Line data Source code
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 <stdlib.h> // qsort()
14 :
15 : #include "./aom_config.h"
16 : #include "./aom_dsp_rtcd.h"
17 : #include "./aom_scale_rtcd.h"
18 : #include "./av1_rtcd.h"
19 :
20 : #include "aom/aom_codec.h"
21 : #include "aom_dsp/aom_dsp_common.h"
22 : #include "aom_dsp/bitreader.h"
23 : #include "aom_dsp/bitreader_buffer.h"
24 : #include "aom_dsp/binary_codes_reader.h"
25 : #include "aom_mem/aom_mem.h"
26 : #include "aom_ports/mem.h"
27 : #include "aom_ports/mem_ops.h"
28 : #include "aom_scale/aom_scale.h"
29 : #include "aom_util/aom_thread.h"
30 :
31 : #if CONFIG_BITSTREAM_DEBUG
32 : #include "aom_util/debug_util.h"
33 : #endif // CONFIG_BITSTREAM_DEBUG
34 :
35 : #include "av1/common/alloccommon.h"
36 : #if CONFIG_CDEF
37 : #include "av1/common/cdef.h"
38 : #include "av1/common/clpf.h"
39 : #endif
40 : #if CONFIG_INSPECTION
41 : #include "av1/decoder/inspection.h"
42 : #endif
43 : #include "av1/common/common.h"
44 : #include "av1/common/entropy.h"
45 : #include "av1/common/entropymode.h"
46 : #include "av1/common/entropymv.h"
47 : #include "av1/common/idct.h"
48 : #include "av1/common/pred_common.h"
49 : #include "av1/common/quant_common.h"
50 : #include "av1/common/reconinter.h"
51 : #include "av1/common/reconintra.h"
52 : #include "av1/common/seg_common.h"
53 : #include "av1/common/thread_common.h"
54 : #include "av1/common/tile_common.h"
55 :
56 : #include "av1/decoder/decodeframe.h"
57 : #include "av1/decoder/decodemv.h"
58 : #include "av1/decoder/decoder.h"
59 : #if CONFIG_LV_MAP
60 : #include "av1/decoder/decodetxb.h"
61 : #endif
62 : #include "av1/decoder/detokenize.h"
63 : #include "av1/decoder/dsubexp.h"
64 :
65 : #if CONFIG_WARPED_MOTION || CONFIG_GLOBAL_MOTION
66 : #include "av1/common/warped_motion.h"
67 : #endif // CONFIG_WARPED_MOTION || CONFIG_GLOBAL_MOTION
68 :
69 : #define MAX_AV1_HEADER_SIZE 80
70 : #define ACCT_STR __func__
71 :
72 : #if CONFIG_PVQ
73 : #include "av1/common/partition.h"
74 : #include "av1/common/pvq.h"
75 : #include "av1/common/scan.h"
76 : #include "av1/decoder/decint.h"
77 : #include "av1/decoder/pvq_decoder.h"
78 : #include "av1/encoder/encodemb.h"
79 : #include "av1/encoder/hybrid_fwd_txfm.h"
80 : #endif
81 :
82 : #if CONFIG_CFL
83 : #include "av1/common/cfl.h"
84 : #endif
85 :
86 : static struct aom_read_bit_buffer *init_read_bit_buffer(
87 : AV1Decoder *pbi, struct aom_read_bit_buffer *rb, const uint8_t *data,
88 : const uint8_t *data_end, uint8_t clear_data[MAX_AV1_HEADER_SIZE]);
89 : static int read_compressed_header(AV1Decoder *pbi, const uint8_t *data,
90 : size_t partition_size);
91 : static size_t read_uncompressed_header(AV1Decoder *pbi,
92 : struct aom_read_bit_buffer *rb);
93 :
94 0 : static int is_compound_reference_allowed(const AV1_COMMON *cm) {
95 : #if CONFIG_ONE_SIDED_COMPOUND // Normative in decoder
96 0 : return !frame_is_intra_only(cm);
97 : #else
98 : int i;
99 : if (frame_is_intra_only(cm)) return 0;
100 : for (i = 1; i < INTER_REFS_PER_FRAME; ++i)
101 : if (cm->ref_frame_sign_bias[i + 1] != cm->ref_frame_sign_bias[1]) return 1;
102 :
103 : return 0;
104 : #endif
105 : }
106 :
107 0 : static void setup_compound_reference_mode(AV1_COMMON *cm) {
108 : #if CONFIG_EXT_REFS
109 0 : cm->comp_fwd_ref[0] = LAST_FRAME;
110 0 : cm->comp_fwd_ref[1] = LAST2_FRAME;
111 0 : cm->comp_fwd_ref[2] = LAST3_FRAME;
112 0 : cm->comp_fwd_ref[3] = GOLDEN_FRAME;
113 :
114 0 : cm->comp_bwd_ref[0] = BWDREF_FRAME;
115 0 : cm->comp_bwd_ref[1] = ALTREF_FRAME;
116 : #else
117 : if (cm->ref_frame_sign_bias[LAST_FRAME] ==
118 : cm->ref_frame_sign_bias[GOLDEN_FRAME]) {
119 : cm->comp_fixed_ref = ALTREF_FRAME;
120 : cm->comp_var_ref[0] = LAST_FRAME;
121 : cm->comp_var_ref[1] = GOLDEN_FRAME;
122 : } else if (cm->ref_frame_sign_bias[LAST_FRAME] ==
123 : cm->ref_frame_sign_bias[ALTREF_FRAME]) {
124 : cm->comp_fixed_ref = GOLDEN_FRAME;
125 : cm->comp_var_ref[0] = LAST_FRAME;
126 : cm->comp_var_ref[1] = ALTREF_FRAME;
127 : } else {
128 : cm->comp_fixed_ref = LAST_FRAME;
129 : cm->comp_var_ref[0] = GOLDEN_FRAME;
130 : cm->comp_var_ref[1] = ALTREF_FRAME;
131 : }
132 : #endif // CONFIG_EXT_REFS
133 0 : }
134 :
135 0 : static int read_is_valid(const uint8_t *start, size_t len, const uint8_t *end) {
136 0 : return len != 0 && len <= (size_t)(end - start);
137 : }
138 :
139 0 : static int decode_unsigned_max(struct aom_read_bit_buffer *rb, int max) {
140 0 : const int data = aom_rb_read_literal(rb, get_unsigned_bits(max));
141 0 : return data > max ? max : data;
142 : }
143 :
144 0 : static TX_MODE read_tx_mode(AV1_COMMON *cm, MACROBLOCKD *xd,
145 : struct aom_read_bit_buffer *rb) {
146 0 : int i, all_lossless = 1;
147 : #if CONFIG_TX64X64
148 : TX_MODE tx_mode;
149 : #endif
150 :
151 0 : if (cm->seg.enabled) {
152 0 : for (i = 0; i < MAX_SEGMENTS; ++i) {
153 0 : if (!xd->lossless[i]) {
154 0 : all_lossless = 0;
155 0 : break;
156 : }
157 : }
158 : } else {
159 0 : all_lossless = xd->lossless[0];
160 : }
161 :
162 0 : if (all_lossless) return ONLY_4X4;
163 : #if CONFIG_TX64X64
164 : tx_mode = aom_rb_read_bit(rb) ? TX_MODE_SELECT : aom_rb_read_literal(rb, 2);
165 : if (tx_mode == ALLOW_32X32) tx_mode += aom_rb_read_bit(rb);
166 : return tx_mode;
167 : #else
168 0 : return aom_rb_read_bit(rb) ? TX_MODE_SELECT : aom_rb_read_literal(rb, 2);
169 : #endif // CONFIG_TX64X64
170 : }
171 :
172 : #if !CONFIG_EC_ADAPT
173 : static void read_tx_size_probs(FRAME_CONTEXT *fc, aom_reader *r) {
174 : int i, j, k;
175 : for (i = 0; i < MAX_TX_DEPTH; ++i)
176 : for (j = 0; j < TX_SIZE_CONTEXTS; ++j)
177 : for (k = 0; k < i + 1; ++k)
178 : av1_diff_update_prob(r, &fc->tx_size_probs[i][j][k], ACCT_STR);
179 : }
180 : #endif
181 :
182 : #if !CONFIG_EC_ADAPT
183 : static void read_switchable_interp_probs(FRAME_CONTEXT *fc, aom_reader *r) {
184 : int i, j;
185 : for (j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j) {
186 : for (i = 0; i < SWITCHABLE_FILTERS - 1; ++i)
187 : av1_diff_update_prob(r, &fc->switchable_interp_prob[j][i], ACCT_STR);
188 : }
189 : }
190 : #endif
191 :
192 0 : static void read_inter_mode_probs(FRAME_CONTEXT *fc, aom_reader *r) {
193 : int i;
194 0 : for (i = 0; i < NEWMV_MODE_CONTEXTS; ++i)
195 0 : av1_diff_update_prob(r, &fc->newmv_prob[i], ACCT_STR);
196 0 : for (i = 0; i < ZEROMV_MODE_CONTEXTS; ++i)
197 0 : av1_diff_update_prob(r, &fc->zeromv_prob[i], ACCT_STR);
198 0 : for (i = 0; i < REFMV_MODE_CONTEXTS; ++i)
199 0 : av1_diff_update_prob(r, &fc->refmv_prob[i], ACCT_STR);
200 0 : for (i = 0; i < DRL_MODE_CONTEXTS; ++i)
201 0 : av1_diff_update_prob(r, &fc->drl_prob[i], ACCT_STR);
202 0 : }
203 :
204 : #if CONFIG_EXT_INTER
205 0 : static void read_inter_compound_mode_probs(FRAME_CONTEXT *fc, aom_reader *r) {
206 : int i, j;
207 0 : if (aom_read(r, GROUP_DIFF_UPDATE_PROB, ACCT_STR)) {
208 0 : for (j = 0; j < INTER_MODE_CONTEXTS; ++j) {
209 0 : for (i = 0; i < INTER_COMPOUND_MODES - 1; ++i) {
210 0 : av1_diff_update_prob(r, &fc->inter_compound_mode_probs[j][i], ACCT_STR);
211 : }
212 : }
213 : }
214 0 : }
215 : #endif // CONFIG_EXT_INTER
216 : #if !CONFIG_EC_ADAPT
217 : #if !CONFIG_EXT_TX
218 : static void read_ext_tx_probs(FRAME_CONTEXT *fc, aom_reader *r) {
219 : int i, j, k;
220 : if (aom_read(r, GROUP_DIFF_UPDATE_PROB, ACCT_STR)) {
221 : for (i = TX_4X4; i < EXT_TX_SIZES; ++i) {
222 : for (j = 0; j < TX_TYPES; ++j) {
223 : for (k = 0; k < TX_TYPES - 1; ++k)
224 : av1_diff_update_prob(r, &fc->intra_ext_tx_prob[i][j][k], ACCT_STR);
225 : }
226 : }
227 : }
228 : if (aom_read(r, GROUP_DIFF_UPDATE_PROB, ACCT_STR)) {
229 : for (i = TX_4X4; i < EXT_TX_SIZES; ++i) {
230 : for (k = 0; k < TX_TYPES - 1; ++k)
231 : av1_diff_update_prob(r, &fc->inter_ext_tx_prob[i][k], ACCT_STR);
232 : }
233 : }
234 : }
235 : #endif
236 : #endif
237 :
238 0 : static REFERENCE_MODE read_frame_reference_mode(
239 : const AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
240 0 : if (is_compound_reference_allowed(cm)) {
241 : #if CONFIG_REF_ADAPT
242 : return aom_rb_read_bit(rb) ? REFERENCE_MODE_SELECT : SINGLE_REFERENCE;
243 : #else
244 0 : return aom_rb_read_bit(rb)
245 : ? REFERENCE_MODE_SELECT
246 0 : : (aom_rb_read_bit(rb) ? COMPOUND_REFERENCE : SINGLE_REFERENCE);
247 : #endif // CONFIG_REF_ADAPT
248 : } else {
249 0 : return SINGLE_REFERENCE;
250 : }
251 : }
252 :
253 0 : static void read_frame_reference_mode_probs(AV1_COMMON *cm, aom_reader *r) {
254 0 : FRAME_CONTEXT *const fc = cm->fc;
255 : int i, j;
256 :
257 0 : if (cm->reference_mode == REFERENCE_MODE_SELECT)
258 0 : for (i = 0; i < COMP_INTER_CONTEXTS; ++i)
259 0 : av1_diff_update_prob(r, &fc->comp_inter_prob[i], ACCT_STR);
260 :
261 0 : if (cm->reference_mode != COMPOUND_REFERENCE) {
262 0 : for (i = 0; i < REF_CONTEXTS; ++i) {
263 0 : for (j = 0; j < (SINGLE_REFS - 1); ++j) {
264 0 : av1_diff_update_prob(r, &fc->single_ref_prob[i][j], ACCT_STR);
265 : }
266 : }
267 : }
268 :
269 0 : if (cm->reference_mode != SINGLE_REFERENCE) {
270 0 : for (i = 0; i < REF_CONTEXTS; ++i) {
271 : #if CONFIG_EXT_REFS
272 0 : for (j = 0; j < (FWD_REFS - 1); ++j)
273 0 : av1_diff_update_prob(r, &fc->comp_ref_prob[i][j], ACCT_STR);
274 0 : for (j = 0; j < (BWD_REFS - 1); ++j)
275 0 : av1_diff_update_prob(r, &fc->comp_bwdref_prob[i][j], ACCT_STR);
276 : #else
277 : for (j = 0; j < (COMP_REFS - 1); ++j)
278 : av1_diff_update_prob(r, &fc->comp_ref_prob[i][j], ACCT_STR);
279 : #endif // CONFIG_EXT_REFS
280 : }
281 : }
282 0 : }
283 :
284 0 : static void update_mv_probs(aom_prob *p, int n, aom_reader *r) {
285 : int i;
286 0 : for (i = 0; i < n; ++i) av1_diff_update_prob(r, &p[i], ACCT_STR);
287 0 : }
288 :
289 0 : static void read_mv_probs(nmv_context *ctx, int allow_hp, aom_reader *r) {
290 : int i;
291 :
292 : #if !CONFIG_EC_ADAPT
293 : int j;
294 : update_mv_probs(ctx->joints, MV_JOINTS - 1, r);
295 :
296 : for (i = 0; i < 2; ++i) {
297 : nmv_component *const comp_ctx = &ctx->comps[i];
298 : update_mv_probs(&comp_ctx->sign, 1, r);
299 : update_mv_probs(comp_ctx->classes, MV_CLASSES - 1, r);
300 : update_mv_probs(comp_ctx->class0, CLASS0_SIZE - 1, r);
301 : update_mv_probs(comp_ctx->bits, MV_OFFSET_BITS, r);
302 : }
303 : for (i = 0; i < 2; ++i) {
304 : nmv_component *const comp_ctx = &ctx->comps[i];
305 : for (j = 0; j < CLASS0_SIZE; ++j) {
306 : update_mv_probs(comp_ctx->class0_fp[j], MV_FP_SIZE - 1, r);
307 : }
308 : update_mv_probs(comp_ctx->fp, MV_FP_SIZE - 1, r);
309 : }
310 : #endif // !CONFIG_EC_ADAPT
311 :
312 0 : if (allow_hp) {
313 0 : for (i = 0; i < 2; ++i) {
314 0 : nmv_component *const comp_ctx = &ctx->comps[i];
315 0 : update_mv_probs(&comp_ctx->class0_hp, 1, r);
316 0 : update_mv_probs(&comp_ctx->hp, 1, r);
317 : }
318 : }
319 0 : }
320 :
321 0 : static void inverse_transform_block(MACROBLOCKD *xd, int plane,
322 : const TX_TYPE tx_type,
323 : const TX_SIZE tx_size, uint8_t *dst,
324 : int stride, int16_t scan_line, int eob) {
325 0 : struct macroblockd_plane *const pd = &xd->plane[plane];
326 0 : tran_low_t *const dqcoeff = pd->dqcoeff;
327 0 : av1_inverse_transform_block(xd, dqcoeff, tx_type, tx_size, dst, stride, eob);
328 0 : memset(dqcoeff, 0, (scan_line + 1) * sizeof(dqcoeff[0]));
329 0 : }
330 :
331 : #if CONFIG_PVQ
332 : static int av1_pvq_decode_helper(MACROBLOCKD *xd, tran_low_t *ref_coeff,
333 : tran_low_t *dqcoeff, int16_t *quant, int pli,
334 : int bs, TX_TYPE tx_type, int xdec,
335 : PVQ_SKIP_TYPE ac_dc_coded) {
336 : unsigned int flags; // used for daala's stream analyzer.
337 : int off;
338 : const int is_keyframe = 0;
339 : const int has_dc_skip = 1;
340 : int coeff_shift = 3 - av1_get_tx_scale(bs);
341 : int hbd_downshift = 0;
342 : int rounding_mask;
343 : // DC quantizer for PVQ
344 : int pvq_dc_quant;
345 : int lossless = (quant[0] == 0);
346 : const int blk_size = tx_size_wide[bs];
347 : int eob = 0;
348 : int i;
349 : od_dec_ctx *dec = &xd->daala_dec;
350 : int use_activity_masking = dec->use_activity_masking;
351 : DECLARE_ALIGNED(16, tran_low_t, dqcoeff_pvq[OD_TXSIZE_MAX * OD_TXSIZE_MAX]);
352 : DECLARE_ALIGNED(16, tran_low_t, ref_coeff_pvq[OD_TXSIZE_MAX * OD_TXSIZE_MAX]);
353 :
354 : od_coeff ref_int32[OD_TXSIZE_MAX * OD_TXSIZE_MAX];
355 : od_coeff out_int32[OD_TXSIZE_MAX * OD_TXSIZE_MAX];
356 :
357 : hbd_downshift = xd->bd - 8;
358 :
359 : od_raster_to_coding_order(ref_coeff_pvq, blk_size, tx_type, ref_coeff,
360 : blk_size);
361 :
362 : assert(OD_COEFF_SHIFT >= 4);
363 : if (lossless)
364 : pvq_dc_quant = 1;
365 : else {
366 : if (use_activity_masking)
367 : pvq_dc_quant = OD_MAXI(
368 : 1, (quant[0] << (OD_COEFF_SHIFT - 3) >> hbd_downshift) *
369 : dec->state.pvq_qm_q4[pli][od_qm_get_index(bs, 0)] >>
370 : 4);
371 : else
372 : pvq_dc_quant =
373 : OD_MAXI(1, quant[0] << (OD_COEFF_SHIFT - 3) >> hbd_downshift);
374 : }
375 :
376 : off = od_qm_offset(bs, xdec);
377 :
378 : // copy int16 inputs to int32
379 : for (i = 0; i < blk_size * blk_size; i++) {
380 : ref_int32[i] =
381 : AOM_SIGNED_SHL(ref_coeff_pvq[i], OD_COEFF_SHIFT - coeff_shift) >>
382 : hbd_downshift;
383 : }
384 :
385 : od_pvq_decode(dec, ref_int32, out_int32,
386 : OD_MAXI(1, quant[1] << (OD_COEFF_SHIFT - 3) >> hbd_downshift),
387 : pli, bs, OD_PVQ_BETA[use_activity_masking][pli][bs],
388 : is_keyframe, &flags, ac_dc_coded, dec->state.qm + off,
389 : dec->state.qm_inv + off);
390 :
391 : if (!has_dc_skip || out_int32[0]) {
392 : out_int32[0] =
393 : has_dc_skip + generic_decode(dec->r, &dec->state.adapt->model_dc[pli],
394 : &dec->state.adapt->ex_dc[pli][bs][0], 2,
395 : "dc:mag");
396 : if (out_int32[0]) out_int32[0] *= aom_read_bit(dec->r, "dc:sign") ? -1 : 1;
397 : }
398 : out_int32[0] = out_int32[0] * pvq_dc_quant + ref_int32[0];
399 :
400 : // copy int32 result back to int16
401 : assert(OD_COEFF_SHIFT > coeff_shift);
402 : rounding_mask = (1 << (OD_COEFF_SHIFT - coeff_shift - 1)) - 1;
403 : for (i = 0; i < blk_size * blk_size; i++) {
404 : out_int32[i] = AOM_SIGNED_SHL(out_int32[i], hbd_downshift);
405 : dqcoeff_pvq[i] = (out_int32[i] + (out_int32[i] < 0) + rounding_mask) >>
406 : (OD_COEFF_SHIFT - coeff_shift);
407 : }
408 :
409 : od_coding_order_to_raster(dqcoeff, blk_size, tx_type, dqcoeff_pvq, blk_size);
410 :
411 : eob = blk_size * blk_size;
412 :
413 : return eob;
414 : }
415 :
416 : static PVQ_SKIP_TYPE read_pvq_skip(AV1_COMMON *cm, MACROBLOCKD *const xd,
417 : int plane, TX_SIZE tx_size) {
418 : // decode ac/dc coded flag. bit0: DC coded, bit1 : AC coded
419 : // NOTE : we don't use 5 symbols for luma here in aom codebase,
420 : // since block partition is taken care of by aom.
421 : // So, only AC/DC skip info is coded
422 : const int ac_dc_coded = aom_read_symbol(
423 : xd->daala_dec.r,
424 : xd->daala_dec.state.adapt->skip_cdf[2 * tx_size + (plane != 0)], 4,
425 : "skip");
426 : if (ac_dc_coded < 0 || ac_dc_coded > 3) {
427 : aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM,
428 : "Invalid PVQ Skip Type");
429 : }
430 : return ac_dc_coded;
431 : }
432 :
433 : static int av1_pvq_decode_helper2(AV1_COMMON *cm, MACROBLOCKD *const xd,
434 : MB_MODE_INFO *const mbmi, int plane, int row,
435 : int col, TX_SIZE tx_size, TX_TYPE tx_type) {
436 : struct macroblockd_plane *const pd = &xd->plane[plane];
437 : // transform block size in pixels
438 : int tx_blk_size = tx_size_wide[tx_size];
439 : int i, j;
440 : tran_low_t *pvq_ref_coeff = pd->pvq_ref_coeff;
441 : const int diff_stride = tx_blk_size;
442 : int16_t *pred = pd->pred;
443 : tran_low_t *const dqcoeff = pd->dqcoeff;
444 : uint8_t *dst;
445 : int eob;
446 : const PVQ_SKIP_TYPE ac_dc_coded = read_pvq_skip(cm, xd, plane, tx_size);
447 :
448 : eob = 0;
449 : dst = &pd->dst.buf[4 * row * pd->dst.stride + 4 * col];
450 :
451 : if (ac_dc_coded) {
452 : int xdec = pd->subsampling_x;
453 : int seg_id = mbmi->segment_id;
454 : int16_t *quant;
455 : FWD_TXFM_PARAM fwd_txfm_param;
456 : // ToDo(yaowu): correct this with optimal number from decoding process.
457 : const int max_scan_line = tx_size_2d[tx_size];
458 : #if CONFIG_HIGHBITDEPTH
459 : if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
460 : for (j = 0; j < tx_blk_size; j++)
461 : for (i = 0; i < tx_blk_size; i++)
462 : pred[diff_stride * j + i] =
463 : CONVERT_TO_SHORTPTR(dst)[pd->dst.stride * j + i];
464 : } else {
465 : #endif
466 : for (j = 0; j < tx_blk_size; j++)
467 : for (i = 0; i < tx_blk_size; i++)
468 : pred[diff_stride * j + i] = dst[pd->dst.stride * j + i];
469 : #if CONFIG_HIGHBITDEPTH
470 : }
471 : #endif
472 :
473 : fwd_txfm_param.tx_type = tx_type;
474 : fwd_txfm_param.tx_size = tx_size;
475 : fwd_txfm_param.lossless = xd->lossless[seg_id];
476 :
477 : #if CONFIG_HIGHBITDEPTH
478 : if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
479 : fwd_txfm_param.bd = xd->bd;
480 : av1_highbd_fwd_txfm(pred, pvq_ref_coeff, diff_stride, &fwd_txfm_param);
481 : } else {
482 : #endif // CONFIG_HIGHBITDEPTH
483 : av1_fwd_txfm(pred, pvq_ref_coeff, diff_stride, &fwd_txfm_param);
484 : #if CONFIG_HIGHBITDEPTH
485 : }
486 : #endif // CONFIG_HIGHBITDEPTH
487 :
488 : quant = &pd->seg_dequant[seg_id][0]; // aom's quantizer
489 :
490 : eob = av1_pvq_decode_helper(xd, pvq_ref_coeff, dqcoeff, quant, plane,
491 : tx_size, tx_type, xdec, ac_dc_coded);
492 :
493 : inverse_transform_block(xd, plane, tx_type, tx_size, dst, pd->dst.stride,
494 : max_scan_line, eob);
495 : }
496 :
497 : return eob;
498 : }
499 : #endif
500 :
501 0 : static int get_block_idx(const MACROBLOCKD *xd, int plane, int row, int col) {
502 0 : const int bsize = xd->mi[0]->mbmi.sb_type;
503 0 : const struct macroblockd_plane *pd = &xd->plane[plane];
504 : #if CONFIG_CB4X4
505 : #if CONFIG_CHROMA_2X2
506 : const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
507 : #else
508 0 : const BLOCK_SIZE plane_bsize =
509 0 : AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
510 : #endif // CONFIG_CHROMA_2X2
511 : #else
512 : const BLOCK_SIZE plane_bsize =
513 : get_plane_block_size(AOMMAX(BLOCK_8X8, bsize), pd);
514 : #endif
515 0 : const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
516 0 : const TX_SIZE tx_size = get_tx_size(plane, xd);
517 0 : const uint8_t txh_unit = tx_size_high_unit[tx_size];
518 0 : return row * max_blocks_wide + col * txh_unit;
519 : }
520 :
521 : #if CONFIG_DPCM_INTRA
522 : static void process_block_dpcm_vert(TX_SIZE tx_size, TX_TYPE_1D tx_type_1d,
523 : const tran_low_t *dqcoeff, uint8_t *dst,
524 : int dst_stride) {
525 : const int tx1d_width = tx_size_wide[tx_size];
526 : const int tx1d_height = tx_size_high[tx_size];
527 : dpcm_inv_txfm_add_func inverse_tx =
528 : av1_get_dpcm_inv_txfm_add_func(tx1d_width);
529 : for (int r = 0; r < tx1d_height; ++r) {
530 : if (r > 0) memcpy(dst, dst - dst_stride, tx1d_width * sizeof(dst[0]));
531 : inverse_tx(dqcoeff, 1, tx_type_1d, dst);
532 : dqcoeff += tx1d_width;
533 : dst += dst_stride;
534 : }
535 : }
536 :
537 : static void process_block_dpcm_horz(TX_SIZE tx_size, TX_TYPE_1D tx_type_1d,
538 : const tran_low_t *dqcoeff, uint8_t *dst,
539 : int dst_stride) {
540 : const int tx1d_width = tx_size_wide[tx_size];
541 : const int tx1d_height = tx_size_high[tx_size];
542 : dpcm_inv_txfm_add_func inverse_tx =
543 : av1_get_dpcm_inv_txfm_add_func(tx1d_height);
544 : tran_low_t tx_buff[64];
545 : for (int c = 0; c < tx1d_width; ++c, ++dqcoeff, ++dst) {
546 : for (int r = 0; r < tx1d_height; ++r) {
547 : if (c > 0) dst[r * dst_stride] = dst[r * dst_stride - 1];
548 : tx_buff[r] = dqcoeff[r * tx1d_width];
549 : }
550 : inverse_tx(tx_buff, dst_stride, tx_type_1d, dst);
551 : }
552 : }
553 :
554 : #if CONFIG_HIGHBITDEPTH
555 : static void hbd_process_block_dpcm_vert(TX_SIZE tx_size, TX_TYPE_1D tx_type_1d,
556 : int bd, const tran_low_t *dqcoeff,
557 : uint8_t *dst8, int dst_stride) {
558 : uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
559 : const int tx1d_width = tx_size_wide[tx_size];
560 : const int tx1d_height = tx_size_high[tx_size];
561 : hbd_dpcm_inv_txfm_add_func inverse_tx =
562 : av1_get_hbd_dpcm_inv_txfm_add_func(tx1d_width);
563 : for (int r = 0; r < tx1d_height; ++r) {
564 : if (r > 0) memcpy(dst, dst - dst_stride, tx1d_width * sizeof(dst[0]));
565 : inverse_tx(dqcoeff, 1, tx_type_1d, bd, dst);
566 : dqcoeff += tx1d_width;
567 : dst += dst_stride;
568 : }
569 : }
570 :
571 : static void hbd_process_block_dpcm_horz(TX_SIZE tx_size, TX_TYPE_1D tx_type_1d,
572 : int bd, const tran_low_t *dqcoeff,
573 : uint8_t *dst8, int dst_stride) {
574 : uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
575 : const int tx1d_width = tx_size_wide[tx_size];
576 : const int tx1d_height = tx_size_high[tx_size];
577 : hbd_dpcm_inv_txfm_add_func inverse_tx =
578 : av1_get_hbd_dpcm_inv_txfm_add_func(tx1d_height);
579 : tran_low_t tx_buff[64];
580 : switch (tx1d_height) {
581 : case 4: inverse_tx = av1_hbd_dpcm_inv_txfm_add_4_c; break;
582 : case 8: inverse_tx = av1_hbd_dpcm_inv_txfm_add_8_c; break;
583 : case 16: inverse_tx = av1_hbd_dpcm_inv_txfm_add_16_c; break;
584 : case 32: inverse_tx = av1_hbd_dpcm_inv_txfm_add_32_c; break;
585 : default: assert(0);
586 : }
587 :
588 : for (int c = 0; c < tx1d_width; ++c, ++dqcoeff, ++dst) {
589 : for (int r = 0; r < tx1d_height; ++r) {
590 : if (c > 0) dst[r * dst_stride] = dst[r * dst_stride - 1];
591 : tx_buff[r] = dqcoeff[r * tx1d_width];
592 : }
593 : inverse_tx(tx_buff, dst_stride, tx_type_1d, bd, dst);
594 : }
595 : }
596 : #endif // CONFIG_HIGHBITDEPTH
597 :
598 : static void inverse_transform_block_dpcm(MACROBLOCKD *xd, int plane,
599 : PREDICTION_MODE mode, TX_SIZE tx_size,
600 : TX_TYPE tx_type, uint8_t *dst,
601 : int dst_stride, int16_t scan_line) {
602 : struct macroblockd_plane *const pd = &xd->plane[plane];
603 : tran_low_t *const dqcoeff = pd->dqcoeff;
604 : TX_TYPE_1D tx_type_1d = DCT_1D;
605 : switch (tx_type) {
606 : case IDTX: tx_type_1d = IDTX_1D; break;
607 : case V_DCT:
608 : assert(mode == H_PRED);
609 : tx_type_1d = DCT_1D;
610 : break;
611 : case H_DCT:
612 : assert(mode == V_PRED);
613 : tx_type_1d = DCT_1D;
614 : break;
615 : default: assert(0);
616 : }
617 : switch (mode) {
618 : case V_PRED:
619 : #if CONFIG_HIGHBITDEPTH
620 : if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
621 : hbd_process_block_dpcm_vert(tx_size, tx_type_1d, xd->bd, dqcoeff, dst,
622 : dst_stride);
623 : } else {
624 : #endif // CONFIG_HIGHBITDEPTH
625 : process_block_dpcm_vert(tx_size, tx_type_1d, dqcoeff, dst, dst_stride);
626 : #if CONFIG_HIGHBITDEPTH
627 : }
628 : #endif // CONFIG_HIGHBITDEPTH
629 : break;
630 : case H_PRED:
631 : #if CONFIG_HIGHBITDEPTH
632 : if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
633 : hbd_process_block_dpcm_horz(tx_size, tx_type_1d, xd->bd, dqcoeff, dst,
634 : dst_stride);
635 : } else {
636 : #endif // CONFIG_HIGHBITDEPTH
637 : process_block_dpcm_horz(tx_size, tx_type_1d, dqcoeff, dst, dst_stride);
638 : #if CONFIG_HIGHBITDEPTH
639 : }
640 : #endif // CONFIG_HIGHBITDEPTH
641 : break;
642 : default: assert(0);
643 : }
644 : memset(dqcoeff, 0, (scan_line + 1) * sizeof(dqcoeff[0]));
645 : }
646 : #endif // CONFIG_DPCM_INTRA
647 :
648 0 : static void predict_and_reconstruct_intra_block(
649 : AV1_COMMON *cm, MACROBLOCKD *const xd, aom_reader *const r,
650 : MB_MODE_INFO *const mbmi, int plane, int row, int col, TX_SIZE tx_size) {
651 0 : PLANE_TYPE plane_type = get_plane_type(plane);
652 0 : const int block_idx = get_block_idx(xd, plane, row, col);
653 : #if CONFIG_PVQ
654 : (void)r;
655 : #endif
656 0 : av1_predict_intra_block_facade(xd, plane, block_idx, col, row, tx_size);
657 :
658 0 : if (!mbmi->skip) {
659 : #if !CONFIG_PVQ
660 0 : struct macroblockd_plane *const pd = &xd->plane[plane];
661 : #if CONFIG_LV_MAP
662 : int16_t max_scan_line = 0;
663 : int eob;
664 : av1_read_coeffs_txb_facade(cm, xd, r, row, col, block_idx, plane,
665 : pd->dqcoeff, &max_scan_line, &eob);
666 : // tx_type will be read out in av1_read_coeffs_txb_facade
667 : TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size);
668 : #else // CONFIG_LV_MAP
669 0 : TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size);
670 0 : const SCAN_ORDER *scan_order = get_scan(cm, tx_size, tx_type, 0);
671 0 : int16_t max_scan_line = 0;
672 0 : const int eob =
673 0 : av1_decode_block_tokens(cm, xd, plane, scan_order, col, row, tx_size,
674 0 : tx_type, &max_scan_line, r, mbmi->segment_id);
675 : #endif // CONFIG_LV_MAP
676 0 : if (eob) {
677 0 : uint8_t *dst =
678 0 : &pd->dst.buf[(row * pd->dst.stride + col) << tx_size_wide_log2[0]];
679 : #if CONFIG_DPCM_INTRA
680 : const int block_raster_idx =
681 : av1_block_index_to_raster_order(tx_size, block_idx);
682 : const PREDICTION_MODE mode = (plane == 0)
683 : ? get_y_mode(xd->mi[0], block_raster_idx)
684 : : mbmi->uv_mode;
685 : if (av1_use_dpcm_intra(plane, mode, tx_type, mbmi)) {
686 : inverse_transform_block_dpcm(xd, plane, mode, tx_size, tx_type, dst,
687 : pd->dst.stride, max_scan_line);
688 : } else {
689 : #endif // CONFIG_DPCM_INTRA
690 0 : inverse_transform_block(xd, plane, tx_type, tx_size, dst,
691 : pd->dst.stride, max_scan_line, eob);
692 : #if CONFIG_DPCM_INTRA
693 : }
694 : #endif // CONFIG_DPCM_INTRA
695 : }
696 : #else
697 : TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size);
698 : av1_pvq_decode_helper2(cm, xd, mbmi, plane, row, col, tx_size, tx_type);
699 : #endif
700 : }
701 : #if CONFIG_CFL
702 : if (plane == AOM_PLANE_Y) {
703 : struct macroblockd_plane *const pd = &xd->plane[plane];
704 : uint8_t *dst =
705 : &pd->dst.buf[(row * pd->dst.stride + col) << tx_size_wide_log2[0]];
706 : cfl_store(xd->cfl, dst, pd->dst.stride, row, col, tx_size);
707 : }
708 : #endif
709 0 : }
710 :
711 : #if CONFIG_VAR_TX && !CONFIG_COEF_INTERLEAVE
712 0 : static void decode_reconstruct_tx(AV1_COMMON *cm, MACROBLOCKD *const xd,
713 : aom_reader *r, MB_MODE_INFO *const mbmi,
714 : int plane, BLOCK_SIZE plane_bsize,
715 : int blk_row, int blk_col, TX_SIZE tx_size,
716 : int *eob_total) {
717 0 : const struct macroblockd_plane *const pd = &xd->plane[plane];
718 0 : const BLOCK_SIZE bsize = txsize_to_bsize[tx_size];
719 0 : const int tx_row = blk_row >> (1 - pd->subsampling_y);
720 0 : const int tx_col = blk_col >> (1 - pd->subsampling_x);
721 0 : const TX_SIZE plane_tx_size =
722 0 : plane ? uv_txsize_lookup[bsize][mbmi->inter_tx_size[tx_row][tx_col]][0][0]
723 0 : : mbmi->inter_tx_size[tx_row][tx_col];
724 : // Scale to match transform block unit.
