1 /*
2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include <assert.h>
12 #include <stdlib.h> // qsort()
13
14 #include "./vp9_rtcd.h"
15 #include "./vpx_dsp_rtcd.h"
16 #include "./vpx_scale_rtcd.h"
17
18 #include "vpx_dsp/bitreader_buffer.h"
19 #include "vpx_dsp/bitreader.h"
20 #include "vpx_dsp/vpx_dsp_common.h"
21 #include "vpx_mem/vpx_mem.h"
22 #include "vpx_ports/mem.h"
23 #include "vpx_ports/mem_ops.h"
24 #include "vpx_scale/vpx_scale.h"
25 #include "vpx_util/vpx_thread.h"
26
27 #include "vp9/common/vp9_alloccommon.h"
28 #include "vp9/common/vp9_common.h"
29 #include "vp9/common/vp9_entropy.h"
30 #include "vp9/common/vp9_entropymode.h"
31 #include "vp9/common/vp9_idct.h"
32 #include "vp9/common/vp9_thread_common.h"
33 #include "vp9/common/vp9_pred_common.h"
34 #include "vp9/common/vp9_quant_common.h"
35 #include "vp9/common/vp9_reconintra.h"
36 #include "vp9/common/vp9_reconinter.h"
37 #include "vp9/common/vp9_seg_common.h"
38 #include "vp9/common/vp9_tile_common.h"
39
40 #include "vp9/decoder/vp9_decodeframe.h"
41 #include "vp9/decoder/vp9_detokenize.h"
42 #include "vp9/decoder/vp9_decodemv.h"
43 #include "vp9/decoder/vp9_decoder.h"
44 #include "vp9/decoder/vp9_dsubexp.h"
45
46 #define MAX_VP9_HEADER_SIZE 80
47
is_compound_reference_allowed(const VP9_COMMON * cm)48 static int is_compound_reference_allowed(const VP9_COMMON *cm) {
49 int i;
50 for (i = 1; i < REFS_PER_FRAME; ++i)
51 if (cm->ref_frame_sign_bias[i + 1] != cm->ref_frame_sign_bias[1]) return 1;
52
53 return 0;
54 }
55
setup_compound_reference_mode(VP9_COMMON * cm)56 static void setup_compound_reference_mode(VP9_COMMON *cm) {
57 if (cm->ref_frame_sign_bias[LAST_FRAME] ==
58 cm->ref_frame_sign_bias[GOLDEN_FRAME]) {
59 cm->comp_fixed_ref = ALTREF_FRAME;
60 cm->comp_var_ref[0] = LAST_FRAME;
61 cm->comp_var_ref[1] = GOLDEN_FRAME;
62 } else if (cm->ref_frame_sign_bias[LAST_FRAME] ==
63 cm->ref_frame_sign_bias[ALTREF_FRAME]) {
64 cm->comp_fixed_ref = GOLDEN_FRAME;
65 cm->comp_var_ref[0] = LAST_FRAME;
66 cm->comp_var_ref[1] = ALTREF_FRAME;
67 } else {
68 cm->comp_fixed_ref = LAST_FRAME;
69 cm->comp_var_ref[0] = GOLDEN_FRAME;
70 cm->comp_var_ref[1] = ALTREF_FRAME;
71 }
72 }
73
read_is_valid(const uint8_t * start,size_t len,const uint8_t * end)74 static int read_is_valid(const uint8_t *start, size_t len, const uint8_t *end) {
75 return len != 0 && len <= (size_t)(end - start);
76 }
77
decode_unsigned_max(struct vpx_read_bit_buffer * rb,int max)78 static int decode_unsigned_max(struct vpx_read_bit_buffer *rb, int max) {
79 const int data = vpx_rb_read_literal(rb, get_unsigned_bits(max));
80 return data > max ? max : data;
81 }
82
read_tx_mode(vpx_reader * r)83 static TX_MODE read_tx_mode(vpx_reader *r) {
84 TX_MODE tx_mode = vpx_read_literal(r, 2);
85 if (tx_mode == ALLOW_32X32) tx_mode += vpx_read_bit(r);
86 return tx_mode;
87 }
88
read_tx_mode_probs(struct tx_probs * tx_probs,vpx_reader * r)89 static void read_tx_mode_probs(struct tx_probs *tx_probs, vpx_reader *r) {
90 int i, j;
91
92 for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
93 for (j = 0; j < TX_SIZES - 3; ++j)
94 vp9_diff_update_prob(r, &tx_probs->p8x8[i][j]);
95
96 for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
97 for (j = 0; j < TX_SIZES - 2; ++j)
98 vp9_diff_update_prob(r, &tx_probs->p16x16[i][j]);
99
100 for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
101 for (j = 0; j < TX_SIZES - 1; ++j)
102 vp9_diff_update_prob(r, &tx_probs->p32x32[i][j]);
103 }
104
read_switchable_interp_probs(FRAME_CONTEXT * fc,vpx_reader * r)105 static void read_switchable_interp_probs(FRAME_CONTEXT *fc, vpx_reader *r) {
106 int i, j;
107 for (j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j)
108 for (i = 0; i < SWITCHABLE_FILTERS - 1; ++i)
109 vp9_diff_update_prob(r, &fc->switchable_interp_prob[j][i]);
110 }
111
read_inter_mode_probs(FRAME_CONTEXT * fc,vpx_reader * r)112 static void read_inter_mode_probs(FRAME_CONTEXT *fc, vpx_reader *r) {
113 int i, j;
114 for (i = 0; i < INTER_MODE_CONTEXTS; ++i)
115 for (j = 0; j < INTER_MODES - 1; ++j)
116 vp9_diff_update_prob(r, &fc->inter_mode_probs[i][j]);
117 }
118
read_frame_reference_mode(const VP9_COMMON * cm,vpx_reader * r)119 static REFERENCE_MODE read_frame_reference_mode(const VP9_COMMON *cm,
120 vpx_reader *r) {
121 if (is_compound_reference_allowed(cm)) {
122 return vpx_read_bit(r)
123 ? (vpx_read_bit(r) ? REFERENCE_MODE_SELECT : COMPOUND_REFERENCE)
124 : SINGLE_REFERENCE;
125 } else {
126 return SINGLE_REFERENCE;
127 }
128 }
129
read_frame_reference_mode_probs(VP9_COMMON * cm,vpx_reader * r)130 static void read_frame_reference_mode_probs(VP9_COMMON *cm, vpx_reader *r) {
131 FRAME_CONTEXT *const fc = cm->fc;
132 int i;
133
134 if (cm->reference_mode == REFERENCE_MODE_SELECT)
135 for (i = 0; i < COMP_INTER_CONTEXTS; ++i)
136 vp9_diff_update_prob(r, &fc->comp_inter_prob[i]);
137
138 if (cm->reference_mode != COMPOUND_REFERENCE)
139 for (i = 0; i < REF_CONTEXTS; ++i) {
