_count should be 1. |
4548 | 1.65k | assert(cm->cur_frame->ref_count == 1); |
4549 | | // assign_frame_buffer_p() decrements ref_count directly rather than |
4550 | | // call decrease_ref_count(). If cm->cur_frame->raw_frame_buffer has |
4551 | | // already been allocated, it will not be released by |
4552 | | // assign_frame_buffer_p()! |
4553 | 1.65k | assert(!cm->cur_frame->raw_frame_buffer.data); |
4554 | 1.65k | assign_frame_buffer_p(&cm->cur_frame, frame_to_show); |
4555 | 1.65k | pbi->reset_decoder_state = frame_to_show->frame_type == KEY_FRAME; |
4556 | 1.65k | unlock_buffer_pool(pool); |
4557 | | |
4558 | 1.65k | cm->lf.filter_level[0] = 0; |
4559 | 1.65k | cm->lf.filter_level[1] = 0; |
4560 | 1.65k | cm->show_frame = 1; |
4561 | 1.65k | current_frame->order_hint = frame_to_show->order_hint; |
4562 | | |
4563 | | // Section 6.8.2: It is a requirement of bitstream conformance that when |
4564 | | // show_existing_frame is used to show a previous frame, that the value |
4565 | | // of showable_frame for the previous frame was equal to 1. |
4566 | 1.65k | if (!frame_to_show->showable_frame) { |
4567 | 866 | aom_internal_error(&pbi->error, AOM_CODEC_UNSUP_BITSTREAM, |
4568 | 866 | "Buffer does not contain a showable frame"); |
4569 | 866 | } |
4570 | | // Section 6.8.2: It is a requirement of bitstream conformance that when |
4571 | | // show_existing_frame is used to show a previous frame with |
4572 | | // RefFrameType[ frame_to_show_map_idx ] equal to KEY_FRAME, that the |
4573 | | // frame is output via the show_existing_frame mechanism at most once. |
4574 | 1.65k | if (pbi->reset_decoder_state) frame_to_show->showable_frame = 0; |
4575 | | |
4576 | 1.65k | cm->film_grain_params = frame_to_show->film_grain_params; |
4577 | | |
4578 | 1.65k | if (pbi->reset_decoder_state) { |
4579 | 167 | show_existing_frame_reset(pbi, existing_frame_idx); |
4580 | 1.48k | } else { |
4581 | 1.48k | current_frame->refresh_frame_flags = 0; |
4582 | 1.48k | } |
4583 | | |
4584 | 1.65k | return 0; |
4585 | 1.65k | } |
4586 | | |
4587 | 223k | current_frame->frame_type = (FRAME_TYPE)aom_rb_read_literal(rb, 2); |
4588 | 223k | if (pbi->sequence_header_changed) { |
4589 | 16.1k | if (current_frame->frame_type == KEY_FRAME) { |
4590 | | // This is the start of a new coded video sequence. |
4591 | 15.7k | pbi->sequence_header_changed = 0; |
4592 | 15.7k | pbi->decoding_first_frame = 1; |
4593 | 15.7k | reset_frame_buffers(cm); |
4594 | 15.7k | } else { |
4595 | 463 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
4596 | 463 | "Sequence header has changed without a keyframe."); |
4597 | 463 | } |
4598 | 16.1k | } |
4599 | | |
4600 | 223k | cm->show_frame = aom_rb_read_bit(rb); |
4601 | 223k | if (cm->show_frame == 0) pbi->is_arf_frame_present = 1; |
4602 | 223k | if (cm->show_frame == 0 && cm->current_frame.frame_type == KEY_FRAME) |
4603 | 40.2k | pbi->is_fwd_kf_present = 1; |
4604 | 223k | if (cm->current_frame.frame_type == S_FRAME) { |
4605 | 159 | sframe_info->is_s_frame = 1; |
4606 | 159 | sframe_info->is_s_frame_at_altref = cm->show_frame ? 0 : 1; |
4607 | 159 | } |
4608 | 223k | if (seq_params->still_picture && |
4609 | 223k | (current_frame->frame_type != KEY_FRAME || !cm->show_frame)) { |
4610 | 167 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
4611 | 167 | "Still pictures must be coded as shown keyframes"); |
4612 | 167 | } |
4613 | 223k | cm->showable_frame = current_frame->frame_type != KEY_FRAME; |
4614 | 223k | if (cm->show_frame) { |
4615 | 160k | if (seq_params->decoder_model_info_present_flag && |
4616 | 160k | seq_params->timing_info.equal_picture_interval == 0) |
4617 | 1.00k | read_temporal_point_info(cm, rb); |
4618 | 160k | } else { |
4619 | | // See if this frame can be used as show_existing_frame in future |
4620 | 63.2k | cm->showable_frame = aom_rb_read_bit(rb); |
4621 | 63.2k | } |
4622 | 223k | cm->cur_frame->showable_frame = cm->showable_frame; |
4623 | 223k | features->error_resilient_mode = |
4624 | 223k | frame_is_sframe(cm) || |
4625 | 223k | (current_frame->frame_type == KEY_FRAME && cm->show_frame) |
4626 | 223k | ? 1 |
4627 | 223k | : aom_rb_read_bit(rb); |
4628 | 223k | } |
4629 | | |
4630 | 257k | if (current_frame->frame_type == KEY_FRAME && cm->show_frame) { |
4631 | | /* All frames need to be marked as not valid for referencing */ |
4632 | 528k | for (int i = 0; i < REF_FRAMES; i++) { |
4633 | 469k | pbi->valid_for_referencing[i] = 0; |
4634 | 469k | } |
4635 | 58.6k | } |
4636 | 257k | features->disable_cdf_update = aom_rb_read_bit(rb); |
4637 | 257k | if (seq_params->force_screen_content_tools == 2) { |
4638 | 219k | features->allow_screen_content_tools = aom_rb_read_bit(rb); |
4639 | 219k | } else { |
4640 | 38.6k | features->allow_screen_content_tools = |
4641 | 38.6k | seq_params->force_screen_content_tools; |
4642 | 38.6k | } |
4643 | | |
4644 | 257k | if (features->allow_screen_content_tools) { |
4645 | 83.7k | if (seq_params->force_integer_mv == 2) { |
4646 | 66.3k | features->cur_frame_force_integer_mv = aom_rb_read_bit(rb); |
4647 | 66.3k | } else { |
4648 | 17.4k | features->cur_frame_force_integer_mv = seq_params->force_integer_mv; |