725 0 : const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
726 0 : const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
727 :
728 0 : if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
729 :
730 0 : if (tx_size == plane_tx_size) {
731 0 : PLANE_TYPE plane_type = get_plane_type(plane);
732 0 : int block_idx = get_block_idx(xd, plane, blk_row, blk_col);
733 : #if CONFIG_LV_MAP
734 : int16_t max_scan_line = 0;
735 : int eob;
736 : av1_read_coeffs_txb_facade(cm, xd, r, blk_row, blk_col, block_idx, plane,
737 : pd->dqcoeff, &max_scan_line, &eob);
738 : // tx_type will be read out in av1_read_coeffs_txb_facade
739 : TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, plane_tx_size);
740 : #else // CONFIG_LV_MAP
741 0 : TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, plane_tx_size);
742 0 : const SCAN_ORDER *sc = get_scan(cm, plane_tx_size, tx_type, 1);
743 0 : int16_t max_scan_line = 0;
744 0 : const int eob = av1_decode_block_tokens(
745 : cm, xd, plane, sc, blk_col, blk_row, plane_tx_size, tx_type,
746 0 : &max_scan_line, r, mbmi->segment_id);
747 : #endif // CONFIG_LV_MAP
748 0 : inverse_transform_block(xd, plane, tx_type, plane_tx_size,
749 0 : &pd->dst.buf[(blk_row * pd->dst.stride + blk_col)
750 0 : << tx_size_wide_log2[0]],
751 : pd->dst.stride, max_scan_line, eob);
752 0 : *eob_total += eob;
753 : } else {
754 0 : const TX_SIZE sub_txs = sub_tx_size_map[tx_size];
755 0 : const int bsl = tx_size_wide_unit[sub_txs];
756 0 : assert(sub_txs < tx_size);
757 : int i;
758 :
759 0 : assert(bsl > 0);
760 :
761 0 : for (i = 0; i < 4; ++i) {
762 0 : const int offsetr = blk_row + (i >> 1) * bsl;
763 0 : const int offsetc = blk_col + (i & 0x01) * bsl;
764 :
765 0 : if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue;
766 :
767 0 : decode_reconstruct_tx(cm, xd, r, mbmi, plane, plane_bsize, offsetr,
768 : offsetc, sub_txs, eob_total);
769 : }
770 : }
771 : }
772 : #endif // CONFIG_VAR_TX
773 :
774 : #if !CONFIG_VAR_TX || CONFIG_SUPERTX || CONFIG_COEF_INTERLEAVE || \
775 : (!CONFIG_VAR_TX && CONFIG_EXT_TX && CONFIG_RECT_TX)
776 : static int reconstruct_inter_block(AV1_COMMON *cm, MACROBLOCKD *const xd,
777 : aom_reader *const r, int segment_id,
778 : int plane, int row, int col,
779 : TX_SIZE tx_size) {
780 : PLANE_TYPE plane_type = get_plane_type(plane);
781 : int block_idx = get_block_idx(xd, plane, row, col);
782 : #if CONFIG_PVQ
783 : int eob;
784 : (void)r;
785 : (void)segment_id;
786 : #else
787 : struct macroblockd_plane *const pd = &xd->plane[plane];
788 : #endif
789 :
790 : #if !CONFIG_PVQ
791 : #if CONFIG_LV_MAP
792 : (void)segment_id;
793 : int16_t max_scan_line = 0;
794 : int eob;
795 : av1_read_coeffs_txb_facade(cm, xd, r, row, col, block_idx, plane, pd->dqcoeff,
796 : &max_scan_line, &eob);
797 : // tx_type will be read out in av1_read_coeffs_txb_facade
798 : TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size);
799 : #else // CONFIG_LV_MAP
800 : int16_t max_scan_line = 0;
801 : TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size);
802 : const SCAN_ORDER *scan_order = get_scan(cm, tx_size, tx_type, 1);
803 : const int eob =
804 : av1_decode_block_tokens(cm, xd, plane, scan_order, col, row, tx_size,
805 : tx_type, &max_scan_line, r, segment_id);
806 : #endif // CONFIG_LV_MAP
807 : uint8_t *dst =
808 : &pd->dst.buf[(row * pd->dst.stride + col) << tx_size_wide_log2[0]];
809 : if (eob)
810 : inverse_transform_block(xd, plane, tx_type, tx_size, dst, pd->dst.stride,
811 : max_scan_line, eob);
812 : #else
813 : TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size);
814 : eob = av1_pvq_decode_helper2(cm, xd, &xd->mi[0]->mbmi, plane, row, col,
815 : tx_size, tx_type);
816 : #endif
817 : return eob;
818 : }
819 : #endif // !CONFIG_VAR_TX || CONFIG_SUPER_TX
820 :
821 0 : static void set_offsets(AV1_COMMON *const cm, MACROBLOCKD *const xd,
822 : BLOCK_SIZE bsize, int mi_row, int mi_col, int bw,
823 : int bh, int x_mis, int y_mis) {
824 0 : const int offset = mi_row * cm->mi_stride + mi_col;
825 : int x, y;
826 0 : const TileInfo *const tile = &xd->tile;
827 :
828 0 : xd->mi = cm->mi_grid_visible + offset;
829 0 : xd->mi[0] = &cm->mi[offset];
830 : // TODO(slavarnway): Generate sb_type based on bwl and bhl, instead of
831 : // passing bsize from decode_partition().
832 0 : xd->mi[0]->mbmi.sb_type = bsize;
833 : #if CONFIG_RD_DEBUG
834 : xd->mi[0]->mbmi.mi_row = mi_row;
835 : xd->mi[0]->mbmi.mi_col = mi_col;
836 : #endif
837 0 : for (y = 0; y < y_mis; ++y)
838 0 : for (x = !y; x < x_mis; ++x) xd->mi[y * cm->mi_stride + x] = xd->mi[0];
839 :
840 0 : set_plane_n4(xd, bw, bh);
841 0 : set_skip_context(xd, mi_row, mi_col);
842 :
843 : #if CONFIG_VAR_TX
844 0 : xd->max_tx_size = max_txsize_lookup[bsize];
845 : #endif
846 :
847 : // Distance of Mb to the various image edges. These are specified to 8th pel
848 : // as they are always compared to values that are in 1/8th pel units
849 0 : set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw,
850 : #if CONFIG_DEPENDENT_HORZTILES
851 : cm->dependent_horz_tiles,
852 : #endif // CONFIG_DEPENDENT_HORZTILES
853 : cm->mi_rows, cm->mi_cols);
854 :
855 0 : av1_setup_dst_planes(xd->plane, bsize, get_frame_new_buffer(cm), mi_row,
856 : mi_col);
857 0 : }
858 :
859 : #if CONFIG_SUPERTX
860 : static MB_MODE_INFO *set_offsets_extend(AV1_COMMON *const cm,
861 : MACROBLOCKD *const xd,
862 : const TileInfo *const tile,
863 : BLOCK_SIZE bsize_pred, int mi_row_pred,
864 : int mi_col_pred, int mi_row_ori,
865 : int mi_col_ori) {
866 : // Used in supertx
867 : // (mi_row_ori, mi_col_ori): location for mv
868 : // (mi_row_pred, mi_col_pred, bsize_pred): region to predict
869 : const int bw = mi_size_wide[bsize_pred];
870 : const int bh = mi_size_high[bsize_pred];
871 : const int offset = mi_row_ori * cm->mi_stride + mi_col_ori;
872 : xd->mi = cm->mi_grid_visible + offset;
873 : xd->mi[0] = cm->mi + offset;
874 : set_mi_row_col(xd, tile, mi_row_pred, bh, mi_col_pred, bw,
875 : #if CONFIG_DEPENDENT_HORZTILES
876 : cm->dependent_horz_tiles,
877 : #endif // CONFIG_DEPENDENT_HORZTILES
878 : cm->mi_rows, cm->mi_cols);
879 :
880 : xd->up_available = (mi_row_ori > tile->mi_row_start);
881 : xd->left_available = (mi_col_ori > tile->mi_col_start);
882 :
883 : set_plane_n4(xd, bw, bh);
884 :
885 : return &xd->mi[0]->mbmi;
886 : }
887 :
888 : #if CONFIG_SUPERTX
889 : static MB_MODE_INFO *set_mb_offsets(AV1_COMMON *const cm, MACROBLOCKD *const xd,
890 : BLOCK_SIZE bsize, int mi_row, int mi_col,
891 : int bw, int bh, int x_mis, int y_mis) {
892 : const int offset = mi_row * cm->mi_stride + mi_col;
893 : const TileInfo *const tile = &xd->tile;
894 : int x, y;
895 :
896 : xd->mi = cm->mi_grid_visible + offset;
897 : xd->mi[0] = cm->mi + offset;
898 : xd->mi[0]->mbmi.sb_type = bsize;
899 : for (y = 0; y < y_mis; ++y)
900 : for (x = !y; x < x_mis; ++x) xd->mi[y * cm->mi_stride + x] = xd->mi[0];
901 :
902 : set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw,
903 : #if CONFIG_DEPENDENT_HORZTILES
904 : cm->dependent_horz_tiles,
905 : #endif // CONFIG_DEPENDENT_HORZTILES
906 : cm->mi_rows, cm->mi_cols);
907 : return &xd->mi[0]->mbmi;
908 : }
909 : #endif
910 :
911 : static void set_offsets_topblock(AV1_COMMON *const cm, MACROBLOCKD *const xd,
912 : const TileInfo *const tile, BLOCK_SIZE bsize,
913 : int mi_row, int mi_col) {
914 : const int bw = mi_size_wide[bsize];
915 : const int bh = mi_size_high[bsize];
916 : const int offset = mi_row * cm->mi_stride + mi_col;
917 :
918 : xd->mi = cm->mi_grid_visible + offset;
919 : xd->mi[0] = cm->mi + offset;
920 :
921 : set_plane_n4(xd, bw, bh);
922 :
923 : set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw,
924 : #if CONFIG_DEPENDENT_HORZTILES
925 : cm->dependent_horz_tiles,
926 : #endif // CONFIG_DEPENDENT_HORZTILES
927 : cm->mi_rows, cm->mi_cols);
928 :
929 : av1_setup_dst_planes(xd->plane, bsize, get_frame_new_buffer(cm), mi_row,
930 : mi_col);
931 : }
932 :
933 : static void set_param_topblock(AV1_COMMON *const cm, MACROBLOCKD *const xd,
934 : BLOCK_SIZE bsize, int mi_row, int mi_col,
935 : int txfm, int skip) {
936 : const int bw = mi_size_wide[bsize];
937 : const int bh = mi_size_high[bsize];
938 : const int x_mis = AOMMIN(bw, cm->mi_cols - mi_col);
939 : const int y_mis = AOMMIN(bh, cm->mi_rows - mi_row);
940 : const int offset = mi_row * cm->mi_stride + mi_col;
941 : int x, y;
942 :
943 : xd->mi = cm->mi_grid_visible + offset;
944 : xd->mi[0] = cm->mi + offset;
945 :
946 : for (y = 0; y < y_mis; ++y)
947 : for (x = 0; x < x_mis; ++x) {
948 : xd->mi[y * cm->mi_stride + x]->mbmi.skip = skip;
949 : xd->mi[y * cm->mi_stride + x]->mbmi.tx_type = txfm;
950 : }
951 : #if CONFIG_VAR_TX
952 : xd->above_txfm_context = cm->above_txfm_context + mi_col;
953 : xd->left_txfm_context =
954 : xd->left_txfm_context_buffer + (mi_row & MAX_MIB_MASK);
955 : set_txfm_ctxs(xd->mi[0]->mbmi.tx_size, bw, bh, skip, xd);
956 : #endif
957 : }
958 :
959 : static void set_ref(AV1_COMMON *const cm, MACROBLOCKD *const xd, int idx,
960 : int mi_row, int mi_col) {
961 : MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
962 : RefBuffer *ref_buffer = &cm->frame_refs[mbmi->ref_frame[idx] - LAST_FRAME];
963 : xd->block_refs[idx] = ref_buffer;
964 : if (!av1_is_valid_scale(&ref_buffer->sf))
965 : aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
966 : "Invalid scale factors");
967 : av1_setup_pre_planes(xd, idx, ref_buffer->buf, mi_row, mi_col,
968 : &ref_buffer->sf);
969 : aom_merge_corrupted_flag(&xd->corrupted, ref_buffer->buf->corrupted);
970 : }
971 :
972 : static void dec_predict_b_extend(
973 : AV1Decoder *const pbi, MACROBLOCKD *const xd, const TileInfo *const tile,
974 : int block, int mi_row_ori, int mi_col_ori, int mi_row_pred, int mi_col_pred,
975 : int mi_row_top, int mi_col_top, uint8_t *dst_buf[3], int dst_stride[3],
976 : BLOCK_SIZE bsize_top, BLOCK_SIZE bsize_pred, int b_sub8x8, int bextend) {
977 : // Used in supertx
978 : // (mi_row_ori, mi_col_ori): location for mv
979 : // (mi_row_pred, mi_col_pred, bsize_pred): region to predict
980 : // (mi_row_top, mi_col_top, bsize_top): region of the top partition size
981 : // block: sub location of sub8x8 blocks
982 : // b_sub8x8: 1: ori is sub8x8; 0: ori is not sub8x8
983 : // bextend: 1: region to predict is an extension of ori; 0: not
984 : int r = (mi_row_pred - mi_row_top) * MI_SIZE;
985 : int c = (mi_col_pred - mi_col_top) * MI_SIZE;
986 : const int mi_width_top = mi_size_wide[bsize_top];
987 : const int mi_height_top = mi_size_high[bsize_top];
988 : MB_MODE_INFO *mbmi;
989 : AV1_COMMON *const cm = &pbi->common;
990 :
991 : if (mi_row_pred < mi_row_top || mi_col_pred < mi_col_top ||
992 : mi_row_pred >= mi_row_top + mi_height_top ||
993 : mi_col_pred >= mi_col_top + mi_width_top || mi_row_pred >= cm->mi_rows ||
994 : mi_col_pred >= cm->mi_cols)
995 : return;
996 :
997 : mbmi = set_offsets_extend(cm, xd, tile, bsize_pred, mi_row_pred, mi_col_pred,
998 : mi_row_ori, mi_col_ori);
999 : set_ref(cm, xd, 0, mi_row_pred, mi_col_pred);
1000 : if (has_second_ref(&xd->mi[0]->mbmi))
1001 : set_ref(cm, xd, 1, mi_row_pred, mi_col_pred);
1002 :
1003 : if (!bextend) mbmi->tx_size = max_txsize_lookup[bsize_top];
1004 :
1005 : xd->plane[0].dst.stride = dst_stride[0];
1006 : xd->plane[1].dst.stride = dst_stride[1];
1007 : xd->plane[2].dst.stride = dst_stride[2];
1008 : xd->plane[0].dst.buf = dst_buf[0] +
1009 : (r >> xd->plane[0].subsampling_y) * dst_stride[0] +
1010 : (c >> xd->plane[0].subsampling_x);
1011 : xd->plane[1].dst.buf = dst_buf[1] +
1012 : (r >> xd->plane[1].subsampling_y) * dst_stride[1] +
1013 : (c >> xd->plane[1].subsampling_x);
1014 : xd->plane[2].dst.buf = dst_buf[2] +
1015 : (r >> xd->plane[2].subsampling_y) * dst_stride[2] +
1016 : (c >> xd->plane[2].subsampling_x);
1017 :
1018 : if (!b_sub8x8)
1019 : av1_build_inter_predictors_sb_extend(&pbi->common, xd,
1020 : #if CONFIG_EXT_INTER
1021 : mi_row_ori, mi_col_ori,
1022 : #endif // CONFIG_EXT_INTER
1023 : mi_row_pred, mi_col_pred, bsize_pred);
1024 : else
1025 : av1_build_inter_predictors_sb_sub8x8_extend(&pbi->common, xd,
1026 : #if CONFIG_EXT_INTER
1027 : mi_row_ori, mi_col_ori,
1028 : #endif // CONFIG_EXT_INTER
1029 : mi_row_pred, mi_col_pred,
1030 : bsize_pred, block);
1031 : }
1032 :
1033 : static void dec_extend_dir(AV1Decoder *const pbi, MACROBLOCKD *const xd,
1034 : const TileInfo *const tile, int block,
1035 : BLOCK_SIZE bsize, BLOCK_SIZE top_bsize, int mi_row,
1036 : int mi_col, int mi_row_top, int mi_col_top,
1037 : uint8_t *dst_buf[3], int dst_stride[3], int dir) {
1038 : // dir: 0-lower, 1-upper, 2-left, 3-right
1039 : // 4-lowerleft, 5-upperleft, 6-lowerright, 7-upperright
1040 : const int mi_width = mi_size_wide[bsize];
1041 : const int mi_height = mi_size_high[bsize];
1042 : int xss = xd->plane[1].subsampling_x;
1043 : int yss = xd->plane[1].subsampling_y;
1044 : #if CONFIG_CB4X4
1045 : const int unify_bsize = 1;
1046 : #else
1047 : const int unify_bsize = 0;
1048 : #endif
1049 : int b_sub8x8 = (bsize < BLOCK_8X8) && !unify_bsize ? 1 : 0;
1050 : BLOCK_SIZE extend_bsize;
1051 : int mi_row_pred, mi_col_pred;
1052 :
1053 : int wide_unit, high_unit;
1054 : int i, j;
1055 : int ext_offset = 0;
1056 :
1057 : if (dir == 0 || dir == 1) {
1058 : extend_bsize =
1059 : (mi_width == mi_size_wide[BLOCK_8X8] || bsize < BLOCK_8X8 || xss < yss)
1060 : ? BLOCK_8X8
1061 : : BLOCK_16X8;
1062 : #if CONFIG_CB4X4
1063 : if (bsize < BLOCK_8X8) {
1064 : extend_bsize = BLOCK_4X4;
1065 : ext_offset = mi_size_wide[BLOCK_8X8];
1066 : }
1067 : #endif
1068 :
1069 : wide_unit = mi_size_wide[extend_bsize];
1070 : high_unit = mi_size_high[extend_bsize];
1071 :
1072 : mi_row_pred = mi_row + ((dir == 0) ? mi_height : -(mi_height + ext_offset));
1073 : mi_col_pred = mi_col;
1074 :
1075 : for (j = 0; j < mi_height + ext_offset; j += high_unit)
1076 : for (i = 0; i < mi_width + ext_offset; i += wide_unit)
1077 : dec_predict_b_extend(pbi, xd, tile, block, mi_row, mi_col,
1078 : mi_row_pred + j, mi_col_pred + i, mi_row_top,
1079 : mi_col_top, dst_buf, dst_stride, top_bsize,
1080 : extend_bsize, b_sub8x8, 1);
1081 : } else if (dir == 2 || dir == 3) {
1082 : extend_bsize =
1083 : (mi_height == mi_size_high[BLOCK_8X8] || bsize < BLOCK_8X8 || yss < xss)
1084 : ? BLOCK_8X8
1085 : : BLOCK_8X16;
1086 : #if CONFIG_CB4X4
1087 : if (bsize < BLOCK_8X8) {
1088 : extend_bsize = BLOCK_4X4;
1089 : ext_offset = mi_size_wide[BLOCK_8X8];
1090 : }
1091 : #endif
1092 :
1093 : wide_unit = mi_size_wide[extend_bsize];
1094 : high_unit = mi_size_high[extend_bsize];
1095 :
1096 : mi_row_pred = mi_row;
1097 : mi_col_pred = mi_col + ((dir == 3) ? mi_width : -(mi_width + ext_offset));
1098 :
1099 : for (j = 0; j < mi_height + ext_offset; j += high_unit)
1100 : for (i = 0; i < mi_width + ext_offset; i += wide_unit)
1101 : dec_predict_b_extend(pbi, xd, tile, block, mi_row, mi_col,
1102 : mi_row_pred + j, mi_col_pred + i, mi_row_top,
1103 : mi_col_top, dst_buf, dst_stride, top_bsize,
1104 : extend_bsize, b_sub8x8, 1);
1105 : } else {
1106 : extend_bsize = BLOCK_8X8;
1107 : #if CONFIG_CB4X4
1108 : if (bsize < BLOCK_8X8) {
1109 : extend_bsize = BLOCK_4X4;
1110 : ext_offset = mi_size_wide[BLOCK_8X8];
1111 : }
1112 : #endif
1113 : wide_unit = mi_size_wide[extend_bsize];
1114 : high_unit = mi_size_high[extend_bsize];
1115 :
1116 : mi_row_pred = mi_row + ((dir == 4 || dir == 6) ? mi_height
1117 : : -(mi_height + ext_offset));
1118 : mi_col_pred =
1119 : mi_col + ((dir == 6 || dir == 7) ? mi_width : -(mi_width + ext_offset));
1120 :
1121 : for (j = 0; j < mi_height + ext_offset; j += high_unit)
1122 : for (i = 0; i < mi_width + ext_offset; i += wide_unit)
1123 : dec_predict_b_extend(pbi, xd, tile, block, mi_row, mi_col,
1124 : mi_row_pred + j, mi_col_pred + i, mi_row_top,
1125 : mi_col_top, dst_buf, dst_stride, top_bsize,
1126 : extend_bsize, b_sub8x8, 1);
1127 : }
1128 : }
1129 :
1130 : static void dec_extend_all(AV1Decoder *const pbi, MACROBLOCKD *const xd,
1131 : const TileInfo *const tile, int block,
1132 : BLOCK_SIZE bsize, BLOCK_SIZE top_bsize, int mi_row,
1133 : int mi_col, int mi_row_top, int mi_col_top,
1134 : uint8_t *dst_buf[3], int dst_stride[3]) {
1135 : for (int i = 0; i < 8; ++i) {
1136 : dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col,
1137 : mi_row_top, mi_col_top, dst_buf, dst_stride, i);
1138 : }
1139 : }
1140 :
1141 : static void dec_predict_sb_complex(AV1Decoder *const pbi, MACROBLOCKD *const xd,
1142 : const TileInfo *const tile, int mi_row,
1143 : int mi_col, int mi_row_top, int mi_col_top,
1144 : BLOCK_SIZE bsize, BLOCK_SIZE top_bsize,
1145 : uint8_t *dst_buf[3], int dst_stride[3]) {
1146 : const AV1_COMMON *const cm = &pbi->common;
1147 : const int hbs = mi_size_wide[bsize] / 2;
1148 : const PARTITION_TYPE partition = get_partition(cm, mi_row, mi_col, bsize);
1149 : const BLOCK_SIZE subsize = get_subsize(bsize, partition);
1150 : #if CONFIG_EXT_PARTITION_TYPES
1151 : const BLOCK_SIZE bsize2 = get_subsize(bsize, PARTITION_SPLIT);
1152 : #endif
1153 : int i;
1154 : const int mi_offset = mi_row * cm->mi_stride + mi_col;
1155 : uint8_t *dst_buf1[3], *dst_buf2[3], *dst_buf3[3];
1156 : #if CONFIG_CB4X4
1157 : const int unify_bsize = 1;
1158 : #else
1159 : const int unify_bsize = 0;
1160 : #endif
1161 :
1162 : DECLARE_ALIGNED(16, uint8_t, tmp_buf1[MAX_MB_PLANE * MAX_TX_SQUARE * 2]);
1163 : DECLARE_ALIGNED(16, uint8_t, tmp_buf2[MAX_MB_PLANE * MAX_TX_SQUARE * 2]);
1164 : DECLARE_ALIGNED(16, uint8_t, tmp_buf3[MAX_MB_PLANE * MAX_TX_SQUARE * 2]);
1165 : int dst_stride1[3] = { MAX_TX_SIZE, MAX_TX_SIZE, MAX_TX_SIZE };
1166 : int dst_stride2[3] = { MAX_TX_SIZE, MAX_TX_SIZE, MAX_TX_SIZE };
1167 : int dst_stride3[3] = { MAX_TX_SIZE, MAX_TX_SIZE, MAX_TX_SIZE };
1168 :
1169 : #if CONFIG_HIGHBITDEPTH
1170 : if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
1171 : int len = sizeof(uint16_t);
1172 : dst_buf1[0] = CONVERT_TO_BYTEPTR(tmp_buf1);
1173 : dst_buf1[1] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_TX_SQUARE * len);
1174 : dst_buf1[2] = CONVERT_TO_BYTEPTR(tmp_buf1 + 2 * MAX_TX_SQUARE * len);
1175 : dst_buf2[0] = CONVERT_TO_BYTEPTR(tmp_buf2);
1176 : dst_buf2[1] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_TX_SQUARE * len);
1177 : dst_buf2[2] = CONVERT_TO_BYTEPTR(tmp_buf2 + 2 * MAX_TX_SQUARE * len);
1178 : dst_buf3[0] = CONVERT_TO_BYTEPTR(tmp_buf3);
1179 : dst_buf3[1] = CONVERT_TO_BYTEPTR(tmp_buf3 + MAX_TX_SQUARE * len);
1180 : dst_buf3[2] = CONVERT_TO_BYTEPTR(tmp_buf3 + 2 * MAX_TX_SQUARE * len);
1181 : } else {
1182 : #endif
1183 : dst_buf1[0] = tmp_buf1;
1184 : dst_buf1[1] = tmp_buf1 + MAX_TX_SQUARE;
1185 : dst_buf1[2] = tmp_buf1 + 2 * MAX_TX_SQUARE;
1186 : dst_buf2[0] = tmp_buf2;
1187 : dst_buf2[1] = tmp_buf2 + MAX_TX_SQUARE;
1188 : dst_buf2[2] = tmp_buf2 + 2 * MAX_TX_SQUARE;
1189 : dst_buf3[0] = tmp_buf3;
1190 : dst_buf3[1] = tmp_buf3 + MAX_TX_SQUARE;
1191 : dst_buf3[2] = tmp_buf3 + 2 * MAX_TX_SQUARE;
1192 : #if CONFIG_HIGHBITDEPTH
1193 : }
1194 : #endif
1195 :
1196 : if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;
1197 :
1198 : xd->mi = cm->mi_grid_visible + mi_offset;
1199 : xd->mi[0] = cm->mi + mi_offset;
1200 :
1201 : for (i = 0; i < MAX_MB_PLANE; i++) {
1202 : xd->plane[i].dst.buf = dst_buf[i];
1203 : xd->plane[i].dst.stride = dst_stride[i];
1204 : }
1205 :
1206 : switch (partition) {
1207 : case PARTITION_NONE:
1208 : assert(bsize < top_bsize);
1209 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
1210 : mi_row_top, mi_col_top, dst_buf, dst_stride,
1211 : top_bsize, bsize, 0, 0);
1212 : dec_extend_all(pbi, xd, tile, 0, bsize, top_bsize, mi_row, mi_col,
1213 : mi_row_top, mi_col_top, dst_buf, dst_stride);
1214 : break;
1215 : case PARTITION_HORZ:
1216 : if (bsize == BLOCK_8X8 && !unify_bsize) {
1217 : // For sub8x8, predict in 8x8 unit
1218 : // First half
1219 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
1220 : mi_row_top, mi_col_top, dst_buf, dst_stride,
1221 : top_bsize, BLOCK_8X8, 1, 0);
1222 : if (bsize < top_bsize)
1223 : dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
1224 : mi_row_top, mi_col_top, dst_buf, dst_stride);
1225 :
1226 : // Second half
1227 : dec_predict_b_extend(pbi, xd, tile, 2, mi_row, mi_col, mi_row, mi_col,
1228 : mi_row_top, mi_col_top, dst_buf1, dst_stride1,
1229 : top_bsize, BLOCK_8X8, 1, 1);
1230 : if (bsize < top_bsize)
1231 : dec_extend_all(pbi, xd, tile, 2, subsize, top_bsize, mi_row, mi_col,
1232 : mi_row_top, mi_col_top, dst_buf1, dst_stride1);
1233 :
1234 : // weighted average to smooth the boundary
1235 : xd->plane[0].dst.buf = dst_buf[0];
1236 : xd->plane[0].dst.stride = dst_stride[0];
1237 : av1_build_masked_inter_predictor_complex(
1238 : xd, dst_buf[0], dst_stride[0], dst_buf1[0], dst_stride1[0], mi_row,
1239 : mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ,
1240 : 0);
1241 : } else {
1242 : // First half
1243 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
1244 : mi_row_top, mi_col_top, dst_buf, dst_stride,
1245 : top_bsize, subsize, 0, 0);
1246 : if (bsize < top_bsize)
1247 : dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
1248 : mi_row_top, mi_col_top, dst_buf, dst_stride);
1249 : else
1250 : dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
1251 : mi_row_top, mi_col_top, dst_buf, dst_stride, 0);
1252 :
1253 : if (mi_row + hbs < cm->mi_rows) {
1254 : // Second half
1255 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col,
1256 : mi_row + hbs, mi_col, mi_row_top, mi_col_top,
1257 : dst_buf1, dst_stride1, top_bsize, subsize, 0, 0);
1258 : if (bsize < top_bsize)
1259 : dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs,
1260 : mi_col, mi_row_top, mi_col_top, dst_buf1,
1261 : dst_stride1);
1262 : else
1263 : dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs,
1264 : mi_col, mi_row_top, mi_col_top, dst_buf1,
1265 : dst_stride1, 1);
1266 :
1267 : // weighted average to smooth the boundary
1268 : for (i = 0; i < MAX_MB_PLANE; i++) {
1269 : xd->plane[i].dst.buf = dst_buf[i];
1270 : xd->plane[i].dst.stride = dst_stride[i];
1271 : av1_build_masked_inter_predictor_complex(
1272 : xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i],
1273 : mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
1274 : PARTITION_HORZ, i);
1275 : }
1276 : }
1277 : }
1278 : break;
1279 : case PARTITION_VERT:
1280 : if (bsize == BLOCK_8X8 && !unify_bsize) {
1281 : // First half
1282 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
1283 : mi_row_top, mi_col_top, dst_buf, dst_stride,
1284 : top_bsize, BLOCK_8X8, 1, 0);
1285 : if (bsize < top_bsize)
1286 : dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
1287 : mi_row_top, mi_col_top, dst_buf, dst_stride);
1288 :
1289 : // Second half
1290 : dec_predict_b_extend(pbi, xd, tile, 1, mi_row, mi_col, mi_row, mi_col,
1291 : mi_row_top, mi_col_top, dst_buf1, dst_stride1,
1292 : top_bsize, BLOCK_8X8, 1, 1);
1293 : if (bsize < top_bsize)
1294 : dec_extend_all(pbi, xd, tile, 1, subsize, top_bsize, mi_row, mi_col,
1295 : mi_row_top, mi_col_top, dst_buf1, dst_stride1);
1296 :
1297 : // Smooth
1298 : xd->plane[0].dst.buf = dst_buf[0];
1299 : xd->plane[0].dst.stride = dst_stride[0];
1300 : av1_build_masked_inter_predictor_complex(
1301 : xd, dst_buf[0], dst_stride[0], dst_buf1[0], dst_stride1[0], mi_row,
1302 : mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT,
1303 : 0);
1304 : } else {
1305 : // First half
1306 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
1307 : mi_row_top, mi_col_top, dst_buf, dst_stride,
1308 : top_bsize, subsize, 0, 0);
1309 : if (bsize < top_bsize)
1310 : dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
1311 : mi_row_top, mi_col_top, dst_buf, dst_stride);
1312 : else
1313 : dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
1314 : mi_row_top, mi_col_top, dst_buf, dst_stride, 3);
1315 :
1316 : // Second half
1317 : if (mi_col + hbs < cm->mi_cols) {
1318 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row,
1319 : mi_col + hbs, mi_row_top, mi_col_top, dst_buf1,
1320 : dst_stride1, top_bsize, subsize, 0, 0);
1321 : if (bsize < top_bsize)
1322 : dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
1323 : mi_col + hbs, mi_row_top, mi_col_top, dst_buf1,
1324 : dst_stride1);
1325 : else
1326 : dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
1327 : mi_col + hbs, mi_row_top, mi_col_top, dst_buf1,
1328 : dst_stride1, 2);
1329 :
1330 : // Smooth
1331 : for (i = 0; i < MAX_MB_PLANE; i++) {
1332 : xd->plane[i].dst.buf = dst_buf[i];
1333 : xd->plane[i].dst.stride = dst_stride[i];
1334 : av1_build_masked_inter_predictor_complex(
1335 : xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i],
1336 : mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
1337 : PARTITION_VERT, i);
1338 : }
1339 : }
1340 : }
1341 : break;
1342 : case PARTITION_SPLIT:
1343 : if (bsize == BLOCK_8X8 && !unify_bsize) {
1344 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
1345 : mi_row_top, mi_col_top, dst_buf, dst_stride,
1346 : top_bsize, BLOCK_8X8, 1, 0);
1347 : dec_predict_b_extend(pbi, xd, tile, 1, mi_row, mi_col, mi_row, mi_col,
1348 : mi_row_top, mi_col_top, dst_buf1, dst_stride1,
1349 : top_bsize, BLOCK_8X8, 1, 1);
1350 : dec_predict_b_extend(pbi, xd, tile, 2, mi_row, mi_col, mi_row, mi_col,
1351 : mi_row_top, mi_col_top, dst_buf2, dst_stride2,
1352 : top_bsize, BLOCK_8X8, 1, 1);
1353 : dec_predict_b_extend(pbi, xd, tile, 3, mi_row, mi_col, mi_row, mi_col,
1354 : mi_row_top, mi_col_top, dst_buf3, dst_stride3,
1355 : top_bsize, BLOCK_8X8, 1, 1);
1356 : if (bsize < top_bsize) {
1357 : dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
1358 : mi_row_top, mi_col_top, dst_buf, dst_stride);
1359 : dec_extend_all(pbi, xd, tile, 1, subsize, top_bsize, mi_row, mi_col,
1360 : mi_row_top, mi_col_top, dst_buf1, dst_stride1);
1361 : dec_extend_all(pbi, xd, tile, 2, subsize, top_bsize, mi_row, mi_col,
1362 : mi_row_top, mi_col_top, dst_buf2, dst_stride2);
1363 : dec_extend_all(pbi, xd, tile, 3, subsize, top_bsize, mi_row, mi_col,
1364 : mi_row_top, mi_col_top, dst_buf3, dst_stride3);
1365 : }
1366 : } else {
1367 : dec_predict_sb_complex(pbi, xd, tile, mi_row, mi_col, mi_row_top,
1368 : mi_col_top, subsize, top_bsize, dst_buf,
1369 : dst_stride);
1370 : if (mi_row < cm->mi_rows && mi_col + hbs < cm->mi_cols)
1371 : dec_predict_sb_complex(pbi, xd, tile, mi_row, mi_col + hbs,
1372 : mi_row_top, mi_col_top, subsize, top_bsize,
1373 : dst_buf1, dst_stride1);
1374 : if (mi_row + hbs < cm->mi_rows && mi_col < cm->mi_cols)
1375 : dec_predict_sb_complex(pbi, xd, tile, mi_row + hbs, mi_col,
1376 : mi_row_top, mi_col_top, subsize, top_bsize,
1377 : dst_buf2, dst_stride2);
1378 : if (mi_row + hbs < cm->mi_rows && mi_col + hbs < cm->mi_cols)
1379 : dec_predict_sb_complex(pbi, xd, tile, mi_row + hbs, mi_col + hbs,
1380 : mi_row_top, mi_col_top, subsize, top_bsize,
1381 : dst_buf3, dst_stride3);
1382 : }
1383 : for (i = 0; i < MAX_MB_PLANE; i++) {
1384 : #if !CONFIG_CB4X4
1385 : if (bsize == BLOCK_8X8 && i != 0)
1386 : continue; // Skip <4x4 chroma smoothing
1387 : #endif