140 vp9_diff_update_prob(r, &fc->single_ref_prob[i][0]);
141 vp9_diff_update_prob(r, &fc->single_ref_prob[i][1]);
142 }
143
144 if (cm->reference_mode != SINGLE_REFERENCE)
145 for (i = 0; i < REF_CONTEXTS; ++i)
146 vp9_diff_update_prob(r, &fc->comp_ref_prob[i]);
147 }
148
update_mv_probs(vpx_prob * p,int n,vpx_reader * r)149 static void update_mv_probs(vpx_prob *p, int n, vpx_reader *r) {
150 int i;
151 for (i = 0; i < n; ++i)
152 if (vpx_read(r, MV_UPDATE_PROB)) p[i] = (vpx_read_literal(r, 7) << 1) | 1;
153 }
154
read_mv_probs(nmv_context * ctx,int allow_hp,vpx_reader * r)155 static void read_mv_probs(nmv_context *ctx, int allow_hp, vpx_reader *r) {
156 int i, j;
157
158 update_mv_probs(ctx->joints, MV_JOINTS - 1, r);
159
160 for (i = 0; i < 2; ++i) {
161 nmv_component *const comp_ctx = &ctx->comps[i];
162 update_mv_probs(&comp_ctx->sign, 1, r);
163 update_mv_probs(comp_ctx->classes, MV_CLASSES - 1, r);
164 update_mv_probs(comp_ctx->class0, CLASS0_SIZE - 1, r);
165 update_mv_probs(comp_ctx->bits, MV_OFFSET_BITS, r);
166 }
167
168 for (i = 0; i < 2; ++i) {
169 nmv_component *const comp_ctx = &ctx->comps[i];
170 for (j = 0; j < CLASS0_SIZE; ++j)
171 update_mv_probs(comp_ctx->class0_fp[j], MV_FP_SIZE - 1, r);
172 update_mv_probs(comp_ctx->fp, 3, r);
173 }
174
175 if (allow_hp) {
176 for (i = 0; i < 2; ++i) {
177 nmv_component *const comp_ctx = &ctx->comps[i];
178 update_mv_probs(&comp_ctx->class0_hp, 1, r);
179 update_mv_probs(&comp_ctx->hp, 1, r);
180 }
181 }
182 }
183
inverse_transform_block_inter(MACROBLOCKD * xd,int plane,const TX_SIZE tx_size,uint8_t * dst,int stride,int eob)184 static void inverse_transform_block_inter(MACROBLOCKD *xd, int plane,
185 const TX_SIZE tx_size, uint8_t *dst,
186 int stride, int eob) {
187 struct macroblockd_plane *const pd = &xd->plane[plane];
188 tran_low_t *const dqcoeff = pd->dqcoeff;
189 assert(eob > 0);
190 #if CONFIG_VP9_HIGHBITDEPTH
191 if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
192 uint16_t *const dst16 = CONVERT_TO_SHORTPTR(dst);
193 if (xd->lossless) {
194 vp9_highbd_iwht4x4_add(dqcoeff, dst16, stride, eob, xd->bd);
195 } else {
196 switch (tx_size) {
197 case TX_4X4:
198 vp9_highbd_idct4x4_add(dqcoeff, dst16, stride, eob, xd->bd);
199 break;
200 case TX_8X8:
201 vp9_highbd_idct8x8_add(dqcoeff, dst16, stride, eob, xd->bd);
202 break;
203 case TX_16X16:
204 vp9_highbd_idct16x16_add(dqcoeff, dst16, stride, eob, xd->bd);
205 break;
206 case TX_32X32:
207 vp9_highbd_idct32x32_add(dqcoeff, dst16, stride, eob, xd->bd);
208 break;
209 default: assert(0 && "Invalid transform size");
210 }
211 }
212 } else {
213 if (xd->lossless) {
214 vp9_iwht4x4_add(dqcoeff, dst, stride, eob);
215 } else {
216 switch (tx_size) {
217 case TX_4X4: vp9_idct4x4_add(dqcoeff, dst, stride, eob); break;
218 case TX_8X8: vp9_idct8x8_add(dqcoeff, dst, stride, eob); break;
219 case TX_16X16: vp9_idct16x16_add(dqcoeff, dst, stride, eob); break;
220 case TX_32X32: vp9_idct32x32_add(dqcoeff, dst, stride, eob); break;
221 default: assert(0 && "Invalid transform size"); return;
222 }
223 }
224 }
225 #else
226 if (xd->lossless) {
227 vp9_iwht4x4_add(dqcoeff, dst, stride, eob);
228 } else {
229 switch (tx_size) {
230 case TX_4X4: vp9_idct4x4_add(dqcoeff, dst, stride, eob); break;
231 case TX_8X8: vp9_idct8x8_add(dqcoeff, dst, stride, eob); break;
232 case TX_16X16: vp9_idct16x16_add(dqcoeff, dst, stride, eob); break;
233 case TX_32X32: vp9_idct32x32_add(dqcoeff, dst, stride, eob); break;
234 default: assert(0 && "Invalid transform size"); return;
235 }
236 }
237 #endif // CONFIG_VP9_HIGHBITDEPTH
238
239 if (eob == 1) {
240 dqcoeff[0] = 0;
241 } else {
242 if (tx_size <= TX_16X16 && eob <= 10)
243 memset(dqcoeff, 0, 4 * (4 << tx_size) * sizeof(dqcoeff[0]));
244 else if (tx_size == TX_32X32 && eob <= 34)
245 memset(dqcoeff, 0, 256 * sizeof(dqcoeff[0]));
246 else
247 memset(dqcoeff, 0, (16 << (tx_size << 1)) * sizeof(dqcoeff[0]));
248 }
249 }
250
inverse_transform_block_intra(MACROBLOCKD * xd,int plane,const TX_TYPE tx_type,const TX_SIZE tx_size,uint8_t * dst,int stride,int eob)251 static void inverse_transform_block_intra(MACROBLOCKD *xd, int plane,
252 const TX_TYPE tx_type,
253 const TX_SIZE tx_size, uint8_t *dst,
254 int stride, int eob) {
255 struct macroblockd_plane *const pd = &xd->plane[plane];
256 tran_low_t *const dqcoeff = pd->dqcoeff;
257 assert(eob > 0);
258 #if CONFIG_VP9_HIGHBITDEPTH
259 if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
260 uint16_t *const dst16 = CONVERT_TO_SHORTPTR(dst);
261 if (xd->lossless) {
262 vp9_highbd_iwht4x4_add(dqcoeff, dst16, stride, eob, xd->bd);
263 } else {
264 switch (tx_size) {
265 case TX_4X4:
266 vp9_highbd_iht4x4_add(tx_type, dqcoeff, dst16, stride, eob, xd->bd);
267 break;
268 case TX_8X8:
269 vp9_highbd_iht8x8_add(tx_type, dqcoeff, dst16, stride, eob, xd->bd);
270 break;
271 case TX_16X16:
272 vp9_highbd_iht16x16_add(tx_type, dqcoeff, dst16, stride, eob, xd->bd);
273 break;