4649 | 17.4k | } |
4650 | 174k | } else { |
4651 | 174k | features->cur_frame_force_integer_mv = 0; |
4652 | 174k | } |
4653 | | |
4654 | 257k | int frame_size_override_flag = 0; |
4655 | 257k | features->allow_intrabc = 0; |
4656 | 257k | features->primary_ref_frame = PRIMARY_REF_NONE; |
4657 | | |
4658 | 257k | if (!seq_params->reduced_still_picture_hdr) { |
4659 | 221k | if (seq_params->frame_id_numbers_present_flag) { |
4660 | 29.2k | int frame_id_length = seq_params->frame_id_length; |
4661 | 29.2k | int diff_len = seq_params->delta_frame_id_length; |
4662 | 29.2k | int prev_frame_id = 0; |
4663 | 29.2k | int have_prev_frame_id = |
4664 | 29.2k | !pbi->decoding_first_frame && |
4665 | 29.2k | !(current_frame->frame_type == KEY_FRAME && cm->show_frame); |
4666 | 29.2k | if (have_prev_frame_id) { |
4667 | 6.64k | prev_frame_id = cm->current_frame_id; |
4668 | 6.64k | } |
4669 | 29.2k | cm->current_frame_id = aom_rb_read_literal(rb, frame_id_length); |
4670 | | |
4671 | 29.2k | if (have_prev_frame_id) { |
4672 | 6.64k | int diff_frame_id; |
4673 | 6.64k | if (cm->current_frame_id > prev_frame_id) { |
4674 | 2.66k | diff_frame_id = cm->current_frame_id - prev_frame_id; |
4675 | 3.97k | } else { |
4676 | 3.97k | diff_frame_id = |
4677 | 3.97k | (1 << frame_id_length) + cm->current_frame_id - prev_frame_id; |
4678 | 3.97k | } |
4679 | | /* Check current_frame_id for conformance */ |
4680 | 6.64k | if (prev_frame_id == cm->current_frame_id || |
4681 | 6.64k | diff_frame_id >= (1 << (frame_id_length - 1))) { |
4682 | 3.86k | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
4683 | 3.86k | "Invalid value of current_frame_id"); |
4684 | 3.86k | } |
4685 | 6.64k | } |
4686 | | /* Check if some frames need to be marked as not valid for referencing */ |
4687 | 232k | for (int i = 0; i < REF_FRAMES; i++) { |
4688 | 203k | if (cm->current_frame_id - (1 << diff_len) > 0) { |
4689 | 101k | if (cm->ref_frame_id[i] > cm->current_frame_id || |
4690 | 101k | cm->ref_frame_id[i] < cm->current_frame_id - (1 << diff_len)) |
4691 | 52.1k | pbi->valid_for_referencing[i] = 0; |
4692 | 101k | } else { |
4693 | 101k | if (cm->ref_frame_id[i] > cm->current_frame_id && |
4694 | 101k | cm->ref_frame_id[i] < (1 << frame_id_length) + |
4695 | 51.7k | cm->current_frame_id - (1 << diff_len)) |
4696 | 21.8k | pbi->valid_for_referencing[i] = 0; |
4697 | 101k | } |
4698 | 203k | } |
4699 | 29.2k | } |
4700 | | |
4701 | 221k | frame_size_override_flag = frame_is_sframe(cm) ? 1 : aom_rb_read_bit(rb); |
4702 | | |
4703 | 221k | current_frame->order_hint = aom_rb_read_literal( |
4704 | 221k | rb, seq_params->order_hint_info.order_hint_bits_minus_1 + 1); |
4705 | | |
4706 | 221k | if (seq_params->order_hint_info.enable_order_hint) |
4707 | 198k | current_frame->frame_number = current_frame->order_hint; |
4708 | | |
4709 | 221k | if (!features->error_resilient_mode && !frame_is_intra_only(cm)) { |
4710 | 123k | features->primary_ref_frame = aom_rb_read_literal(rb, PRIMARY_REF_BITS); |
4711 | 123k | } |
4712 | 221k | } |
4713 | | |
4714 | 257k | if (seq_params->decoder_model_info_present_flag) { |
4715 | 4.07k | pbi->buffer_removal_time_present = aom_rb_read_bit(rb); |
4716 | 4.07k | if (pbi->buffer_removal_time_present) { |
4717 | 1.63k | for (int op_num = 0; |
4718 | 8.10k | op_num < seq_params->operating_points_cnt_minus_1 + 1; op_num++) { |
4719 | 6.47k | if (seq_params->op_params[op_num].decoder_model_param_present_flag) { |
4720 | 2.55k | if (seq_params->operating_point_idc[op_num] == 0 || |
4721 | 2.55k | (((seq_params->operating_point_idc[op_num] >> |
4722 | 2.13k | cm->temporal_layer_id) & |
4723 | 2.13k | 0x1) && |
4724 | 2.13k | ((seq_params->operating_point_idc[op_num] >> |
4725 | 1.22k | (cm->spatial_layer_id + 8)) & |
4726 | 1.22k | 0x1))) { |
4727 | 1.18k | cm->buffer_removal_times[op_num] = aom_rb_read_unsigned_literal( |
4728 | 1.18k | rb, seq_params->decoder_model_info.buffer_removal_time_length); |
4729 | 1.36k | } else { |
4730 | 1.36k | cm->buffer_removal_times[op_num] = 0; |
4731 | 1.36k | } |
4732 | 3.92k | } else { |
4733 | 3.92k | cm->buffer_removal_times[op_num] = 0; |
4734 | 3.92k | } |
4735 | 6.47k | } |
4736 | 1.63k | } |
4737 | 4.07k | } |
4738 | 257k | if (current_frame->frame_type == KEY_FRAME) { |
4739 | 98.3k | if (!cm->show_frame) { // unshown keyframe (forward keyframe) |
4740 | 39.7k | current_frame->refresh_frame_flags = aom_rb_read_literal(rb, REF_FRAMES); |
4741 | 58.6k | } else { // shown keyframe |
4742 | 58.6k | current_frame->refresh_frame_flags = (1 << REF_FRAMES) - 1; |
4743 | 58.6k | } |
4744 | | |
4745 | 786k | for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) { |
4746 | 688k | cm->remapped_ref_idx[i] = INVALID_IDX; |
4747 | 688k | } |
4748 | 98.3k | if (pbi->need_resync) { |
4749 | 69.7k | reset_ref_frame_map(cm); |
4750 | 69.7k | pbi->need_resync = 0; |
4751 | 69.7k | } |
4752 | 159k | } else { |
4753 | 159k | if (current_frame->frame_type == INTRA_ONLY_FRAME) { |
4754 | 24.7k | current_frame->refresh_frame_flags = aom_rb_read_literal(rb, REF_FRAMES); |
4755 | 24.7k | if (current_frame->refresh_frame_flags == 0xFF) { |