1388 : if (mi_row < cm->mi_rows && mi_col + hbs < cm->mi_cols) {
1389 : av1_build_masked_inter_predictor_complex(
1390 : xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i],
1391 : mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
1392 : PARTITION_VERT, i);
1393 : if (mi_row + hbs < cm->mi_rows) {
1394 : av1_build_masked_inter_predictor_complex(
1395 : xd, dst_buf2[i], dst_stride2[i], dst_buf3[i], dst_stride3[i],
1396 : mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
1397 : PARTITION_VERT, i);
1398 : av1_build_masked_inter_predictor_complex(
1399 : xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i],
1400 : mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
1401 : PARTITION_HORZ, i);
1402 : }
1403 : } else if (mi_row + hbs < cm->mi_rows && mi_col < cm->mi_cols) {
1404 : av1_build_masked_inter_predictor_complex(
1405 : xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i],
1406 : mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
1407 : PARTITION_HORZ, i);
1408 : }
1409 : }
1410 : break;
1411 : #if CONFIG_EXT_PARTITION_TYPES
1412 : case PARTITION_HORZ_A:
1413 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
1414 : mi_row_top, mi_col_top, dst_buf, dst_stride,
1415 : top_bsize, bsize2, 0, 0);
1416 : dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col,
1417 : mi_row_top, mi_col_top, dst_buf, dst_stride);
1418 :
1419 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row,
1420 : mi_col + hbs, mi_row_top, mi_col_top, dst_buf1,
1421 : dst_stride1, top_bsize, bsize2, 0, 0);
1422 : dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col + hbs,
1423 : mi_row_top, mi_col_top, dst_buf1, dst_stride1);
1424 :
1425 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col, mi_row + hbs,
1426 : mi_col, mi_row_top, mi_col_top, dst_buf2,
1427 : dst_stride2, top_bsize, subsize, 0, 0);
1428 : if (bsize < top_bsize)
1429 : dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs,
1430 : mi_col, mi_row_top, mi_col_top, dst_buf2, dst_stride2);
1431 : else
1432 : dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs,
1433 : mi_col, mi_row_top, mi_col_top, dst_buf2, dst_stride2,
1434 : 1);
1435 :
1436 : for (i = 0; i < MAX_MB_PLANE; i++) {
1437 : xd->plane[i].dst.buf = dst_buf[i];
1438 : xd->plane[i].dst.stride = dst_stride[i];
1439 : av1_build_masked_inter_predictor_complex(
1440 : xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row,
1441 : mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT,
1442 : i);
1443 : }
1444 : for (i = 0; i < MAX_MB_PLANE; i++) {
1445 : av1_build_masked_inter_predictor_complex(
1446 : xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i], mi_row,
1447 : mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ,
1448 : i);
1449 : }
1450 : break;
1451 : case PARTITION_VERT_A:
1452 :
1453 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
1454 : mi_row_top, mi_col_top, dst_buf, dst_stride,
1455 : top_bsize, bsize2, 0, 0);
1456 : dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col,
1457 : mi_row_top, mi_col_top, dst_buf, dst_stride);
1458 :
1459 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col, mi_row + hbs,
1460 : mi_col, mi_row_top, mi_col_top, dst_buf1,
1461 : dst_stride1, top_bsize, bsize2, 0, 0);
1462 : dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row + hbs, mi_col,
1463 : mi_row_top, mi_col_top, dst_buf1, dst_stride1);
1464 :
1465 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row,
1466 : mi_col + hbs, mi_row_top, mi_col_top, dst_buf2,
1467 : dst_stride2, top_bsize, subsize, 0, 0);
1468 : if (bsize < top_bsize)
1469 : dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
1470 : mi_col + hbs, mi_row_top, mi_col_top, dst_buf2,
1471 : dst_stride2);
1472 : else
1473 : dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
1474 : mi_col + hbs, mi_row_top, mi_col_top, dst_buf2,
1475 : dst_stride2, 2);
1476 :
1477 : for (i = 0; i < MAX_MB_PLANE; i++) {
1478 : xd->plane[i].dst.buf = dst_buf[i];
1479 : xd->plane[i].dst.stride = dst_stride[i];
1480 : av1_build_masked_inter_predictor_complex(
1481 : xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row,
1482 : mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ,
1483 : i);
1484 : }
1485 : for (i = 0; i < MAX_MB_PLANE; i++) {
1486 : av1_build_masked_inter_predictor_complex(
1487 : xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i], mi_row,
1488 : mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT,
1489 : i);
1490 : }
1491 : break;
1492 : case PARTITION_HORZ_B:
1493 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
1494 : mi_row_top, mi_col_top, dst_buf, dst_stride,
1495 : top_bsize, subsize, 0, 0);
1496 : if (bsize < top_bsize)
1497 : dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
1498 : mi_row_top, mi_col_top, dst_buf, dst_stride);
1499 : else
1500 : dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
1501 : mi_row_top, mi_col_top, dst_buf, dst_stride, 0);
1502 :
1503 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col, mi_row + hbs,
1504 : mi_col, mi_row_top, mi_col_top, dst_buf1,
1505 : dst_stride1, top_bsize, bsize2, 0, 0);
1506 : dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row + hbs, mi_col,
1507 : mi_row_top, mi_col_top, dst_buf1, dst_stride1);
1508 :
1509 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col + hbs,
1510 : mi_row + hbs, mi_col + hbs, mi_row_top, mi_col_top,
1511 : dst_buf2, dst_stride2, top_bsize, bsize2, 0, 0);
1512 : dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row + hbs,
1513 : mi_col + hbs, mi_row_top, mi_col_top, dst_buf2,
1514 : dst_stride2);
1515 :
1516 : for (i = 0; i < MAX_MB_PLANE; i++) {
1517 : xd->plane[i].dst.buf = dst_buf1[i];
1518 : xd->plane[i].dst.stride = dst_stride1[i];
1519 : av1_build_masked_inter_predictor_complex(
1520 : xd, dst_buf1[i], dst_stride1[i], dst_buf2[i], dst_stride2[i],
1521 : mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
1522 : PARTITION_VERT, i);
1523 : }
1524 : for (i = 0; i < MAX_MB_PLANE; i++) {
1525 : xd->plane[i].dst.buf = dst_buf[i];
1526 : xd->plane[i].dst.stride = dst_stride[i];
1527 : av1_build_masked_inter_predictor_complex(
1528 : xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row,
1529 : mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ,
1530 : i);
1531 : }
1532 : break;
1533 : case PARTITION_VERT_B:
1534 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
1535 : mi_row_top, mi_col_top, dst_buf, dst_stride,
1536 : top_bsize, subsize, 0, 0);
1537 : if (bsize < top_bsize)
1538 : dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
1539 : mi_row_top, mi_col_top, dst_buf, dst_stride);
1540 : else
1541 : dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
1542 : mi_row_top, mi_col_top, dst_buf, dst_stride, 3);
1543 :
1544 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row,
1545 : mi_col + hbs, mi_row_top, mi_col_top, dst_buf1,
1546 : dst_stride1, top_bsize, bsize2, 0, 0);
1547 : dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col + hbs,
1548 : mi_row_top, mi_col_top, dst_buf1, dst_stride1);
1549 :
1550 : dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col + hbs,
1551 : mi_row + hbs, mi_col + hbs, mi_row_top, mi_col_top,
1552 : dst_buf2, dst_stride2, top_bsize, bsize2, 0, 0);
1553 : dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row + hbs,
1554 : mi_col + hbs, mi_row_top, mi_col_top, dst_buf2,
1555 : dst_stride2);
1556 :
1557 : for (i = 0; i < MAX_MB_PLANE; i++) {
1558 : xd->plane[i].dst.buf = dst_buf1[i];
1559 : xd->plane[i].dst.stride = dst_stride1[i];
1560 : av1_build_masked_inter_predictor_complex(
1561 : xd, dst_buf1[i], dst_stride1[i], dst_buf2[i], dst_stride2[i],
1562 : mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
1563 : PARTITION_HORZ, i);
1564 : }
1565 : for (i = 0; i < MAX_MB_PLANE; i++) {
1566 : xd->plane[i].dst.buf = dst_buf[i];
1567 : xd->plane[i].dst.stride = dst_stride[i];
1568 : av1_build_masked_inter_predictor_complex(
1569 : xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row,
1570 : mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT,
1571 : i);
1572 : }
1573 : break;
1574 : #endif // CONFIG_EXT_PARTITION_TYPES
1575 : default: assert(0);
1576 : }
1577 : }
1578 :
1579 : static void set_segment_id_supertx(const AV1_COMMON *const cm, int mi_row,
1580 : int mi_col, BLOCK_SIZE bsize) {
1581 : const struct segmentation *seg = &cm->seg;
1582 : const int miw = AOMMIN(mi_size_wide[bsize], cm->mi_cols - mi_col);
1583 : const int mih = AOMMIN(mi_size_high[bsize], cm->mi_rows - mi_row);
1584 : const int mi_offset = mi_row * cm->mi_stride + mi_col;
1585 : MODE_INFO **const mip = cm->mi_grid_visible + mi_offset;
1586 : int r, c;
1587 : int seg_id_supertx = MAX_SEGMENTS;
1588 :
1589 : if (!seg->enabled) {
1590 : seg_id_supertx = 0;
1591 : } else {
1592 : // Find the minimum segment_id
1593 : for (r = 0; r < mih; r++)
1594 : for (c = 0; c < miw; c++)
1595 : seg_id_supertx =
1596 : AOMMIN(mip[r * cm->mi_stride + c]->mbmi.segment_id, seg_id_supertx);
1597 : assert(0 <= seg_id_supertx && seg_id_supertx < MAX_SEGMENTS);
1598 : }
1599 :
1600 : // Assign the the segment_id back to segment_id_supertx
1601 : for (r = 0; r < mih; r++)
1602 : for (c = 0; c < miw; c++)
1603 : mip[r * cm->mi_stride + c]->mbmi.segment_id_supertx = seg_id_supertx;
1604 : }
1605 : #endif // CONFIG_SUPERTX
1606 :
1607 0 : static void decode_mbmi_block(AV1Decoder *const pbi, MACROBLOCKD *const xd,
1608 : #if CONFIG_SUPERTX
1609 : int supertx_enabled,
1610 : #endif // CONFIG_SUPERTX
1611 : int mi_row, int mi_col, aom_reader *r,
1612 : #if CONFIG_EXT_PARTITION_TYPES
1613 : PARTITION_TYPE partition,
1614 : #endif // CONFIG_EXT_PARTITION_TYPES
1615 : BLOCK_SIZE bsize) {
1616 0 : AV1_COMMON *const cm = &pbi->common;
1617 0 : const int bw = mi_size_wide[bsize];
1618 0 : const int bh = mi_size_high[bsize];
1619 0 : const int x_mis = AOMMIN(bw, cm->mi_cols - mi_col);
1620 0 : const int y_mis = AOMMIN(bh, cm->mi_rows - mi_row);
1621 :
1622 : #if CONFIG_ACCOUNTING
1623 : aom_accounting_set_context(&pbi->accounting, mi_col, mi_row);
1624 : #endif
1625 : #if CONFIG_SUPERTX
1626 : if (supertx_enabled) {
1627 : set_mb_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis);
1628 : } else {
1629 : set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis);
1630 : }
1631 : #if CONFIG_EXT_PARTITION_TYPES
1632 : xd->mi[0]->mbmi.partition = partition;
1633 : #endif
1634 : av1_read_mode_info(pbi, xd, supertx_enabled, mi_row, mi_col, r, x_mis, y_mis);
1635 : #else
1636 0 : set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis);
1637 : #if CONFIG_EXT_PARTITION_TYPES
1638 : xd->mi[0]->mbmi.partition = partition;
1639 : #endif
1640 0 : av1_read_mode_info(pbi, xd, mi_row, mi_col, r, x_mis, y_mis);
1641 : #endif // CONFIG_SUPERTX
1642 :
1643 0 : if (bsize >= BLOCK_8X8 && (cm->subsampling_x || cm->subsampling_y)) {
1644 0 : const BLOCK_SIZE uv_subsize =
1645 0 : ss_size_lookup[bsize][cm->subsampling_x][cm->subsampling_y];
1646 0 : if (uv_subsize == BLOCK_INVALID)
1647 0 : aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME,
1648 : "Invalid block size.");
1649 : }
1650 :
1651 : #if CONFIG_SUPERTX
1652 : xd->mi[0]->mbmi.segment_id_supertx = MAX_SEGMENTS;
1653 : #endif // CONFIG_SUPERTX
1654 :
1655 0 : int reader_corrupted_flag = aom_reader_has_error(r);
1656 0 : aom_merge_corrupted_flag(&xd->corrupted, reader_corrupted_flag);
1657 0 : }
1658 :
1659 0 : static void decode_token_and_recon_block(AV1Decoder *const pbi,
1660 : MACROBLOCKD *const xd, int mi_row,
1661 : int mi_col, aom_reader *r,
1662 : BLOCK_SIZE bsize) {
1663 0 : AV1_COMMON *const cm = &pbi->common;
1664 0 : const int bw = mi_size_wide[bsize];
1665 0 : const int bh = mi_size_high[bsize];
1666 0 : const int x_mis = AOMMIN(bw, cm->mi_cols - mi_col);
1667 0 : const int y_mis = AOMMIN(bh, cm->mi_rows - mi_row);
1668 :
1669 0 : set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis);
1670 0 : MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
1671 :
1672 : #if CONFIG_DELTA_Q
1673 0 : if (cm->delta_q_present_flag) {
1674 : int i;
1675 0 : for (i = 0; i < MAX_SEGMENTS; i++) {
1676 : #if CONFIG_EXT_DELTA_Q
1677 0 : xd->plane[0].seg_dequant[i][0] =
1678 0 : av1_dc_quant(av1_get_qindex(&cm->seg, i, xd->current_qindex),
1679 : cm->y_dc_delta_q, cm->bit_depth);
1680 0 : xd->plane[0].seg_dequant[i][1] = av1_ac_quant(
1681 0 : av1_get_qindex(&cm->seg, i, xd->current_qindex), 0, cm->bit_depth);
1682 0 : xd->plane[1].seg_dequant[i][0] =
1683 0 : av1_dc_quant(av1_get_qindex(&cm->seg, i, xd->current_qindex),
1684 : cm->uv_dc_delta_q, cm->bit_depth);
1685 0 : xd->plane[1].seg_dequant[i][1] =
1686 0 : av1_ac_quant(av1_get_qindex(&cm->seg, i, xd->current_qindex),
1687 : cm->uv_ac_delta_q, cm->bit_depth);
1688 0 : xd->plane[2].seg_dequant[i][0] =
1689 0 : av1_dc_quant(av1_get_qindex(&cm->seg, i, xd->current_qindex),
1690 : cm->uv_dc_delta_q, cm->bit_depth);
1691 0 : xd->plane[2].seg_dequant[i][1] =
1692 0 : av1_ac_quant(av1_get_qindex(&cm->seg, i, xd->current_qindex),
1693 : cm->uv_ac_delta_q, cm->bit_depth);
1694 : #else
1695 : xd->plane[0].seg_dequant[i][0] =
1696 : av1_dc_quant(xd->current_qindex, cm->y_dc_delta_q, cm->bit_depth);
1697 : xd->plane[0].seg_dequant[i][1] =
1698 : av1_ac_quant(xd->current_qindex, 0, cm->bit_depth);
1699 : xd->plane[1].seg_dequant[i][0] =
1700 : av1_dc_quant(xd->current_qindex, cm->uv_dc_delta_q, cm->bit_depth);
1701 : xd->plane[1].seg_dequant[i][1] =
1702 : av1_ac_quant(xd->current_qindex, cm->uv_ac_delta_q, cm->bit_depth);
1703 : xd->plane[2].seg_dequant[i][0] =
1704 : av1_dc_quant(xd->current_qindex, cm->uv_dc_delta_q, cm->bit_depth);
1705 : xd->plane[2].seg_dequant[i][1] =
1706 : av1_ac_quant(xd->current_qindex, cm->uv_ac_delta_q, cm->bit_depth);
1707 : #endif
1708 : }
1709 : }
1710 : #endif
1711 :
1712 : #if CONFIG_CB4X4
1713 0 : if (mbmi->skip) av1_reset_skip_context(xd, mi_row, mi_col, bsize);
1714 : #else
1715 : if (mbmi->skip) {
1716 : av1_reset_skip_context(xd, mi_row, mi_col, AOMMAX(BLOCK_8X8, bsize));
1717 : }
1718 : #endif
1719 :
1720 : #if CONFIG_COEF_INTERLEAVE
1721 : {
1722 : const struct macroblockd_plane *const pd_y = &xd->plane[0];
1723 : const struct macroblockd_plane *const pd_c = &xd->plane[1];
1724 : const TX_SIZE tx_log2_y = mbmi->tx_size;
1725 : const TX_SIZE tx_log2_c = get_uv_tx_size(mbmi, pd_c);
1726 : const int tx_sz_y = (1 << tx_log2_y);
1727 : const int tx_sz_c = (1 << tx_log2_c);
1728 : const int num_4x4_w_y = pd_y->n4_w;
1729 : const int num_4x4_h_y = pd_y->n4_h;
1730 : const int num_4x4_w_c = pd_c->n4_w;
1731 : const int num_4x4_h_c = pd_c->n4_h;
1732 : const int max_4x4_w_y = get_max_4x4_size(num_4x4_w_y, xd->mb_to_right_edge,
1733 : pd_y->subsampling_x);
1734 : const int max_4x4_h_y = get_max_4x4_size(num_4x4_h_y, xd->mb_to_bottom_edge,
1735 : pd_y->subsampling_y);
1736 : const int max_4x4_w_c = get_max_4x4_size(num_4x4_w_c, xd->mb_to_right_edge,
1737 : pd_c->subsampling_x);
1738 : const int max_4x4_h_c = get_max_4x4_size(num_4x4_h_c, xd->mb_to_bottom_edge,
1739 : pd_c->subsampling_y);
1740 :
1741 : // The max_4x4_w/h may be smaller than tx_sz under some corner cases,
1742 : // i.e. when the SB is splitted by tile boundaries.
1743 : const int tu_num_w_y = (max_4x4_w_y + tx_sz_y - 1) / tx_sz_y;
1744 : const int tu_num_h_y = (max_4x4_h_y + tx_sz_y - 1) / tx_sz_y;
1745 : const int tu_num_w_c = (max_4x4_w_c + tx_sz_c - 1) / tx_sz_c;
1746 : const int tu_num_h_c = (max_4x4_h_c + tx_sz_c - 1) / tx_sz_c;
1747 : const int tu_num_c = tu_num_w_c * tu_num_h_c;
1748 :
1749 : if (!is_inter_block(mbmi)) {
1750 : int tu_idx_c = 0;
1751 : int row_y, col_y, row_c, col_c;
1752 : int plane;
1753 :
1754 : #if CONFIG_PALETTE
1755 : for (plane = 0; plane <= 1; ++plane) {
1756 : if (mbmi->palette_mode_info.palette_size[plane])
1757 : av1_decode_palette_tokens(xd, plane, r);
1758 : }
1759 : #endif
1760 :
1761 : for (row_y = 0; row_y < tu_num_h_y; row_y++) {
1762 : for (col_y = 0; col_y < tu_num_w_y; col_y++) {
1763 : // luma
1764 : predict_and_reconstruct_intra_block(
1765 : cm, xd, r, mbmi, 0, row_y * tx_sz_y, col_y * tx_sz_y, tx_log2_y);
1766 : // chroma
1767 : if (tu_idx_c < tu_num_c) {
1768 : row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
1769 : col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
1770 : predict_and_reconstruct_intra_block(cm, xd, r, mbmi, 1, row_c,
1771 : col_c, tx_log2_c);
1772 : predict_and_reconstruct_intra_block(cm, xd, r, mbmi, 2, row_c,
1773 : col_c, tx_log2_c);
1774 : tu_idx_c++;
1775 : }
1776 : }
1777 : }
1778 :
1779 : // In 422 case, it's possilbe that Chroma has more TUs than Luma
1780 : while (tu_idx_c < tu_num_c) {
1781 : row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
1782 : col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
1783 : predict_and_reconstruct_intra_block(cm, xd, r, mbmi, 1, row_c, col_c,
1784 : tx_log2_c);
1785 : predict_and_reconstruct_intra_block(cm, xd, r, mbmi, 2, row_c, col_c,
1786 : tx_log2_c);
1787 : tu_idx_c++;
1788 : }
1789 : } else {
1790 : // Prediction
1791 : av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL,
1792 : AOMMAX(bsize, BLOCK_8X8));
1793 :
1794 : // Reconstruction
1795 : if (!mbmi->skip) {
1796 : int eobtotal = 0;
1797 : int tu_idx_c = 0;
1798 : int row_y, col_y, row_c, col_c;
1799 :
1800 : for (row_y = 0; row_y < tu_num_h_y; row_y++) {
1801 : for (col_y = 0; col_y < tu_num_w_y; col_y++) {
1802 : // luma
1803 : eobtotal += reconstruct_inter_block(cm, xd, r, mbmi->segment_id, 0,
1804 : row_y * tx_sz_y,
1805 : col_y * tx_sz_y, tx_log2_y);
1806 : // chroma
1807 : if (tu_idx_c < tu_num_c) {
1808 : row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
1809 : col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
1810 : eobtotal += reconstruct_inter_block(cm, xd, r, mbmi->segment_id,
1811 : 1, row_c, col_c, tx_log2_c);
1812 : eobtotal += reconstruct_inter_block(cm, xd, r, mbmi->segment_id,
1813 : 2, row_c, col_c, tx_log2_c);
1814 : tu_idx_c++;
1815 : }
1816 : }
1817 : }
1818 :
1819 : // In 422 case, it's possilbe that Chroma has more TUs than Luma
1820 : while (tu_idx_c < tu_num_c) {
1821 : row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
1822 : col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
1823 : eobtotal += reconstruct_inter_block(cm, xd, r, mbmi->segment_id, 1,
1824 : row_c, col_c, tx_log2_c);
1825 : eobtotal += reconstruct_inter_block(cm, xd, r, mbmi->segment_id, 2,
1826 : row_c, col_c, tx_log2_c);
1827 : tu_idx_c++;
1828 : }
1829 :
1830 : // TODO(CONFIG_COEF_INTERLEAVE owners): bring eob == 0 corner case
1831 : // into line with the defaut configuration
1832 : if (bsize >= BLOCK_8X8 && eobtotal == 0) mbmi->skip = 1;
1833 : }
1834 : }
1835 : }
1836 : #else // CONFIG_COEF_INTERLEAVE
1837 0 : if (!is_inter_block(mbmi)) {
1838 : int plane;
1839 : #if CONFIG_PALETTE
1840 0 : for (plane = 0; plane <= 1; ++plane) {
1841 0 : if (mbmi->palette_mode_info.palette_size[plane])
1842 0 : av1_decode_palette_tokens(xd, plane, r);
1843 : }
1844 : #endif // CONFIG_PALETTE
1845 0 : for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
1846 0 : const struct macroblockd_plane *const pd = &xd->plane[plane];
1847 0 : const TX_SIZE tx_size = get_tx_size(plane, xd);
1848 0 : const int stepr = tx_size_high_unit[tx_size];
1849 0 : const int stepc = tx_size_wide_unit[tx_size];
1850 : #if CONFIG_CB4X4
1851 : #if CONFIG_CHROMA_2X2
1852 : const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
1853 : #else
1854 0 : const BLOCK_SIZE plane_bsize =
1855 0 : AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
1856 : #endif // CONFIG_CHROMA_2X2
1857 : #else
1858 : const BLOCK_SIZE plane_bsize =
1859 : get_plane_block_size(AOMMAX(BLOCK_8X8, bsize), pd);
1860 : #endif
1861 : int row, col;
1862 0 : const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
1863 0 : const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
1864 : #if CONFIG_CB4X4
1865 0 : if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
1866 : pd->subsampling_y))
1867 0 : continue;
1868 : #endif
1869 :
1870 0 : for (row = 0; row < max_blocks_high; row += stepr)
1871 0 : for (col = 0; col < max_blocks_wide; col += stepc)
1872 0 : predict_and_reconstruct_intra_block(cm, xd, r, mbmi, plane, row, col,
1873 : tx_size);
1874 : }
1875 : } else {
1876 : int ref;
1877 :
1878 0 : for (ref = 0; ref < 1 + has_second_ref(mbmi); ++ref) {
1879 0 : const MV_REFERENCE_FRAME frame = mbmi->ref_frame[ref];
1880 0 : if (frame < LAST_FRAME) {
1881 : #if CONFIG_INTRABC
1882 : assert(is_intrabc_block(mbmi));
1883 : assert(frame == INTRA_FRAME);
1884 : assert(ref == 0);
1885 : #else
1886 0 : assert(0);
1887 : #endif // CONFIG_INTRABC
1888 : } else {
1889 0 : RefBuffer *ref_buf = &cm->frame_refs[frame - LAST_FRAME];
1890 :
1891 0 : xd->block_refs[ref] = ref_buf;
1892 0 : if ((!av1_is_valid_scale(&ref_buf->sf)))
1893 0 : aom_internal_error(xd->error_info, AOM_CODEC_UNSUP_BITSTREAM,
1894 : "Reference frame has invalid dimensions");
1895 0 : av1_setup_pre_planes(xd, ref, ref_buf->buf, mi_row, mi_col,
1896 0 : &ref_buf->sf);
1897 : }
1898 : }
1899 :
1900 : #if CONFIG_CB4X4
1901 0 : av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL, bsize);
1902 : #else
1903 : av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL,
1904 : AOMMAX(bsize, BLOCK_8X8));
1905 : #endif
1906 :
1907 : #if CONFIG_MOTION_VAR
1908 0 : if (mbmi->motion_mode == OBMC_CAUSAL) {
1909 : #if CONFIG_NCOBMC
1910 : av1_build_ncobmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
1911 : #else
1912 0 : av1_build_obmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
1913 : #endif
1914 : }
1915 : #endif // CONFIG_MOTION_VAR
1916 :
1917 : // Reconstruction
1918 0 : if (!mbmi->skip) {
1919 0 : int eobtotal = 0;
1920 : int plane;
1921 :
1922 0 : for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
1923 0 : const struct macroblockd_plane *const pd = &xd->plane[plane];
1924 : #if CONFIG_CB4X4
1925 : #if CONFIG_CHROMA_2X2
1926 : const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
1927 : #else
1928 0 : const BLOCK_SIZE plane_bsize =
1929 0 : AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
1930 : #endif // CONFIG_CHROMA_2X2
1931 : #else
1932 : const BLOCK_SIZE plane_bsize =
1933 : get_plane_block_size(AOMMAX(BLOCK_8X8, bsize), pd);
1934 : #endif
1935 0 : const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
1936 0 : const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
1937 : int row, col;
1938 :
1939 : #if CONFIG_CB4X4
1940 0 : if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
1941 : pd->subsampling_y))
1942 0 : continue;
1943 : #endif
1944 :
1945 : #if CONFIG_VAR_TX
1946 0 : const TX_SIZE max_tx_size = get_vartx_max_txsize(mbmi, plane_bsize);
1947 0 : const int bh_var_tx = tx_size_high_unit[max_tx_size];
1948 0 : const int bw_var_tx = tx_size_wide_unit[max_tx_size];
1949 0 : for (row = 0; row < max_blocks_high; row += bh_var_tx)
1950 0 : for (col = 0; col < max_blocks_wide; col += bw_var_tx)
1951 0 : decode_reconstruct_tx(cm, xd, r, mbmi, plane, plane_bsize, row, col,
1952 : max_tx_size, &eobtotal);
1953 : #else
1954 : const TX_SIZE tx_size = get_tx_size(plane, xd);
1955 : const int stepr = tx_size_high_unit[tx_size];
1956 : const int stepc = tx_size_wide_unit[tx_size];
1957 : for (row = 0; row < max_blocks_high; row += stepr)
1958 : for (col = 0; col < max_blocks_wide; col += stepc)
1959 : eobtotal += reconstruct_inter_block(cm, xd, r, mbmi->segment_id,
1960 : plane, row, col, tx_size);
1961 : #endif
1962 : }
1963 : }
1964 : }
1965 : #endif // CONFIG_COEF_INTERLEAVE
1966 :
1967 0 : int reader_corrupted_flag = aom_reader_has_error(r);
1968 0 : aom_merge_corrupted_flag(&xd->corrupted, reader_corrupted_flag);
1969 0 : }
1970 :
1971 : #if CONFIG_NCOBMC && CONFIG_MOTION_VAR
1972 : static void detoken_and_recon_sb(AV1Decoder *const pbi, MACROBLOCKD *const xd,
1973 : int mi_row, int mi_col, aom_reader *r,
1974 : BLOCK_SIZE bsize) {
1975 : AV1_COMMON *const cm = &pbi->common;
1976 : const int hbs = mi_size_wide[bsize] >> 1;
1977 : #if CONFIG_CB4X4
1978 : const int unify_bsize = 1;
1979 : #else
1980 : const int unify_bsize = 0;
1981 : #endif
1982 : #if CONFIG_EXT_PARTITION_TYPES
1983 : BLOCK_SIZE bsize2 = get_subsize(bsize, PARTITION_SPLIT);
1984 : #endif
1985 : PARTITION_TYPE partition;
1986 : BLOCK_SIZE subsize;
1987 : const int has_rows = (mi_row + hbs) < cm->mi_rows;
1988 : const int has_cols = (mi_col + hbs) < cm->mi_cols;
1989 :
1990 : if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;
1991 :
1992 : partition = get_partition(cm, mi_row, mi_col, bsize);
1993 : subsize = subsize_lookup[partition][bsize];
1994 :
1995 : if (!hbs && !unify_bsize) {
1996 : xd->bmode_blocks_wl = 1 >> !!(partition & PARTITION_VERT);
1997 : xd->bmode_blocks_hl = 1 >> !!(partition & PARTITION_HORZ);
1998 : decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
1999 : } else {
2000 : switch (partition) {
2001 : case PARTITION_NONE:
2002 : decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, bsize);
2003 : break;
2004 : case PARTITION_HORZ:
2005 : decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
2006 : if (has_rows)
2007 : decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col, r,
2008 : subsize);
2009 : break;
2010 : case PARTITION_VERT:
2011 : decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
2012 : if (has_cols)
2013 : decode_token_and_recon_block(pbi, xd, mi_row, mi_col + hbs, r,
2014 : subsize);
2015 : break;
2016 : case PARTITION_SPLIT:
2017 : detoken_and_recon_sb(pbi, xd, mi_row, mi_col, r, subsize);
2018 : detoken_and_recon_sb(pbi, xd, mi_row, mi_col + hbs, r, subsize);
2019 : detoken_and_recon_sb(pbi, xd, mi_row + hbs, mi_col, r, subsize);
2020 : detoken_and_recon_sb(pbi, xd, mi_row + hbs, mi_col + hbs, r, subsize);
2021 : break;
2022 : #if CONFIG_EXT_PARTITION_TYPES
2023 : case PARTITION_HORZ_A:
2024 : decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, bsize2);
2025 : decode_token_and_recon_block(pbi, xd, mi_row, mi_col + hbs, r, bsize2);
2026 : decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col, r, subsize);
2027 : break;
2028 : case PARTITION_HORZ_B:
2029 : decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
2030 : decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col, r, bsize2);
2031 : decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col + hbs, r,
2032 : bsize2);
2033 : break;
2034 : case PARTITION_VERT_A:
2035 : decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, bsize2);
2036 : decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col, r, bsize2);
2037 : decode_token_and_recon_block(pbi, xd, mi_row, mi_col + hbs, r, subsize);
2038 : break;
2039 : case PARTITION_VERT_B:
2040 : decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
2041 : decode_token_and_recon_block(pbi, xd, mi_row, mi_col + hbs, r, bsize2);
2042 : decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col + hbs, r,
2043 : bsize2);
2044 : break;
2045 : #endif
2046 : default: assert(0 && "Invalid partition type");
2047 : }
2048 : }
2049 : }
2050 : #endif
2051 :
2052 0 : static void decode_block(AV1Decoder *const pbi, MACROBLOCKD *const xd,