274 case TX_32X32:
275 vp9_highbd_idct32x32_add(dqcoeff, dst16, stride, eob, xd->bd);
276 break;
277 default: assert(0 && "Invalid transform size");
278 }
279 }
280 } else {
281 if (xd->lossless) {
282 vp9_iwht4x4_add(dqcoeff, dst, stride, eob);
283 } else {
284 switch (tx_size) {
285 case TX_4X4: vp9_iht4x4_add(tx_type, dqcoeff, dst, stride, eob); break;
286 case TX_8X8: vp9_iht8x8_add(tx_type, dqcoeff, dst, stride, eob); break;
287 case TX_16X16:
288 vp9_iht16x16_add(tx_type, dqcoeff, dst, stride, eob);
289 break;
290 case TX_32X32: vp9_idct32x32_add(dqcoeff, dst, stride, eob); break;
291 default: assert(0 && "Invalid transform size"); return;
292 }
293 }
294 }
295 #else
296 if (xd->lossless) {
297 vp9_iwht4x4_add(dqcoeff, dst, stride, eob);
298 } else {
299 switch (tx_size) {
300 case TX_4X4: vp9_iht4x4_add(tx_type, dqcoeff, dst, stride, eob); break;
301 case TX_8X8: vp9_iht8x8_add(tx_type, dqcoeff, dst, stride, eob); break;
302 case TX_16X16:
303 vp9_iht16x16_add(tx_type, dqcoeff, dst, stride, eob);
304 break;
305 case TX_32X32: vp9_idct32x32_add(dqcoeff, dst, stride, eob); break;
306 default: assert(0 && "Invalid transform size"); return;
307 }
308 }
309 #endif // CONFIG_VP9_HIGHBITDEPTH
310
311 if (eob == 1) {
312 dqcoeff[0] = 0;
313 } else {
314 if (tx_type == DCT_DCT && tx_size <= TX_16X16 && eob <= 10)
315 memset(dqcoeff, 0, 4 * (4 << tx_size) * sizeof(dqcoeff[0]));
316 else if (tx_size == TX_32X32 && eob <= 34)
317 memset(dqcoeff, 0, 256 * sizeof(dqcoeff[0]));
318 else
319 memset(dqcoeff, 0, (16 << (tx_size << 1)) * sizeof(dqcoeff[0]));
320 }
321 }
322
predict_and_reconstruct_intra_block(TileWorkerData * twd,MODE_INFO * const mi,int plane,int row,int col,TX_SIZE tx_size)323 static void predict_and_reconstruct_intra_block(TileWorkerData *twd,
324 MODE_INFO *const mi, int plane,
325 int row, int col,
326 TX_SIZE tx_size) {
327 MACROBLOCKD *const xd = &twd->xd;
328 struct macroblockd_plane *const pd = &xd->plane[plane];
329 PREDICTION_MODE mode = (plane == 0) ? mi->mode : mi->uv_mode;
330 uint8_t *dst;
331 dst = &pd->dst.buf[4 * row * pd->dst.stride + 4 * col];
332
333 if (mi->sb_type < BLOCK_8X8)
334 if (plane == 0) mode = xd->mi[0]->bmi[(row << 1) + col].as_mode;
335
336 vp9_predict_intra_block(xd, pd->n4_wl, tx_size, mode, dst, pd->dst.stride,
337 dst, pd->dst.stride, col, row, plane);
338
339 if (!mi->skip) {
340 const TX_TYPE tx_type =
341 (plane || xd->lossless) ? DCT_DCT : intra_mode_to_tx_type_lookup[mode];
342 const scan_order *sc = (plane || xd->lossless)
343 ? &vp9_default_scan_orders[tx_size]
344 : &vp9_scan_orders[tx_size][tx_type];
345 const int eob = vp9_decode_block_tokens(twd, plane, sc, col, row, tx_size,
346 mi->segment_id);
347 if (eob > 0) {
348 inverse_transform_block_intra(xd, plane, tx_type, tx_size, dst,
349 pd->dst.stride, eob);
350 }
351 }
352 }
353
reconstruct_inter_block(TileWorkerData * twd,MODE_INFO * const mi,int plane,int row,int col,TX_SIZE tx_size)354 static int reconstruct_inter_block(TileWorkerData *twd, MODE_INFO *const mi,
355 int plane, int row, int col,
356 TX_SIZE tx_size) {
357 MACROBLOCKD *const xd = &twd->xd;
358 struct macroblockd_plane *const pd = &xd->plane[plane];
359 const scan_order *sc = &vp9_default_scan_orders[tx_size];
360 const int eob = vp9_decode_block_tokens(twd, plane, sc, col, row, tx_size,
361 mi->segment_id);
362
363 if (eob > 0) {
364 inverse_transform_block_inter(
365 xd, plane, tx_size, &pd->dst.buf[4 * row * pd->dst.stride + 4 * col],
366 pd->dst.stride, eob);
367 }
368 return eob;
369 }
370
build_mc_border(const uint8_t * src,int src_stride,uint8_t * dst,int dst_stride,int x,int y,int b_w,int b_h,int w,int h)371 static void build_mc_border(const uint8_t *src, int src_stride, uint8_t *dst,
372 int dst_stride, int x, int y, int b_w, int b_h,
373 int w, int h) {
374 // Get a pointer to the start of the real data for this row.
375 const uint8_t *ref_row = src - x - y * src_stride;
376
377 if (y >= h)
378 ref_row += (h - 1) * src_stride;
379 else if (y > 0)
380 ref_row += y * src_stride;
381
382 do {
383 int right = 0, copy;
384 int left = x < 0 ? -x : 0;
385
386 if (left > b_w) left = b_w;
387
388 if (x + b_w > w) right = x + b_w - w;
389
390 if (right > b_w) right = b_w;
391
392 copy = b_w - left - right;
393
394 if (left) memset(dst, ref_row[0], left);
395
396 if (copy) memcpy(dst + left, ref_row + x + left, copy);
397
398 if (right) memset(dst + left + copy, ref_row[w - 1], right);
399
400 dst += dst_stride;
401 ++y;
402
403 if (y > 0 && y < h) ref_row += src_stride;
404 } while (--b_h);
405 }
406
407 #if CONFIG_VP9_HIGHBITDEPTH
high_build_mc_border(const uint8_t * src8,int src_stride,uint16_t * dst,int dst_stride,int x,int y,int b_w,int b_h,int w,int h)408 static void high_build_mc_border(const uint8_t *src8, int src_stride,
409 uint16_t *dst, int dst_stride, int x, int y,
410 int b_w, int b_h, int w, int h) {
411 // Get a pointer to the start of the real data for this row.