4756 | 632 | aom_internal_error(&pbi->error, AOM_CODEC_UNSUP_BITSTREAM, |
4757 | 632 | "Intra only frames cannot have refresh flags 0xFF"); |
4758 | 632 | } |
4759 | 24.7k | if (pbi->need_resync) { |
4760 | 21.5k | reset_ref_frame_map(cm); |
4761 | 21.5k | pbi->need_resync = 0; |
4762 | 21.5k | } |
4763 | 134k | } else if (pbi->need_resync != 1) { /* Skip if need resync */ |
4764 | 100k | current_frame->refresh_frame_flags = |
4765 | 100k | frame_is_sframe(cm) ? 0xFF : aom_rb_read_literal(rb, REF_FRAMES); |
4766 | 100k | } |
4767 | 159k | } |
4768 | | |
4769 | 257k | if (!frame_is_intra_only(cm) || current_frame->refresh_frame_flags != 0xFF) { |
4770 | | // Read all ref frame order hints if error_resilient_mode == 1 |
4771 | 192k | if (features->error_resilient_mode && |
4772 | 192k | seq_params->order_hint_info.enable_order_hint) { |
4773 | 221k | for (int ref_idx = 0; ref_idx < REF_FRAMES; ref_idx++) { |
4774 | | // Read order hint from bit stream |
4775 | 196k | unsigned int order_hint = aom_rb_read_literal( |
4776 | 196k | rb, seq_params->order_hint_info.order_hint_bits_minus_1 + 1); |
4777 | | // Get buffer |
4778 | 196k | RefCntBuffer *buf = cm->ref_frame_map[ref_idx]; |
4779 | 196k | if (buf == NULL || order_hint != buf->order_hint) { |
4780 | 182k | if (buf != NULL) { |
4781 | 35.6k | lock_buffer_pool(pool); |
4782 | 35.6k | decrease_ref_count(buf, pool); |
4783 | 35.6k | unlock_buffer_pool(pool); |
4784 | 35.6k | cm->ref_frame_map[ref_idx] = NULL; |
4785 | 35.6k | } |
4786 | | // If no corresponding buffer exists, allocate a new buffer with all |
4787 | | // pixels set to neutral grey. |
4788 | 182k | int buf_idx = get_free_fb(cm); |
4789 | 182k | if (buf_idx == INVALID_IDX) { |
4790 | 0 | aom_internal_error(&pbi->error, AOM_CODEC_MEM_ERROR, |
4791 | 0 | "Unable to find free frame buffer"); |
4792 | 0 | } |
4793 | 182k | buf = &frame_bufs[buf_idx]; |
4794 | 182k | lock_buffer_pool(pool); |
4795 | 182k | if (aom_realloc_frame_buffer( |
4796 | 182k | &buf->buf, seq_params->max_frame_width, |
4797 | 182k | seq_params->max_frame_height, seq_params->subsampling_x, |
4798 | 182k | seq_params->subsampling_y, seq_params->use_highbitdepth, |
4799 | 182k | AOM_BORDER_IN_PIXELS, features->byte_alignment, |
4800 | 182k | &buf->raw_frame_buffer, pool->get_fb_cb, pool->cb_priv, false, |
4801 | 182k | 0)) { |
4802 | 186 | decrease_ref_count(buf, pool); |
4803 | 186 | unlock_buffer_pool(pool); |
4804 | 186 | aom_internal_error(&pbi->error, AOM_CODEC_MEM_ERROR, |
4805 | 186 | "Failed to allocate frame buffer"); |
4806 | 186 | } |
4807 | 182k | unlock_buffer_pool(pool); |
4808 | | // According to the specification, valid bitstreams are required to |
4809 | | // never use missing reference frames so the filling process for |
4810 | | // missing frames is not normatively defined and RefValid for missing |
4811 | | // frames is set to 0. |
4812 | | |
4813 | | // To make libaom more robust when the bitstream has been corrupted |
4814 | | // by the loss of some frames of data, this code adds a neutral grey |
4815 | | // buffer in place of missing frames, i.e. |
4816 | | // |
4817 | 182k | set_planes_to_neutral_grey(seq_params, &buf->buf, 0); |
4818 | | // |
4819 | | // and allows the frames to be used for referencing, i.e. |
4820 | | // |
4821 | 182k | pbi->valid_for_referencing[ref_idx] = 1; |
4822 | | // |
4823 | | // Please note such behavior is not normative and other decoders may |
4824 | | // use a different approach. |
4825 | 182k | cm->ref_frame_map[ref_idx] = buf; |
4826 | 182k | buf->order_hint = order_hint; |
4827 | 182k | } |
4828 | 196k | } |
4829 | 24.7k | } |
4830 | 192k | } |
4831 | | |
4832 | 257k | if (current_frame->frame_type == KEY_FRAME) { |
4833 | 98.3k | setup_frame_size(cm, frame_size_override_flag, rb); |
4834 | | |
4835 | 98.3k | if (features->allow_screen_content_tools && !av1_superres_scaled(cm)) |
4836 | 48.5k | features->allow_intrabc = aom_rb_read_bit(rb); |
4837 | 98.3k | features->allow_ref_frame_mvs = 0; |
4838 | 98.3k | cm->prev_frame = NULL; |
4839 | 159k | } else { |
4840 | 159k | features->allow_ref_frame_mvs = 0; |
4841 | | |
4842 | 159k | if (current_frame->frame_type == INTRA_ONLY_FRAME) { |
4843 | 23.9k | cm->cur_frame->film_grain_params_present = |
4844 | 23.9k | seq_params->film_grain_params_present; |
4845 | 23.9k | setup_frame_size(cm, frame_size_override_flag, rb); |
4846 | 23.9k | if (features->allow_screen_content_tools && !av1_superres_scaled(cm)) |
4847 | 18.2k | features->allow_intrabc = aom_rb_read_bit(rb); |
4848 | | |
4849 | 135k | } else if (pbi->need_resync != 1) { /* Skip if need resync */ |
4850 | 100k | int frame_refs_short_signaling = 0; |
4851 | | // Frame refs short signaling is off when error resilient mode is on. |
4852 | 100k | if (seq_params->order_hint_info.enable_order_hint) |
4853 | 100k | frame_refs_short_signaling = aom_rb_read_bit(rb); |
4854 | | |
4855 | 100k | if (frame_refs_short_signaling) { |
4856 | | // == LAST_FRAME == |
4857 | 59.7k | const int lst_ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2); |
4858 | 59.7k | const RefCntBuffer *const lst_buf = cm->ref_frame_map[lst_ref]; |