2053 : #if CONFIG_SUPERTX
2054 : int supertx_enabled,
2055 : #endif // CONFIG_SUPERTX
2056 : int mi_row, int mi_col, aom_reader *r,
2057 : #if CONFIG_EXT_PARTITION_TYPES
2058 : PARTITION_TYPE partition,
2059 : #endif // CONFIG_EXT_PARTITION_TYPES
2060 : BLOCK_SIZE bsize) {
2061 0 : decode_mbmi_block(pbi, xd,
2062 : #if CONFIG_SUPERTX
2063 : supertx_enabled,
2064 : #endif
2065 : mi_row, mi_col, r,
2066 : #if CONFIG_EXT_PARTITION_TYPES
2067 : partition,
2068 : #endif
2069 : bsize);
2070 : #if !(CONFIG_MOTION_VAR && CONFIG_NCOBMC)
2071 : #if CONFIG_SUPERTX
2072 : if (!supertx_enabled)
2073 : #endif // CONFIG_SUPERTX
2074 0 : decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, bsize);
2075 : #endif
2076 0 : }
2077 :
2078 0 : static PARTITION_TYPE read_partition(AV1_COMMON *cm, MACROBLOCKD *xd,
2079 : int mi_row, int mi_col, aom_reader *r,
2080 : int has_rows, int has_cols,
2081 : BLOCK_SIZE bsize) {
2082 : #if CONFIG_UNPOISON_PARTITION_CTX
2083 : const int ctx =
2084 : partition_plane_context(xd, mi_row, mi_col, has_rows, has_cols, bsize);
2085 : const aom_prob *const probs =
2086 : ctx < PARTITION_CONTEXTS ? cm->fc->partition_prob[ctx] : NULL;
2087 : FRAME_COUNTS *const counts = ctx < PARTITION_CONTEXTS ? xd->counts : NULL;
2088 : #else
2089 0 : const int ctx = partition_plane_context(xd, mi_row, mi_col, bsize);
2090 0 : const aom_prob *const probs = cm->fc->partition_prob[ctx];
2091 0 : FRAME_COUNTS *const counts = xd->counts;
2092 : #endif
2093 : PARTITION_TYPE p;
2094 : #if CONFIG_EC_ADAPT
2095 0 : FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
2096 : (void)cm;
2097 : #else
2098 : FRAME_CONTEXT *ec_ctx = cm->fc;
2099 : #endif
2100 :
2101 0 : aom_cdf_prob *partition_cdf = (ctx >= 0) ? ec_ctx->partition_cdf[ctx] : NULL;
2102 :
2103 0 : if (has_rows && has_cols)
2104 : #if CONFIG_EXT_PARTITION_TYPES
2105 : if (bsize <= BLOCK_8X8)
2106 : p = (PARTITION_TYPE)aom_read_symbol(r, partition_cdf, PARTITION_TYPES,
2107 : ACCT_STR);
2108 : else
2109 : p = (PARTITION_TYPE)aom_read_symbol(r, partition_cdf, EXT_PARTITION_TYPES,
2110 : ACCT_STR);
2111 : #else
2112 0 : p = (PARTITION_TYPE)aom_read_symbol(r, partition_cdf, PARTITION_TYPES,
2113 : ACCT_STR);
2114 : #endif // CONFIG_EXT_PARTITION_TYPES
2115 0 : else if (!has_rows && has_cols)
2116 0 : p = aom_read(r, probs[1], ACCT_STR) ? PARTITION_SPLIT : PARTITION_HORZ;
2117 0 : else if (has_rows && !has_cols)
2118 0 : p = aom_read(r, probs[2], ACCT_STR) ? PARTITION_SPLIT : PARTITION_VERT;
2119 : else
2120 0 : p = PARTITION_SPLIT;
2121 :
2122 0 : if (counts) ++counts->partition[ctx][p];
2123 :
2124 0 : return p;
2125 : }
2126 :
2127 : #if CONFIG_SUPERTX
2128 : static int read_skip(AV1_COMMON *cm, const MACROBLOCKD *xd, int segment_id,
2129 : aom_reader *r) {
2130 : if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) {
2131 : return 1;
2132 : } else {
2133 : const int ctx = av1_get_skip_context(xd);
2134 : const int skip = aom_read(r, cm->fc->skip_probs[ctx], ACCT_STR);
2135 : FRAME_COUNTS *counts = xd->counts;
2136 : if (counts) ++counts->skip[ctx][skip];
2137 : return skip;
2138 : }
2139 : }
2140 : #endif // CONFIG_SUPERTX
2141 :
2142 : // TODO(slavarnway): eliminate bsize and subsize in future commits
2143 0 : static void decode_partition(AV1Decoder *const pbi, MACROBLOCKD *const xd,
2144 : #if CONFIG_SUPERTX
2145 : int supertx_enabled,
2146 : #endif
2147 : int mi_row, int mi_col, aom_reader *r,
2148 : BLOCK_SIZE bsize, int n4x4_l2) {
2149 0 : AV1_COMMON *const cm = &pbi->common;
2150 0 : const int n8x8_l2 = n4x4_l2 - 1;
2151 0 : const int num_8x8_wh = mi_size_wide[bsize];
2152 0 : const int hbs = num_8x8_wh >> 1;
2153 : #if CONFIG_CB4X4
2154 0 : const int unify_bsize = 1;
2155 : #else
2156 : const int unify_bsize = 0;
2157 : #endif
2158 : PARTITION_TYPE partition;
2159 : BLOCK_SIZE subsize;
2160 : #if CONFIG_EXT_PARTITION_TYPES
2161 : BLOCK_SIZE bsize2 = get_subsize(bsize, PARTITION_SPLIT);
2162 : #endif
2163 0 : const int has_rows = (mi_row + hbs) < cm->mi_rows;
2164 0 : const int has_cols = (mi_col + hbs) < cm->mi_cols;
2165 : #if CONFIG_SUPERTX
2166 : const int read_token = !supertx_enabled;
2167 : int skip = 0;
2168 : TX_SIZE supertx_size = max_txsize_lookup[bsize];
2169 : const TileInfo *const tile = &xd->tile;
2170 : int txfm = DCT_DCT;
2171 : #endif // CONFIG_SUPERTX
2172 :
2173 0 : if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;
2174 :
2175 0 : partition = (bsize < BLOCK_8X8) ? PARTITION_NONE
2176 0 : : read_partition(cm, xd, mi_row, mi_col, r,
2177 : has_rows, has_cols, bsize);
2178 0 : subsize = subsize_lookup[partition][bsize]; // get_subsize(bsize, partition);
2179 :
2180 : #if CONFIG_PVQ
2181 : assert(partition < PARTITION_TYPES);
2182 : assert(subsize < BLOCK_SIZES);
2183 : #endif
2184 : #if CONFIG_SUPERTX
2185 : if (!frame_is_intra_only(cm) && partition != PARTITION_NONE &&
2186 : bsize <= MAX_SUPERTX_BLOCK_SIZE && !supertx_enabled && !xd->lossless[0]) {
2187 : const int supertx_context = partition_supertx_context_lookup[partition];
2188 : supertx_enabled = aom_read(
2189 : r, cm->fc->supertx_prob[supertx_context][supertx_size], ACCT_STR);
2190 : if (xd->counts)
2191 : xd->counts->supertx[supertx_context][supertx_size][supertx_enabled]++;
2192 : #if CONFIG_VAR_TX
2193 : if (supertx_enabled) xd->supertx_size = supertx_size;
2194 : #endif
2195 : }
2196 : #endif // CONFIG_SUPERTX
2197 0 : if (!hbs && !unify_bsize) {
2198 : // calculate bmode block dimensions (log 2)
2199 0 : xd->bmode_blocks_wl = 1 >> !!(partition & PARTITION_VERT);
2200 0 : xd->bmode_blocks_hl = 1 >> !!(partition & PARTITION_HORZ);
2201 0 : decode_block(pbi, xd,
2202 : #if CONFIG_SUPERTX
2203 : supertx_enabled,
2204 : #endif // CONFIG_SUPERTX
2205 : mi_row, mi_col, r,
2206 : #if CONFIG_EXT_PARTITION_TYPES
2207 : partition,
2208 : #endif // CONFIG_EXT_PARTITION_TYPES
2209 : subsize);
2210 : } else {
2211 0 : switch (partition) {
2212 : case PARTITION_NONE:
2213 0 : decode_block(pbi, xd,
2214 : #if CONFIG_SUPERTX
2215 : supertx_enabled,
2216 : #endif // CONFIG_SUPERTX
2217 : mi_row, mi_col, r,
2218 : #if CONFIG_EXT_PARTITION_TYPES
2219 : partition,
2220 : #endif // CONFIG_EXT_PARTITION_TYPES
2221 : subsize);
2222 0 : break;
2223 : case PARTITION_HORZ:
2224 0 : decode_block(pbi, xd,
2225 : #if CONFIG_SUPERTX
2226 : supertx_enabled,
2227 : #endif // CONFIG_SUPERTX
2228 : mi_row, mi_col, r,
2229 : #if CONFIG_EXT_PARTITION_TYPES
2230 : partition,
2231 : #endif // CONFIG_EXT_PARTITION_TYPES
2232 : subsize);
2233 0 : if (has_rows)
2234 0 : decode_block(pbi, xd,
2235 : #if CONFIG_SUPERTX
2236 : supertx_enabled,
2237 : #endif // CONFIG_SUPERTX
2238 : mi_row + hbs, mi_col, r,
2239 : #if CONFIG_EXT_PARTITION_TYPES
2240 : partition,
2241 : #endif // CONFIG_EXT_PARTITION_TYPES
2242 : subsize);
2243 0 : break;
2244 : case PARTITION_VERT:
2245 0 : decode_block(pbi, xd,
2246 : #if CONFIG_SUPERTX
2247 : supertx_enabled,
2248 : #endif // CONFIG_SUPERTX
2249 : mi_row, mi_col, r,
2250 : #if CONFIG_EXT_PARTITION_TYPES
2251 : partition,
2252 : #endif // CONFIG_EXT_PARTITION_TYPES
2253 : subsize);
2254 0 : if (has_cols)
2255 0 : decode_block(pbi, xd,
2256 : #if CONFIG_SUPERTX
2257 : supertx_enabled,
2258 : #endif // CONFIG_SUPERTX
2259 : mi_row, mi_col + hbs, r,
2260 : #if CONFIG_EXT_PARTITION_TYPES
2261 : partition,
2262 : #endif // CONFIG_EXT_PARTITION_TYPES
2263 : subsize);
2264 0 : break;
2265 : case PARTITION_SPLIT:
2266 0 : decode_partition(pbi, xd,
2267 : #if CONFIG_SUPERTX
2268 : supertx_enabled,
2269 : #endif // CONFIG_SUPERTX
2270 : mi_row, mi_col, r, subsize, n8x8_l2);
2271 0 : decode_partition(pbi, xd,
2272 : #if CONFIG_SUPERTX
2273 : supertx_enabled,
2274 : #endif // CONFIG_SUPERTX
2275 : mi_row, mi_col + hbs, r, subsize, n8x8_l2);
2276 0 : decode_partition(pbi, xd,
2277 : #if CONFIG_SUPERTX
2278 : supertx_enabled,
2279 : #endif // CONFIG_SUPERTX
2280 : mi_row + hbs, mi_col, r, subsize, n8x8_l2);
2281 0 : decode_partition(pbi, xd,
2282 : #if CONFIG_SUPERTX
2283 : supertx_enabled,
2284 : #endif // CONFIG_SUPERTX
2285 : mi_row + hbs, mi_col + hbs, r, subsize, n8x8_l2);
2286 0 : break;
2287 : #if CONFIG_EXT_PARTITION_TYPES
2288 : case PARTITION_HORZ_A:
2289 : decode_block(pbi, xd,
2290 : #if CONFIG_SUPERTX
2291 : supertx_enabled,
2292 : #endif
2293 : mi_row, mi_col, r, partition, bsize2);
2294 : decode_block(pbi, xd,
2295 : #if CONFIG_SUPERTX
2296 : supertx_enabled,
2297 : #endif
2298 : mi_row, mi_col + hbs, r, partition, bsize2);
2299 : decode_block(pbi, xd,
2300 : #if CONFIG_SUPERTX
2301 : supertx_enabled,
2302 : #endif
2303 : mi_row + hbs, mi_col, r, partition, subsize);
2304 : break;
2305 : case PARTITION_HORZ_B:
2306 : decode_block(pbi, xd,
2307 : #if CONFIG_SUPERTX
2308 : supertx_enabled,
2309 : #endif
2310 : mi_row, mi_col, r, partition, subsize);
2311 : decode_block(pbi, xd,
2312 : #if CONFIG_SUPERTX
2313 : supertx_enabled,
2314 : #endif
2315 : mi_row + hbs, mi_col, r, partition, bsize2);
2316 : decode_block(pbi, xd,
2317 : #if CONFIG_SUPERTX
2318 : supertx_enabled,
2319 : #endif
2320 : mi_row + hbs, mi_col + hbs, r, partition, bsize2);
2321 : break;
2322 : case PARTITION_VERT_A:
2323 : decode_block(pbi, xd,
2324 : #if CONFIG_SUPERTX
2325 : supertx_enabled,
2326 : #endif
2327 : mi_row, mi_col, r, partition, bsize2);
2328 : decode_block(pbi, xd,
2329 : #if CONFIG_SUPERTX
2330 : supertx_enabled,
2331 : #endif
2332 : mi_row + hbs, mi_col, r, partition, bsize2);
2333 : decode_block(pbi, xd,
2334 : #if CONFIG_SUPERTX
2335 : supertx_enabled,
2336 : #endif
2337 : mi_row, mi_col + hbs, r, partition, subsize);
2338 : break;
2339 : case PARTITION_VERT_B:
2340 : decode_block(pbi, xd,
2341 : #if CONFIG_SUPERTX
2342 : supertx_enabled,
2343 : #endif
2344 : mi_row, mi_col, r, partition, subsize);
2345 : decode_block(pbi, xd,
2346 : #if CONFIG_SUPERTX
2347 : supertx_enabled,
2348 : #endif
2349 : mi_row, mi_col + hbs, r, partition, bsize2);
2350 : decode_block(pbi, xd,
2351 : #if CONFIG_SUPERTX
2352 : supertx_enabled,
2353 : #endif
2354 : mi_row + hbs, mi_col + hbs, r, partition, bsize2);
2355 : break;
2356 : #endif
2357 0 : default: assert(0 && "Invalid partition type");
2358 : }
2359 : }
2360 :
2361 : #if CONFIG_SUPERTX
2362 : if (supertx_enabled && read_token) {
2363 : uint8_t *dst_buf[3];
2364 : int dst_stride[3], i;
2365 : int offset = mi_row * cm->mi_stride + mi_col;
2366 :
2367 : set_segment_id_supertx(cm, mi_row, mi_col, bsize);
2368 :
2369 : #if CONFIG_DELTA_Q
2370 : if (cm->delta_q_present_flag) {
2371 : for (i = 0; i < MAX_SEGMENTS; i++) {
2372 : xd->plane[0].seg_dequant[i][0] =
2373 : av1_dc_quant(xd->current_qindex, cm->y_dc_delta_q, cm->bit_depth);
2374 : xd->plane[0].seg_dequant[i][1] =
2375 : av1_ac_quant(xd->current_qindex, 0, cm->bit_depth);
2376 : xd->plane[1].seg_dequant[i][0] =
2377 : av1_dc_quant(xd->current_qindex, cm->uv_dc_delta_q, cm->bit_depth);
2378 : xd->plane[1].seg_dequant[i][1] =
2379 : av1_ac_quant(xd->current_qindex, cm->uv_ac_delta_q, cm->bit_depth);
2380 : xd->plane[2].seg_dequant[i][0] =
2381 : av1_dc_quant(xd->current_qindex, cm->uv_dc_delta_q, cm->bit_depth);
2382 : xd->plane[2].seg_dequant[i][1] =
2383 : av1_ac_quant(xd->current_qindex, cm->uv_ac_delta_q, cm->bit_depth);
2384 : }
2385 : }
2386 : #endif
2387 :
2388 : xd->mi = cm->mi_grid_visible + offset;
2389 : xd->mi[0] = cm->mi + offset;
2390 : set_mi_row_col(xd, tile, mi_row, mi_size_high[bsize], mi_col,
2391 : mi_size_wide[bsize],
2392 : #if CONFIG_DEPENDENT_HORZTILES
2393 : cm->dependent_horz_tiles,
2394 : #endif // CONFIG_DEPENDENT_HORZTILES
2395 : cm->mi_rows, cm->mi_cols);
2396 : set_skip_context(xd, mi_row, mi_col);
2397 : skip = read_skip(cm, xd, xd->mi[0]->mbmi.segment_id_supertx, r);
2398 : if (skip) {
2399 : av1_reset_skip_context(xd, mi_row, mi_col, bsize);
2400 : } else {
2401 : #if CONFIG_EXT_TX
2402 : if (get_ext_tx_types(supertx_size, bsize, 1, cm->reduced_tx_set_used) >
2403 : 1) {
2404 : const int eset =
2405 : get_ext_tx_set(supertx_size, bsize, 1, cm->reduced_tx_set_used);
2406 : if (eset > 0) {
2407 : txfm = aom_read_tree(r, av1_ext_tx_inter_tree[eset],
2408 : cm->fc->inter_ext_tx_prob[eset][supertx_size],
2409 : ACCT_STR);
2410 : if (xd->counts) ++xd->counts->inter_ext_tx[eset][supertx_size][txfm];
2411 : }
2412 : }
2413 : #else
2414 : if (supertx_size < TX_32X32) {
2415 : txfm = aom_read_tree(r, av1_ext_tx_tree,
2416 : cm->fc->inter_ext_tx_prob[supertx_size], ACCT_STR);
2417 : if (xd->counts) ++xd->counts->inter_ext_tx[supertx_size][txfm];
2418 : }
2419 : #endif // CONFIG_EXT_TX
2420 : }
2421 :
2422 : av1_setup_dst_planes(xd->plane, bsize, get_frame_new_buffer(cm), mi_row,
2423 : mi_col);
2424 : for (i = 0; i < MAX_MB_PLANE; i++) {
2425 : dst_buf[i] = xd->plane[i].dst.buf;
2426 : dst_stride[i] = xd->plane[i].dst.stride;
2427 : }
2428 : dec_predict_sb_complex(pbi, xd, tile, mi_row, mi_col, mi_row, mi_col, bsize,
2429 : bsize, dst_buf, dst_stride);
2430 :
2431 : if (!skip) {
2432 : int eobtotal = 0;
2433 : MB_MODE_INFO *mbmi;
2434 : set_offsets_topblock(cm, xd, tile, bsize, mi_row, mi_col);
2435 : mbmi = &xd->mi[0]->mbmi;
2436 : mbmi->tx_type = txfm;
2437 : assert(mbmi->segment_id_supertx != MAX_SEGMENTS);
2438 : for (i = 0; i < MAX_MB_PLANE; ++i) {
2439 : const struct macroblockd_plane *const pd = &xd->plane[i];
2440 : int row, col;
2441 : const TX_SIZE tx_size = get_tx_size(i, xd);
2442 : const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
2443 : const int stepr = tx_size_high_unit[tx_size];
2444 : const int stepc = tx_size_wide_unit[tx_size];
2445 : const int max_blocks_wide = max_block_wide(xd, plane_bsize, i);
2446 : const int max_blocks_high = max_block_high(xd, plane_bsize, i);
2447 :
2448 : for (row = 0; row < max_blocks_high; row += stepr)
2449 : for (col = 0; col < max_blocks_wide; col += stepc)
2450 : eobtotal += reconstruct_inter_block(
2451 : cm, xd, r, mbmi->segment_id_supertx, i, row, col, tx_size);
2452 : }
2453 : if ((unify_bsize || !(subsize < BLOCK_8X8)) && eobtotal == 0) skip = 1;
2454 : }
2455 : set_param_topblock(cm, xd, bsize, mi_row, mi_col, txfm, skip);
2456 : }
2457 : #endif // CONFIG_SUPERTX
2458 :
2459 : #if CONFIG_EXT_PARTITION_TYPES
2460 : update_ext_partition_context(xd, mi_row, mi_col, subsize, bsize, partition);
2461 : #else
2462 : // update partition context
2463 0 : if (bsize >= BLOCK_8X8 &&
2464 0 : (bsize == BLOCK_8X8 || partition != PARTITION_SPLIT))
2465 0 : update_partition_context(xd, mi_row, mi_col, subsize, bsize);
2466 : #endif // CONFIG_EXT_PARTITION_TYPES
2467 :
2468 : #if CONFIG_CDEF
2469 0 : if (bsize == cm->sb_size) {
2470 0 : if (!sb_all_skip(cm, mi_row, mi_col)) {
2471 0 : cm->mi_grid_visible[mi_row * cm->mi_stride + mi_col]->mbmi.cdef_strength =
2472 0 : aom_read_literal(r, cm->cdef_bits, ACCT_STR);
2473 : } else {
2474 0 : cm->mi_grid_visible[mi_row * cm->mi_stride + mi_col]->mbmi.cdef_strength =
2475 : -1;
2476 : }
2477 : }
2478 : #endif // CONFIG_CDEF
2479 : }
2480 :
2481 0 : static void setup_bool_decoder(const uint8_t *data, const uint8_t *data_end,
2482 : const size_t read_size,
2483 : struct aom_internal_error_info *error_info,
2484 : aom_reader *r,
2485 : #if CONFIG_ANS && ANS_MAX_SYMBOLS
2486 : int window_size,
2487 : #endif // CONFIG_ANS && ANS_MAX_SYMBOLS
2488 : aom_decrypt_cb decrypt_cb, void *decrypt_state) {
2489 : // Validate the calculated partition length. If the buffer
2490 : // described by the partition can't be fully read, then restrict
2491 : // it to the portion that can be (for EC mode) or throw an error.
2492 0 : if (!read_is_valid(data, read_size, data_end))
2493 0 : aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME,
2494 : "Truncated packet or corrupt tile length");
2495 :
2496 : #if CONFIG_ANS && ANS_MAX_SYMBOLS
2497 : r->window_size = window_size;
2498 : #endif
2499 0 : if (aom_reader_init(r, data, read_size, decrypt_cb, decrypt_state))
2500 0 : aom_internal_error(error_info, AOM_CODEC_MEM_ERROR,
2501 : "Failed to allocate bool decoder %d", 1);
2502 0 : }
2503 :
2504 : #if !CONFIG_PVQ && !CONFIG_EC_ADAPT && !CONFIG_LV_MAP
2505 : static void read_coef_probs_common(av1_coeff_probs_model *coef_probs,
2506 : aom_reader *r) {
2507 : int i, j, k, l, m;
2508 : #if CONFIG_EC_ADAPT
2509 : const int node_limit = UNCONSTRAINED_NODES - 1;
2510 : #else
2511 : const int node_limit = UNCONSTRAINED_NODES;
2512 : #endif
2513 :
2514 : if (aom_read_bit(r, ACCT_STR))
2515 : for (i = 0; i < PLANE_TYPES; ++i)
2516 : for (j = 0; j < REF_TYPES; ++j)
2517 : for (k = 0; k < COEF_BANDS; ++k)
2518 : for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l)
2519 : for (m = 0; m < node_limit; ++m)
2520 : av1_diff_update_prob(r, &coef_probs[i][j][k][l][m], ACCT_STR);
2521 : }
2522 :
2523 : static void read_coef_probs(FRAME_CONTEXT *fc, TX_MODE tx_mode, aom_reader *r) {
2524 : const TX_SIZE max_tx_size = tx_mode_to_biggest_tx_size[tx_mode];
2525 : TX_SIZE tx_size;
2526 : for (tx_size = 0; tx_size <= max_tx_size; ++tx_size)
2527 : read_coef_probs_common(fc->coef_probs[tx_size], r);
2528 : }
2529 : #endif
2530 :
2531 0 : static void setup_segmentation(AV1_COMMON *const cm,
2532 : struct aom_read_bit_buffer *rb) {
2533 0 : struct segmentation *const seg = &cm->seg;
2534 : int i, j;
2535 :
2536 0 : seg->update_map = 0;
2537 0 : seg->update_data = 0;
2538 :
2539 0 : seg->enabled = aom_rb_read_bit(rb);
2540 0 : if (!seg->enabled) return;
2541 :
2542 : // Segmentation map update
2543 0 : if (frame_is_intra_only(cm) || cm->error_resilient_mode) {
2544 0 : seg->update_map = 1;
2545 : } else {
2546 0 : seg->update_map = aom_rb_read_bit(rb);
2547 : }
2548 0 : if (seg->update_map) {
2549 0 : if (frame_is_intra_only(cm) || cm->error_resilient_mode) {
2550 0 : seg->temporal_update = 0;
2551 : } else {
2552 0 : seg->temporal_update = aom_rb_read_bit(rb);
2553 : }
2554 : }
2555 :
2556 : // Segmentation data update
2557 0 : seg->update_data = aom_rb_read_bit(rb);
2558 0 : if (seg->update_data) {
2559 0 : seg->abs_delta = aom_rb_read_bit(rb);
2560 :
2561 0 : av1_clearall_segfeatures(seg);
2562 :
2563 0 : for (i = 0; i < MAX_SEGMENTS; i++) {
2564 0 : for (j = 0; j < SEG_LVL_MAX; j++) {
2565 0 : int data = 0;
2566 0 : const int feature_enabled = aom_rb_read_bit(rb);
2567 0 : if (feature_enabled) {
2568 0 : av1_enable_segfeature(seg, i, j);
2569 0 : data = decode_unsigned_max(rb, av1_seg_feature_data_max(j));
2570 0 : if (av1_is_segfeature_signed(j))
2571 0 : data = aom_rb_read_bit(rb) ? -data : data;
2572 : }
2573 0 : av1_set_segdata(seg, i, j, data);
2574 : }
2575 : }
2576 : }
2577 : }
2578 :
2579 : #if CONFIG_LOOP_RESTORATION
2580 : static void decode_restoration_mode(AV1_COMMON *cm,
2581 : struct aom_read_bit_buffer *rb) {
2582 : int p;
2583 : RestorationInfo *rsi = &cm->rst_info[0];
2584 : if (aom_rb_read_bit(rb)) {
2585 : rsi->frame_restoration_type =
2586 : aom_rb_read_bit(rb) ? RESTORE_SGRPROJ : RESTORE_WIENER;
2587 : } else {
2588 : rsi->frame_restoration_type =
2589 : aom_rb_read_bit(rb) ? RESTORE_SWITCHABLE : RESTORE_NONE;
2590 : }
2591 : for (p = 1; p < MAX_MB_PLANE; ++p) {
2592 : rsi = &cm->rst_info[p];
2593 : if (aom_rb_read_bit(rb)) {
2594 : rsi->frame_restoration_type =
2595 : aom_rb_read_bit(rb) ? RESTORE_SGRPROJ : RESTORE_WIENER;
2596 : } else {
2597 : rsi->frame_restoration_type = RESTORE_NONE;
2598 : }
2599 : }
2600 :
2601 : cm->rst_info[0].restoration_tilesize = RESTORATION_TILESIZE_MAX;
2602 : cm->rst_info[1].restoration_tilesize = RESTORATION_TILESIZE_MAX;
2603 : cm->rst_info[2].restoration_tilesize = RESTORATION_TILESIZE_MAX;
2604 : if (cm->rst_info[0].frame_restoration_type != RESTORE_NONE ||
2605 : cm->rst_info[1].frame_restoration_type != RESTORE_NONE ||
2606 : cm->rst_info[2].frame_restoration_type != RESTORE_NONE) {
2607 : rsi = &cm->rst_info[0];
2608 : rsi->restoration_tilesize >>= aom_rb_read_bit(rb);
2609 : if (rsi->restoration_tilesize != RESTORATION_TILESIZE_MAX) {
2610 : rsi->restoration_tilesize >>= aom_rb_read_bit(rb);
2611 : }
2612 : cm->rst_info[1].restoration_tilesize = cm->rst_info[0].restoration_tilesize;
2613 : cm->rst_info[2].restoration_tilesize = cm->rst_info[0].restoration_tilesize;
2614 : }
2615 : }
2616 :
2617 : static void read_wiener_filter(WienerInfo *wiener_info,
2618 : WienerInfo *ref_wiener_info, aom_reader *rb) {
2619 : wiener_info->vfilter[0] = wiener_info->vfilter[WIENER_WIN - 1] =
2620 : aom_read_primitive_refsubexpfin(
2621 : rb, WIENER_FILT_TAP0_MAXV - WIENER_FILT_TAP0_MINV + 1,
2622 : WIENER_FILT_TAP0_SUBEXP_K,
2623 : ref_wiener_info->vfilter[0] - WIENER_FILT_TAP0_MINV, ACCT_STR) +
2624 : WIENER_FILT_TAP0_MINV;
2625 : wiener_info->vfilter[1] = wiener_info->vfilter[WIENER_WIN - 2] =
2626 : aom_read_primitive_refsubexpfin(
2627 : rb, WIENER_FILT_TAP1_MAXV - WIENER_FILT_TAP1_MINV + 1,
2628 : WIENER_FILT_TAP1_SUBEXP_K,
2629 : ref_wiener_info->vfilter[1] - WIENER_FILT_TAP1_MINV, ACCT_STR) +
2630 : WIENER_FILT_TAP1_MINV;
2631 : wiener_info->vfilter[2] = wiener_info->vfilter[WIENER_WIN - 3] =
2632 : aom_read_primitive_refsubexpfin(
2633 : rb, WIENER_FILT_TAP2_MAXV - WIENER_FILT_TAP2_MINV + 1,
2634 : WIENER_FILT_TAP2_SUBEXP_K,
2635 : ref_wiener_info->vfilter[2] - WIENER_FILT_TAP2_MINV, ACCT_STR) +
2636 : WIENER_FILT_TAP2_MINV;
2637 : // The central element has an implicit +WIENER_FILT_STEP
2638 : wiener_info->vfilter[WIENER_HALFWIN] =
2639 : -2 * (wiener_info->vfilter[0] + wiener_info->vfilter[1] +
2640 : wiener_info->vfilter[2]);
2641 :
2642 : wiener_info->hfilter[0] = wiener_info->hfilter[WIENER_WIN - 1] =
2643 : aom_read_primitive_refsubexpfin(
2644 : rb, WIENER_FILT_TAP0_MAXV - WIENER_FILT_TAP0_MINV + 1,
2645 : WIENER_FILT_TAP0_SUBEXP_K,
2646 : ref_wiener_info->hfilter[0] - WIENER_FILT_TAP0_MINV, ACCT_STR) +
2647 : WIENER_FILT_TAP0_MINV;
2648 : wiener_info->hfilter[1] = wiener_info->hfilter[WIENER_WIN - 2] =
2649 : aom_read_primitive_refsubexpfin(
2650 : rb, WIENER_FILT_TAP1_MAXV - WIENER_FILT_TAP1_MINV + 1,
2651 : WIENER_FILT_TAP1_SUBEXP_K,
2652 : ref_wiener_info->hfilter[1] - WIENER_FILT_TAP1_MINV, ACCT_STR) +
2653 : WIENER_FILT_TAP1_MINV;
2654 : wiener_info->hfilter[2] = wiener_info->hfilter[WIENER_WIN - 3] =
2655 : aom_read_primitive_refsubexpfin(
2656 : rb, WIENER_FILT_TAP2_MAXV - WIENER_FILT_TAP2_MINV + 1,
2657 : WIENER_FILT_TAP2_SUBEXP_K,
2658 : ref_wiener_info->hfilter[2] - WIENER_FILT_TAP2_MINV, ACCT_STR) +
2659 : WIENER_FILT_TAP2_MINV;
2660 : // The central element has an implicit +WIENER_FILT_STEP
2661 : wiener_info->hfilter[WIENER_HALFWIN] =
2662 : -2 * (wiener_info->hfilter[0] + wiener_info->hfilter[1] +
2663 : wiener_info->hfilter[2]);
2664 : memcpy(ref_wiener_info, wiener_info, sizeof(*wiener_info));
2665 : }
2666 :
2667 : static void read_sgrproj_filter(SgrprojInfo *sgrproj_info,
2668 : SgrprojInfo *ref_sgrproj_info, aom_reader *rb) {
2669 : sgrproj_info->ep = aom_read_literal(rb, SGRPROJ_PARAMS_BITS, ACCT_STR);
2670 : sgrproj_info->xqd[0] =
2671 : aom_read_primitive_refsubexpfin(
2672 : rb, SGRPROJ_PRJ_MAX0 - SGRPROJ_PRJ_MIN0 + 1, SGRPROJ_PRJ_SUBEXP_K,
2673 : ref_sgrproj_info->xqd[0] - SGRPROJ_PRJ_MIN0, ACCT_STR) +
2674 : SGRPROJ_PRJ_MIN0;
2675 : sgrproj_info->xqd[1] =
2676 : aom_read_primitive_refsubexpfin(
2677 : rb, SGRPROJ_PRJ_MAX1 - SGRPROJ_PRJ_MIN1 + 1, SGRPROJ_PRJ_SUBEXP_K,
2678 : ref_sgrproj_info->xqd[1] - SGRPROJ_PRJ_MIN1, ACCT_STR) +
2679 : SGRPROJ_PRJ_MIN1;
2680 : memcpy(ref_sgrproj_info, sgrproj_info, sizeof(*sgrproj_info));
2681 : }
2682 :
2683 : static void decode_restoration(AV1_COMMON *cm, aom_reader *rb) {
2684 : int i, p;
2685 : SgrprojInfo ref_sgrproj_info;
2686 : WienerInfo ref_wiener_info;
2687 : set_default_wiener(&ref_wiener_info);
2688 : set_default_sgrproj(&ref_sgrproj_info);
2689 : const int ntiles = av1_get_rest_ntiles(cm->width, cm->height,
2690 : cm->rst_info[0].restoration_tilesize,
2691 : NULL, NULL, NULL, NULL);
2692 : const int ntiles_uv = av1_get_rest_ntiles(
2693 : ROUND_POWER_OF_TWO(cm->width, cm->subsampling_x),
2694 : ROUND_POWER_OF_TWO(cm->height, cm->subsampling_y),
2695 : cm->rst_info[1].restoration_tilesize, NULL, NULL, NULL, NULL);
2696 : RestorationInfo *rsi = &cm->rst_info[0];
2697 : if (rsi->frame_restoration_type != RESTORE_NONE) {
2698 : if (rsi->frame_restoration_type == RESTORE_SWITCHABLE) {
2699 : for (i = 0; i < ntiles; ++i) {
2700 : rsi->restoration_type[i] =
2701 : aom_read_tree(rb, av1_switchable_restore_tree,
2702 : cm->fc->switchable_restore_prob, ACCT_STR);
2703 : if (rsi->restoration_type[i] == RESTORE_WIENER) {
2704 : read_wiener_filter(&rsi->wiener_info[i], &ref_wiener_info, rb);
2705 : } else if (rsi->restoration_type[i] == RESTORE_SGRPROJ) {
2706 : read_sgrproj_filter(&rsi->sgrproj_info[i], &ref_sgrproj_info, rb);
2707 : }
2708 : }
2709 : } else if (rsi->frame_restoration_type == RESTORE_WIENER) {
2710 : for (i = 0; i < ntiles; ++i) {
2711 : if (aom_read(rb, RESTORE_NONE_WIENER_PROB, ACCT_STR)) {
2712 : rsi->restoration_type[i] = RESTORE_WIENER;
2713 : read_wiener_filter(&rsi->wiener_info[i], &ref_wiener_info, rb);
2714 : } else {
2715 : rsi->restoration_type[i] = RESTORE_NONE;
2716 : }
2717 : }
2718 : } else if (rsi->frame_restoration_type == RESTORE_SGRPROJ) {
2719 : for (i = 0; i < ntiles; ++i) {
2720 : if (aom_read(rb, RESTORE_NONE_SGRPROJ_PROB, ACCT_STR)) {
2721 : rsi->restoration_type[i] = RESTORE_SGRPROJ;
2722 : read_sgrproj_filter(&rsi->sgrproj_info[i], &ref_sgrproj_info, rb);
2723 : } else {
2724 : rsi->restoration_type[i] = RESTORE_NONE;
2725 : }
2726 : }
2727 : }
2728 : }
2729 : for (p = 1; p < MAX_MB_PLANE; ++p) {
2730 : set_default_wiener(&ref_wiener_info);
2731 : set_default_sgrproj(&ref_sgrproj_info);
2732 : rsi = &cm->rst_info[p];
2733 : if (rsi->frame_restoration_type == RESTORE_WIENER) {
2734 : for (i = 0; i < ntiles_uv; ++i) {
2735 : if (ntiles_uv > 1)
2736 : rsi->restoration_type[i] =
2737 : aom_read(rb, RESTORE_NONE_WIENER_PROB, ACCT_STR) ? RESTORE_WIENER
2738 : : RESTORE_NONE;
2739 : else
2740 : rsi->restoration_type[i] = RESTORE_WIENER;
2741 : if (rsi->restoration_type[i] == RESTORE_WIENER) {
2742 : read_wiener_filter(&rsi->wiener_info[i], &ref_wiener_info, rb);
2743 : }
2744 : }
2745 : } else if (rsi->frame_restoration_type == RESTORE_SGRPROJ) {
2746 : for (i = 0; i < ntiles_uv; ++i) {
2747 : if (ntiles_uv > 1)
2748 : rsi->restoration_type[i] =
2749 : aom_read(rb, RESTORE_NONE_SGRPROJ_PROB, ACCT_STR)
2750 : ? RESTORE_SGRPROJ
2751 : : RESTORE_NONE;
2752 : else
2753 : rsi->restoration_type[i] = RESTORE_SGRPROJ;
2754 : if (rsi->restoration_type[i] == RESTORE_SGRPROJ) {
2755 : read_sgrproj_filter(&rsi->sgrproj_info[i], &ref_sgrproj_info, rb);
2756 : }
2757 : }
2758 : } else if (rsi->frame_restoration_type != RESTORE_NONE) {
2759 : assert(0);
2760 : }
2761 : }
2762 : }
2763 : #endif // CONFIG_LOOP_RESTORATION
2764 :
2765 0 : static void setup_loopfilter(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
2766 0 : struct loopfilter *lf = &cm->lf;
2767 0 : lf->filter_level = aom_rb_read_literal(rb, 6);
2768 0 : lf->sharpness_level = aom_rb_read_literal(rb, 3);
2769 :
2770 : // Read in loop filter deltas applied at the MB level based on mode or ref
2771 : // frame.