412 const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
413 const uint16_t *ref_row = src - x - y * src_stride;
414
415 if (y >= h)
416 ref_row += (h - 1) * src_stride;
417 else if (y > 0)
418 ref_row += y * src_stride;
419
420 do {
421 int right = 0, copy;
422 int left = x < 0 ? -x : 0;
423
424 if (left > b_w) left = b_w;
425
426 if (x + b_w > w) right = x + b_w - w;
427
428 if (right > b_w) right = b_w;
429
430 copy = b_w - left - right;
431
432 if (left) vpx_memset16(dst, ref_row[0], left);
433
434 if (copy) memcpy(dst + left, ref_row + x + left, copy * sizeof(uint16_t));
435
436 if (right) vpx_memset16(dst + left + copy, ref_row[w - 1], right);
437
438 dst += dst_stride;
439 ++y;
440
441 if (y > 0 && y < h) ref_row += src_stride;
442 } while (--b_h);
443 }
444 #endif // CONFIG_VP9_HIGHBITDEPTH
445
446 #if CONFIG_VP9_HIGHBITDEPTH
extend_and_predict(const uint8_t * buf_ptr1,int pre_buf_stride,int x0,int y0,int b_w,int b_h,int frame_width,int frame_height,int border_offset,uint8_t * const dst,int dst_buf_stride,int subpel_x,int subpel_y,const InterpKernel * kernel,const struct scale_factors * sf,MACROBLOCKD * xd,int w,int h,int ref,int xs,int ys)447 static void extend_and_predict(const uint8_t *buf_ptr1, int pre_buf_stride,
448 int x0, int y0, int b_w, int b_h,
449 int frame_width, int frame_height,
450 int border_offset, uint8_t *const dst,
451 int dst_buf_stride, int subpel_x, int subpel_y,
452 const InterpKernel *kernel,
453 const struct scale_factors *sf, MACROBLOCKD *xd,
454 int w, int h, int ref, int xs, int ys) {
455 DECLARE_ALIGNED(16, uint16_t, mc_buf_high[80 * 2 * 80 * 2]);
456
457 if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
458 high_build_mc_border(buf_ptr1, pre_buf_stride, mc_buf_high, b_w, x0, y0,
459 b_w, b_h, frame_width, frame_height);
460 highbd_inter_predictor(mc_buf_high + border_offset, b_w,
461 CONVERT_TO_SHORTPTR(dst), dst_buf_stride, subpel_x,
462 subpel_y, sf, w, h, ref, kernel, xs, ys, xd->bd);
463 } else {
464 build_mc_border(buf_ptr1, pre_buf_stride, (uint8_t *)mc_buf_high, b_w, x0,
465 y0, b_w, b_h, frame_width, frame_height);
466 inter_predictor(((uint8_t *)mc_buf_high) + border_offset, b_w, dst,
467 dst_buf_stride, subpel_x, subpel_y, sf, w, h, ref, kernel,
468 xs, ys);
469 }
470 }
471 #else
extend_and_predict(const uint8_t * buf_ptr1,int pre_buf_stride,int x0,int y0,int b_w,int b_h,int frame_width,int frame_height,int border_offset,uint8_t * const dst,int dst_buf_stride,int subpel_x,int subpel_y,const InterpKernel * kernel,const struct scale_factors * sf,int w,int h,int ref,int xs,int ys)472 static void extend_and_predict(const uint8_t *buf_ptr1, int pre_buf_stride,
473 int x0, int y0, int b_w, int b_h,
474 int frame_width, int frame_height,
475 int border_offset, uint8_t *const dst,
476 int dst_buf_stride, int subpel_x, int subpel_y,
477 const InterpKernel *kernel,
478 const struct scale_factors *sf, int w, int h,
479 int ref, int xs, int ys) {
480 DECLARE_ALIGNED(16, uint8_t, mc_buf[80 * 2 * 80 * 2]);
481 const uint8_t *buf_ptr;
482
483 build_mc_border(buf_ptr1, pre_buf_stride, mc_buf, b_w, x0, y0, b_w, b_h,
484 frame_width, frame_height);
485 buf_ptr = mc_buf + border_offset;
486
487 inter_predictor(buf_ptr, b_w, dst, dst_buf_stride, subpel_x, subpel_y, sf, w,
488 h, ref, kernel, xs, ys);
489 }
490 #endif // CONFIG_VP9_HIGHBITDEPTH
491
dec_build_inter_predictors(VPxWorker * const worker,MACROBLOCKD * xd,int plane,int bw,int bh,int x,int y,int w,int h,int mi_x,int mi_y,const InterpKernel * kernel,const struct scale_factors * sf,struct buf_2d * pre_buf,struct buf_2d * dst_buf,const MV * mv,RefCntBuffer * ref_frame_buf,int is_scaled,int ref)492 static void dec_build_inter_predictors(
493 VPxWorker *const worker, MACROBLOCKD *xd, int plane, int bw, int bh, int x,
494 int y, int w, int h, int mi_x, int mi_y, const InterpKernel *kernel,
495 const struct scale_factors *sf, struct buf_2d *pre_buf,
496 struct buf_2d *dst_buf, const MV *mv, RefCntBuffer *ref_frame_buf,
497 int is_scaled, int ref) {
498 struct macroblockd_plane *const pd = &xd->plane[plane];
499 uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x;
500 MV32 scaled_mv;
501 int xs, ys, x0, y0, x0_16, y0_16, frame_width, frame_height, buf_stride,
502 subpel_x, subpel_y;
503 uint8_t *ref_frame, *buf_ptr;
504
505 // Get reference frame pointer, width and height.
506 if (plane == 0) {
507 frame_width = ref_frame_buf->buf.y_crop_width;
508 frame_height = ref_frame_buf->buf.y_crop_height;
509 ref_frame = ref_frame_buf->buf.y_buffer;
510 } else {
511 frame_width = ref_frame_buf->buf.uv_crop_width;
512 frame_height = ref_frame_buf->buf.uv_crop_height;
513 ref_frame =
514 plane == 1 ? ref_frame_buf->buf.u_buffer : ref_frame_buf->buf.v_buffer;
515 }
516
517 if (is_scaled) {
518 const MV mv_q4 = clamp_mv_to_umv_border_sb(
519 xd, mv, bw, bh, pd->subsampling_x, pd->subsampling_y);
520 // Co-ordinate of containing block to pixel precision.
521 int x_start = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x));
522 int y_start = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y));
523 #if 0 // CONFIG_BETTER_HW_COMPATIBILITY
524 assert(xd->mi[0]->sb_type != BLOCK_4X8 &&
525 xd->mi[0]->sb_type != BLOCK_8X4);
526 assert(mv_q4.row == mv->row * (1 << (1 - pd->subsampling_y)) &&
527 mv_q4.col == mv->col * (1 << (1 - pd->subsampling_x)));
528 #endif
529 // Co-ordinate of the block to 1/16th pixel precision.
530 x0_16 = (x_start + x) << SUBPEL_BITS;
531 y0_16 = (y_start + y) << SUBPEL_BITS;
532
533 // Co-ordinate of current block in reference frame
534 // to 1/16th pixel precision.
535 x0_16 = sf->scale_value_x(x0_16, sf);
536 y0_16 = sf->scale_value_y(y0_16, sf);
537
538 // Map the top left corner of the block into the reference frame.
539 x0 = sf->scale_value_x(x_start + x, sf);
540 y0 = sf->scale_value_y(y_start + y, sf);
541
542 // Scale the MV and incorporate the sub-pixel offset of the block
543 // in the reference frame.