4859 | | |
4860 | | // == GOLDEN_FRAME == |
4861 | 59.7k | const int gld_ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2); |
4862 | 59.7k | const RefCntBuffer *const gld_buf = cm->ref_frame_map[gld_ref]; |
4863 | | |
4864 | | // Most of the time, streams start with a keyframe. In that case, |
4865 | | // ref_frame_map will have been filled in at that point and will not |
4866 | | // contain any NULLs. However, streams are explicitly allowed to start |
4867 | | // with an intra-only frame, so long as they don't then signal a |
4868 | | // reference to a slot that hasn't been set yet. That's what we are |
4869 | | // checking here. |
4870 | 59.7k | if (lst_buf == NULL) |
4871 | 38 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
4872 | 38 | "Inter frame requests nonexistent reference"); |
4873 | 59.7k | if (gld_buf == NULL) |
4874 | 35 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
4875 | 35 | "Inter frame requests nonexistent reference"); |
4876 | | |
4877 | 59.7k | av1_set_frame_refs(cm, cm->remapped_ref_idx, lst_ref, gld_ref); |
4878 | 59.7k | } |
4879 | | |
4880 | 801k | for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) { |
4881 | 700k | int ref = 0; |
4882 | 700k | if (!frame_refs_short_signaling) { |
4883 | 285k | ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2); |
4884 | | |
4885 | | // Most of the time, streams start with a keyframe. In that case, |
4886 | | // ref_frame_map will have been filled in at that point and will not |
4887 | | // contain any NULLs. However, streams are explicitly allowed to start |
4888 | | // with an intra-only frame, so long as they don't then signal a |
4889 | | // reference to a slot that hasn't been set yet. That's what we are |
4890 | | // checking here. |
4891 | 285k | if (cm->ref_frame_map[ref] == NULL) |
4892 | 283 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
4893 | 283 | "Inter frame requests nonexistent reference"); |
4894 | 285k | cm->remapped_ref_idx[i] = ref; |
4895 | 414k | } else { |
4896 | 414k | ref = cm->remapped_ref_idx[i]; |
4897 | 414k | } |
4898 | | // Check valid for referencing |
4899 | 700k | if (pbi->valid_for_referencing[ref] == 0) |
4900 | 109 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
4901 | 109 | "Reference frame not valid for referencing"); |
4902 | | |
4903 | 700k | cm->ref_frame_sign_bias[LAST_FRAME + i] = 0; |
4904 | | |
4905 | 700k | if (seq_params->frame_id_numbers_present_flag) { |
4906 | 38 | int frame_id_length = seq_params->frame_id_length; |
4907 | 38 | int diff_len = seq_params->delta_frame_id_length; |
4908 | 38 | int delta_frame_id_minus_1 = aom_rb_read_literal(rb, diff_len); |
4909 | 38 | int ref_frame_id = |
4910 | 38 | ((cm->current_frame_id - (delta_frame_id_minus_1 + 1) + |
4911 | 38 | (1 << frame_id_length)) % |
4912 | 38 | (1 << frame_id_length)); |
4913 | | // Compare values derived from delta_frame_id_minus_1 and |
4914 | | // refresh_frame_flags. |
4915 | 38 | if (ref_frame_id != cm->ref_frame_id[ref]) |
4916 | 26 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
4917 | 26 | "Reference buffer frame ID mismatch"); |
4918 | 38 | } |
4919 | 700k | } |
4920 | | |
4921 | 100k | if (!features->error_resilient_mode && frame_size_override_flag) { |
4922 | 86.7k | setup_frame_size_with_refs(cm, rb); |
4923 | 86.7k | } else { |
4924 | 14.1k | setup_frame_size(cm, frame_size_override_flag, rb); |
4925 | 14.1k | } |
4926 | | |
4927 | 100k | if (features->cur_frame_force_integer_mv) { |
4928 | 1.79k | features->allow_high_precision_mv = 0; |
4929 | 99.0k | } else { |
4930 | 99.0k | features->allow_high_precision_mv = aom_rb_read_bit(rb); |
4931 | 99.0k | } |
4932 | 100k | features->interp_filter = read_frame_interp_filter(rb); |
4933 | 100k | features->switchable_motion_mode = aom_rb_read_bit(rb); |
4934 | 100k | } |
4935 | | |
4936 | 159k | cm->prev_frame = get_primary_ref_frame_buf(cm); |
4937 | 159k | if (features->primary_ref_frame != PRIMARY_REF_NONE && |
4938 | 159k | get_primary_ref_frame_buf(cm) == NULL) { |
4939 | 19.5k | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
4940 | 19.5k | "Reference frame containing this frame's initial " |
4941 | 19.5k | "frame context is unavailable."); |
4942 | 19.5k | } |
4943 | | |
4944 | 159k | if (!(current_frame->frame_type == INTRA_ONLY_FRAME) && |
4945 | 159k | pbi->need_resync != 1) { |
4946 | 95.6k | if (frame_might_allow_ref_frame_mvs(cm)) |
4947 | 91.0k | features->allow_ref_frame_mvs = aom_rb_read_bit(rb); |
4948 | 4.66k | else |
4949 | 4.66k | features->allow_ref_frame_mvs = 0; |
4950 | | |
4951 | 746k | for (int i = LAST_FRAME; i <= ALTREF_FRAME; ++i) { |
4952 | 651k | const RefCntBuffer *const ref_buf = get_ref_frame_buf(cm, i); |
4953 | 651k | struct scale_factors *const ref_scale_factors = |
4954 | 651k | get_ref_scale_factors(cm, i); |
4955 | 651k | av1_setup_scale_factors_for_frame( |
4956 | 651k | ref_scale_factors, ref_buf->buf.y_crop_width, |
4957 | 651k | ref_buf->buf.y_crop_height, cm->width, cm->height); |
4958 | 651k | if ((!av1_is_valid_scale(ref_scale_factors))) |
4959 | 4.49k | aom_internal_error(&pbi->error, AOM_CODEC_UNSUP_BITSTREAM, |