2772 0 : lf->mode_ref_delta_update = 0;
2773 :
2774 0 : lf->mode_ref_delta_enabled = aom_rb_read_bit(rb);
2775 0 : if (lf->mode_ref_delta_enabled) {
2776 0 : lf->mode_ref_delta_update = aom_rb_read_bit(rb);
2777 0 : if (lf->mode_ref_delta_update) {
2778 : int i;
2779 :
2780 0 : for (i = 0; i < TOTAL_REFS_PER_FRAME; i++)
2781 0 : if (aom_rb_read_bit(rb))
2782 0 : lf->ref_deltas[i] = aom_rb_read_inv_signed_literal(rb, 6);
2783 :
2784 0 : for (i = 0; i < MAX_MODE_LF_DELTAS; i++)
2785 0 : if (aom_rb_read_bit(rb))
2786 0 : lf->mode_deltas[i] = aom_rb_read_inv_signed_literal(rb, 6);
2787 : }
2788 : }
2789 0 : }
2790 :
2791 : #if CONFIG_CDEF
2792 0 : static void setup_cdef(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
2793 : int i;
2794 0 : cm->cdef_dering_damping = aom_rb_read_literal(rb, 1) + 5;
2795 0 : cm->cdef_clpf_damping = aom_rb_read_literal(rb, 2) + 3;
2796 0 : cm->cdef_bits = aom_rb_read_literal(rb, 2);
2797 0 : cm->nb_cdef_strengths = 1 << cm->cdef_bits;
2798 0 : for (i = 0; i < cm->nb_cdef_strengths; i++) {
2799 0 : cm->cdef_strengths[i] = aom_rb_read_literal(rb, CDEF_STRENGTH_BITS);
2800 0 : cm->cdef_uv_strengths[i] = aom_rb_read_literal(rb, CDEF_STRENGTH_BITS);
2801 : }
2802 0 : }
2803 : #endif // CONFIG_CDEF
2804 :
2805 0 : static INLINE int read_delta_q(struct aom_read_bit_buffer *rb) {
2806 0 : return aom_rb_read_bit(rb) ? aom_rb_read_inv_signed_literal(rb, 6) : 0;
2807 : }
2808 :
2809 0 : static void setup_quantization(AV1_COMMON *const cm,
2810 : struct aom_read_bit_buffer *rb) {
2811 0 : cm->base_qindex = aom_rb_read_literal(rb, QINDEX_BITS);
2812 0 : cm->y_dc_delta_q = read_delta_q(rb);
2813 0 : cm->uv_dc_delta_q = read_delta_q(rb);
2814 0 : cm->uv_ac_delta_q = read_delta_q(rb);
2815 0 : cm->dequant_bit_depth = cm->bit_depth;
2816 : #if CONFIG_AOM_QM
2817 : cm->using_qmatrix = aom_rb_read_bit(rb);
2818 : if (cm->using_qmatrix) {
2819 : cm->min_qmlevel = aom_rb_read_literal(rb, QM_LEVEL_BITS);
2820 : cm->max_qmlevel = aom_rb_read_literal(rb, QM_LEVEL_BITS);
2821 : } else {
2822 : cm->min_qmlevel = 0;
2823 : cm->max_qmlevel = 0;
2824 : }
2825 : #endif
2826 0 : }
2827 :
2828 : // Build y/uv dequant values based on segmentation.
2829 0 : static void setup_segmentation_dequant(AV1_COMMON *const cm) {
2830 : #if CONFIG_AOM_QM
2831 : const int using_qm = cm->using_qmatrix;
2832 : const int minqm = cm->min_qmlevel;
2833 : const int maxqm = cm->max_qmlevel;
2834 : #endif
2835 : // When segmentation is disabled, only the first value is used. The
2836 : // remaining are don't cares.
2837 0 : const int max_segments = cm->seg.enabled ? MAX_SEGMENTS : 1;
2838 0 : for (int i = 0; i < max_segments; ++i) {
2839 0 : const int qindex = av1_get_qindex(&cm->seg, i, cm->base_qindex);
2840 0 : cm->y_dequant[i][0] = av1_dc_quant(qindex, cm->y_dc_delta_q, cm->bit_depth);
2841 0 : cm->y_dequant[i][1] = av1_ac_quant(qindex, 0, cm->bit_depth);
2842 0 : cm->uv_dequant[i][0] =
2843 0 : av1_dc_quant(qindex, cm->uv_dc_delta_q, cm->bit_depth);
2844 0 : cm->uv_dequant[i][1] =
2845 0 : av1_ac_quant(qindex, cm->uv_ac_delta_q, cm->bit_depth);
2846 : #if CONFIG_AOM_QM
2847 : const int lossless = qindex == 0 && cm->y_dc_delta_q == 0 &&
2848 : cm->uv_dc_delta_q == 0 && cm->uv_ac_delta_q == 0;
2849 : // NB: depends on base index so there is only 1 set per frame
2850 : // No quant weighting when lossless or signalled not using QM
2851 : const int qmlevel = (lossless || using_qm == 0)
2852 : ? NUM_QM_LEVELS - 1
2853 : : aom_get_qmlevel(cm->base_qindex, minqm, maxqm);
2854 : for (int j = 0; j < TX_SIZES_ALL; ++j) {
2855 : cm->y_iqmatrix[i][1][j] = aom_iqmatrix(cm, qmlevel, 0, j, 1);
2856 : cm->y_iqmatrix[i][0][j] = aom_iqmatrix(cm, qmlevel, 0, j, 0);
2857 : cm->uv_iqmatrix[i][1][j] = aom_iqmatrix(cm, qmlevel, 1, j, 1);
2858 : cm->uv_iqmatrix[i][0][j] = aom_iqmatrix(cm, qmlevel, 1, j, 0);
2859 : }
2860 : #endif // CONFIG_AOM_QM
2861 : #if CONFIG_NEW_QUANT
2862 : for (int dq = 0; dq < QUANT_PROFILES; dq++) {
2863 : for (int b = 0; b < COEF_BANDS; ++b) {
2864 : av1_get_dequant_val_nuq(cm->y_dequant[i][b != 0], b,
2865 : cm->y_dequant_nuq[i][dq][b], NULL, dq);
2866 : av1_get_dequant_val_nuq(cm->uv_dequant[i][b != 0], b,
2867 : cm->uv_dequant_nuq[i][dq][b], NULL, dq);
2868 : }
2869 : }
2870 : #endif // CONFIG_NEW_QUANT
2871 : }
2872 0 : }
2873 :
2874 0 : static InterpFilter read_frame_interp_filter(struct aom_read_bit_buffer *rb) {
2875 0 : return aom_rb_read_bit(rb) ? SWITCHABLE
2876 0 : : aom_rb_read_literal(rb, LOG_SWITCHABLE_FILTERS);
2877 : }
2878 :
2879 0 : static void setup_render_size(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
2880 0 : cm->render_width = cm->width;
2881 0 : cm->render_height = cm->height;
2882 0 : if (aom_rb_read_bit(rb))
2883 0 : av1_read_frame_size(rb, &cm->render_width, &cm->render_height);
2884 0 : }
2885 :
2886 : #if CONFIG_FRAME_SUPERRES
2887 : // TODO(afergs): make "struct aom_read_bit_buffer *const rb"?
2888 : static void setup_superres_size(AV1_COMMON *const cm,
2889 : struct aom_read_bit_buffer *rb, int *width,
2890 : int *height) {
2891 : // TODO(afergs): Save input resolution - it's the upscaled resolution
2892 : if (aom_rb_read_bit(rb)) {
2893 : cm->superres_scale_numerator =
2894 : (uint8_t)aom_rb_read_literal(rb, SUPERRES_SCALE_BITS);
2895 : cm->superres_scale_numerator += SUPERRES_SCALE_NUMERATOR_MIN;
2896 : // Don't edit cm->width or cm->height directly, or the buffers won't get
2897 : // resized correctly
2898 : // TODO(afergs): Should the render resolution not be modified? It's the same
2899 : // by default (ie. when it isn't sent)...
2900 : // resize_context_buffers() will change cm->width to equal cm->render_width,
2901 : // then they'll be the same again
2902 : *width = *width * cm->superres_scale_numerator / SUPERRES_SCALE_DENOMINATOR;
2903 : *height =
2904 : *width * cm->superres_scale_numerator / SUPERRES_SCALE_DENOMINATOR;
2905 : } else {
2906 : // 1:1 scaling - ie. no scaling, scale not provided
2907 : cm->superres_scale_numerator = SUPERRES_SCALE_DENOMINATOR;
2908 : }
2909 : }
2910 : #endif // CONFIG_FRAME_SUPERRES
2911 :
2912 0 : static void resize_mv_buffer(AV1_COMMON *cm) {
2913 0 : aom_free(cm->cur_frame->mvs);
2914 0 : cm->cur_frame->mi_rows = cm->mi_rows;
2915 0 : cm->cur_frame->mi_cols = cm->mi_cols;
2916 0 : CHECK_MEM_ERROR(cm, cm->cur_frame->mvs,
2917 : (MV_REF *)aom_calloc(cm->mi_rows * cm->mi_cols,
2918 : sizeof(*cm->cur_frame->mvs)));
2919 0 : }
2920 :
2921 0 : static void resize_context_buffers(AV1_COMMON *cm, int width, int height) {
2922 : #if CONFIG_SIZE_LIMIT
2923 0 : if (width > DECODE_WIDTH_LIMIT || height > DECODE_HEIGHT_LIMIT)
2924 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
2925 : "Dimensions of %dx%d beyond allowed size of %dx%d.",
2926 : width, height, DECODE_WIDTH_LIMIT, DECODE_HEIGHT_LIMIT);
2927 : #endif
2928 0 : if (cm->width != width || cm->height != height) {
2929 0 : const int new_mi_rows =
2930 0 : ALIGN_POWER_OF_TWO(height, MI_SIZE_LOG2) >> MI_SIZE_LOG2;
2931 0 : const int new_mi_cols =
2932 0 : ALIGN_POWER_OF_TWO(width, MI_SIZE_LOG2) >> MI_SIZE_LOG2;
2933 :
2934 : // Allocations in av1_alloc_context_buffers() depend on individual
2935 : // dimensions as well as the overall size.
2936 0 : if (new_mi_cols > cm->mi_cols || new_mi_rows > cm->mi_rows) {
2937 0 : if (av1_alloc_context_buffers(cm, width, height))
2938 0 : aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
2939 : "Failed to allocate context buffers");
2940 : } else {
2941 0 : av1_set_mb_mi(cm, width, height);
2942 : }
2943 0 : av1_init_context_buffers(cm);
2944 0 : cm->width = width;
2945 0 : cm->height = height;
2946 : }
2947 0 : if (cm->cur_frame->mvs == NULL || cm->mi_rows > cm->cur_frame->mi_rows ||
2948 0 : cm->mi_cols > cm->cur_frame->mi_cols) {
2949 0 : resize_mv_buffer(cm);
2950 : }
2951 0 : }
2952 :
2953 0 : static void setup_frame_size(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
2954 : int width, height;
2955 0 : BufferPool *const pool = cm->buffer_pool;
2956 0 : av1_read_frame_size(rb, &width, &height);
2957 0 : setup_render_size(cm, rb);
2958 : #if CONFIG_FRAME_SUPERRES
2959 : setup_superres_size(cm, rb, &width, &height);
2960 : #endif // CONFIG_FRAME_SUPERRES
2961 0 : resize_context_buffers(cm, width, height);
2962 :
2963 0 : lock_buffer_pool(pool);
2964 0 : if (aom_realloc_frame_buffer(
2965 : get_frame_new_buffer(cm), cm->width, cm->height, cm->subsampling_x,
2966 : cm->subsampling_y,
2967 : #if CONFIG_HIGHBITDEPTH
2968 : cm->use_highbitdepth,
2969 : #endif
2970 : AOM_BORDER_IN_PIXELS, cm->byte_alignment,
2971 0 : &pool->frame_bufs[cm->new_fb_idx].raw_frame_buffer, pool->get_fb_cb,
2972 : pool->cb_priv)) {
2973 0 : unlock_buffer_pool(pool);
2974 0 : aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
2975 : "Failed to allocate frame buffer");
2976 : }
2977 0 : unlock_buffer_pool(pool);
2978 :
2979 0 : pool->frame_bufs[cm->new_fb_idx].buf.subsampling_x = cm->subsampling_x;
2980 0 : pool->frame_bufs[cm->new_fb_idx].buf.subsampling_y = cm->subsampling_y;
2981 0 : pool->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth;
2982 0 : pool->frame_bufs[cm->new_fb_idx].buf.color_space = cm->color_space;
2983 0 : pool->frame_bufs[cm->new_fb_idx].buf.color_range = cm->color_range;
2984 0 : pool->frame_bufs[cm->new_fb_idx].buf.render_width = cm->render_width;
2985 0 : pool->frame_bufs[cm->new_fb_idx].buf.render_height = cm->render_height;
2986 0 : }
2987 :
2988 0 : static INLINE int valid_ref_frame_img_fmt(aom_bit_depth_t ref_bit_depth,
2989 : int ref_xss, int ref_yss,
2990 : aom_bit_depth_t this_bit_depth,
2991 : int this_xss, int this_yss) {
2992 0 : return ref_bit_depth == this_bit_depth && ref_xss == this_xss &&
2993 : ref_yss == this_yss;
2994 : }
2995 :
2996 0 : static void setup_frame_size_with_refs(AV1_COMMON *cm,
2997 : struct aom_read_bit_buffer *rb) {
2998 : int width, height;
2999 0 : int found = 0, i;
3000 0 : int has_valid_ref_frame = 0;
3001 0 : BufferPool *const pool = cm->buffer_pool;
3002 0 : for (i = 0; i < INTER_REFS_PER_FRAME; ++i) {
3003 0 : if (aom_rb_read_bit(rb)) {
3004 0 : YV12_BUFFER_CONFIG *const buf = cm->frame_refs[i].buf;
3005 0 : width = buf->y_crop_width;
3006 0 : height = buf->y_crop_height;
3007 0 : cm->render_width = buf->render_width;
3008 0 : cm->render_height = buf->render_height;
3009 0 : found = 1;
3010 0 : break;
3011 : }
3012 : }
3013 :
3014 0 : if (!found) {
3015 0 : av1_read_frame_size(rb, &width, &height);
3016 0 : setup_render_size(cm, rb);
3017 : #if CONFIG_FRAME_SUPERRES
3018 : setup_superres_size(cm, rb, &width, &height);
3019 : #endif // CONFIG_FRAME_SUPERRES
3020 : }
3021 :
3022 0 : if (width <= 0 || height <= 0)
3023 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
3024 : "Invalid frame size");
3025 :
3026 : // Check to make sure at least one of frames that this frame references
3027 : // has valid dimensions.
3028 0 : for (i = 0; i < INTER_REFS_PER_FRAME; ++i) {
3029 0 : RefBuffer *const ref_frame = &cm->frame_refs[i];
3030 0 : has_valid_ref_frame |=
3031 0 : valid_ref_frame_size(ref_frame->buf->y_crop_width,
3032 0 : ref_frame->buf->y_crop_height, width, height);
3033 : }
3034 0 : if (!has_valid_ref_frame)
3035 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
3036 : "Referenced frame has invalid size");
3037 0 : for (i = 0; i < INTER_REFS_PER_FRAME; ++i) {
3038 0 : RefBuffer *const ref_frame = &cm->frame_refs[i];
3039 0 : if (!valid_ref_frame_img_fmt(ref_frame->buf->bit_depth,
3040 0 : ref_frame->buf->subsampling_x,
3041 0 : ref_frame->buf->subsampling_y, cm->bit_depth,
3042 : cm->subsampling_x, cm->subsampling_y))
3043 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
3044 : "Referenced frame has incompatible color format");
3045 : }
3046 :
3047 0 : resize_context_buffers(cm, width, height);
3048 :
3049 0 : lock_buffer_pool(pool);
3050 0 : if (aom_realloc_frame_buffer(
3051 : get_frame_new_buffer(cm), cm->width, cm->height, cm->subsampling_x,
3052 : cm->subsampling_y,
3053 : #if CONFIG_HIGHBITDEPTH
3054 : cm->use_highbitdepth,
3055 : #endif
3056 : AOM_BORDER_IN_PIXELS, cm->byte_alignment,
3057 0 : &pool->frame_bufs[cm->new_fb_idx].raw_frame_buffer, pool->get_fb_cb,
3058 : pool->cb_priv)) {
3059 0 : unlock_buffer_pool(pool);
3060 0 : aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
3061 : "Failed to allocate frame buffer");
3062 : }
3063 0 : unlock_buffer_pool(pool);
3064 :
3065 0 : pool->frame_bufs[cm->new_fb_idx].buf.subsampling_x = cm->subsampling_x;
3066 0 : pool->frame_bufs[cm->new_fb_idx].buf.subsampling_y = cm->subsampling_y;
3067 0 : pool->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth;
3068 0 : pool->frame_bufs[cm->new_fb_idx].buf.color_space = cm->color_space;
3069 0 : pool->frame_bufs[cm->new_fb_idx].buf.color_range = cm->color_range;
3070 0 : pool->frame_bufs[cm->new_fb_idx].buf.render_width = cm->render_width;
3071 0 : pool->frame_bufs[cm->new_fb_idx].buf.render_height = cm->render_height;
3072 0 : }
3073 :
3074 0 : static void read_tile_info(AV1Decoder *const pbi,
3075 : struct aom_read_bit_buffer *const rb) {
3076 0 : AV1_COMMON *const cm = &pbi->common;
3077 : #if CONFIG_EXT_TILE
3078 : cm->tile_encoding_mode = aom_rb_read_literal(rb, 1);
3079 : // Read the tile width/height
3080 : #if CONFIG_EXT_PARTITION
3081 : if (cm->sb_size == BLOCK_128X128) {
3082 : cm->tile_width = aom_rb_read_literal(rb, 5) + 1;
3083 : cm->tile_height = aom_rb_read_literal(rb, 5) + 1;
3084 : } else
3085 : #endif // CONFIG_EXT_PARTITION
3086 : {
3087 : cm->tile_width = aom_rb_read_literal(rb, 6) + 1;
3088 : cm->tile_height = aom_rb_read_literal(rb, 6) + 1;
3089 : }
3090 :
3091 : #if CONFIG_LOOPFILTERING_ACROSS_TILES
3092 : cm->loop_filter_across_tiles_enabled = aom_rb_read_bit(rb);
3093 : #endif // CONFIG_LOOPFILTERING_ACROSS_TILES
3094 :
3095 : cm->tile_width <<= cm->mib_size_log2;
3096 : cm->tile_height <<= cm->mib_size_log2;
3097 :
3098 : cm->tile_width = AOMMIN(cm->tile_width, cm->mi_cols);
3099 : cm->tile_height = AOMMIN(cm->tile_height, cm->mi_rows);
3100 :
3101 : // Get the number of tiles
3102 : cm->tile_cols = 1;
3103 : while (cm->tile_cols * cm->tile_width < cm->mi_cols) ++cm->tile_cols;
3104 :
3105 : cm->tile_rows = 1;
3106 : while (cm->tile_rows * cm->tile_height < cm->mi_rows) ++cm->tile_rows;
3107 :
3108 : if (cm->tile_cols * cm->tile_rows > 1) {
3109 : // Read the number of bytes used to store tile size
3110 : pbi->tile_col_size_bytes = aom_rb_read_literal(rb, 2) + 1;
3111 : pbi->tile_size_bytes = aom_rb_read_literal(rb, 2) + 1;
3112 : }
3113 :
3114 : #if CONFIG_DEPENDENT_HORZTILES
3115 : if (cm->tile_rows <= 1)
3116 : cm->dependent_horz_tiles = aom_rb_read_bit(rb);
3117 : else
3118 : cm->dependent_horz_tiles = 0;
3119 : #endif
3120 : #else
3121 : int min_log2_tile_cols, max_log2_tile_cols, max_ones;
3122 0 : av1_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);
3123 :
3124 : // columns
3125 0 : max_ones = max_log2_tile_cols - min_log2_tile_cols;
3126 0 : cm->log2_tile_cols = min_log2_tile_cols;
3127 0 : while (max_ones-- && aom_rb_read_bit(rb)) cm->log2_tile_cols++;
3128 :
3129 0 : if (cm->log2_tile_cols > 6)
3130 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
3131 : "Invalid number of tile columns");
3132 :
3133 : // rows
3134 0 : cm->log2_tile_rows = aom_rb_read_bit(rb);
3135 0 : if (cm->log2_tile_rows) cm->log2_tile_rows += aom_rb_read_bit(rb);
3136 : #if CONFIG_DEPENDENT_HORZTILES
3137 : if (cm->log2_tile_rows != 0)
3138 : cm->dependent_horz_tiles = aom_rb_read_bit(rb);
3139 : else
3140 : cm->dependent_horz_tiles = 0;
3141 : #endif
3142 : #if CONFIG_LOOPFILTERING_ACROSS_TILES
3143 0 : cm->loop_filter_across_tiles_enabled = aom_rb_read_bit(rb);
3144 : #endif // CONFIG_LOOPFILTERING_ACROSS_TILES
3145 :
3146 0 : cm->tile_cols = 1 << cm->log2_tile_cols;
3147 0 : cm->tile_rows = 1 << cm->log2_tile_rows;
3148 :
3149 0 : cm->tile_width = ALIGN_POWER_OF_TWO(cm->mi_cols, MAX_MIB_SIZE_LOG2);
3150 0 : cm->tile_width >>= cm->log2_tile_cols;
3151 0 : cm->tile_height = ALIGN_POWER_OF_TWO(cm->mi_rows, MAX_MIB_SIZE_LOG2);
3152 0 : cm->tile_height >>= cm->log2_tile_rows;
3153 :
3154 : // round to integer multiples of superblock size
3155 0 : cm->tile_width = ALIGN_POWER_OF_TWO(cm->tile_width, MAX_MIB_SIZE_LOG2);
3156 0 : cm->tile_height = ALIGN_POWER_OF_TWO(cm->tile_height, MAX_MIB_SIZE_LOG2);
3157 :
3158 : // tile size magnitude
3159 : #if !CONFIG_TILE_GROUPS
3160 : if (cm->tile_rows > 1 || cm->tile_cols > 1)
3161 : #endif
3162 0 : pbi->tile_size_bytes = aom_rb_read_literal(rb, 2) + 1;
3163 : #endif // CONFIG_EXT_TILE
3164 :
3165 : #if CONFIG_TILE_GROUPS
3166 : // Store an index to the location of the tile group information
3167 0 : pbi->tg_size_bit_offset = rb->bit_offset;
3168 0 : pbi->tg_size = 1 << (cm->log2_tile_rows + cm->log2_tile_cols);
3169 0 : if (cm->log2_tile_rows + cm->log2_tile_cols > 0) {
3170 0 : pbi->tg_start =
3171 0 : aom_rb_read_literal(rb, cm->log2_tile_rows + cm->log2_tile_cols);
3172 0 : pbi->tg_size =
3173 0 : 1 + aom_rb_read_literal(rb, cm->log2_tile_rows + cm->log2_tile_cols);
3174 : }
3175 : #endif
3176 0 : }
3177 :
3178 0 : static int mem_get_varsize(const uint8_t *src, int sz) {
3179 0 : switch (sz) {
3180 0 : case 1: return src[0];
3181 0 : case 2: return mem_get_le16(src);
3182 0 : case 3: return mem_get_le24(src);
3183 0 : case 4: return mem_get_le32(src);
3184 0 : default: assert("Invalid size" && 0); return -1;
3185 : }
3186 : }
3187 :
3188 : #if CONFIG_EXT_TILE
3189 : // Reads the next tile returning its size and adjusting '*data' accordingly
3190 : // based on 'is_last'.
3191 : static void get_tile_buffer(const uint8_t *const data_end,
3192 : struct aom_internal_error_info *error_info,
3193 : const uint8_t **data, aom_decrypt_cb decrypt_cb,
3194 : void *decrypt_state,
3195 : TileBufferDec (*const tile_buffers)[MAX_TILE_COLS],
3196 : int tile_size_bytes, int col, int row,
3197 : unsigned int tile_encoding_mode) {
3198 : size_t size;
3199 :
3200 : size_t copy_size = 0;
3201 : const uint8_t *copy_data = NULL;
3202 :
3203 : if (!read_is_valid(*data, tile_size_bytes, data_end))
3204 : aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME,
3205 : "Truncated packet or corrupt tile length");
3206 : if (decrypt_cb) {
3207 : uint8_t be_data[4];
3208 : decrypt_cb(decrypt_state, *data, be_data, tile_size_bytes);
3209 :
3210 : // Only read number of bytes in cm->tile_size_bytes.
3211 : size = mem_get_varsize(be_data, tile_size_bytes);
3212 : } else {
3213 : size = mem_get_varsize(*data, tile_size_bytes);
3214 : }
3215 :
3216 : // If cm->tile_encoding_mode = 1 (i.e. TILE_VR), then the top bit of the tile
3217 : // header indicates copy mode.
3218 : if (tile_encoding_mode && (size >> (tile_size_bytes * 8 - 1)) == 1) {
3219 : // The remaining bits in the top byte signal the row offset
3220 : int offset = (size >> (tile_size_bytes - 1) * 8) & 0x7f;
3221 :
3222 : // Currently, only use tiles in same column as reference tiles.
3223 : copy_data = tile_buffers[row - offset][col].data;
3224 : copy_size = tile_buffers[row - offset][col].size;
3225 : size = 0;
3226 : }
3227 :
3228 : *data += tile_size_bytes;
3229 :
3230 : if (size > (size_t)(data_end - *data))
3231 : aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME,
3232 : "Truncated packet or corrupt tile size");
3233 :
3234 : if (size > 0) {
3235 : tile_buffers[row][col].data = *data;
3236 : tile_buffers[row][col].size = size;
3237 : } else {
3238 : tile_buffers[row][col].data = copy_data;
3239 : tile_buffers[row][col].size = copy_size;
3240 : }
3241 :
3242 : *data += size;
3243 :
3244 : tile_buffers[row][col].raw_data_end = *data;
3245 : }
3246 :
3247 : static void get_tile_buffers(
3248 : AV1Decoder *pbi, const uint8_t *data, const uint8_t *data_end,
3249 : TileBufferDec (*const tile_buffers)[MAX_TILE_COLS]) {
3250 : AV1_COMMON *const cm = &pbi->common;
3251 : const int tile_cols = cm->tile_cols;
3252 : const int tile_rows = cm->tile_rows;
3253 : const int have_tiles = tile_cols * tile_rows > 1;
3254 :
3255 : if (!have_tiles) {
3256 : const size_t tile_size = data_end - data;
3257 : tile_buffers[0][0].data = data;
3258 : tile_buffers[0][0].size = tile_size;
3259 : tile_buffers[0][0].raw_data_end = NULL;
3260 : } else {
3261 : // We locate only the tile buffers that are required, which are the ones
3262 : // specified by pbi->dec_tile_col and pbi->dec_tile_row. Also, we always
3263 : // need the last (bottom right) tile buffer, as we need to know where the
3264 : // end of the compressed frame buffer is for proper superframe decoding.
3265 :
3266 : const uint8_t *tile_col_data_end[MAX_TILE_COLS];
3267 : const uint8_t *const data_start = data;
3268 :
3269 : const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows);
3270 : const int single_row = pbi->dec_tile_row >= 0;
3271 : const int tile_rows_start = single_row ? dec_tile_row : 0;
3272 : const int tile_rows_end = single_row ? tile_rows_start + 1 : tile_rows;
3273 : const int dec_tile_col = AOMMIN(pbi->dec_tile_col, tile_cols);
3274 : const int single_col = pbi->dec_tile_col >= 0;
3275 : const int tile_cols_start = single_col ? dec_tile_col : 0;
3276 : const int tile_cols_end = single_col ? tile_cols_start + 1 : tile_cols;
3277 :
3278 : const int tile_col_size_bytes = pbi->tile_col_size_bytes;
3279 : const int tile_size_bytes = pbi->tile_size_bytes;
3280 :
3281 : size_t tile_col_size;
3282 : int r, c;
3283 :
3284 : // Read tile column sizes for all columns (we need the last tile buffer)
3285 : for (c = 0; c < tile_cols; ++c) {
3286 : const int is_last = c == tile_cols - 1;
3287 : if (!is_last) {
3288 : tile_col_size = mem_get_varsize(data, tile_col_size_bytes);
3289 : data += tile_col_size_bytes;
3290 : tile_col_data_end[c] = data + tile_col_size;
3291 : } else {
3292 : tile_col_size = data_end - data;
3293 : tile_col_data_end[c] = data_end;
3294 : }
3295 : data += tile_col_size;
3296 : }
3297 :
3298 : data = data_start;
3299 :
3300 : // Read the required tile sizes.