544 scaled_mv = vp9_scale_mv(&mv_q4, mi_x + x, mi_y + y, sf);
545 xs = sf->x_step_q4;
546 ys = sf->y_step_q4;
547 } else {
548 // Co-ordinate of containing block to pixel precision.
549 x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
550 y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
551
552 // Co-ordinate of the block to 1/16th pixel precision.
553 x0_16 = x0 << SUBPEL_BITS;
554 y0_16 = y0 << SUBPEL_BITS;
555
556 scaled_mv.row = mv->row * (1 << (1 - pd->subsampling_y));
557 scaled_mv.col = mv->col * (1 << (1 - pd->subsampling_x));
558 xs = ys = 16;
559 }
560 subpel_x = scaled_mv.col & SUBPEL_MASK;
561 subpel_y = scaled_mv.row & SUBPEL_MASK;
562
563 // Calculate the top left corner of the best matching block in the
564 // reference frame.
565 x0 += scaled_mv.col >> SUBPEL_BITS;
566 y0 += scaled_mv.row >> SUBPEL_BITS;
567 x0_16 += scaled_mv.col;
568 y0_16 += scaled_mv.row;
569
570 // Get reference block pointer.
571 buf_ptr = ref_frame + y0 * pre_buf->stride + x0;
572 buf_stride = pre_buf->stride;
573
574 // Do border extension if there is motion or the
575 // width/height is not a multiple of 8 pixels.
576 if (is_scaled || scaled_mv.col || scaled_mv.row || (frame_width & 0x7) ||
577 (frame_height & 0x7)) {
578 int y1 = ((y0_16 + (h - 1) * ys) >> SUBPEL_BITS) + 1;
579
580 // Get reference block bottom right horizontal coordinate.
581 int x1 = ((x0_16 + (w - 1) * xs) >> SUBPEL_BITS) + 1;
582 int x_pad = 0, y_pad = 0;
583
584 if (subpel_x || (sf->x_step_q4 != SUBPEL_SHIFTS)) {
585 x0 -= VP9_INTERP_EXTEND - 1;
586 x1 += VP9_INTERP_EXTEND;
587 x_pad = 1;
588 }
589
590 if (subpel_y || (sf->y_step_q4 != SUBPEL_SHIFTS)) {
591 y0 -= VP9_INTERP_EXTEND - 1;
592 y1 += VP9_INTERP_EXTEND;
593 y_pad = 1;
594 }
595
596 // Wait until reference block is ready. Pad 7 more pixels as last 7
597 // pixels of each superblock row can be changed by next superblock row.
598 if (worker != NULL)
599 vp9_frameworker_wait(worker, ref_frame_buf, VPXMAX(0, (y1 + 7))
600 << (plane == 0 ? 0 : 1));
601
602 // Skip border extension if block is inside the frame.
603 if (x0 < 0 || x0 > frame_width - 1 || x1 < 0 || x1 > frame_width - 1 ||
604 y0 < 0 || y0 > frame_height - 1 || y1 < 0 || y1 > frame_height - 1) {
605 // Extend the border.
606 const uint8_t *const buf_ptr1 = ref_frame + y0 * buf_stride + x0;
607 const int b_w = x1 - x0 + 1;
608 const int b_h = y1 - y0 + 1;
609 const int border_offset = y_pad * 3 * b_w + x_pad * 3;
610
611 extend_and_predict(buf_ptr1, buf_stride, x0, y0, b_w, b_h, frame_width,
612 frame_height, border_offset, dst, dst_buf->stride,
613 subpel_x, subpel_y, kernel, sf,
614 #if CONFIG_VP9_HIGHBITDEPTH
615 xd,
616 #endif
617 w, h, ref, xs, ys);
618 return;
619 }
620 } else {
621 // Wait until reference block is ready. Pad 7 more pixels as last 7
622 // pixels of each superblock row can be changed by next superblock row.
623 if (worker != NULL) {
624 const int y1 = (y0_16 + (h - 1) * ys) >> SUBPEL_BITS;
625 vp9_frameworker_wait(worker, ref_frame_buf, VPXMAX(0, (y1 + 7))
626 << (plane == 0 ? 0 : 1));
627 }
628 }
629 #if CONFIG_VP9_HIGHBITDEPTH
630 if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
631 highbd_inter_predictor(CONVERT_TO_SHORTPTR(buf_ptr), buf_stride,
632 CONVERT_TO_SHORTPTR(dst), dst_buf->stride, subpel_x,
633 subpel_y, sf, w, h, ref, kernel, xs, ys, xd->bd);
634 } else {
635 inter_predictor(buf_ptr, buf_stride, dst, dst_buf->stride, subpel_x,
636 subpel_y, sf, w, h, ref, kernel, xs, ys);
637 }
638 #else
639 inter_predictor(buf_ptr, buf_stride, dst, dst_buf->stride, subpel_x, subpel_y,
640 sf, w, h, ref, kernel, xs, ys);
641 #endif // CONFIG_VP9_HIGHBITDEPTH
642 }
643
dec_build_inter_predictors_sb(VP9Decoder * const pbi,MACROBLOCKD * xd,int mi_row,int mi_col)644 static void dec_build_inter_predictors_sb(VP9Decoder *const pbi,
645 MACROBLOCKD *xd, int mi_row,
646 int mi_col) {
647 int plane;
648 const int mi_x = mi_col * MI_SIZE;
649 const int mi_y = mi_row * MI_SIZE;
650 const MODE_INFO *mi = xd->mi[0];
651 const InterpKernel *kernel = vp9_filter_kernels[mi->interp_filter];
652 const BLOCK_SIZE sb_type = mi->sb_type;
653 const int is_compound = has_second_ref(mi);
654 int ref;
655 int is_scaled;
656 VPxWorker *const fwo =
657 pbi->frame_parallel_decode ? pbi->frame_worker_owner : NULL;
658
659 for (ref = 0; ref < 1 + is_compound; ++ref) {
660 const MV_REFERENCE_FRAME frame = mi->ref_frame[ref];
661 RefBuffer *ref_buf = &pbi->common.frame_refs[frame - LAST_FRAME];
662 const struct scale_factors *const sf = &ref_buf->sf;
663 const int idx = ref_buf->idx;
664 BufferPool *const pool = pbi->common.buffer_pool;
665 RefCntBuffer *const ref_frame_buf = &pool->frame_bufs[idx];
666
667 if (!vp9_is_valid_scale(sf))
668 vpx_internal_error(xd->error_info, VPX_CODEC_UNSUP_BITSTREAM,
669 "Reference frame has invalid dimensions");
670
671 is_scaled = vp9_is_scaled(sf);
672 vp9_setup_pre_planes(xd, ref, ref_buf->buf, mi_row, mi_col,
673 is_scaled ? sf : NULL);
674 xd->block_refs[ref] = ref_buf;
675
676 if (sb_type < BLOCK_8X8) {
677 for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
678 struct macroblockd_plane *const pd = &xd->plane[plane];
679 struct buf_2d *const dst_buf = &pd->dst;
680 const int num_4x4_w = pd->n4_w;
681 const int num_4x4_h = pd->n4_h;
682 const int n4w_x4 = 4 * num_4x4_w;
683 const int n4h_x4 = 4 * num_4x4_h;
684 struct buf_2d *const pre_buf = &pd->pre[ref];
685 int i = 0, x, y;
686 for (y = 0; y < num_4x4_h; ++y) {
687 for (x = 0; x < num_4x4_w; ++x) {
688 const MV mv = average_split_mvs(pd, mi, ref, i++);
689 dec_build_inter_predictors(fwo, xd, plane, n4w_x4, n4h_x4, 4 * x,
690 4 * y, 4, 4, mi_x, mi_y, kernel, sf,
691 pre_buf, dst_buf, &mv, ref_frame_buf,