4960 | 4.49k | "Reference frame has invalid dimensions"); |
4961 | 651k | } |
4962 | 95.6k | } |
4963 | 159k | } |
4964 | | |
4965 | 257k | av1_setup_frame_buf_refs(cm); |
4966 | | |
4967 | 257k | av1_setup_frame_sign_bias(cm); |
4968 | | |
4969 | 257k | cm->cur_frame->frame_type = current_frame->frame_type; |
4970 | | |
4971 | 257k | update_ref_frame_id(pbi); |
4972 | | |
4973 | 257k | const int might_bwd_adapt = !(seq_params->reduced_still_picture_hdr) && |
4974 | 257k | !(features->disable_cdf_update); |
4975 | 257k | if (might_bwd_adapt) { |
4976 | 162k | features->refresh_frame_context = aom_rb_read_bit(rb) |
4977 | 162k | ? REFRESH_FRAME_CONTEXT_DISABLED |
4978 | 162k | : REFRESH_FRAME_CONTEXT_BACKWARD; |
4979 | 162k | } else { |
4980 | 95.5k | features->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED; |
4981 | 95.5k | } |
4982 | | |
4983 | 257k | cm->cur_frame->buf.bit_depth = seq_params->bit_depth; |
4984 | 257k | cm->cur_frame->buf.color_primaries = seq_params->color_primaries; |
4985 | 257k | cm->cur_frame->buf.transfer_characteristics = |
4986 | 257k | seq_params->transfer_characteristics; |
4987 | 257k | cm->cur_frame->buf.matrix_coefficients = seq_params->matrix_coefficients; |
4988 | 257k | cm->cur_frame->buf.monochrome = seq_params->monochrome; |
4989 | 257k | cm->cur_frame->buf.chroma_sample_position = |
4990 | 257k | seq_params->chroma_sample_position; |
4991 | 257k | cm->cur_frame->buf.color_range = seq_params->color_range; |
4992 | 257k | cm->cur_frame->buf.render_width = cm->render_width; |
4993 | 257k | cm->cur_frame->buf.render_height = cm->render_height; |
4994 | | |
4995 | 257k | if (pbi->need_resync) { |
4996 | 8.09k | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
4997 | 8.09k | "Keyframe / intra-only frame required to reset decoder" |
4998 | 8.09k | " state"); |
4999 | 8.09k | } |
5000 | | |
5001 | 257k | if (features->allow_intrabc) { |
5002 | | // Set parameters corresponding to no filtering. |
5003 | 35.4k | struct loopfilter *lf = &cm->lf; |
5004 | 35.4k | lf->filter_level[0] = 0; |
5005 | 35.4k | lf->filter_level[1] = 0; |
5006 | 35.4k | cm->cdef_info.cdef_bits = 0; |
5007 | 35.4k | cm->cdef_info.cdef_strengths[0] = 0; |
5008 | 35.4k | cm->cdef_info.nb_cdef_strengths = 1; |
5009 | 35.4k | cm->cdef_info.cdef_uv_strengths[0] = 0; |
5010 | 35.4k | cm->rst_info[0].frame_restoration_type = RESTORE_NONE; |
5011 | 35.4k | cm->rst_info[1].frame_restoration_type = RESTORE_NONE; |
5012 | 35.4k | cm->rst_info[2].frame_restoration_type = RESTORE_NONE; |
5013 | 35.4k | } |
5014 | | |
5015 | 257k | read_tile_info(pbi, rb); |
5016 | 257k | if (!av1_is_min_tile_width_satisfied(cm)) { |
5017 | 357 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
5018 | 357 | "Minimum tile width requirement not satisfied"); |
5019 | 357 | } |
5020 | | |
5021 | 257k | CommonQuantParams *const quant_params = &cm->quant_params; |
5022 | 257k | setup_quantization(quant_params, av1_num_planes(cm), |
5023 | 257k | cm->seq_params->separate_uv_delta_q, rb); |
5024 | 257k | xd->bd = (int)seq_params->bit_depth; |
5025 | | |
5026 | 257k | CommonContexts *const above_contexts = &cm->above_contexts; |
5027 | 257k | if (above_contexts->num_planes < av1_num_planes(cm) || |
5028 | 257k | above_contexts->num_mi_cols < cm->mi_params.mi_cols || |
5029 | 257k | above_contexts->num_tile_rows < cm->tiles.rows) { |
5030 | 18.3k | av1_free_above_context_buffers(above_contexts); |
5031 | 18.3k | if (av1_alloc_above_context_buffers(above_contexts, cm->tiles.rows, |
5032 | 18.3k | cm->mi_params.mi_cols, |
5033 | 18.3k | av1_num_planes(cm))) { |
5034 | 0 | aom_internal_error(&pbi->error, AOM_CODEC_MEM_ERROR, |
5035 | 0 | "Failed to allocate context buffers"); |
5036 | 0 | } |
5037 | 18.3k | } |
5038 | | |
5039 | 257k | if (features->primary_ref_frame == PRIMARY_REF_NONE) { |
5040 | 122k | av1_setup_past_independence(cm); |
5041 | 122k | } |
5042 | | |
5043 | 257k | setup_segmentation(cm, rb); |
5044 | | |
5045 | 257k | cm->delta_q_info.delta_q_res = 1; |
5046 | 257k | cm->delta_q_info.delta_lf_res = 1; |
5047 | 257k | cm->delta_q_info.delta_lf_present_flag = 0; |
5048 | 257k | cm->delta_q_info.delta_lf_multi = 0; |
5049 | 257k | cm->delta_q_info.delta_q_present_flag = |
5050 | 257k | quant_params->base_qindex > 0 ? aom_rb_read_bit(rb) : 0; |
5051 | 257k | if (cm->delta_q_info.delta_q_present_flag) { |
5052 | 18.9k | xd->current_base_qindex = quant_params->base_qindex; |
5053 | 18.9k | cm->delta_q_info.delta_q_res = 1 << aom_rb_read_literal(rb, 2); |
5054 | 18.9k | if (!features->allow_intrabc) |
5055 | 16.2k | cm->delta_q_info.delta_lf_present_flag = aom_rb_read_bit(rb); |
5056 | 18.9k | if (cm->delta_q_info.delta_lf_present_flag) { |
5057 | 9.11k | cm->delta_q_info.delta_lf_res = 1 << aom_rb_read_literal(rb, 2); |
5058 | 9.11k | cm->delta_q_info.delta_lf_multi = aom_rb_read_bit(rb); |
5059 | 9.11k | av1_reset_loop_filter_delta(xd, av1_num_planes(cm)); |
5060 | 9.11k | } |
5061 | 18.9k | } |
5062 | | |
5063 | 257k | xd->cur_frame_force_integer_mv = features->cur_frame_force_integer_mv; |
5064 | | |
5065 | 1.90M | for (int i = 0; i < MAX_SEGMENTS; ++i) { |