3301 : for (c = tile_cols_start; c < tile_cols_end; ++c) {
3302 : const int is_last = c == tile_cols - 1;
3303 :
3304 : if (c > 0) data = tile_col_data_end[c - 1];
3305 :
3306 : if (!is_last) data += tile_col_size_bytes;
3307 :
3308 : // Get the whole of the last column, otherwise stop at the required tile.
3309 : for (r = 0; r < (is_last ? tile_rows : tile_rows_end); ++r) {
3310 : tile_buffers[r][c].col = c;
3311 :
3312 : get_tile_buffer(tile_col_data_end[c], &pbi->common.error, &data,
3313 : pbi->decrypt_cb, pbi->decrypt_state, tile_buffers,
3314 : tile_size_bytes, c, r, cm->tile_encoding_mode);
3315 : }
3316 : }
3317 :
3318 : // If we have not read the last column, then read it to get the last tile.
3319 : if (tile_cols_end != tile_cols) {
3320 : c = tile_cols - 1;
3321 :
3322 : data = tile_col_data_end[c - 1];
3323 :
3324 : for (r = 0; r < tile_rows; ++r) {
3325 : tile_buffers[r][c].col = c;
3326 :
3327 : get_tile_buffer(tile_col_data_end[c], &pbi->common.error, &data,
3328 : pbi->decrypt_cb, pbi->decrypt_state, tile_buffers,
3329 : tile_size_bytes, c, r, cm->tile_encoding_mode);
3330 : }
3331 : }
3332 : }
3333 : }
3334 : #else
3335 : // Reads the next tile returning its size and adjusting '*data' accordingly
3336 : // based on 'is_last'.
3337 0 : static void get_tile_buffer(const uint8_t *const data_end,
3338 : const int tile_size_bytes, int is_last,
3339 : struct aom_internal_error_info *error_info,
3340 : const uint8_t **data, aom_decrypt_cb decrypt_cb,
3341 : void *decrypt_state, TileBufferDec *const buf) {
3342 : size_t size;
3343 :
3344 0 : if (!is_last) {
3345 0 : if (!read_is_valid(*data, tile_size_bytes, data_end))
3346 0 : aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME,
3347 : "Truncated packet or corrupt tile length");
3348 :
3349 0 : if (decrypt_cb) {
3350 : uint8_t be_data[4];
3351 0 : decrypt_cb(decrypt_state, *data, be_data, tile_size_bytes);
3352 0 : size = mem_get_varsize(be_data, tile_size_bytes);
3353 : } else {
3354 0 : size = mem_get_varsize(*data, tile_size_bytes);
3355 : }
3356 0 : *data += tile_size_bytes;
3357 :
3358 0 : if (size > (size_t)(data_end - *data))
3359 0 : aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME,
3360 : "Truncated packet or corrupt tile size");
3361 : } else {
3362 0 : size = data_end - *data;
3363 : }
3364 :
3365 0 : buf->data = *data;
3366 0 : buf->size = size;
3367 :
3368 0 : *data += size;
3369 0 : }
3370 :
3371 0 : static void get_tile_buffers(
3372 : AV1Decoder *pbi, const uint8_t *data, const uint8_t *data_end,
3373 : TileBufferDec (*const tile_buffers)[MAX_TILE_COLS]) {
3374 0 : AV1_COMMON *const cm = &pbi->common;
3375 : #if CONFIG_TILE_GROUPS
3376 : int r, c;
3377 0 : const int tile_cols = cm->tile_cols;
3378 0 : const int tile_rows = cm->tile_rows;
3379 0 : int tc = 0;
3380 0 : int first_tile_in_tg = 0;
3381 : struct aom_read_bit_buffer rb_tg_hdr;
3382 : uint8_t clear_data[MAX_AV1_HEADER_SIZE];
3383 0 : const int num_tiles = tile_rows * tile_cols;
3384 0 : const int num_bits = OD_ILOG(num_tiles) - 1;
3385 0 : const size_t hdr_size = pbi->uncomp_hdr_size + pbi->first_partition_size;
3386 0 : const int tg_size_bit_offset = pbi->tg_size_bit_offset;
3387 : #if CONFIG_DEPENDENT_HORZTILES
3388 : int tile_group_start_col = 0;
3389 : int tile_group_start_row = 0;
3390 : #endif
3391 :
3392 0 : for (r = 0; r < tile_rows; ++r) {
3393 0 : for (c = 0; c < tile_cols; ++c, ++tc) {
3394 0 : TileBufferDec *const buf = &tile_buffers[r][c];
3395 0 : const int is_last = (r == tile_rows - 1) && (c == tile_cols - 1);
3396 0 : const size_t hdr_offset = (tc && tc == first_tile_in_tg) ? hdr_size : 0;
3397 :
3398 0 : buf->col = c;
3399 0 : if (hdr_offset) {
3400 0 : init_read_bit_buffer(pbi, &rb_tg_hdr, data, data_end, clear_data);
3401 0 : rb_tg_hdr.bit_offset = tg_size_bit_offset;
3402 0 : if (num_tiles) {
3403 0 : pbi->tg_start = aom_rb_read_literal(&rb_tg_hdr, num_bits);
3404 0 : pbi->tg_size = 1 + aom_rb_read_literal(&rb_tg_hdr, num_bits);
3405 : #if CONFIG_DEPENDENT_HORZTILES
3406 : tile_group_start_row = r;
3407 : tile_group_start_col = c;
3408 : #endif
3409 : }
3410 : }
3411 0 : first_tile_in_tg += tc == first_tile_in_tg ? pbi->tg_size : 0;
3412 0 : data += hdr_offset;
3413 0 : get_tile_buffer(data_end, pbi->tile_size_bytes, is_last,
3414 : &pbi->common.error, &data, pbi->decrypt_cb,
3415 : pbi->decrypt_state, buf);
3416 : #if CONFIG_DEPENDENT_HORZTILES
3417 : cm->tile_group_start_row[r][c] = tile_group_start_row;
3418 : cm->tile_group_start_col[r][c] = tile_group_start_col;
3419 : #endif
3420 : }
3421 : }
3422 : #else
3423 : int r, c;
3424 : const int tile_cols = cm->tile_cols;
3425 : const int tile_rows = cm->tile_rows;
3426 :
3427 : for (r = 0; r < tile_rows; ++r) {
3428 : for (c = 0; c < tile_cols; ++c) {
3429 : const int is_last = (r == tile_rows - 1) && (c == tile_cols - 1);
3430 : TileBufferDec *const buf = &tile_buffers[r][c];
3431 : buf->col = c;
3432 : get_tile_buffer(data_end, pbi->tile_size_bytes, is_last, &cm->error,
3433 : &data, pbi->decrypt_cb, pbi->decrypt_state, buf);
3434 : }
3435 : }
3436 : #endif
3437 0 : }
3438 : #endif // CONFIG_EXT_TILE
3439 :
3440 : #if CONFIG_PVQ
3441 : static void daala_dec_init(AV1_COMMON *const cm, daala_dec_ctx *daala_dec,
3442 : aom_reader *r) {
3443 : daala_dec->r = r;
3444 :
3445 : // TODO(yushin) : activity masking info needs be signaled by a bitstream
3446 : daala_dec->use_activity_masking = AV1_PVQ_ENABLE_ACTIVITY_MASKING;
3447 :
3448 : #if !CONFIG_DAALA_DIST
3449 : daala_dec->use_activity_masking = 0;
3450 : #endif
3451 :
3452 : if (daala_dec->use_activity_masking)
3453 : daala_dec->qm = OD_HVS_QM;
3454 : else
3455 : daala_dec->qm = OD_FLAT_QM;
3456 :
3457 : od_init_qm(daala_dec->state.qm, daala_dec->state.qm_inv,
3458 : daala_dec->qm == OD_HVS_QM ? OD_QM8_Q4_HVS : OD_QM8_Q4_FLAT);
3459 :
3460 : if (daala_dec->use_activity_masking) {
3461 : int pli;
3462 : int use_masking = daala_dec->use_activity_masking;
3463 : int segment_id = 0;
3464 : int qindex = av1_get_qindex(&cm->seg, segment_id, cm->base_qindex);
3465 :
3466 : for (pli = 0; pli < MAX_MB_PLANE; pli++) {
3467 : int i;
3468 : int q;
3469 :
3470 : q = qindex;
3471 : if (q <= OD_DEFAULT_QMS[use_masking][0][pli].interp_q << OD_COEFF_SHIFT) {
3472 : od_interp_qm(&daala_dec->state.pvq_qm_q4[pli][0], q,
3473 : &OD_DEFAULT_QMS[use_masking][0][pli], NULL);
3474 : } else {
3475 : i = 0;
3476 : while (OD_DEFAULT_QMS[use_masking][i + 1][pli].qm_q4 != NULL &&
3477 : q > OD_DEFAULT_QMS[use_masking][i + 1][pli].interp_q
3478 : << OD_COEFF_SHIFT) {
3479 : i++;
3480 : }
3481 : od_interp_qm(&daala_dec->state.pvq_qm_q4[pli][0], q,
3482 : &OD_DEFAULT_QMS[use_masking][i][pli],
3483 : &OD_DEFAULT_QMS[use_masking][i + 1][pli]);
3484 : }
3485 : }
3486 : }
3487 : }
3488 : #endif // #if CONFIG_PVQ
3489 :
3490 0 : static const uint8_t *decode_tiles(AV1Decoder *pbi, const uint8_t *data,
3491 : const uint8_t *data_end) {
3492 0 : AV1_COMMON *const cm = &pbi->common;
3493 0 : const AVxWorkerInterface *const winterface = aom_get_worker_interface();
3494 0 : const int tile_cols = cm->tile_cols;
3495 0 : const int tile_rows = cm->tile_rows;
3496 0 : const int n_tiles = tile_cols * tile_rows;
3497 0 : TileBufferDec(*const tile_buffers)[MAX_TILE_COLS] = pbi->tile_buffers;
3498 : #if CONFIG_EXT_TILE
3499 : const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows);
3500 : const int single_row = pbi->dec_tile_row >= 0;
3501 : const int tile_rows_start = single_row ? dec_tile_row : 0;
3502 : const int tile_rows_end = single_row ? dec_tile_row + 1 : tile_rows;
3503 : const int dec_tile_col = AOMMIN(pbi->dec_tile_col, tile_cols);
3504 : const int single_col = pbi->dec_tile_col >= 0;
3505 : const int tile_cols_start = single_col ? dec_tile_col : 0;
3506 : const int tile_cols_end = single_col ? tile_cols_start + 1 : tile_cols;
3507 : const int inv_col_order = pbi->inv_tile_order && !single_col;
3508 : const int inv_row_order = pbi->inv_tile_order && !single_row;
3509 : #else
3510 0 : const int tile_rows_start = 0;
3511 0 : const int tile_rows_end = tile_rows;
3512 0 : const int tile_cols_start = 0;
3513 0 : const int tile_cols_end = tile_cols;
3514 0 : const int inv_col_order = pbi->inv_tile_order;
3515 0 : const int inv_row_order = pbi->inv_tile_order;
3516 : #endif // CONFIG_EXT_TILE
3517 : int tile_row, tile_col;
3518 :
3519 0 : if (cm->lf.filter_level && !cm->skip_loop_filter &&
3520 0 : pbi->lf_worker.data1 == NULL) {
3521 0 : CHECK_MEM_ERROR(cm, pbi->lf_worker.data1,
3522 : aom_memalign(32, sizeof(LFWorkerData)));
3523 0 : pbi->lf_worker.hook = (AVxWorkerHook)av1_loop_filter_worker;
3524 0 : if (pbi->max_threads > 1 && !winterface->reset(&pbi->lf_worker)) {
3525 0 : aom_internal_error(&cm->error, AOM_CODEC_ERROR,
3526 : "Loop filter thread creation failed");
3527 : }
3528 : }
3529 :
3530 0 : if (cm->lf.filter_level && !cm->skip_loop_filter) {
3531 0 : LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1;
3532 : // Be sure to sync as we might be resuming after a failed frame decode.
3533 0 : winterface->sync(&pbi->lf_worker);
3534 0 : av1_loop_filter_data_reset(lf_data, get_frame_new_buffer(cm), cm,
3535 0 : pbi->mb.plane);
3536 : }
3537 :
3538 0 : assert(tile_rows <= MAX_TILE_ROWS);
3539 0 : assert(tile_cols <= MAX_TILE_COLS);
3540 :
3541 0 : get_tile_buffers(pbi, data, data_end, tile_buffers);
3542 :
3543 0 : if (pbi->tile_data == NULL || n_tiles != pbi->allocated_tiles) {
3544 0 : aom_free(pbi->tile_data);
3545 0 : CHECK_MEM_ERROR(cm, pbi->tile_data,
3546 : aom_memalign(32, n_tiles * (sizeof(*pbi->tile_data))));
3547 0 : pbi->allocated_tiles = n_tiles;
3548 : }
3549 : #if CONFIG_ACCOUNTING
3550 : if (pbi->acct_enabled) {
3551 : aom_accounting_reset(&pbi->accounting);
3552 : }
3553 : #endif
3554 : // Load all tile information into tile_data.
3555 0 : for (tile_row = tile_rows_start; tile_row < tile_rows_end; ++tile_row) {
3556 0 : for (tile_col = tile_cols_start; tile_col < tile_cols_end; ++tile_col) {
3557 0 : const TileBufferDec *const buf = &tile_buffers[tile_row][tile_col];
3558 0 : TileData *const td = pbi->tile_data + tile_cols * tile_row + tile_col;
3559 :
3560 0 : td->cm = cm;
3561 0 : td->xd = pbi->mb;
3562 0 : td->xd.corrupted = 0;
3563 0 : td->xd.counts =
3564 0 : cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD
3565 : ? &cm->counts
3566 0 : : NULL;
3567 0 : av1_zero(td->dqcoeff);
3568 : #if CONFIG_PVQ
3569 : av1_zero(td->pvq_ref_coeff);
3570 : #endif
3571 0 : av1_tile_init(&td->xd.tile, td->cm, tile_row, tile_col);
3572 0 : setup_bool_decoder(buf->data, data_end, buf->size, &cm->error,
3573 : &td->bit_reader,
3574 : #if CONFIG_ANS && ANS_MAX_SYMBOLS
3575 : 1 << cm->ans_window_size_log2,
3576 : #endif // CONFIG_ANS && ANS_MAX_SYMBOLS
3577 : pbi->decrypt_cb, pbi->decrypt_state);
3578 : #if CONFIG_ACCOUNTING
3579 : if (pbi->acct_enabled) {
3580 : td->bit_reader.accounting = &pbi->accounting;
3581 : } else {
3582 : td->bit_reader.accounting = NULL;
3583 : }
3584 : #endif
3585 0 : av1_init_macroblockd(cm, &td->xd,
3586 : #if CONFIG_PVQ
3587 : td->pvq_ref_coeff,
3588 : #endif
3589 : #if CONFIG_CFL
3590 : &td->cfl,
3591 : #endif
3592 0 : td->dqcoeff);
3593 :
3594 : #if CONFIG_EC_ADAPT
3595 : // Initialise the tile context from the frame context
3596 0 : td->tctx = *cm->fc;
3597 0 : td->xd.tile_ctx = &td->tctx;
3598 : #endif
3599 :
3600 : #if CONFIG_PVQ
3601 : daala_dec_init(cm, &td->xd.daala_dec, &td->bit_reader);
3602 : td->xd.daala_dec.state.adapt = &td->tctx.pvq_context;
3603 : #endif
3604 :
3605 : #if CONFIG_PALETTE
3606 0 : td->xd.plane[0].color_index_map = td->color_index_map[0];
3607 0 : td->xd.plane[1].color_index_map = td->color_index_map[1];
3608 : #endif // CONFIG_PALETTE
3609 : }
3610 : }
3611 :
3612 0 : for (tile_row = tile_rows_start; tile_row < tile_rows_end; ++tile_row) {
3613 0 : const int row = inv_row_order ? tile_rows - 1 - tile_row : tile_row;
3614 0 : int mi_row = 0;
3615 : TileInfo tile_info;
3616 :
3617 0 : av1_tile_set_row(&tile_info, cm, row);
3618 :
3619 0 : for (tile_col = tile_cols_start; tile_col < tile_cols_end; ++tile_col) {
3620 0 : const int col = inv_col_order ? tile_cols - 1 - tile_col : tile_col;
3621 0 : TileData *const td = pbi->tile_data + tile_cols * row + col;
3622 : #if CONFIG_ACCOUNTING
3623 : if (pbi->acct_enabled) {
3624 : td->bit_reader.accounting->last_tell_frac =
3625 : aom_reader_tell_frac(&td->bit_reader);
3626 : }
3627 : #endif
3628 :
3629 0 : av1_tile_set_col(&tile_info, cm, col);
3630 :
3631 : #if CONFIG_DEPENDENT_HORZTILES
3632 : #if CONFIG_TILE_GROUPS
3633 : av1_tile_set_tg_boundary(&tile_info, cm, tile_row, tile_col);
3634 : if (!cm->dependent_horz_tiles || tile_row == 0 ||
3635 : tile_info.tg_horz_boundary) {
3636 : #else
3637 : if (!cm->dependent_horz_tiles || tile_row == 0) {
3638 : #endif
3639 : av1_zero_above_context(cm, tile_info.mi_col_start,
3640 : tile_info.mi_col_end);
3641 : }
3642 : #else
3643 0 : av1_zero_above_context(cm, tile_info.mi_col_start, tile_info.mi_col_end);
3644 : #endif
3645 :
3646 0 : for (mi_row = tile_info.mi_row_start; mi_row < tile_info.mi_row_end;
3647 0 : mi_row += cm->mib_size) {
3648 : int mi_col;
3649 :
3650 0 : av1_zero_left_context(&td->xd);
3651 :
3652 0 : for (mi_col = tile_info.mi_col_start; mi_col < tile_info.mi_col_end;
3653 0 : mi_col += cm->mib_size) {
3654 0 : av1_update_boundary_info(cm, &tile_info, mi_row, mi_col);
3655 0 : decode_partition(pbi, &td->xd,
3656 : #if CONFIG_SUPERTX
3657 : 0,
3658 : #endif // CONFIG_SUPERTX
3659 0 : mi_row, mi_col, &td->bit_reader, cm->sb_size,
3660 0 : b_width_log2_lookup[cm->sb_size]);
3661 : #if CONFIG_NCOBMC && CONFIG_MOTION_VAR
3662 : detoken_and_recon_sb(pbi, &td->xd, mi_row, mi_col, &td->bit_reader,
3663 : cm->sb_size);
3664 : #endif
3665 : }
3666 0 : aom_merge_corrupted_flag(&pbi->mb.corrupted, td->xd.corrupted);
3667 0 : if (pbi->mb.corrupted)
3668 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
3669 : "Failed to decode tile data");
3670 : }
3671 : }
3672 :
3673 0 : assert(mi_row > 0);
3674 :
3675 : // when Parallel deblocking is enabled, deblocking should not
3676 : // be interleaved with decoding. Instead, deblocking should be done
3677 : // after the entire frame is decoded.
3678 : #if !CONFIG_VAR_TX && !CONFIG_PARALLEL_DEBLOCKING && !CONFIG_CB4X4
3679 : // Loopfilter one tile row.
3680 : // Note: If out-of-order tile decoding is used(for example, inv_row_order
3681 : // = 1), the loopfiltering has be done after all tile rows are decoded.
3682 : if (!inv_row_order && cm->lf.filter_level && !cm->skip_loop_filter) {
3683 : LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1;
3684 : const int lf_start = AOMMAX(0, tile_info.mi_row_start - cm->mib_size);
3685 : const int lf_end = tile_info.mi_row_end - cm->mib_size;
3686 :
3687 : // Delay the loopfilter if the first tile row is only
3688 : // a single superblock high.
3689 : if (lf_end <= 0) continue;
3690 :
3691 : // Decoding has completed. Finish up the loop filter in this thread.
3692 : if (tile_info.mi_row_end >= cm->mi_rows) continue;
3693 :
3694 : winterface->sync(&pbi->lf_worker);
3695 : lf_data->start = lf_start;
3696 : lf_data->stop = lf_end;
3697 : if (pbi->max_threads > 1) {
3698 : winterface->launch(&pbi->lf_worker);
3699 : } else {
3700 : winterface->execute(&pbi->lf_worker);
3701 : }
3702 : }
3703 : #endif // !CONFIG_VAR_TX && !CONFIG_PARALLEL_DEBLOCKING
3704 :
3705 : // After loopfiltering, the last 7 row pixels in each superblock row may
3706 : // still be changed by the longest loopfilter of the next superblock row.
3707 0 : if (cm->frame_parallel_decode)
3708 0 : av1_frameworker_broadcast(pbi->cur_buf, mi_row << cm->mib_size_log2);
3709 : }
3710 :
3711 : #if CONFIG_VAR_TX || CONFIG_CB4X4
3712 : // Loopfilter the whole frame.
3713 0 : av1_loop_filter_frame(get_frame_new_buffer(cm), cm, &pbi->mb,
3714 : cm->lf.filter_level, 0, 0);
3715 : #else
3716 : #if CONFIG_PARALLEL_DEBLOCKING
3717 : // Loopfilter all rows in the frame in the frame.
3718 : if (cm->lf.filter_level && !cm->skip_loop_filter) {
3719 : LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1;
3720 : winterface->sync(&pbi->lf_worker);
3721 : lf_data->start = 0;
3722 : lf_data->stop = cm->mi_rows;
3723 : winterface->execute(&pbi->lf_worker);
3724 : }
3725 : #else
3726 : // Loopfilter remaining rows in the frame.
3727 : if (cm->lf.filter_level && !cm->skip_loop_filter) {
3728 : LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1;
3729 : winterface->sync(&pbi->lf_worker);
3730 : lf_data->start = lf_data->stop;
3731 : lf_data->stop = cm->mi_rows;
3732 : winterface->execute(&pbi->lf_worker);
3733 : }
3734 : #endif // CONFIG_PARALLEL_DEBLOCKING
3735 : #endif // CONFIG_VAR_TX
3736 0 : if (cm->frame_parallel_decode)
3737 0 : av1_frameworker_broadcast(pbi->cur_buf, INT_MAX);
3738 :
3739 : #if CONFIG_EXT_TILE
3740 : if (n_tiles == 1) {
3741 : #if CONFIG_ANS
3742 : return data_end;
3743 : #else
3744 : // Find the end of the single tile buffer
3745 : return aom_reader_find_end(&pbi->tile_data->bit_reader);
3746 : #endif // CONFIG_ANS
3747 : } else {
3748 : // Return the end of the last tile buffer
3749 : return tile_buffers[tile_rows - 1][tile_cols - 1].raw_data_end;
3750 : }
3751 : #else
3752 : #if CONFIG_ANS
3753 : return data_end;
3754 : #else
3755 : {
3756 : // Get last tile data.
3757 0 : TileData *const td = pbi->tile_data + tile_cols * tile_rows - 1;
3758 0 : return aom_reader_find_end(&td->bit_reader);
3759 : }
3760 : #endif // CONFIG_ANS
3761 : #endif // CONFIG_EXT_TILE
3762 : }
3763 :
3764 0 : static int tile_worker_hook(TileWorkerData *const tile_data,
3765 : const TileInfo *const tile) {
3766 0 : AV1Decoder *const pbi = tile_data->pbi;
3767 0 : const AV1_COMMON *const cm = &pbi->common;
3768 : int mi_row, mi_col;
3769 :
3770 0 : if (setjmp(tile_data->error_info.jmp)) {
3771 0 : tile_data->error_info.setjmp = 0;
3772 0 : aom_merge_corrupted_flag(&tile_data->xd.corrupted, 1);
3773 0 : return 0;
3774 : }
3775 :
3776 0 : tile_data->error_info.setjmp = 1;
3777 0 : tile_data->xd.error_info = &tile_data->error_info;
3778 : #if CONFIG_DEPENDENT_HORZTILES
3779 : #if CONFIG_TILE_GROUPS
3780 : if (!cm->dependent_horz_tiles || tile->tg_horz_boundary) {
3781 : #else
3782 : if (!cm->dependent_horz_tiles) {
3783 : #endif
3784 : av1_zero_above_context(&pbi->common, tile->mi_col_start, tile->mi_col_end);
3785 : }
3786 : #else
3787 0 : av1_zero_above_context(&pbi->common, tile->mi_col_start, tile->mi_col_end);
3788 : #endif
3789 :
3790 0 : for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end;
3791 0 : mi_row += cm->mib_size) {
3792 0 : av1_zero_left_context(&tile_data->xd);
3793 :
3794 0 : for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
3795 0 : mi_col += cm->mib_size) {
3796 0 : decode_partition(pbi, &tile_data->xd,
3797 : #if CONFIG_SUPERTX
3798 : 0,
3799 : #endif
3800 0 : mi_row, mi_col, &tile_data->bit_reader, cm->sb_size,
3801 0 : b_width_log2_lookup[cm->sb_size]);
3802 : #if CONFIG_NCOBMC && CONFIG_MOTION_VAR
3803 : detoken_and_recon_sb(pbi, &tile_data->xd, mi_row, mi_col,
3804 : &tile_data->bit_reader, cm->sb_size);
3805 : #endif
3806 : }
3807 : }
3808 0 : return !tile_data->xd.corrupted;
3809 : }
3810 :
3811 : // sorts in descending order
3812 0 : static int compare_tile_buffers(const void *a, const void *b) {
3813 0 : const TileBufferDec *const buf1 = (const TileBufferDec *)a;
3814 0 : const TileBufferDec *const buf2 = (const TileBufferDec *)b;
3815 0 : return (int)(buf2->size - buf1->size);
3816 : }
3817 :
3818 0 : static const uint8_t *decode_tiles_mt(AV1Decoder *pbi, const uint8_t *data,
3819 : const uint8_t *data_end) {
3820 0 : AV1_COMMON *const cm = &pbi->common;
3821 0 : const AVxWorkerInterface *const winterface = aom_get_worker_interface();
3822 0 : const int tile_cols = cm->tile_cols;
3823 0 : const int tile_rows = cm->tile_rows;
3824 0 : const int num_workers = AOMMIN(pbi->max_threads & ~1, tile_cols);
3825 0 : TileBufferDec(*const tile_buffers)[MAX_TILE_COLS] = pbi->tile_buffers;
3826 : #if CONFIG_EXT_TILE
3827 : const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows);
3828 : const int single_row = pbi->dec_tile_row >= 0;
3829 : const int tile_rows_start = single_row ? dec_tile_row : 0;
3830 : const int tile_rows_end = single_row ? dec_tile_row + 1 : tile_rows;
3831 : const int dec_tile_col = AOMMIN(pbi->dec_tile_col, tile_cols);
3832 : const int single_col = pbi->dec_tile_col >= 0;
3833 : const int tile_cols_start = single_col ? dec_tile_col : 0;
3834 : const int tile_cols_end = single_col ? tile_cols_start + 1 : tile_cols;
3835 : #else
3836 0 : const int tile_rows_start = 0;
3837 0 : const int tile_rows_end = tile_rows;
3838 0 : const int tile_cols_start = 0;
3839 0 : const int tile_cols_end = tile_cols;
3840 : #endif // CONFIG_EXT_TILE
3841 : int tile_row, tile_col;
3842 : int i;
3843 :
3844 : #if !(CONFIG_ANS || CONFIG_EXT_TILE)
3845 0 : int final_worker = -1;
3846 : #endif // !(CONFIG_ANS || CONFIG_EXT_TILE)
3847 :
3848 0 : assert(tile_rows <= MAX_TILE_ROWS);
3849 0 : assert(tile_cols <= MAX_TILE_COLS);
3850 :
3851 0 : assert(tile_cols * tile_rows > 1);
3852 :
3853 : // TODO(jzern): See if we can remove the restriction of passing in max
3854 : // threads to the decoder.
3855 0 : if (pbi->num_tile_workers == 0) {
3856 0 : const int num_threads = pbi->max_threads & ~1;
3857 0 : CHECK_MEM_ERROR(cm, pbi->tile_workers,
3858 : aom_malloc(num_threads * sizeof(*pbi->tile_workers)));
3859 : // Ensure tile data offsets will be properly aligned. This may fail on
3860 : // platforms without DECLARE_ALIGNED().
3861 : assert((sizeof(*pbi->tile_worker_data) % 16) == 0);
3862 0 : CHECK_MEM_ERROR(
3863 : cm, pbi->tile_worker_data,
3864 : aom_memalign(32, num_threads * sizeof(*pbi->tile_worker_data)));
3865 0 : CHECK_MEM_ERROR(cm, pbi->tile_worker_info,
3866 : aom_malloc(num_threads * sizeof(*pbi->tile_worker_info)));
3867 0 : for (i = 0; i < num_threads; ++i) {
3868 0 : AVxWorker *const worker = &pbi->tile_workers[i];
3869 0 : ++pbi->num_tile_workers;
3870 :
3871 0 : winterface->init(worker);
3872 0 : if (i < num_threads - 1 && !winterface->reset(worker)) {
3873 0 : aom_internal_error(&cm->error, AOM_CODEC_ERROR,
3874 : "Tile decoder thread creation failed");
3875 : }
3876 : }
3877 : }
3878 :
3879 : // Reset tile decoding hook
3880 0 : for (i = 0; i < num_workers; ++i) {
3881 0 : AVxWorker *const worker = &pbi->tile_workers[i];
3882 0 : winterface->sync(worker);
3883 0 : worker->hook = (AVxWorkerHook)tile_worker_hook;
3884 0 : worker->data1 = &pbi->tile_worker_data[i];
3885 0 : worker->data2 = &pbi->tile_worker_info[i];
3886 : }
3887 :
3888 : // Initialize thread frame counts.
3889 0 : if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
3890 0 : for (i = 0; i < num_workers; ++i) {
3891 0 : TileWorkerData *const twd = (TileWorkerData *)pbi->tile_workers[i].data1;
3892 0 : av1_zero(twd->counts);
3893 : }
3894 : }
3895 :
3896 : // Load tile data into tile_buffers
3897 0 : get_tile_buffers(pbi, data, data_end, tile_buffers);
3898 :
3899 0 : for (tile_row = tile_rows_start; tile_row < tile_rows_end; ++tile_row) {
3900 : // Sort the buffers in this tile row based on size in descending order.
3901 0 : qsort(&tile_buffers[tile_row][tile_cols_start],
3902 0 : tile_cols_end - tile_cols_start, sizeof(tile_buffers[0][0]),
3903 : compare_tile_buffers);
3904 :
3905 : // Rearrange the tile buffers in this tile row such that per-tile group
3906 : // the largest, and presumably the most difficult tile will be decoded in
3907 : // the main thread. This should help minimize the number of instances
3908 : // where the main thread is waiting for a worker to complete.
3909 : {
3910 : int group_start;
3911 0 : for (group_start = tile_cols_start; group_start < tile_cols_end;
3912 0 : group_start += num_workers) {
3913 0 : const int group_end = AOMMIN(group_start + num_workers, tile_cols);
3914 0 : const TileBufferDec largest = tile_buffers[tile_row][group_start];
3915 0 : memmove(&tile_buffers[tile_row][group_start],
3916 0 : &tile_buffers[tile_row][group_start + 1],
3917 0 : (group_end - group_start - 1) * sizeof(tile_buffers[0][0]));
3918 0 : tile_buffers[tile_row][group_end - 1] = largest;
3919 : }
3920 : }
3921 :
3922 0 : for (tile_col = tile_cols_start; tile_col < tile_cols_end;) {
3923 : // Launch workers for individual columns
3924 0 : for (i = 0; i < num_workers && tile_col < tile_cols_end;
3925 0 : ++i, ++tile_col) {
3926 0 : TileBufferDec *const buf = &tile_buffers[tile_row][tile_col];
3927 0 : AVxWorker *const worker = &pbi->tile_workers[i];
3928 0 : TileWorkerData *const twd = (TileWorkerData *)worker->data1;
3929 0 : TileInfo *const tile_info = (TileInfo *)worker->data2;
3930 :
3931 0 : twd->pbi = pbi;
3932 0 : twd->xd = pbi->mb;
3933 0 : twd->xd.corrupted = 0;
3934 0 : twd->xd.counts =
3935 0 : cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD
3936 : ? &twd->counts
3937 0 : : NULL;
3938 0 : av1_zero(twd->dqcoeff);
3939 0 : av1_tile_init(tile_info, cm, tile_row, buf->col);
3940 0 : av1_tile_init(&twd->xd.tile, cm, tile_row, buf->col);
3941 0 : setup_bool_decoder(buf->data, data_end, buf->size, &cm->error,
3942 : &twd->bit_reader,
3943 : #if CONFIG_ANS && ANS_MAX_SYMBOLS
3944 : 1 << cm->ans_window_size_log2,
3945 : #endif // CONFIG_ANS && ANS_MAX_SYMBOLS
3946 : pbi->decrypt_cb, pbi->decrypt_state);
3947 0 : av1_init_macroblockd(cm, &twd->xd,
3948 : #if CONFIG_PVQ
3949 : twd->pvq_ref_coeff,
3950 : #endif
3951 : #if CONFIG_CFL
3952 : &twd->cfl,
3953 : #endif
3954 0 : twd->dqcoeff);
3955 : #if CONFIG_PVQ
3956 : daala_dec_init(cm, &twd->xd.daala_dec, &twd->bit_reader);
3957 : twd->xd.daala_dec.state.adapt = &twd->tctx.pvq_context;
3958 : #endif
3959 : #if CONFIG_EC_ADAPT
3960 : // Initialise the tile context from the frame context
3961 0 : twd->tctx = *cm->fc;
3962 0 : twd->xd.tile_ctx = &twd->tctx;
3963 : #endif
3964 : #if CONFIG_PALETTE
3965 0 : twd->xd.plane[0].color_index_map = twd->color_index_map[0];
3966 0 : twd->xd.plane[1].color_index_map = twd->color_index_map[1];
3967 : #endif // CONFIG_PALETTE
3968 :
3969 0 : worker->had_error = 0;
3970 0 : if (i == num_workers - 1 || tile_col == tile_cols_end - 1) {
3971 0 : winterface->execute(worker);
3972 : } else {
3973 0 : winterface->launch(worker);
3974 : }
3975 :
3976 : #if !(CONFIG_ANS || CONFIG_EXT_TILE)
3977 0 : if (tile_row == tile_rows - 1 && buf->col == tile_cols - 1) {
3978 0 : final_worker = i;
3979 : }
3980 : #endif // !(CONFIG_ANS || CONFIG_EXT_TILE)
3981 : }
3982 :
3983 : // Sync all workers
3984 0 : for (; i > 0; --i) {
3985 0 : AVxWorker *const worker = &pbi->tile_workers[i - 1];
3986 : // TODO(jzern): The tile may have specific error data associated with
3987 : // its aom_internal_error_info which could be propagated to the main
3988 : // info in cm. Additionally once the threads have been synced and an
3989 : // error is detected, there's no point in continuing to decode tiles.