692 is_scaled, ref);
693 }
694 }
695 }
696 } else {
697 const MV mv = mi->mv[ref].as_mv;
698 for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
699 struct macroblockd_plane *const pd = &xd->plane[plane];
700 struct buf_2d *const dst_buf = &pd->dst;
701 const int num_4x4_w = pd->n4_w;
702 const int num_4x4_h = pd->n4_h;
703 const int n4w_x4 = 4 * num_4x4_w;
704 const int n4h_x4 = 4 * num_4x4_h;
705 struct buf_2d *const pre_buf = &pd->pre[ref];
706 dec_build_inter_predictors(fwo, xd, plane, n4w_x4, n4h_x4, 0, 0, n4w_x4,
707 n4h_x4, mi_x, mi_y, kernel, sf, pre_buf,
708 dst_buf, &mv, ref_frame_buf, is_scaled, ref);
709 }
710 }
711 }
712 }
713
dec_reset_skip_context(MACROBLOCKD * xd)714 static INLINE void dec_reset_skip_context(MACROBLOCKD *xd) {
715 int i;
716 for (i = 0; i < MAX_MB_PLANE; i++) {
717 struct macroblockd_plane *const pd = &xd->plane[i];
718 memset(pd->above_context, 0, sizeof(ENTROPY_CONTEXT) * pd->n4_w);
719 memset(pd->left_context, 0, sizeof(ENTROPY_CONTEXT) * pd->n4_h);
720 }
721 }
722
set_plane_n4(MACROBLOCKD * const xd,int bw,int bh,int bwl,int bhl)723 static void set_plane_n4(MACROBLOCKD *const xd, int bw, int bh, int bwl,
724 int bhl) {
725 int i;
726 for (i = 0; i < MAX_MB_PLANE; i++) {
727 xd->plane[i].n4_w = (bw << 1) >> xd->plane[i].subsampling_x;
728 xd->plane[i].n4_h = (bh << 1) >> xd->plane[i].subsampling_y;
729 xd->plane[i].n4_wl = bwl - xd->plane[i].subsampling_x;
730 xd->plane[i].n4_hl = bhl - xd->plane[i].subsampling_y;
731 }
732 }
733
set_offsets(VP9_COMMON * const cm,MACROBLOCKD * const xd,BLOCK_SIZE bsize,int mi_row,int mi_col,int bw,int bh,int x_mis,int y_mis,int bwl,int bhl)734 static MODE_INFO *set_offsets(VP9_COMMON *const cm, MACROBLOCKD *const xd,
735 BLOCK_SIZE bsize, int mi_row, int mi_col, int bw,
736 int bh, int x_mis, int y_mis, int bwl, int bhl) {
737 const int offset = mi_row * cm->mi_stride + mi_col;
738 int x, y;
739 const TileInfo *const tile = &xd->tile;
740
741 xd->mi = cm->mi_grid_visible + offset;
742 xd->mi[0] = &cm->mi[offset];
743 // TODO(slavarnway): Generate sb_type based on bwl and bhl, instead of
744 // passing bsize from decode_partition().
745 xd->mi[0]->sb_type = bsize;
746 for (y = 0; y < y_mis; ++y)
747 for (x = !y; x < x_mis; ++x) {
748 xd->mi[y * cm->mi_stride + x] = xd->mi[0];
749 }
750
751 set_plane_n4(xd, bw, bh, bwl, bhl);
752
753 set_skip_context(xd, mi_row, mi_col);
754
755 // Distance of Mb to the various image edges. These are specified to 8th pel
756 // as they are always compared to values that are in 1/8th pel units
757 set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
758
759 vp9_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
760 return xd->mi[0];
761 }
762
decode_block(TileWorkerData * twd,VP9Decoder * const pbi,int mi_row,int mi_col,BLOCK_SIZE bsize,int bwl,int bhl)763 static void decode_block(TileWorkerData *twd, VP9Decoder *const pbi, int mi_row,
764 int mi_col, BLOCK_SIZE bsize, int bwl, int bhl) {
765 VP9_COMMON *const cm = &pbi->common;
766 const int less8x8 = bsize < BLOCK_8X8;
767 const int bw = 1 << (bwl - 1);
768 const int bh = 1 << (bhl - 1);
769 const int x_mis = VPXMIN(bw, cm->mi_cols - mi_col);
770 const int y_mis = VPXMIN(bh, cm->mi_rows - mi_row);
771 vpx_reader *r = &twd->bit_reader;
772 MACROBLOCKD *const xd = &twd->xd;
773
774 MODE_INFO *mi = set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis,
775 y_mis, bwl, bhl);
776
777 if (bsize >= BLOCK_8X8 && (cm->subsampling_x || cm->subsampling_y)) {
778 const BLOCK_SIZE uv_subsize =
779 ss_size_lookup[bsize][cm->subsampling_x][cm->subsampling_y];
780 if (uv_subsize == BLOCK_INVALID)
781 vpx_internal_error(xd->error_info, VPX_CODEC_CORRUPT_FRAME,
782 "Invalid block size.");
783 }
784
785 vp9_read_mode_info(twd, pbi, mi_row, mi_col, x_mis, y_mis);
786
787 if (mi->skip) {
788 dec_reset_skip_context(xd);
789 }
790
791 if (!is_inter_block(mi)) {
792 int plane;
793 for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
794 const struct macroblockd_plane *const pd = &xd->plane[plane];
795 const TX_SIZE tx_size = plane ? get_uv_tx_size(mi, pd) : mi->tx_size;
796 const int num_4x4_w = pd->n4_w;
797 const int num_4x4_h = pd->n4_h;
798 const int step = (1 << tx_size);
799 int row, col;
800 const int max_blocks_wide =
801 num_4x4_w + (xd->mb_to_right_edge >= 0
802 ? 0
803 : xd->mb_to_right_edge >> (5 + pd->subsampling_x));
804 const int max_blocks_high =
805 num_4x4_h + (xd->mb_to_bottom_edge >= 0
806 ? 0
807 : xd->mb_to_bottom_edge >> (5 + pd->subsampling_y));
808
809 xd->max_blocks_wide = xd->mb_to_right_edge >= 0 ? 0 : max_blocks_wide;
810 xd->max_blocks_high = xd->mb_to_bottom_edge >= 0 ? 0 : max_blocks_high;
811
812 for (row = 0; row < max_blocks_high; row += step)
813 for (col = 0; col < max_blocks_wide; col += step)
814 predict_and_reconstruct_intra_block(twd, mi, plane, row, col,
815 tx_size);
816 }
817 } else {
818 // Prediction
819 dec_build_inter_predictors_sb(pbi, xd, mi_row, mi_col);
820
821 // Reconstruction
822 if (!mi->skip) {
823 int eobtotal = 0;
824 int plane;
825
826 for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
827 const struct macroblockd_plane *const pd = &xd->plane[plane];
828 const TX_SIZE tx_size = plane ? get_uv_tx_size(mi, pd) : mi->tx_size;
829 const int num_4x4_w = pd->n4_w;
830 const int num_4x4_h = pd->n4_h;
831 const int step = (1 << tx_size);
832 int row, col;
833 const int max_blocks_wide =
834 num_4x4_w + (xd->mb_to_right_edge >= 0
835 ? 0
836 : xd->mb_to_right_edge >> (5 + pd->subsampling_x));
837 const int max_blocks_high =
838 num_4x4_h +
839 (xd->mb_to_bottom_edge >= 0
840 ? 0
841 : xd->mb_to_bottom_edge >> (5 + pd->subsampling_y));
842