5066 | 1.65M | const int qindex = av1_get_qindex(&cm->seg, i, quant_params->base_qindex); |
5067 | 1.65M | xd->lossless[i] = |
5068 | 1.65M | qindex == 0 && quant_params->y_dc_delta_q == 0 && |
5069 | 1.65M | quant_params->u_dc_delta_q == 0 && quant_params->u_ac_delta_q == 0 && |
5070 | 1.65M | quant_params->v_dc_delta_q == 0 && quant_params->v_ac_delta_q == 0; |
5071 | 1.65M | xd->qindex[i] = qindex; |
5072 | 1.65M | } |
5073 | 257k | features->coded_lossless = is_coded_lossless(cm, xd); |
5074 | 257k | features->all_lossless = features->coded_lossless && !av1_superres_scaled(cm); |
5075 | 257k | setup_segmentation_dequant(cm, xd); |
5076 | 257k | if (features->coded_lossless) { |
5077 | 23.5k | cm->lf.filter_level[0] = 0; |
5078 | 23.5k | cm->lf.filter_level[1] = 0; |
5079 | 23.5k | } |
5080 | 257k | if (features->coded_lossless || !seq_params->enable_cdef) { |
5081 | 87.2k | cm->cdef_info.cdef_bits = 0; |
5082 | 87.2k | cm->cdef_info.cdef_strengths[0] = 0; |
5083 | 87.2k | cm->cdef_info.cdef_uv_strengths[0] = 0; |
5084 | 87.2k | } |
5085 | 257k | if (features->all_lossless || !seq_params->enable_restoration) { |
5086 | 75.9k | cm->rst_info[0].frame_restoration_type = RESTORE_NONE; |
5087 | 75.9k | cm->rst_info[1].frame_restoration_type = RESTORE_NONE; |
5088 | 75.9k | cm->rst_info[2].frame_restoration_type = RESTORE_NONE; |
5089 | 75.9k | } |
5090 | 257k | setup_loopfilter(cm, rb); |
5091 | | |
5092 | 257k | if (!features->coded_lossless && seq_params->enable_cdef) { |
5093 | 119k | setup_cdef(cm, rb); |
5094 | 119k | } |
5095 | 257k | if (!features->all_lossless && seq_params->enable_restoration) { |
5096 | 129k | decode_restoration_mode(cm, rb); |
5097 | 129k | } |
5098 | | |
5099 | 257k | features->tx_mode = read_tx_mode(rb, features->coded_lossless); |
5100 | 257k | current_frame->reference_mode = read_frame_reference_mode(cm, rb); |
5101 | | |
5102 | 257k | av1_setup_skip_mode_allowed(cm); |
5103 | 257k | current_frame->skip_mode_info.skip_mode_flag = |
5104 | 257k | current_frame->skip_mode_info.skip_mode_allowed ? aom_rb_read_bit(rb) : 0; |
5105 | | |
5106 | 257k | if (frame_might_allow_warped_motion(cm)) |
5107 | 80.7k | features->allow_warped_motion = aom_rb_read_bit(rb); |
5108 | 177k | else |
5109 | 177k | features->allow_warped_motion = 0; |
5110 | | |
5111 | 257k | features->reduced_tx_set_used = aom_rb_read_bit(rb); |
5112 | | |
5113 | 257k | if (features->allow_ref_frame_mvs && !frame_might_allow_ref_frame_mvs(cm)) { |
5114 | 0 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
5115 | 0 | "Frame wrongly requests reference frame MVs"); |
5116 | 0 | } |
5117 | | |
5118 | 257k | if (!frame_is_intra_only(cm)) read_global_motion(cm, rb); |
5119 | | |
5120 | 257k | cm->cur_frame->film_grain_params_present = |
5121 | 257k | seq_params->film_grain_params_present; |
5122 | 257k | read_film_grain(cm, rb); |
5123 | | |
5124 | 257k | #if EXT_TILE_DEBUG |
5125 | 257k | if (pbi->ext_tile_debug && cm->tiles.large_scale) { |
5126 | 13.5k | read_ext_tile_info(pbi, rb); |
5127 | 13.5k | av1_set_single_tile_decoding_mode(cm); |
5128 | 13.5k | } |
5129 | 257k | #endif // EXT_TILE_DEBUG |
5130 | 257k | return 0; |
5131 | 266k | } |
5132 | | |
5133 | | struct aom_read_bit_buffer *av1_init_read_bit_buffer( |
5134 | | AV1Decoder *pbi, struct aom_read_bit_buffer *rb, const uint8_t *data, |
5135 | 534k | const uint8_t *data_end) { |
5136 | 534k | rb->bit_offset = 0; |
5137 | 534k | rb->error_handler = error_handler; |
5138 | 534k | rb->error_handler_data = &pbi->common; |
5139 | 534k | rb->bit_buffer = data; |
5140 | 534k | rb->bit_buffer_end = data_end; |
5141 | 534k | return rb; |
5142 | 534k | } |
5143 | | |
5144 | | void av1_read_frame_size(struct aom_read_bit_buffer *rb, int num_bits_width, |
5145 | 84.6k | int num_bits_height, int *width, int *height) { |
5146 | 84.6k | *width = aom_rb_read_literal(rb, num_bits_width) + 1; |
5147 | 84.6k | *height = aom_rb_read_literal(rb, num_bits_height) + 1; |
5148 | 84.6k | } |
5149 | | |
5150 | 188k | BITSTREAM_PROFILE av1_read_profile(struct aom_read_bit_buffer *rb) { |
5151 | 188k | int profile = aom_rb_read_literal(rb, PROFILE_BITS); |
5152 | 188k | return (BITSTREAM_PROFILE)profile; |
5153 | 188k | } |
5154 | | |
5155 | 42.1k | static AOM_INLINE void superres_post_decode(AV1Decoder *pbi) { |
5156 | 42.1k | AV1_COMMON *const cm = &pbi->common; |
5157 | 42.1k | BufferPool *const pool = cm->buffer_pool; |
5158 | | |
5159 | 42.1k | if (!av1_superres_scaled(cm)) return; |
5160 | 10.3k | assert(!cm->features.all_lossless); |
5161 | | |
5162 | 10.3k | av1_superres_upscale(cm, pool, 0); |
5163 | 10.3k | } |
5164 | | |
5165 | | uint32_t av1_decode_frame_headers_and_setup(AV1Decoder *pbi, |
5166 | | struct aom_read_bit_buffer *rb, |
5167 | 266k | int trailing_bits_present) { |
5168 | 266k | AV1_COMMON *const cm = &pbi->common; |
5169 | 266k | const int num_planes = av1_num_planes(cm); |
5170 | 266k | MACROBLOCKD *const xd = &pbi->dcb.xd; |
5171 | | |
5172 | | #if CONFIG_BITSTREAM_DEBUG |
5173 | | if (cm->seq_params->order_hint_info.enable_order_hint) { |
5174 | | aom_bitstream_queue_set_frame_read(cm->current_frame.order_hint * 2 + |
5175 | | cm->show_frame); |