3990 0 : pbi->mb.corrupted |= !winterface->sync(worker);
3991 : }
3992 : }
3993 : }
3994 :
3995 : // Accumulate thread frame counts.
3996 0 : if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
3997 0 : for (i = 0; i < num_workers; ++i) {
3998 0 : TileWorkerData *const twd = (TileWorkerData *)pbi->tile_workers[i].data1;
3999 0 : av1_accumulate_frame_counts(&cm->counts, &twd->counts);
4000 : }
4001 : }
4002 :
4003 : #if CONFIG_EXT_TILE
4004 : // Return the end of the last tile buffer
4005 : return tile_buffers[tile_rows - 1][tile_cols - 1].raw_data_end;
4006 : #else
4007 : #if CONFIG_ANS
4008 : return data_end;
4009 : #else
4010 0 : assert(final_worker != -1);
4011 : {
4012 0 : TileWorkerData *const twd =
4013 0 : (TileWorkerData *)pbi->tile_workers[final_worker].data1;
4014 0 : return aom_reader_find_end(&twd->bit_reader);
4015 : }
4016 : #endif // CONFIG_ANS
4017 : #endif // CONFIG_EXT_TILE
4018 : }
4019 :
4020 0 : static void error_handler(void *data) {
4021 0 : AV1_COMMON *const cm = (AV1_COMMON *)data;
4022 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, "Truncated packet");
4023 0 : }
4024 :
4025 0 : static void read_bitdepth_colorspace_sampling(AV1_COMMON *cm,
4026 : struct aom_read_bit_buffer *rb) {
4027 0 : if (cm->profile >= PROFILE_2) {
4028 0 : cm->bit_depth = aom_rb_read_bit(rb) ? AOM_BITS_12 : AOM_BITS_10;
4029 : } else {
4030 0 : cm->bit_depth = AOM_BITS_8;
4031 : }
4032 :
4033 : #if CONFIG_HIGHBITDEPTH
4034 0 : cm->use_highbitdepth = cm->bit_depth > AOM_BITS_8 || !CONFIG_LOWBITDEPTH;
4035 : #endif
4036 :
4037 0 : cm->color_space = aom_rb_read_literal(rb, 3);
4038 0 : if (cm->color_space != AOM_CS_SRGB) {
4039 : // [16,235] (including xvycc) vs [0,255] range
4040 0 : cm->color_range = aom_rb_read_bit(rb);
4041 0 : if (cm->profile == PROFILE_1 || cm->profile == PROFILE_3) {
4042 0 : cm->subsampling_x = aom_rb_read_bit(rb);
4043 0 : cm->subsampling_y = aom_rb_read_bit(rb);
4044 0 : if (cm->subsampling_x == 1 && cm->subsampling_y == 1)
4045 0 : aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
4046 : "4:2:0 color not supported in profile 1 or 3");
4047 0 : if (aom_rb_read_bit(rb))
4048 0 : aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
4049 : "Reserved bit set");
4050 : } else {
4051 0 : cm->subsampling_y = cm->subsampling_x = 1;
4052 : }
4053 : } else {
4054 0 : if (cm->profile == PROFILE_1 || cm->profile == PROFILE_3) {
4055 : // Note if colorspace is SRGB then 4:4:4 chroma sampling is assumed.
4056 : // 4:2:2 or 4:4:0 chroma sampling is not allowed.
4057 0 : cm->subsampling_y = cm->subsampling_x = 0;
4058 0 : if (aom_rb_read_bit(rb))
4059 0 : aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
4060 : "Reserved bit set");
4061 : } else {
4062 0 : aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
4063 : "4:4:4 color not supported in profile 0 or 2");
4064 : }
4065 : }
4066 0 : }
4067 :
4068 : #if CONFIG_REFERENCE_BUFFER
4069 0 : void read_sequence_header(SequenceHeader *seq_params) {
4070 : /* Placeholder for actually reading from the bitstream */
4071 0 : seq_params->frame_id_numbers_present_flag = FRAME_ID_NUMBERS_PRESENT_FLAG;
4072 0 : seq_params->frame_id_length_minus7 = FRAME_ID_LENGTH_MINUS7;
4073 0 : seq_params->delta_frame_id_length_minus2 = DELTA_FRAME_ID_LENGTH_MINUS2;
4074 0 : }
4075 : #endif
4076 :
4077 : #if CONFIG_EXT_INTER
4078 0 : static void read_compound_tools(AV1_COMMON *cm,
4079 : struct aom_read_bit_buffer *rb) {
4080 : (void)cm;
4081 : (void)rb;
4082 : #if CONFIG_INTERINTRA
4083 0 : if (!frame_is_intra_only(cm) && cm->reference_mode != COMPOUND_REFERENCE) {
4084 0 : cm->allow_interintra_compound = aom_rb_read_bit(rb);
4085 : } else {
4086 0 : cm->allow_interintra_compound = 0;
4087 : }
4088 : #endif // CONFIG_INTERINTRA
4089 : #if CONFIG_WEDGE || CONFIG_COMPOUND_SEGMENT
4090 0 : if (!frame_is_intra_only(cm) && cm->reference_mode != SINGLE_REFERENCE) {
4091 0 : cm->allow_masked_compound = aom_rb_read_bit(rb);
4092 : } else {
4093 0 : cm->allow_masked_compound = 0;
4094 : }
4095 : #endif // CONFIG_WEDGE || CONFIG_COMPOUND_SEGMENT
4096 0 : }
4097 : #endif // CONFIG_EXT_INTER
4098 :
4099 0 : static size_t read_uncompressed_header(AV1Decoder *pbi,
4100 : struct aom_read_bit_buffer *rb) {
4101 0 : AV1_COMMON *const cm = &pbi->common;
4102 0 : MACROBLOCKD *const xd = &pbi->mb;
4103 0 : BufferPool *const pool = cm->buffer_pool;
4104 0 : RefCntBuffer *const frame_bufs = pool->frame_bufs;
4105 0 : int i, mask, ref_index = 0;
4106 : size_t sz;
4107 :
4108 : #if CONFIG_REFERENCE_BUFFER
4109 : /* TODO: Move outside frame loop or inside key-frame branch */
4110 0 : read_sequence_header(&pbi->seq_params);
4111 : #endif
4112 :
4113 0 : cm->last_frame_type = cm->frame_type;
4114 0 : cm->last_intra_only = cm->intra_only;
4115 :
4116 : #if CONFIG_EXT_REFS
4117 : // NOTE: By default all coded frames to be used as a reference
4118 0 : cm->is_reference_frame = 1;
4119 : #endif // CONFIG_EXT_REFS
4120 :
4121 0 : if (aom_rb_read_literal(rb, 2) != AOM_FRAME_MARKER)
4122 0 : aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
4123 : "Invalid frame marker");
4124 :
4125 0 : cm->profile = av1_read_profile(rb);
4126 :
4127 0 : const BITSTREAM_PROFILE MAX_SUPPORTED_PROFILE =
4128 : CONFIG_HIGHBITDEPTH ? MAX_PROFILES : PROFILE_2;
4129 :
4130 0 : if (cm->profile >= MAX_SUPPORTED_PROFILE)
4131 0 : aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
4132 : "Unsupported bitstream profile");
4133 :
4134 0 : cm->show_existing_frame = aom_rb_read_bit(rb);
4135 :
4136 0 : if (cm->show_existing_frame) {
4137 : // Show an existing frame directly.
4138 0 : const int existing_frame_idx = aom_rb_read_literal(rb, 3);
4139 0 : const int frame_to_show = cm->ref_frame_map[existing_frame_idx];
4140 : #if CONFIG_REFERENCE_BUFFER
4141 0 : if (pbi->seq_params.frame_id_numbers_present_flag) {
4142 0 : int frame_id_length = pbi->seq_params.frame_id_length_minus7 + 7;
4143 0 : int display_frame_id = aom_rb_read_literal(rb, frame_id_length);
4144 : /* Compare display_frame_id with ref_frame_id and check valid for
4145 : * referencing */
4146 0 : if (display_frame_id != cm->ref_frame_id[existing_frame_idx] ||
4147 0 : cm->valid_for_referencing[existing_frame_idx] == 0)
4148 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
4149 : "Reference buffer frame ID mismatch");
4150 : }
4151 : #endif
4152 0 : lock_buffer_pool(pool);
4153 0 : if (frame_to_show < 0 || frame_bufs[frame_to_show].ref_count < 1) {
4154 0 : unlock_buffer_pool(pool);
4155 0 : aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
4156 : "Buffer %d does not contain a decoded frame",
4157 : frame_to_show);
4158 : }
4159 0 : ref_cnt_fb(frame_bufs, &cm->new_fb_idx, frame_to_show);
4160 0 : unlock_buffer_pool(pool);
4161 :
4162 0 : cm->lf.filter_level = 0;
4163 0 : cm->show_frame = 1;
4164 0 : pbi->refresh_frame_flags = 0;
4165 :
4166 0 : if (cm->frame_parallel_decode) {
4167 0 : for (i = 0; i < REF_FRAMES; ++i)
4168 0 : cm->next_ref_frame_map[i] = cm->ref_frame_map[i];
4169 : }
4170 :
4171 0 : return 0;
4172 : }
4173 :
4174 0 : cm->frame_type = (FRAME_TYPE)aom_rb_read_bit(rb);
4175 0 : cm->show_frame = aom_rb_read_bit(rb);
4176 0 : cm->error_resilient_mode = aom_rb_read_bit(rb);
4177 : #if CONFIG_REFERENCE_BUFFER
4178 0 : if (pbi->seq_params.frame_id_numbers_present_flag) {
4179 0 : int frame_id_length = pbi->seq_params.frame_id_length_minus7 + 7;
4180 0 : int diff_len = pbi->seq_params.delta_frame_id_length_minus2 + 2;
4181 0 : int prev_frame_id = 0;
4182 0 : if (cm->frame_type != KEY_FRAME) {
4183 0 : prev_frame_id = cm->current_frame_id;
4184 : }
4185 0 : cm->current_frame_id = aom_rb_read_literal(rb, frame_id_length);
4186 :
4187 0 : if (cm->frame_type != KEY_FRAME) {
4188 : int diff_frame_id;
4189 0 : if (cm->current_frame_id > prev_frame_id) {
4190 0 : diff_frame_id = cm->current_frame_id - prev_frame_id;
4191 : } else {
4192 0 : diff_frame_id =
4193 0 : (1 << frame_id_length) + cm->current_frame_id - prev_frame_id;
4194 : }
4195 : /* Check current_frame_id for conformance */
4196 0 : if (prev_frame_id == cm->current_frame_id ||
4197 0 : diff_frame_id >= (1 << (frame_id_length - 1))) {
4198 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
4199 : "Invalid value of current_frame_id");
4200 : }
4201 : }
4202 : /* Check if some frames need to be marked as not valid for referencing */
4203 0 : for (i = 0; i < REF_FRAMES; i++) {
4204 0 : if (cm->frame_type == KEY_FRAME) {
4205 0 : cm->valid_for_referencing[i] = 0;
4206 0 : } else if (cm->current_frame_id - (1 << diff_len) > 0) {
4207 0 : if (cm->ref_frame_id[i] > cm->current_frame_id ||
4208 0 : cm->ref_frame_id[i] < cm->current_frame_id - (1 << diff_len))
4209 0 : cm->valid_for_referencing[i] = 0;
4210 : } else {
4211 0 : if (cm->ref_frame_id[i] > cm->current_frame_id &&
4212 0 : cm->ref_frame_id[i] <
4213 0 : (1 << frame_id_length) + cm->current_frame_id - (1 << diff_len))
4214 0 : cm->valid_for_referencing[i] = 0;
4215 : }
4216 : }
4217 : }
4218 : #endif
4219 0 : if (cm->frame_type == KEY_FRAME) {
4220 0 : if (!av1_read_sync_code(rb))
4221 0 : aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
4222 : "Invalid frame sync code");
4223 :
4224 0 : read_bitdepth_colorspace_sampling(cm, rb);
4225 0 : pbi->refresh_frame_flags = (1 << REF_FRAMES) - 1;
4226 :
4227 0 : for (i = 0; i < INTER_REFS_PER_FRAME; ++i) {
4228 0 : cm->frame_refs[i].idx = INVALID_IDX;
4229 0 : cm->frame_refs[i].buf = NULL;
4230 : }
4231 :
4232 0 : setup_frame_size(cm, rb);
4233 0 : if (pbi->need_resync) {
4234 0 : memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
4235 0 : pbi->need_resync = 0;
4236 : }
4237 : #if CONFIG_ANS && ANS_MAX_SYMBOLS
4238 : cm->ans_window_size_log2 = aom_rb_read_literal(rb, 4) + 8;
4239 : #endif // CONFIG_ANS && ANS_MAX_SYMBOLS
4240 : #if CONFIG_PALETTE || CONFIG_INTRABC
4241 0 : cm->allow_screen_content_tools = aom_rb_read_bit(rb);
4242 : #endif // CONFIG_PALETTE || CONFIG_INTRABC
4243 : #if CONFIG_TEMPMV_SIGNALING
4244 0 : cm->use_prev_frame_mvs = 0;
4245 : #endif
4246 : } else {
4247 0 : cm->intra_only = cm->show_frame ? 0 : aom_rb_read_bit(rb);
4248 : #if CONFIG_PALETTE || CONFIG_INTRABC
4249 0 : if (cm->intra_only) cm->allow_screen_content_tools = aom_rb_read_bit(rb);
4250 : #endif // CONFIG_PALETTE || CONFIG_INTRABC
4251 : #if CONFIG_TEMPMV_SIGNALING
4252 0 : if (cm->intra_only || cm->error_resilient_mode) cm->use_prev_frame_mvs = 0;
4253 : #endif
4254 0 : if (cm->error_resilient_mode) {
4255 0 : cm->reset_frame_context = RESET_FRAME_CONTEXT_ALL;
4256 : } else {
4257 0 : if (cm->intra_only) {
4258 0 : cm->reset_frame_context = aom_rb_read_bit(rb)
4259 : ? RESET_FRAME_CONTEXT_ALL
4260 0 : : RESET_FRAME_CONTEXT_CURRENT;
4261 : } else {
4262 0 : cm->reset_frame_context = aom_rb_read_bit(rb)
4263 : ? RESET_FRAME_CONTEXT_CURRENT
4264 0 : : RESET_FRAME_CONTEXT_NONE;
4265 0 : if (cm->reset_frame_context == RESET_FRAME_CONTEXT_CURRENT)
4266 0 : cm->reset_frame_context = aom_rb_read_bit(rb)
4267 : ? RESET_FRAME_CONTEXT_ALL
4268 0 : : RESET_FRAME_CONTEXT_CURRENT;
4269 : }
4270 : }
4271 :
4272 0 : if (cm->intra_only) {
4273 0 : if (!av1_read_sync_code(rb))
4274 0 : aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
4275 : "Invalid frame sync code");
4276 :
4277 0 : read_bitdepth_colorspace_sampling(cm, rb);
4278 :
4279 0 : pbi->refresh_frame_flags = aom_rb_read_literal(rb, REF_FRAMES);
4280 0 : setup_frame_size(cm, rb);
4281 0 : if (pbi->need_resync) {
4282 0 : memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
4283 0 : pbi->need_resync = 0;
4284 : }
4285 : #if CONFIG_ANS && ANS_MAX_SYMBOLS
4286 : cm->ans_window_size_log2 = aom_rb_read_literal(rb, 4) + 8;
4287 : #endif
4288 0 : } else if (pbi->need_resync != 1) { /* Skip if need resync */
4289 0 : pbi->refresh_frame_flags = aom_rb_read_literal(rb, REF_FRAMES);
4290 :
4291 : #if CONFIG_EXT_REFS
4292 0 : if (!pbi->refresh_frame_flags) {
4293 : // NOTE: "pbi->refresh_frame_flags == 0" indicates that the coded frame
4294 : // will not be used as a reference
4295 0 : cm->is_reference_frame = 0;
4296 : }
4297 : #endif // CONFIG_EXT_REFS
4298 :
4299 0 : for (i = 0; i < INTER_REFS_PER_FRAME; ++i) {
4300 0 : const int ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2);
4301 0 : const int idx = cm->ref_frame_map[ref];
4302 0 : RefBuffer *const ref_frame = &cm->frame_refs[i];
4303 0 : ref_frame->idx = idx;
4304 0 : ref_frame->buf = &frame_bufs[idx].buf;
4305 0 : cm->ref_frame_sign_bias[LAST_FRAME + i] = aom_rb_read_bit(rb);
4306 : #if CONFIG_REFERENCE_BUFFER
4307 0 : if (pbi->seq_params.frame_id_numbers_present_flag) {
4308 0 : int frame_id_length = pbi->seq_params.frame_id_length_minus7 + 7;
4309 0 : int diff_len = pbi->seq_params.delta_frame_id_length_minus2 + 2;
4310 0 : int delta_frame_id_minus1 = aom_rb_read_literal(rb, diff_len);
4311 0 : int ref_frame_id =
4312 0 : ((cm->current_frame_id - (delta_frame_id_minus1 + 1) +
4313 0 : (1 << frame_id_length)) %
4314 0 : (1 << frame_id_length));
4315 : /* Compare values derived from delta_frame_id_minus1 and
4316 : * refresh_frame_flags. Also, check valid for referencing */
4317 0 : if (ref_frame_id != cm->ref_frame_id[ref] ||
4318 0 : cm->valid_for_referencing[ref] == 0)
4319 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
4320 : "Reference buffer frame ID mismatch");
4321 : }
4322 : #endif
4323 : }
4324 :
4325 : #if CONFIG_FRAME_SIZE
4326 : if (cm->error_resilient_mode == 0) {
4327 : setup_frame_size_with_refs(cm, rb);
4328 : } else {
4329 : setup_frame_size(cm, rb);
4330 : }
4331 : #else
4332 0 : setup_frame_size_with_refs(cm, rb);
4333 : #endif
4334 :
4335 0 : cm->allow_high_precision_mv = aom_rb_read_bit(rb);
4336 0 : cm->interp_filter = read_frame_interp_filter(rb);
4337 : #if CONFIG_TEMPMV_SIGNALING
4338 0 : if (!cm->error_resilient_mode) {
4339 0 : cm->use_prev_frame_mvs = aom_rb_read_bit(rb);
4340 : }
4341 : #endif
4342 0 : for (i = 0; i < INTER_REFS_PER_FRAME; ++i) {
4343 0 : RefBuffer *const ref_buf = &cm->frame_refs[i];
4344 : #if CONFIG_HIGHBITDEPTH
4345 0 : av1_setup_scale_factors_for_frame(
4346 0 : &ref_buf->sf, ref_buf->buf->y_crop_width,
4347 0 : ref_buf->buf->y_crop_height, cm->width, cm->height,
4348 : cm->use_highbitdepth);
4349 : #else
4350 : av1_setup_scale_factors_for_frame(
4351 : &ref_buf->sf, ref_buf->buf->y_crop_width,
4352 : ref_buf->buf->y_crop_height, cm->width, cm->height);
4353 : #endif
4354 : }
4355 : }
4356 : }
4357 : #if CONFIG_TEMPMV_SIGNALING
4358 0 : cm->cur_frame->intra_only = cm->frame_type == KEY_FRAME || cm->intra_only;
4359 : #endif
4360 :
4361 : #if CONFIG_REFERENCE_BUFFER
4362 0 : if (pbi->seq_params.frame_id_numbers_present_flag) {
4363 : /* If bitmask is set, update reference frame id values and
4364 : mark frames as valid for reference */
4365 0 : int refresh_frame_flags =
4366 0 : cm->frame_type == KEY_FRAME ? 0xFF : pbi->refresh_frame_flags;
4367 0 : for (i = 0; i < REF_FRAMES; i++) {
4368 0 : if ((refresh_frame_flags >> i) & 1) {
4369 0 : cm->ref_frame_id[i] = cm->current_frame_id;
4370 0 : cm->valid_for_referencing[i] = 1;
4371 : }
4372 : }
4373 : }
4374 : #endif
4375 :
4376 0 : get_frame_new_buffer(cm)->bit_depth = cm->bit_depth;
4377 0 : get_frame_new_buffer(cm)->color_space = cm->color_space;
4378 0 : get_frame_new_buffer(cm)->color_range = cm->color_range;
4379 0 : get_frame_new_buffer(cm)->render_width = cm->render_width;
4380 0 : get_frame_new_buffer(cm)->render_height = cm->render_height;
4381 :
4382 0 : if (pbi->need_resync) {
4383 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
4384 : "Keyframe / intra-only frame required to reset decoder"
4385 : " state");
4386 : }
4387 :
4388 0 : if (!cm->error_resilient_mode) {
4389 0 : cm->refresh_frame_context = aom_rb_read_bit(rb)
4390 : ? REFRESH_FRAME_CONTEXT_FORWARD
4391 0 : : REFRESH_FRAME_CONTEXT_BACKWARD;
4392 : } else {
4393 0 : cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_FORWARD;
4394 : }
4395 :
4396 : // This flag will be overridden by the call to av1_setup_past_independence
4397 : // below, forcing the use of context 0 for those frame types.
4398 0 : cm->frame_context_idx = aom_rb_read_literal(rb, FRAME_CONTEXTS_LOG2);
4399 :
4400 : // Generate next_ref_frame_map.
4401 0 : lock_buffer_pool(pool);
4402 0 : for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
4403 0 : if (mask & 1) {
4404 0 : cm->next_ref_frame_map[ref_index] = cm->new_fb_idx;
4405 0 : ++frame_bufs[cm->new_fb_idx].ref_count;
4406 : } else {
4407 0 : cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index];
4408 : }
4409 : // Current thread holds the reference frame.
4410 0 : if (cm->ref_frame_map[ref_index] >= 0)
4411 0 : ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count;
4412 0 : ++ref_index;
4413 : }
4414 :
4415 0 : for (; ref_index < REF_FRAMES; ++ref_index) {
4416 0 : cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index];
4417 :
4418 : // Current thread holds the reference frame.
4419 0 : if (cm->ref_frame_map[ref_index] >= 0)
4420 0 : ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count;
4421 : }
4422 0 : unlock_buffer_pool(pool);
4423 0 : pbi->hold_ref_buf = 1;
4424 :
4425 0 : if (frame_is_intra_only(cm) || cm->error_resilient_mode)
4426 0 : av1_setup_past_independence(cm);
4427 :
4428 : #if CONFIG_EXT_PARTITION
4429 : set_sb_size(cm, aom_rb_read_bit(rb) ? BLOCK_128X128 : BLOCK_64X64);
4430 : #else
4431 0 : set_sb_size(cm, BLOCK_64X64);
4432 : #endif // CONFIG_EXT_PARTITION
4433 :
4434 0 : setup_loopfilter(cm, rb);
4435 : #if CONFIG_CDEF
4436 0 : setup_cdef(cm, rb);
4437 : #endif
4438 : #if CONFIG_LOOP_RESTORATION
4439 : decode_restoration_mode(cm, rb);
4440 : #endif // CONFIG_LOOP_RESTORATION
4441 0 : setup_quantization(cm, rb);
4442 0 : xd->bd = (int)cm->bit_depth;
4443 :
4444 : #if CONFIG_Q_ADAPT_PROBS
4445 : av1_default_coef_probs(cm);
4446 : if (cm->frame_type == KEY_FRAME || cm->error_resilient_mode ||
4447 : cm->reset_frame_context == RESET_FRAME_CONTEXT_ALL) {
4448 : for (i = 0; i < FRAME_CONTEXTS; ++i) cm->frame_contexts[i] = *cm->fc;
4449 : } else if (cm->reset_frame_context == RESET_FRAME_CONTEXT_CURRENT) {
4450 : cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
4451 : }
4452 : #endif // CONFIG_Q_ADAPT_PROBS
4453 :
4454 0 : setup_segmentation(cm, rb);
4455 :
4456 : #if CONFIG_DELTA_Q
4457 : {
4458 0 : struct segmentation *const seg = &cm->seg;
4459 0 : int segment_quantizer_active = 0;
4460 0 : for (i = 0; i < MAX_SEGMENTS; i++) {
4461 0 : if (segfeature_active(seg, i, SEG_LVL_ALT_Q)) {
4462 0 : segment_quantizer_active = 1;
4463 : }
4464 : }
4465 :
4466 0 : cm->delta_q_res = 1;
4467 : #if CONFIG_EXT_DELTA_Q
4468 0 : cm->delta_lf_res = 1;
4469 : #endif
4470 0 : if (segment_quantizer_active == 0 && cm->base_qindex > 0) {
4471 0 : cm->delta_q_present_flag = aom_rb_read_bit(rb);
4472 : } else {
4473 0 : cm->delta_q_present_flag = 0;
4474 : }
4475 0 : if (cm->delta_q_present_flag) {
4476 0 : xd->prev_qindex = cm->base_qindex;
4477 0 : cm->delta_q_res = 1 << aom_rb_read_literal(rb, 2);
4478 : #if CONFIG_EXT_DELTA_Q
4479 0 : if (segment_quantizer_active) {
4480 0 : assert(seg->abs_delta == SEGMENT_DELTADATA);
4481 : }
4482 0 : cm->delta_lf_present_flag = aom_rb_read_bit(rb);
4483 0 : if (cm->delta_lf_present_flag) {
4484 0 : xd->prev_delta_lf_from_base = 0;
4485 0 : cm->delta_lf_res = 1 << aom_rb_read_literal(rb, 2);
4486 : } else {
4487 0 : cm->delta_lf_present_flag = 0;
4488 : }
4489 : #endif // CONFIG_EXT_DELTA_Q
4490 : }
4491 : }
4492 : #endif
4493 :
4494 0 : for (i = 0; i < MAX_SEGMENTS; ++i) {
4495 0 : const int qindex = cm->seg.enabled
4496 0 : ? av1_get_qindex(&cm->seg, i, cm->base_qindex)
4497 0 : : cm->base_qindex;
4498 0 : xd->lossless[i] = qindex == 0 && cm->y_dc_delta_q == 0 &&
4499 0 : cm->uv_dc_delta_q == 0 && cm->uv_ac_delta_q == 0;
4500 0 : xd->qindex[i] = qindex;
4501 : }
4502 :
4503 0 : setup_segmentation_dequant(cm);
4504 0 : cm->tx_mode = read_tx_mode(cm, xd, rb);
4505 0 : cm->reference_mode = read_frame_reference_mode(cm, rb);
4506 : #if CONFIG_EXT_INTER
4507 0 : read_compound_tools(cm, rb);
4508 : #endif // CONFIG_EXT_INTER
4509 :
4510 : #if CONFIG_EXT_TX
4511 0 : cm->reduced_tx_set_used = aom_rb_read_bit(rb);
4512 : #endif // CONFIG_EXT_TX
4513 :
4514 0 : read_tile_info(pbi, rb);
4515 0 : sz = aom_rb_read_literal(rb, 16);
4516 :
4517 0 : if (sz == 0)
4518 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
4519 : "Invalid header size");
4520 0 : return sz;
4521 : }
4522 :
4523 : #if CONFIG_EXT_TX
4524 : #if !CONFIG_EC_ADAPT
4525 : static void read_ext_tx_probs(FRAME_CONTEXT *fc, aom_reader *r) {
4526 : int i, j, k;
4527 : int s;
4528 : for (s = 1; s < EXT_TX_SETS_INTER; ++s) {
4529 : if (aom_read(r, GROUP_DIFF_UPDATE_PROB, ACCT_STR)) {
4530 : for (i = TX_4X4; i < EXT_TX_SIZES; ++i) {
4531 : if (!use_inter_ext_tx_for_txsize[s][i]) continue;
4532 : for (j = 0; j < num_ext_tx_set[ext_tx_set_type_inter[s]] - 1; ++j)
4533 : av1_diff_update_prob(r, &fc->inter_ext_tx_prob[s][i][j], ACCT_STR);
4534 : }
4535 : }
4536 : }
4537 :
4538 : for (s = 1; s < EXT_TX_SETS_INTRA; ++s) {
4539 : if (aom_read(r, GROUP_DIFF_UPDATE_PROB, ACCT_STR)) {
4540 : for (i = TX_4X4; i < EXT_TX_SIZES; ++i) {
4541 : if (!use_intra_ext_tx_for_txsize[s][i]) continue;
4542 : for (j = 0; j < INTRA_MODES; ++j)
4543 : for (k = 0; k < num_ext_tx_set[ext_tx_set_type_intra[s]] - 1; ++k)
4544 : av1_diff_update_prob(r, &fc->intra_ext_tx_prob[s][i][j][k],
4545 : ACCT_STR);
4546 : }
4547 : }
4548 : }
4549 : }
4550 : #endif // !CONFIG_EC_ADAPT
4551 : #else
4552 :
4553 : #endif // CONFIG_EXT_TX
4554 : #if CONFIG_SUPERTX
4555 : static void read_supertx_probs(FRAME_CONTEXT *fc, aom_reader *r) {
4556 : int i, j;
4557 : if (aom_read(r, GROUP_DIFF_UPDATE_PROB, ACCT_STR)) {
4558 : for (i = 0; i < PARTITION_SUPERTX_CONTEXTS; ++i) {
4559 : for (j = TX_8X8; j < TX_SIZES; ++j) {
4560 : av1_diff_update_prob(r, &fc->supertx_prob[i][j], ACCT_STR);
4561 : }
4562 : }
4563 : }
4564 : }
4565 : #endif // CONFIG_SUPERTX
4566 :
4567 : #if CONFIG_GLOBAL_MOTION
4568 0 : static void read_global_motion_params(WarpedMotionParams *params,
4569 : WarpedMotionParams *ref_params,
4570 : aom_prob *probs, aom_reader *r,
4571 : int allow_hp) {
4572 0 : TransformationType type =
4573 0 : aom_read_tree(r, av1_global_motion_types_tree, probs, ACCT_STR);
4574 : int trans_bits;
4575 : int trans_dec_factor;
4576 : int trans_prec_diff;
4577 0 : set_default_warp_params(params);
4578 0 : params->wmtype = type;
4579 0 : switch (type) {
4580 : case HOMOGRAPHY:
4581 : case HORTRAPEZOID:
4582 : case VERTRAPEZOID:
4583 0 : if (type != HORTRAPEZOID)
4584 0 : params->wmmat[6] =
4585 0 : aom_read_signed_primitive_refsubexpfin(
4586 : r, GM_ROW3HOMO_MAX + 1, SUBEXPFIN_K,
4587 0 : (ref_params->wmmat[6] >> GM_ROW3HOMO_PREC_DIFF), ACCT_STR) *
4588 : GM_ROW3HOMO_DECODE_FACTOR;
4589 0 : if (type != VERTRAPEZOID)
4590 0 : params->wmmat[7] =
4591 0 : aom_read_signed_primitive_refsubexpfin(
4592 : r, GM_ROW3HOMO_MAX + 1, SUBEXPFIN_K,
4593 0 : (ref_params->wmmat[7] >> GM_ROW3HOMO_PREC_DIFF), ACCT_STR) *
4594 : GM_ROW3HOMO_DECODE_FACTOR;
4595 : case AFFINE:
4596 : case ROTZOOM:
4597 0 : params->wmmat[2] = aom_read_signed_primitive_refsubexpfin(
4598 : r, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
4599 : (ref_params->wmmat[2] >> GM_ALPHA_PREC_DIFF) -
4600 : (1 << GM_ALPHA_PREC_BITS),
4601 : ACCT_STR) *
4602 0 : GM_ALPHA_DECODE_FACTOR +
4603 : (1 << WARPEDMODEL_PREC_BITS);
4604 0 : if (type != VERTRAPEZOID)
4605 0 : params->wmmat[3] =
4606 0 : aom_read_signed_primitive_refsubexpfin(
4607 : r, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
4608 0 : (ref_params->wmmat[3] >> GM_ALPHA_PREC_DIFF), ACCT_STR) *
4609 : GM_ALPHA_DECODE_FACTOR;
4610 0 : if (type >= AFFINE) {
4611 0 : if (type != HORTRAPEZOID)
4612 0 : params->wmmat[4] =
4613 0 : aom_read_signed_primitive_refsubexpfin(
4614 : r, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
4615 0 : (ref_params->wmmat[4] >> GM_ALPHA_PREC_DIFF), ACCT_STR) *
4616 : GM_ALPHA_DECODE_FACTOR;
4617 0 : params->wmmat[5] = aom_read_signed_primitive_refsubexpfin(
4618 : r, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
4619 : (ref_params->wmmat[5] >> GM_ALPHA_PREC_DIFF) -
4620 : (1 << GM_ALPHA_PREC_BITS),
4621 : ACCT_STR) *
4622 0 : GM_ALPHA_DECODE_FACTOR +
4623 : (1 << WARPEDMODEL_PREC_BITS);
4624 : } else {
4625 0 : params->wmmat[4] = -params->wmmat[3];
4626 0 : params->wmmat[5] = params->wmmat[2];
4627 : }
4628 : // fallthrough intended
4629 : case TRANSLATION:
4630 0 : trans_bits = (type == TRANSLATION) ? GM_ABS_TRANS_ONLY_BITS - !allow_hp
4631 0 : : GM_ABS_TRANS_BITS;
4632 0 : trans_dec_factor = (type == TRANSLATION)
4633 0 : ? GM_TRANS_ONLY_DECODE_FACTOR * (1 << !allow_hp)
4634 0 : : GM_TRANS_DECODE_FACTOR;
4635 0 : trans_prec_diff = (type == TRANSLATION)
4636 0 : ? GM_TRANS_ONLY_PREC_DIFF + !allow_hp
4637 0 : : GM_TRANS_PREC_DIFF;
4638 0 : params->wmmat[0] =
4639 0 : aom_read_signed_primitive_refsubexpfin(
4640 : r, (1 << trans_bits) + 1, SUBEXPFIN_K,