843 xd->max_blocks_wide = xd->mb_to_right_edge >= 0 ? 0 : max_blocks_wide;
844 xd->max_blocks_high = xd->mb_to_bottom_edge >= 0 ? 0 : max_blocks_high;
845
846 for (row = 0; row < max_blocks_high; row += step)
847 for (col = 0; col < max_blocks_wide; col += step)
848 eobtotal +=
849 reconstruct_inter_block(twd, mi, plane, row, col, tx_size);
850 }
851
852 if (!less8x8 && eobtotal == 0) mi->skip = 1; // skip loopfilter
853 }
854 }
855
856 xd->corrupted |= vpx_reader_has_error(r);
857
858 if (cm->lf.filter_level) {
859 vp9_build_mask(cm, mi, mi_row, mi_col, bw, bh);
860 }
861 }
862
dec_partition_plane_context(TileWorkerData * twd,int mi_row,int mi_col,int bsl)863 static INLINE int dec_partition_plane_context(TileWorkerData *twd, int mi_row,
864 int mi_col, int bsl) {
865 const PARTITION_CONTEXT *above_ctx = twd->xd.above_seg_context + mi_col;
866 const PARTITION_CONTEXT *left_ctx =
867 twd->xd.left_seg_context + (mi_row & MI_MASK);
868 int above = (*above_ctx >> bsl) & 1, left = (*left_ctx >> bsl) & 1;
869
870 // assert(bsl >= 0);
871
872 return (left * 2 + above) + bsl * PARTITION_PLOFFSET;
873 }
874
dec_update_partition_context(TileWorkerData * twd,int mi_row,int mi_col,BLOCK_SIZE subsize,int bw)875 static INLINE void dec_update_partition_context(TileWorkerData *twd, int mi_row,
876 int mi_col, BLOCK_SIZE subsize,
877 int bw) {
878 PARTITION_CONTEXT *const above_ctx = twd->xd.above_seg_context + mi_col;
879 PARTITION_CONTEXT *const left_ctx =
880 twd->xd.left_seg_context + (mi_row & MI_MASK);
881
882 // update the partition context at the end notes. set partition bits
883 // of block sizes larger than the current one to be one, and partition
884 // bits of smaller block sizes to be zero.
885 memset(above_ctx, partition_context_lookup[subsize].above, bw);
886 memset(left_ctx, partition_context_lookup[subsize].left, bw);
887 }
888
read_partition(TileWorkerData * twd,int mi_row,int mi_col,int has_rows,int has_cols,int bsl)889 static PARTITION_TYPE read_partition(TileWorkerData *twd, int mi_row,
890 int mi_col, int has_rows, int has_cols,
891 int bsl) {
892 const int ctx = dec_partition_plane_context(twd, mi_row, mi_col, bsl);
893 const vpx_prob *const probs = twd->xd.partition_probs[ctx];
894 FRAME_COUNTS *counts = twd->xd.counts;
895 PARTITION_TYPE p;
896 vpx_reader *r = &twd->bit_reader;
897
898 if (has_rows && has_cols)
899 p = (PARTITION_TYPE)vpx_read_tree(r, vp9_partition_tree, probs);
900 else if (!has_rows && has_cols)
901 p = vpx_read(r, probs[1]) ? PARTITION_SPLIT : PARTITION_HORZ;
902 else if (has_rows && !has_cols)
903 p = vpx_read(r, probs[2]) ? PARTITION_SPLIT : PARTITION_VERT;
904 else
905 p = PARTITION_SPLIT;
906
907 if (counts) ++counts->partition[ctx][p];
908
909 return p;
910 }
911
912 // TODO(slavarnway): eliminate bsize and subsize in future commits
decode_partition(TileWorkerData * twd,VP9Decoder * const pbi,int mi_row,int mi_col,BLOCK_SIZE bsize,int n4x4_l2)913 static void decode_partition(TileWorkerData *twd, VP9Decoder *const pbi,
914 int mi_row, int mi_col, BLOCK_SIZE bsize,
915 int n4x4_l2) {
916 VP9_COMMON *const cm = &pbi->common;
917 const int n8x8_l2 = n4x4_l2 - 1;
918 const int num_8x8_wh = 1 << n8x8_l2;
919 const int hbs = num_8x8_wh >> 1;
920 PARTITION_TYPE partition;
921 BLOCK_SIZE subsize;
922 const int has_rows = (mi_row + hbs) < cm->mi_rows;
923 const int has_cols = (mi_col + hbs) < cm->mi_cols;
924 MACROBLOCKD *const xd = &twd->xd;
925
926 if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;
927
928 partition = read_partition(twd, mi_row, mi_col, has_rows, has_cols, n8x8_l2);
929 subsize = subsize_lookup[partition][bsize]; // get_subsize(bsize, partition);
930 if (!hbs) {
931 // calculate bmode block dimensions (log 2)
932 xd->bmode_blocks_wl = 1 >> !!(partition & PARTITION_VERT);
933 xd->bmode_blocks_hl = 1 >> !!(partition & PARTITION_HORZ);
934 decode_block(twd, pbi, mi_row, mi_col, subsize, 1, 1);
935 } else {
936 switch (partition) {
937 case PARTITION_NONE:
938 decode_block(twd, pbi, mi_row, mi_col, subsize, n4x4_l2, n4x4_l2);
939 break;
940 case PARTITION_HORZ:
941 decode_block(twd, pbi, mi_row, mi_col, subsize, n4x4_l2, n8x8_l2);
942 if (has_rows)
943 decode_block(twd, pbi, mi_row + hbs, mi_col, subsize, n4x4_l2,
944 n8x8_l2);
945 break;
946 case PARTITION_VERT:
947 decode_block(twd, pbi, mi_row, mi_col, subsize, n8x8_l2, n4x4_l2);
948 if (has_cols)
949 decode_block(twd, pbi, mi_row, mi_col + hbs, subsize, n8x8_l2,
950 n4x4_l2);
951 break;
952 case PARTITION_SPLIT:
953 decode_partition(twd, pbi, mi_row, mi_col, subsize, n8x8_l2);
954 decode_partition(twd, pbi, mi_row, mi_col + hbs, subsize, n8x8_l2);
955 decode_partition(twd, pbi, mi_row + hbs, mi_col, subsize, n8x8_l2);
956 decode_partition(twd, pbi, mi_row + hbs, mi_col + hbs, subsize,
957 n8x8_l2);
958 break;
959 default: assert(0 && "Invalid partition type");
960 }
961 }
962
963 // update partition context
964 if (bsize >= BLOCK_8X8 &&
965 (bsize == BLOCK_8X8 || partition != PARTITION_SPLIT))
966 dec_update_partition_context(twd, mi_row, mi_col, subsize, num_8x8_wh);
967 }
968
setup_token_decoder(const uint8_t * data,const uint8_t * data_end,size_t read_size,struct vpx_internal_error_info * error_info,vpx_reader * r,vpx_decrypt_cb decrypt_cb,void * decrypt_state)969 static void setup_token_decoder(const uint8_t *data, const uint8_t *data_end,
970 size_t read_size,
971 struct vpx_internal_error_info *error_info,
972 vpx_reader *r, vpx_decrypt_cb decrypt_cb,
973 void *decrypt_state) {
974 // Validate the calculated partition length. If the buffer
975 // described by the partition can't be fully read, then restrict
976 // it to the portion that can be (for EC mode) or throw an error.