5176 | | } else { |
5177 | | // This is currently used in RTC encoding. cm->show_frame is always 1. |
5178 | | assert(cm->show_frame); |
5179 | | aom_bitstream_queue_set_frame_read(cm->current_frame.frame_number); |
5180 | | } |
5181 | | #endif |
5182 | | #if CONFIG_MISMATCH_DEBUG |
5183 | | mismatch_move_frame_idx_r(); |
5184 | | #endif |
5185 | | |
5186 | 2.12M | for (int i = LAST_FRAME; i <= ALTREF_FRAME; ++i) { |
5187 | 1.86M | cm->global_motion[i] = default_warp_params; |
5188 | 1.86M | cm->cur_frame->global_motion[i] = default_warp_params; |
5189 | 1.86M | } |
5190 | 266k | xd->global_motion = cm->global_motion; |
5191 | | |
5192 | 266k | read_uncompressed_header(pbi, rb); |
5193 | | |
5194 | 266k | if (trailing_bits_present) av1_check_trailing_bits(pbi, rb); |
5195 | | |
5196 | 266k | if (!cm->tiles.single_tile_decoding && |
5197 | 266k | (pbi->dec_tile_row >= 0 || pbi->dec_tile_col >= 0)) { |
5198 | 0 | pbi->dec_tile_row = -1; |
5199 | 0 | pbi->dec_tile_col = -1; |
5200 | 0 | } |
5201 | | |
5202 | 266k | const uint32_t uncomp_hdr_size = |
5203 | 266k | (uint32_t)aom_rb_bytes_read(rb); // Size of the uncompressed header |
5204 | 266k | YV12_BUFFER_CONFIG *new_fb = &cm->cur_frame->buf; |
5205 | 266k | xd->cur_buf = new_fb; |
5206 | 266k | if (av1_allow_intrabc(cm)) { |
5207 | 34.8k | av1_setup_scale_factors_for_frame( |
5208 | 34.8k | &cm->sf_identity, xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height, |
5209 | 34.8k | xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height); |
5210 | 34.8k | } |
5211 | | |
5212 | | // Showing a frame directly. |
5213 | 266k | if (cm->show_existing_frame) { |
5214 | 789 | if (pbi->reset_decoder_state) { |
5215 | | // Use the default frame context values. |
5216 | 167 | *cm->fc = *cm->default_frame_context; |
5217 | 167 | if (!cm->fc->initialized) |
5218 | 0 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
5219 | 0 | "Uninitialized entropy context."); |
5220 | 167 | } |
5221 | 789 | return uncomp_hdr_size; |
5222 | 789 | } |
5223 | | |
5224 | 265k | cm->mi_params.setup_mi(&cm->mi_params); |
5225 | | |
5226 | 265k | av1_calculate_ref_frame_side(cm); |
5227 | 265k | if (cm->features.allow_ref_frame_mvs) av1_setup_motion_field(cm); |
5228 | | |
5229 | 265k | av1_setup_block_planes(xd, cm->seq_params->subsampling_x, |
5230 | 265k | cm->seq_params->subsampling_y, num_planes); |
5231 | 265k | if (cm->features.primary_ref_frame == PRIMARY_REF_NONE) { |
5232 | | // use the default frame context values |
5233 | 115k | *cm->fc = *cm->default_frame_context; |
5234 | 149k | } else { |
5235 | 149k | *cm->fc = get_primary_ref_frame_buf(cm)->frame_context; |
5236 | 149k | } |
5237 | 265k | if (!cm->fc->initialized) |
5238 | 40 | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
5239 | 40 | "Uninitialized entropy context."); |
5240 | | |
5241 | 265k | pbi->dcb.corrupted = 0; |
5242 | 265k | return uncomp_hdr_size; |
5243 | 266k | } |
5244 | | |
5245 | | // Once-per-frame initialization |
5246 | 189k | static AOM_INLINE void setup_frame_info(AV1Decoder *pbi) { |
5247 | 189k | AV1_COMMON *const cm = &pbi->common; |
5248 | | |
5249 | 189k | if (cm->rst_info[0].frame_restoration_type != RESTORE_NONE || |
5250 | 189k | cm->rst_info[1].frame_restoration_type != RESTORE_NONE || |
5251 | 189k | cm->rst_info[2].frame_restoration_type != RESTORE_NONE) { |
5252 | 26.3k | av1_alloc_restoration_buffers(cm, /*is_sgr_enabled =*/true); |
5253 | 101k | for (int p = 0; p < av1_num_planes(cm); p++) { |
5254 | 74.8k | av1_alloc_restoration_struct(cm, &cm->rst_info[p], p > 0); |
5255 | 74.8k | } |
5256 | 26.3k | } |
5257 | | |
5258 | 189k | const int use_highbd = cm->seq_params->use_highbitdepth; |
5259 | 189k | const int buf_size = MC_TEMP_BUF_PELS << use_highbd; |
5260 | 189k | if (pbi->td.mc_buf_size != buf_size) { |
5261 | 15.6k | av1_free_mc_tmp_buf(&pbi->td); |
5262 | 15.6k | allocate_mc_tmp_buf(cm, &pbi->td, buf_size, use_highbd); |
5263 | 15.6k | } |
5264 | 189k | } |
5265 | | |
5266 | | void av1_decode_tg_tiles_and_wrapup(AV1Decoder *pbi, const uint8_t *data, |
5267 | | const uint8_t *data_end, |
5268 | | const uint8_t **p_data_end, int start_tile, |
5269 | 189k | int end_tile, int initialize_flag) { |
5270 | 189k | AV1_COMMON *const cm = &pbi->common; |
5271 | 189k | CommonTileParams *const tiles = &cm->tiles; |
5272 | 189k | MACROBLOCKD *const xd = &pbi->dcb.xd; |
5273 | 189k | const int tile_count_tg = end_tile - start_tile + 1; |
5274 | | |
5275 | 189k | xd->error_info = cm->error; |
5276 | 189k | if (initialize_flag) setup_frame_info(pbi); |
5277 | 189k | const int num_planes = av1_num_planes(cm); |
5278 | | |
5279 | 189k | if (pbi->max_threads > 1 && !(tiles->large_scale && !pbi->ext_tile_debug) && |
5280 | 189k | pbi->row_mt) |
5281 | 124k | *p_data_end = |
5282 | 124k | decode_tiles_row_mt(pbi, data, data_end, start_tile, end_tile); |
5283 | 65.5k | else if (pbi->max_threads > 1 && tile_count_tg > 1 && |
5284 | 65.5k | !(tiles->large_scale && !pbi->ext_tile_debug)) |
5285 | 0 | *p_data_end = decode_tiles_mt(pbi, data, data_end, start_tile, end_tile); |
5286 | 65.5k | else |