4641 0 : (ref_params->wmmat[0] >> trans_prec_diff), ACCT_STR) *
4642 : trans_dec_factor;
4643 0 : params->wmmat[1] =
4644 0 : aom_read_signed_primitive_refsubexpfin(
4645 : r, (1 << trans_bits) + 1, SUBEXPFIN_K,
4646 0 : (ref_params->wmmat[1] >> trans_prec_diff), ACCT_STR) *
4647 : trans_dec_factor;
4648 0 : case IDENTITY: break;
4649 0 : default: assert(0);
4650 : }
4651 0 : if (params->wmtype <= AFFINE)
4652 0 : if (!get_shear_params(params)) assert(0);
4653 0 : }
4654 :
4655 0 : static void read_global_motion(AV1_COMMON *cm, aom_reader *r) {
4656 : int frame;
4657 0 : for (frame = LAST_FRAME; frame <= ALTREF_FRAME; ++frame) {
4658 0 : read_global_motion_params(
4659 0 : &cm->global_motion[frame], &cm->prev_frame->global_motion[frame],
4660 0 : cm->fc->global_motion_types_prob, r, cm->allow_high_precision_mv);
4661 : /*
4662 : printf("Dec Ref %d [%d/%d]: %d %d %d %d\n",
4663 : frame, cm->current_video_frame, cm->show_frame,
4664 : cm->global_motion[frame].wmmat[0],
4665 : cm->global_motion[frame].wmmat[1],
4666 : cm->global_motion[frame].wmmat[2],
4667 : cm->global_motion[frame].wmmat[3]);
4668 : */
4669 : }
4670 0 : memcpy(cm->cur_frame->global_motion, cm->global_motion,
4671 : TOTAL_REFS_PER_FRAME * sizeof(WarpedMotionParams));
4672 0 : }
4673 : #endif // CONFIG_GLOBAL_MOTION
4674 :
4675 0 : static int read_compressed_header(AV1Decoder *pbi, const uint8_t *data,
4676 : size_t partition_size) {
4677 0 : AV1_COMMON *const cm = &pbi->common;
4678 : #if CONFIG_SUPERTX
4679 : MACROBLOCKD *const xd = &pbi->mb;
4680 : #endif
4681 0 : FRAME_CONTEXT *const fc = cm->fc;
4682 : aom_reader r;
4683 : int k, i;
4684 : #if !CONFIG_EC_ADAPT || \
4685 : (CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION || CONFIG_EXT_INTER)
4686 : int j;
4687 : #endif
4688 :
4689 : #if CONFIG_ANS && ANS_MAX_SYMBOLS
4690 : r.window_size = 1 << cm->ans_window_size_log2;
4691 : #endif
4692 0 : if (aom_reader_init(&r, data, partition_size, pbi->decrypt_cb,
4693 : pbi->decrypt_state))
4694 0 : aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
4695 : "Failed to allocate bool decoder 0");
4696 :
4697 : #if CONFIG_LOOP_RESTORATION
4698 : if (cm->rst_info[0].frame_restoration_type != RESTORE_NONE ||
4699 : cm->rst_info[1].frame_restoration_type != RESTORE_NONE ||
4700 : cm->rst_info[2].frame_restoration_type != RESTORE_NONE) {
4701 : av1_alloc_restoration_buffers(cm);
4702 : decode_restoration(cm, &r);
4703 : }
4704 : #endif
4705 :
4706 : #if !CONFIG_EC_ADAPT
4707 : if (cm->tx_mode == TX_MODE_SELECT) read_tx_size_probs(fc, &r);
4708 : #endif
4709 : #if CONFIG_EXT_TX && CONFIG_RECT_TX && CONFIG_RECT_TX_EXT
4710 : if (cm->tx_mode == TX_MODE_SELECT)
4711 : av1_diff_update_prob(&r, &fc->quarter_tx_size_prob, ACCT_STR);
4712 : #endif // CONFIG_EXT_TX && CONFIG_RECT_TX && CONFIG_RECT_TX_EXT
4713 :
4714 : #if CONFIG_LV_MAP
4715 : av1_read_txb_probs(fc, cm->tx_mode, &r);
4716 : #else // CONFIG_LV_MAP
4717 : #if !CONFIG_PVQ
4718 : #if !CONFIG_EC_ADAPT
4719 : read_coef_probs(fc, cm->tx_mode, &r);
4720 : #endif // !CONFIG_EC_ADAPT
4721 : #endif // !CONFIG_PVQ
4722 : #endif // CONFIG_LV_MAP
4723 :
4724 : #if CONFIG_VAR_TX
4725 0 : for (k = 0; k < TXFM_PARTITION_CONTEXTS; ++k)
4726 0 : av1_diff_update_prob(&r, &fc->txfm_partition_prob[k], ACCT_STR);
4727 : #endif // CONFIG_VAR_TX
4728 0 : for (k = 0; k < SKIP_CONTEXTS; ++k)
4729 0 : av1_diff_update_prob(&r, &fc->skip_probs[k], ACCT_STR);
4730 :
4731 : #if CONFIG_DELTA_Q && !CONFIG_EC_ADAPT
4732 : #if CONFIG_EXT_DELTA_Q
4733 : if (cm->delta_q_present_flag) {
4734 : for (k = 0; k < DELTA_Q_PROBS; ++k)
4735 : av1_diff_update_prob(&r, &fc->delta_q_prob[k], ACCT_STR);
4736 : }
4737 : if (cm->delta_lf_present_flag) {
4738 : for (k = 0; k < DELTA_LF_PROBS; ++k)
4739 : av1_diff_update_prob(&r, &fc->delta_lf_prob[k], ACCT_STR);
4740 : }
4741 : #else
4742 : for (k = 0; k < DELTA_Q_PROBS; ++k)
4743 : av1_diff_update_prob(&r, &fc->delta_q_prob[k], ACCT_STR);
4744 : #endif
4745 : #endif
4746 :
4747 : #if !CONFIG_EC_ADAPT
4748 : if (cm->seg.enabled && cm->seg.update_map) {
4749 : if (cm->seg.temporal_update) {
4750 : for (k = 0; k < PREDICTION_PROBS; k++)
4751 : av1_diff_update_prob(&r, &cm->fc->seg.pred_probs[k], ACCT_STR);
4752 : }
4753 : for (k = 0; k < MAX_SEGMENTS - 1; k++)
4754 : av1_diff_update_prob(&r, &cm->fc->seg.tree_probs[k], ACCT_STR);
4755 : }
4756 :
4757 : for (j = 0; j < INTRA_MODES; j++) {
4758 : for (i = 0; i < INTRA_MODES - 1; ++i)
4759 : av1_diff_update_prob(&r, &fc->uv_mode_prob[j][i], ACCT_STR);
4760 : }
4761 :
4762 : #if CONFIG_EXT_PARTITION_TYPES
4763 : for (j = 0; j < PARTITION_PLOFFSET; ++j)
4764 : for (i = 0; i < PARTITION_TYPES - 1; ++i)
4765 : av1_diff_update_prob(&r, &fc->partition_prob[j][i], ACCT_STR);
4766 : for (; j < PARTITION_CONTEXTS_PRIMARY; ++j)
4767 : for (i = 0; i < EXT_PARTITION_TYPES - 1; ++i)
4768 : av1_diff_update_prob(&r, &fc->partition_prob[j][i], ACCT_STR);
4769 : #else
4770 : for (j = 0; j < PARTITION_CONTEXTS_PRIMARY; ++j)
4771 : for (i = 0; i < PARTITION_TYPES - 1; ++i)
4772 : av1_diff_update_prob(&r, &fc->partition_prob[j][i], ACCT_STR);
4773 : #endif // CONFIG_EXT_PARTITION_TYPES
4774 :
4775 : #if CONFIG_UNPOISON_PARTITION_CTX
4776 : for (; j < PARTITION_CONTEXTS_PRIMARY + PARTITION_BLOCK_SIZES; ++j)
4777 : av1_diff_update_prob(&r, &fc->partition_prob[j][PARTITION_VERT], ACCT_STR);
4778 : for (; j < PARTITION_CONTEXTS_PRIMARY + 2 * PARTITION_BLOCK_SIZES; ++j)
4779 : av1_diff_update_prob(&r, &fc->partition_prob[j][PARTITION_HORZ], ACCT_STR);
4780 : #endif // CONFIG_UNPOISON_PARTITION_CTX
4781 :
4782 : #if CONFIG_EXT_INTRA && CONFIG_INTRA_INTERP
4783 : for (i = 0; i < INTRA_FILTERS + 1; ++i)
4784 : for (j = 0; j < INTRA_FILTERS - 1; ++j)
4785 : av1_diff_update_prob(&r, &fc->intra_filter_probs[i][j], ACCT_STR);
4786 : #endif // CONFIG_EXT_INTRA && CONFIG_INTRA_INTERP
4787 : #endif // !CONFIG_EC_ADAPT
4788 :
4789 0 : if (frame_is_intra_only(cm)) {
4790 0 : av1_copy(cm->kf_y_prob, av1_kf_y_mode_prob);
4791 0 : av1_copy(cm->fc->kf_y_cdf, av1_kf_y_mode_cdf);
4792 : #if !CONFIG_EC_ADAPT
4793 : for (k = 0; k < INTRA_MODES; k++)
4794 : for (j = 0; j < INTRA_MODES; j++)
4795 : for (i = 0; i < INTRA_MODES - 1; ++i)
4796 : av1_diff_update_prob(&r, &cm->kf_y_prob[k][j][i], ACCT_STR);
4797 : #endif
4798 : #if CONFIG_INTRABC
4799 : if (cm->allow_screen_content_tools) {
4800 : av1_diff_update_prob(&r, &fc->intrabc_prob, ACCT_STR);
4801 : }
4802 : #endif
4803 : } else {
4804 0 : read_inter_mode_probs(fc, &r);
4805 :
4806 : #if CONFIG_EXT_INTER
4807 0 : read_inter_compound_mode_probs(fc, &r);
4808 : #if CONFIG_INTERINTRA
4809 0 : if (cm->reference_mode != COMPOUND_REFERENCE &&
4810 0 : cm->allow_interintra_compound) {
4811 0 : for (i = 0; i < BLOCK_SIZE_GROUPS; i++) {
4812 0 : if (is_interintra_allowed_bsize_group(i)) {
4813 0 : av1_diff_update_prob(&r, &fc->interintra_prob[i], ACCT_STR);
4814 : }
4815 : }
4816 0 : for (i = 0; i < BLOCK_SIZE_GROUPS; i++) {
4817 0 : for (j = 0; j < INTERINTRA_MODES - 1; j++)
4818 0 : av1_diff_update_prob(&r, &fc->interintra_mode_prob[i][j], ACCT_STR);
4819 : }
4820 : #if CONFIG_WEDGE
4821 0 : for (i = 0; i < BLOCK_SIZES; i++) {
4822 0 : if (is_interintra_allowed_bsize(i) && is_interintra_wedge_used(i)) {
4823 0 : av1_diff_update_prob(&r, &fc->wedge_interintra_prob[i], ACCT_STR);
4824 : }
4825 : }
4826 : #endif // CONFIG_WEDGE
4827 : }
4828 : #endif // CONFIG_INTERINTRA
4829 : #if CONFIG_COMPOUND_SEGMENT || CONFIG_WEDGE
4830 0 : if (cm->reference_mode != SINGLE_REFERENCE && cm->allow_masked_compound) {
4831 0 : for (i = 0; i < BLOCK_SIZES; i++) {
4832 0 : for (j = 0; j < COMPOUND_TYPES - 1; j++) {
4833 0 : av1_diff_update_prob(&r, &fc->compound_type_prob[i][j], ACCT_STR);
4834 : }
4835 : }
4836 : }
4837 : #endif // CONFIG_COMPOUND_SEGMENT || CONFIG_WEDGE
4838 : #endif // CONFIG_EXT_INTER
4839 :
4840 : #if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
4841 0 : for (i = BLOCK_8X8; i < BLOCK_SIZES; ++i) {
4842 0 : for (j = 0; j < MOTION_MODES - 1; ++j)
4843 0 : av1_diff_update_prob(&r, &fc->motion_mode_prob[i][j], ACCT_STR);
4844 : }
4845 : #endif // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
4846 :
4847 : #if !CONFIG_EC_ADAPT
4848 : if (cm->interp_filter == SWITCHABLE) read_switchable_interp_probs(fc, &r);
4849 : #endif
4850 :
4851 0 : for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
4852 0 : av1_diff_update_prob(&r, &fc->intra_inter_prob[i], ACCT_STR);
4853 :
4854 0 : if (cm->reference_mode != SINGLE_REFERENCE)
4855 0 : setup_compound_reference_mode(cm);
4856 0 : read_frame_reference_mode_probs(cm, &r);
4857 :
4858 : #if !CONFIG_EC_ADAPT
4859 : for (j = 0; j < BLOCK_SIZE_GROUPS; j++) {
4860 : for (i = 0; i < INTRA_MODES - 1; ++i)
4861 : av1_diff_update_prob(&r, &fc->y_mode_prob[j][i], ACCT_STR);
4862 : }
4863 : #endif
4864 :
4865 0 : for (i = 0; i < NMV_CONTEXTS; ++i)
4866 0 : read_mv_probs(&fc->nmvc[i], cm->allow_high_precision_mv, &r);
4867 : #if !CONFIG_EC_ADAPT
4868 : read_ext_tx_probs(fc, &r);
4869 : #endif // EC_ADAPT
4870 : #if CONFIG_SUPERTX
4871 : if (!xd->lossless[0]) read_supertx_probs(fc, &r);
4872 : #endif
4873 : #if CONFIG_GLOBAL_MOTION
4874 0 : read_global_motion(cm, &r);
4875 : #endif
4876 : }
4877 : #if !CONFIG_EC_ADAPT
4878 : av1_coef_head_cdfs(fc);
4879 : /* Make tail distribution from head */
4880 : av1_coef_pareto_cdfs(fc);
4881 : for (i = 0; i < NMV_CONTEXTS; ++i) av1_set_mv_cdfs(&fc->nmvc[i]);
4882 : av1_set_mode_cdfs(cm);
4883 : #endif // !CONFIG_EC_ADAPT
4884 :
4885 0 : return aom_reader_has_error(&r);
4886 : }
4887 :
4888 : #ifdef NDEBUG
4889 : #define debug_check_frame_counts(cm) (void)0
4890 : #else // !NDEBUG
4891 : // Counts should only be incremented when frame_parallel_decoding_mode and
4892 : // error_resilient_mode are disabled.
4893 0 : static void debug_check_frame_counts(const AV1_COMMON *const cm) {
4894 : FRAME_COUNTS zero_counts;
4895 0 : av1_zero(zero_counts);
4896 0 : assert(cm->refresh_frame_context != REFRESH_FRAME_CONTEXT_BACKWARD ||
4897 : cm->error_resilient_mode);
4898 0 : assert(!memcmp(cm->counts.y_mode, zero_counts.y_mode,
4899 : sizeof(cm->counts.y_mode)));
4900 0 : assert(!memcmp(cm->counts.uv_mode, zero_counts.uv_mode,
4901 : sizeof(cm->counts.uv_mode)));
4902 0 : assert(!memcmp(cm->counts.partition, zero_counts.partition,
4903 : sizeof(cm->counts.partition)));
4904 0 : assert(!memcmp(cm->counts.coef, zero_counts.coef, sizeof(cm->counts.coef)));
4905 0 : assert(!memcmp(cm->counts.eob_branch, zero_counts.eob_branch,
4906 : sizeof(cm->counts.eob_branch)));
4907 0 : assert(!memcmp(cm->counts.blockz_count, zero_counts.blockz_count,
4908 : sizeof(cm->counts.blockz_count)));
4909 0 : assert(!memcmp(cm->counts.switchable_interp, zero_counts.switchable_interp,
4910 : sizeof(cm->counts.switchable_interp)));
4911 0 : assert(!memcmp(cm->counts.inter_mode, zero_counts.inter_mode,
4912 : sizeof(cm->counts.inter_mode)));
4913 : #if CONFIG_EXT_INTER
4914 0 : assert(!memcmp(cm->counts.inter_compound_mode,
4915 : zero_counts.inter_compound_mode,
4916 : sizeof(cm->counts.inter_compound_mode)));
4917 : #if CONFIG_INTERINTRA
4918 0 : assert(!memcmp(cm->counts.interintra, zero_counts.interintra,
4919 : sizeof(cm->counts.interintra)));
4920 : #if CONFIG_WEDGE
4921 0 : assert(!memcmp(cm->counts.wedge_interintra, zero_counts.wedge_interintra,
4922 : sizeof(cm->counts.wedge_interintra)));
4923 : #endif // CONFIG_WEDGE
4924 : #endif // CONFIG_INTERINTRA
4925 0 : assert(!memcmp(cm->counts.compound_interinter,
4926 : zero_counts.compound_interinter,
4927 : sizeof(cm->counts.compound_interinter)));
4928 : #endif // CONFIG_EXT_INTER
4929 : #if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
4930 0 : assert(!memcmp(cm->counts.motion_mode, zero_counts.motion_mode,
4931 : sizeof(cm->counts.motion_mode)));
4932 : #endif // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
4933 0 : assert(!memcmp(cm->counts.intra_inter, zero_counts.intra_inter,
4934 : sizeof(cm->counts.intra_inter)));
4935 0 : assert(!memcmp(cm->counts.comp_inter, zero_counts.comp_inter,
4936 : sizeof(cm->counts.comp_inter)));
4937 0 : assert(!memcmp(cm->counts.single_ref, zero_counts.single_ref,
4938 : sizeof(cm->counts.single_ref)));
4939 0 : assert(!memcmp(cm->counts.comp_ref, zero_counts.comp_ref,
4940 : sizeof(cm->counts.comp_ref)));
4941 : #if CONFIG_EXT_REFS
4942 0 : assert(!memcmp(cm->counts.comp_bwdref, zero_counts.comp_bwdref,
4943 : sizeof(cm->counts.comp_bwdref)));
4944 : #endif // CONFIG_EXT_REFS
4945 0 : assert(!memcmp(&cm->counts.tx_size, &zero_counts.tx_size,
4946 : sizeof(cm->counts.tx_size)));
4947 0 : assert(!memcmp(cm->counts.skip, zero_counts.skip, sizeof(cm->counts.skip)));
4948 0 : assert(
4949 : !memcmp(&cm->counts.mv[0], &zero_counts.mv[0], sizeof(cm->counts.mv[0])));
4950 0 : assert(
4951 : !memcmp(&cm->counts.mv[1], &zero_counts.mv[1], sizeof(cm->counts.mv[0])));
4952 0 : assert(!memcmp(cm->counts.inter_ext_tx, zero_counts.inter_ext_tx,
4953 : sizeof(cm->counts.inter_ext_tx)));
4954 0 : assert(!memcmp(cm->counts.intra_ext_tx, zero_counts.intra_ext_tx,
4955 : sizeof(cm->counts.intra_ext_tx)));
4956 0 : }
4957 : #endif // NDEBUG
4958 :
4959 0 : static struct aom_read_bit_buffer *init_read_bit_buffer(
4960 : AV1Decoder *pbi, struct aom_read_bit_buffer *rb, const uint8_t *data,
4961 : const uint8_t *data_end, uint8_t clear_data[MAX_AV1_HEADER_SIZE]) {
4962 0 : rb->bit_offset = 0;
4963 0 : rb->error_handler = error_handler;
4964 0 : rb->error_handler_data = &pbi->common;
4965 0 : if (pbi->decrypt_cb) {
4966 0 : const int n = (int)AOMMIN(MAX_AV1_HEADER_SIZE, data_end - data);
4967 0 : pbi->decrypt_cb(pbi->decrypt_state, data, clear_data, n);
4968 0 : rb->bit_buffer = clear_data;
4969 0 : rb->bit_buffer_end = clear_data + n;
4970 : } else {
4971 0 : rb->bit_buffer = data;
4972 0 : rb->bit_buffer_end = data_end;
4973 : }
4974 0 : return rb;
4975 : }
4976 :
4977 : //------------------------------------------------------------------------------
4978 :
4979 0 : int av1_read_sync_code(struct aom_read_bit_buffer *const rb) {
4980 0 : return aom_rb_read_literal(rb, 8) == AV1_SYNC_CODE_0 &&
4981 0 : aom_rb_read_literal(rb, 8) == AV1_SYNC_CODE_1 &&
4982 0 : aom_rb_read_literal(rb, 8) == AV1_SYNC_CODE_2;
4983 : }
4984 :
4985 0 : void av1_read_frame_size(struct aom_read_bit_buffer *rb, int *width,
4986 : int *height) {
4987 0 : *width = aom_rb_read_literal(rb, 16) + 1;
4988 0 : *height = aom_rb_read_literal(rb, 16) + 1;
4989 0 : }
4990 :
4991 0 : BITSTREAM_PROFILE av1_read_profile(struct aom_read_bit_buffer *rb) {
4992 0 : int profile = aom_rb_read_bit(rb);
4993 0 : profile |= aom_rb_read_bit(rb) << 1;
4994 0 : if (profile > 2) profile += aom_rb_read_bit(rb);
4995 0 : return (BITSTREAM_PROFILE)profile;
4996 : }
4997 :
4998 : #if CONFIG_EC_ADAPT
4999 0 : static void make_update_tile_list_dec(AV1Decoder *pbi, int tile_rows,
5000 : int tile_cols, FRAME_CONTEXT *ec_ctxs[]) {
5001 : int i;
5002 0 : for (i = 0; i < tile_rows * tile_cols; ++i)
5003 0 : ec_ctxs[i] = &pbi->tile_data[i].tctx;
5004 0 : }
5005 : #endif
5006 :
5007 0 : void av1_decode_frame(AV1Decoder *pbi, const uint8_t *data,
5008 : const uint8_t *data_end, const uint8_t **p_data_end) {
5009 0 : AV1_COMMON *const cm = &pbi->common;
5010 0 : MACROBLOCKD *const xd = &pbi->mb;
5011 : struct aom_read_bit_buffer rb;
5012 0 : int context_updated = 0;
5013 : uint8_t clear_data[MAX_AV1_HEADER_SIZE];
5014 : size_t first_partition_size;
5015 : YV12_BUFFER_CONFIG *new_fb;
5016 : #if CONFIG_EXT_REFS || CONFIG_TEMPMV_SIGNALING
5017 0 : RefBuffer *last_fb_ref_buf = &cm->frame_refs[LAST_FRAME - LAST_FRAME];
5018 : #endif // CONFIG_EXT_REFS || CONFIG_TEMPMV_SIGNALING
5019 :
5020 : #if CONFIG_ADAPT_SCAN
5021 : av1_deliver_eob_threshold(cm, xd);
5022 : #endif
5023 : #if CONFIG_BITSTREAM_DEBUG
5024 : bitstream_queue_set_frame_read(cm->current_video_frame * 2 + cm->show_frame);
5025 : #endif
5026 :
5027 0 : first_partition_size = read_uncompressed_header(
5028 : pbi, init_read_bit_buffer(pbi, &rb, data, data_end, clear_data));
5029 :
5030 : #if CONFIG_EXT_TILE
5031 : // If cm->tile_encoding_mode == TILE_NORMAL, the independent decoding of a
5032 : // single tile or a section of a frame is not allowed.
5033 : if (!cm->tile_encoding_mode &&
5034 : (pbi->dec_tile_row >= 0 || pbi->dec_tile_col >= 0)) {
5035 : pbi->dec_tile_row = -1;
5036 : pbi->dec_tile_col = -1;
5037 : }
5038 : #endif // CONFIG_EXT_TILE
5039 :
5040 : #if CONFIG_TILE_GROUPS
5041 0 : pbi->first_partition_size = first_partition_size;
5042 0 : pbi->uncomp_hdr_size = aom_rb_bytes_read(&rb);
5043 : #endif
5044 0 : new_fb = get_frame_new_buffer(cm);
5045 0 : xd->cur_buf = new_fb;
5046 : #if CONFIG_INTRABC
5047 : #if CONFIG_HIGHBITDEPTH
5048 : av1_setup_scale_factors_for_frame(
5049 : &xd->sf_identity, xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height,
5050 : xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height,
5051 : cm->use_highbitdepth);
5052 : #else
5053 : av1_setup_scale_factors_for_frame(
5054 : &xd->sf_identity, xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height,
5055 : xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height);
5056 : #endif // CONFIG_HIGHBITDEPTH
5057 : #endif // CONFIG_INTRABC
5058 : #if CONFIG_GLOBAL_MOTION
5059 : int i;
5060 0 : for (i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
5061 0 : set_default_warp_params(&cm->global_motion[i]);
5062 0 : set_default_warp_params(&cm->cur_frame->global_motion[i]);
5063 : }
5064 0 : xd->global_motion = cm->global_motion;
5065 : #endif // CONFIG_GLOBAL_MOTION
5066 :
5067 0 : if (!first_partition_size) {
5068 : // showing a frame directly
5069 0 : *p_data_end = data + aom_rb_bytes_read(&rb);
5070 0 : return;
5071 : }
5072 :
5073 0 : data += aom_rb_bytes_read(&rb);
5074 0 : if (!read_is_valid(data, first_partition_size, data_end))
5075 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
5076 : "Truncated packet or corrupt header length");
5077 :
5078 0 : cm->setup_mi(cm);
5079 :
5080 : #if CONFIG_EXT_REFS || CONFIG_TEMPMV_SIGNALING
5081 : // NOTE(zoeliu): As cm->prev_frame can take neither a frame of
5082 : // show_exisiting_frame=1, nor can it take a frame not used as
5083 : // a reference, it is probable that by the time it is being
5084 : // referred to, the frame buffer it originally points to may
5085 : // already get expired and have been reassigned to the current
5086 : // newly coded frame. Hence, we need to check whether this is
5087 : // the case, and if yes, we have 2 choices:
5088 : // (1) Simply disable the use of previous frame mvs; or
5089 : // (2) Have cm->prev_frame point to one reference frame buffer,
5090 : // e.g. LAST_FRAME.
5091 0 : if (!dec_is_ref_frame_buf(pbi, cm->prev_frame)) {
5092 : // Reassign the LAST_FRAME buffer to cm->prev_frame.
5093 0 : cm->prev_frame = last_fb_ref_buf->idx != INVALID_IDX
5094 0 : ? &cm->buffer_pool->frame_bufs[last_fb_ref_buf->idx]
5095 0 : : NULL;
5096 : }
5097 : #endif // CONFIG_EXT_REFS || CONFIG_TEMPMV_SIGNALING
5098 :
5099 : #if CONFIG_TEMPMV_SIGNALING
5100 0 : if (cm->use_prev_frame_mvs) {
5101 0 : assert(!cm->error_resilient_mode && cm->prev_frame &&
5102 : cm->width == last_fb_ref_buf->buf->y_width &&
5103 : cm->height == last_fb_ref_buf->buf->y_height &&
5104 : !cm->prev_frame->intra_only);
5105 : }
5106 : #else
5107 : cm->use_prev_frame_mvs = !cm->error_resilient_mode && cm->prev_frame &&
5108 : cm->width == cm->prev_frame->buf.y_crop_width &&
5109 : cm->height == cm->prev_frame->buf.y_crop_height &&
5110 : !cm->last_intra_only && cm->last_show_frame &&
5111 : (cm->last_frame_type != KEY_FRAME);
5112 : #endif // CONFIG_TEMPMV_SIGNALING
5113 :
5114 0 : av1_setup_block_planes(xd, cm->subsampling_x, cm->subsampling_y);
5115 :
5116 0 : *cm->fc = cm->frame_contexts[cm->frame_context_idx];
5117 0 : cm->pre_fc = &cm->frame_contexts[cm->frame_context_idx];
5118 0 : if (!cm->fc->initialized)
5119 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
5120 : "Uninitialized entropy context.");
5121 :
5122 0 : av1_zero(cm->counts);
5123 :
5124 0 : xd->corrupted = 0;
5125 0 : new_fb->corrupted = read_compressed_header(pbi, data, first_partition_size);
5126 0 : if (new_fb->corrupted)
5127 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
5128 : "Decode failed. Frame data header is corrupted.");
5129 :
5130 0 : if (cm->lf.filter_level && !cm->skip_loop_filter) {
5131 0 : av1_loop_filter_frame_init(cm, cm->lf.filter_level);
5132 : }
5133 :
5134 : // If encoded in frame parallel mode, frame context is ready after decoding
5135 : // the frame header.
5136 0 : if (cm->frame_parallel_decode &&
5137 0 : cm->refresh_frame_context != REFRESH_FRAME_CONTEXT_BACKWARD) {
5138 0 : AVxWorker *const worker = pbi->frame_worker_owner;
5139 0 : FrameWorkerData *const frame_worker_data = worker->data1;
5140 0 : if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_FORWARD) {
5141 0 : context_updated = 1;
5142 0 : cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
5143 : }
5144 0 : av1_frameworker_lock_stats(worker);
5145 0 : pbi->cur_buf->row = -1;
5146 0 : pbi->cur_buf->col = -1;
5147 0 : frame_worker_data->frame_context_ready = 1;
5148 : // Signal the main thread that context is ready.
5149 0 : av1_frameworker_signal_stats(worker);
5150 0 : av1_frameworker_unlock_stats(worker);
5151 : }
5152 :
5153 : if (pbi->max_threads > 1 && !CONFIG_CB4X4 &&
5154 : #if CONFIG_EXT_TILE
5155 : pbi->dec_tile_col < 0 && // Decoding all columns
5156 : #endif // CONFIG_EXT_TILE
5157 : cm->tile_cols > 1) {
5158 : // Multi-threaded tile decoder
5159 : *p_data_end = decode_tiles_mt(pbi, data + first_partition_size, data_end);
5160 : if (!xd->corrupted) {
5161 : if (!cm->skip_loop_filter) {
5162 : // If multiple threads are used to decode tiles, then we use those
5163 : // threads to do parallel loopfiltering.
5164 : av1_loop_filter_frame_mt(new_fb, cm, pbi->mb.plane, cm->lf.filter_level,
5165 : 0, 0, pbi->tile_workers, pbi->num_tile_workers,
5166 : &pbi->lf_row_sync);
5167 : }
5168 : } else {
5169 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
5170 : "Decode failed. Frame data is corrupted.");
5171 : }
5172 : } else {
5173 0 : *p_data_end = decode_tiles(pbi, data + first_partition_size, data_end);
5174 : }
5175 :
5176 : #if CONFIG_CDEF
5177 0 : if (!cm->skip_loop_filter) {
5178 0 : av1_cdef_frame(&pbi->cur_buf->buf, cm, &pbi->mb);
5179 : }
5180 : #endif // CONFIG_CDEF
5181 :
5182 : #if CONFIG_LOOP_RESTORATION
5183 : if (cm->rst_info[0].frame_restoration_type != RESTORE_NONE ||
5184 : cm->rst_info[1].frame_restoration_type != RESTORE_NONE ||
5185 : cm->rst_info[2].frame_restoration_type != RESTORE_NONE) {
5186 : av1_loop_restoration_frame(new_fb, cm, cm->rst_info, 7, 0, NULL);
5187 : }
5188 : #endif // CONFIG_LOOP_RESTORATION
5189 :
5190 0 : if (!xd->corrupted) {
5191 0 : if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
5192 : #if CONFIG_EC_ADAPT
5193 0 : FRAME_CONTEXT **tile_ctxs = aom_malloc(cm->tile_rows * cm->tile_cols *
5194 : sizeof(&pbi->tile_data[0].tctx));
5195 0 : aom_cdf_prob **cdf_ptrs =
5196 0 : aom_malloc(cm->tile_rows * cm->tile_cols *
5197 : sizeof(&pbi->tile_data[0].tctx.partition_cdf[0][0]));
5198 0 : make_update_tile_list_dec(pbi, cm->tile_rows, cm->tile_cols, tile_ctxs);
5199 : #endif
5200 0 : av1_adapt_coef_probs(cm);
5201 0 : av1_adapt_intra_frame_probs(cm);
5202 : #if CONFIG_EC_ADAPT
5203 0 : av1_average_tile_coef_cdfs(pbi->common.fc, tile_ctxs, cdf_ptrs,
5204 0 : cm->tile_rows * cm->tile_cols);
5205 0 : av1_average_tile_intra_cdfs(pbi->common.fc, tile_ctxs, cdf_ptrs,
5206 0 : cm->tile_rows * cm->tile_cols);
5207 : #if CONFIG_PVQ
5208 : av1_average_tile_pvq_cdfs(pbi->common.fc, tile_ctxs,
5209 : cm->tile_rows * cm->tile_cols);
5210 : #endif // CONFIG_PVQ
5211 : #endif // CONFIG_EC_ADAPT
5212 : #if CONFIG_ADAPT_SCAN
5213 : av1_adapt_scan_order(cm);
5214 : #endif // CONFIG_ADAPT_SCAN
5215 :
5216 0 : if (!frame_is_intra_only(cm)) {
5217 0 : av1_adapt_inter_frame_probs(cm);
5218 0 : av1_adapt_mv_probs(cm, cm->allow_high_precision_mv);
5219 : #if CONFIG_EC_ADAPT
5220 0 : av1_average_tile_inter_cdfs(&pbi->common, pbi->common.fc, tile_ctxs,
5221 0 : cdf_ptrs, cm->tile_rows * cm->tile_cols);
5222 0 : av1_average_tile_mv_cdfs(pbi->common.fc, tile_ctxs, cdf_ptrs,
5223 0 : cm->tile_rows * cm->tile_cols);
5224 : #endif
5225 : }
5226 : #if CONFIG_EC_ADAPT
5227 0 : aom_free(tile_ctxs);
5228 0 : aom_free(cdf_ptrs);
5229 : #endif
5230 : } else {
5231 0 : debug_check_frame_counts(cm);
5232 : }
5233 : } else {
5234 0 : aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
5235 : "Decode failed. Frame data is corrupted.");
5236 : }
5237 :
5238 : #if CONFIG_INSPECTION
5239 : if (pbi->inspect_cb != NULL) {
5240 : (*pbi->inspect_cb)(pbi, pbi->inspect_ctx);
5241 : }
5242 : #endif
5243 :
5244 : // Non frame parallel update frame context here.
5245 0 : if (!cm->error_resilient_mode && !context_updated)
5246 0 : cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
5247 : }
|