5287 | 65.5k | *p_data_end = decode_tiles(pbi, data, data_end, start_tile, end_tile); |
5288 | | |
5289 | | // If the bit stream is monochrome, set the U and V buffers to a constant. |
5290 | 189k | if (num_planes < 3) { |
5291 | 5.39k | set_planes_to_neutral_grey(cm->seq_params, xd->cur_buf, 1); |
5292 | 5.39k | } |
5293 | | |
5294 | 189k | if (end_tile != tiles->rows * tiles->cols - 1) { |
5295 | 27 | return; |
5296 | 27 | } |
5297 | | |
5298 | 189k | av1_alloc_cdef_buffers(cm, &pbi->cdef_worker, &pbi->cdef_sync, |
5299 | 189k | pbi->num_workers, 1); |
5300 | 189k | av1_alloc_cdef_sync(cm, &pbi->cdef_sync, pbi->num_workers); |
5301 | | |
5302 | 189k | if (!cm->features.allow_intrabc && !tiles->single_tile_decoding) { |
5303 | 126k | if (cm->lf.filter_level[0] || cm->lf.filter_level[1]) { |
5304 | 35.5k | av1_loop_filter_frame_mt(&cm->cur_frame->buf, cm, &pbi->dcb.xd, 0, |
5305 | 35.5k | num_planes, 0, pbi->tile_workers, |
5306 | 35.5k | pbi->num_workers, &pbi->lf_row_sync, 0); |
5307 | 35.5k | } |
5308 | | |
5309 | 126k | const int do_cdef = |
5310 | 126k | !pbi->skip_loop_filter && !cm->features.coded_lossless && |
5311 | 126k | (cm->cdef_info.cdef_bits || cm->cdef_info.cdef_strengths[0] || |
5312 | 114k | cm->cdef_info.cdef_uv_strengths[0]); |
5313 | 126k | const int do_superres = av1_superres_scaled(cm); |
5314 | 126k | const int optimized_loop_restoration = !do_cdef && !do_superres; |
5315 | 126k | const int do_loop_restoration = |
5316 | 126k | cm->rst_info[0].frame_restoration_type != RESTORE_NONE || |
5317 | 126k | cm->rst_info[1].frame_restoration_type != RESTORE_NONE || |
5318 | 126k | cm->rst_info[2].frame_restoration_type != RESTORE_NONE; |
5319 | | // Frame border extension is not required in the decoder |
5320 | | // as it happens in extend_mc_border(). |
5321 | 126k | int do_extend_border_mt = 0; |
5322 | 126k | if (!optimized_loop_restoration) { |
5323 | 42.1k | if (do_loop_restoration) |
5324 | 17.3k | av1_loop_restoration_save_boundary_lines(&pbi->common.cur_frame->buf, |
5325 | 17.3k | cm, 0); |
5326 | | |
5327 | 42.1k | if (do_cdef) { |
5328 | 34.5k | if (pbi->num_workers > 1) { |
5329 | 28.1k | av1_cdef_frame_mt(cm, &pbi->dcb.xd, pbi->cdef_worker, |
5330 | 28.1k | pbi->tile_workers, &pbi->cdef_sync, |
5331 | 28.1k | pbi->num_workers, av1_cdef_init_fb_row_mt, |
5332 | 28.1k | do_extend_border_mt); |
5333 | 28.1k | } else { |
5334 | 6.42k | av1_cdef_frame(&pbi->common.cur_frame->buf, cm, &pbi->dcb.xd, |
5335 | 6.42k | av1_cdef_init_fb_row); |
5336 | 6.42k | } |
5337 | 34.5k | } |
5338 | | |
5339 | 42.1k | superres_post_decode(pbi); |
5340 | | |
5341 | 42.1k | if (do_loop_restoration) { |
5342 | 17.3k | av1_loop_restoration_save_boundary_lines(&pbi->common.cur_frame->buf, |
5343 | 17.3k | cm, 1); |
5344 | 17.3k | if (pbi->num_workers > 1) { |
5345 | 12.0k | av1_loop_restoration_filter_frame_mt( |
5346 | 12.0k | (YV12_BUFFER_CONFIG *)xd->cur_buf, cm, optimized_loop_restoration, |
5347 | 12.0k | pbi->tile_workers, pbi->num_workers, &pbi->lr_row_sync, |
5348 | 12.0k | &pbi->lr_ctxt, do_extend_border_mt); |
5349 | 12.0k | } else { |
5350 | 5.28k | av1_loop_restoration_filter_frame((YV12_BUFFER_CONFIG *)xd->cur_buf, |
5351 | 5.28k | cm, optimized_loop_restoration, |
5352 | 5.28k | &pbi->lr_ctxt); |
5353 | 5.28k | } |
5354 | 17.3k | } |
5355 | 84.6k | } else { |
5356 | | // In no cdef and no superres case. Provide an optimized version of |
5357 | | // loop_restoration_filter. |
5358 | 84.6k | if (do_loop_restoration) { |
5359 | 3.94k | if (pbi->num_workers > 1) { |
5360 | 1.77k | av1_loop_restoration_filter_frame_mt( |
5361 | 1.77k | (YV12_BUFFER_CONFIG *)xd->cur_buf, cm, optimized_loop_restoration, |
5362 | 1.77k | pbi->tile_workers, pbi->num_workers, &pbi->lr_row_sync, |
5363 | 1.77k | &pbi->lr_ctxt, do_extend_border_mt); |
5364 | 2.17k | } else { |
5365 | 2.17k | av1_loop_restoration_filter_frame((YV12_BUFFER_CONFIG *)xd->cur_buf, |
5366 | 2.17k | cm, optimized_loop_restoration, |
5367 | 2.17k | &pbi->lr_ctxt); |
5368 | 2.17k | } |
5369 | 3.94k | } |
5370 | 84.6k | } |
5371 | 126k | } |
5372 | | |
5373 | 189k | if (!pbi->dcb.corrupted) { |
5374 | 135k | if (cm->features.refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) { |
5375 | 26.5k | assert(pbi->context_update_tile_id < pbi->allocated_tiles); |
5376 | 26.5k | *cm->fc = pbi->tile_data[pbi->context_update_tile_id].tctx; |
5377 | 26.5k | av1_reset_cdf_symbol_counters(cm->fc); |
5378 | 26.5k | } |
5379 | 135k | } else { |
5380 | 54.6k | aom_internal_error(&pbi->error, AOM_CODEC_CORRUPT_FRAME, |
5381 | 54.6k | "Decode failed. Frame data is corrupted."); |
5382 | 54.6k | } |
5383 | | |
5384 | | #if CONFIG_INSPECTION |
5385 | | if (pbi->inspect_cb != NULL) { |
5386 | | (*pbi->inspect_cb)(pbi, pbi->inspect_ctx); |
5387 | | } |
5388 | | #endif |
5389 | | |
5390 | | // Non frame parallel update frame context here. |
5391 | 189k | if (!tiles->large_scale) { |
5392 | 129k | cm->cur_frame->frame_context = *cm->fc; |
5393 | 129k | } |
5394 | | |
5395 | 189k | if (cm->show_frame && !cm->seq_params->order_hint_info.enable_order_hint) { |
5396 | 9.23k | ++cm->current_frame.frame_number; |
5397 | 9.23k | } |
5398 | 189k | } |