977 if (!read_is_valid(data, read_size, data_end))
978 vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME,
979 "Truncated packet or corrupt tile length");
980
981 if (vpx_reader_init(r, data, read_size, decrypt_cb, decrypt_state))
982 vpx_internal_error(error_info, VPX_CODEC_MEM_ERROR,
983 "Failed to allocate bool decoder %d", 1);
984 }
985
read_coef_probs_common(vp9_coeff_probs_model * coef_probs,vpx_reader * r)986 static void read_coef_probs_common(vp9_coeff_probs_model *coef_probs,
987 vpx_reader *r) {
988 int i, j, k, l, m;
989
990 if (vpx_read_bit(r))
991 for (i = 0; i < PLANE_TYPES; ++i)
992 for (j = 0; j < REF_TYPES; ++j)
993 for (k = 0; k < COEF_BANDS; ++k)
994 for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l)
995 for (m = 0; m < UNCONSTRAINED_NODES; ++m)
996 vp9_diff_update_prob(r, &coef_probs[i][j][k][l][m]);
997 }
998
read_coef_probs(FRAME_CONTEXT * fc,TX_MODE tx_mode,vpx_reader * r)999 static void read_coef_probs(FRAME_CONTEXT *fc, TX_MODE tx_mode, vpx_reader *r) {
1000 const TX_SIZE max_tx_size = tx_mode_to_biggest_tx_size[tx_mode];
1001 TX_SIZE tx_size;
1002 for (tx_size = TX_4X4; tx_size <= max_tx_size; ++tx_size)
1003 read_coef_probs_common(fc->coef_probs[tx_size], r);
1004 }
1005
setup_segmentation(struct segmentation * seg,struct vpx_read_bit_buffer * rb)1006 static void setup_segmentation(struct segmentation *seg,
1007 struct vpx_read_bit_buffer *rb) {
1008 int i, j;
1009
1010 seg->update_map = 0;
1011 seg->update_data = 0;
1012
1013 seg->enabled = vpx_rb_read_bit(rb);
1014 if (!seg->enabled) return;
1015
1016 // Segmentation map update
1017 seg->update_map = vpx_rb_read_bit(rb);
1018 if (seg->update_map) {
1019 for (i = 0; i < SEG_TREE_PROBS; i++)
1020 seg->tree_probs[i] =
1021 vpx_rb_read_bit(rb) ? vpx_rb_read_literal(rb, 8) : MAX_PROB;
1022
1023 seg->temporal_update = vpx_rb_read_bit(rb);
1024 if (seg->temporal_update) {
1025 for (i = 0; i < PREDICTION_PROBS; i++)
1026 seg->pred_probs[i] =
1027 vpx_rb_read_bit(rb) ? vpx_rb_read_literal(rb, 8) : MAX_PROB;
1028 } else {
1029 for (i = 0; i < PREDICTION_PROBS; i++) seg->pred_probs[i] = MAX_PROB;
1030 }
1031 }
1032
1033 // Segmentation data update
1034 seg->update_data = vpx_rb_read_bit(rb);
1035 if (seg->update_data) {
1036 seg->abs_delta = vpx_rb_read_bit(rb);
1037
1038 vp9_clearall_segfeatures(seg);
1039
1040 for (i = 0; i < MAX_SEGMENTS; i++) {
1041 for (j = 0; j < SEG_LVL_MAX; j++) {
1042 int data = 0;
1043 const int feature_enabled = vpx_rb_read_bit(rb);
1044 if (feature_enabled) {
1045 vp9_enable_segfeature(seg, i, j);
1046 data = decode_unsigned_max(rb, vp9_seg_feature_data_max(j));
1047 if (vp9_is_segfeature_signed(j))
1048 data = vpx_rb_read_bit(rb) ? -data : data;
1049 }
1050 vp9_set_segdata(seg, i, j, data);
1051 }
1052 }
1053 }
1054 }
1055
setup_loopfilter(struct loopfilter * lf,struct vpx_read_bit_buffer * rb)1056 static void setup_loopfilter(struct loopfilter *lf,
1057 struct vpx_read_bit_buffer *rb) {
1058 lf->filter_level = vpx_rb_read_literal(rb, 6);
1059 lf->sharpness_level = vpx_rb_read_literal(rb, 3);
1060
1061 // Read in loop filter deltas applied at the MB level based on mode or ref
1062 // frame.
1063 lf->mode_ref_delta_update = 0;
1064
1065 lf->mode_ref_delta_enabled = vpx_rb_read_bit(rb);
1066 if (lf->mode_ref_delta_enabled) {
1067 lf->mode_ref_delta_update = vpx_rb_read_bit(rb);
1068 if (lf->mode_ref_delta_update) {
1069 int i;
1070
1071 for (i = 0; i < MAX_REF_LF_DELTAS; i++)
1072 if (vpx_rb_read_bit(rb))
1073 lf->ref_deltas[i] = vpx_rb_read_signed_literal(rb, 6);
1074
1075 for (i = 0; i < MAX_MODE_LF_DELTAS; i++)
1076 if (vpx_rb_read_bit(rb))
1077 lf->mode_deltas[i] = vpx_rb_read_signed_literal(rb, 6);
1078 }
1079 }
1080 }
1081
read_delta_q(struct vpx_read_bit_buffer * rb)1082 static INLINE int read_delta_q(struct vpx_read_bit_buffer *rb) {
1083 return vpx_rb_read_bit(rb) ? vpx_rb_read_signed_literal(rb, 4) : 0;
1084 }
1085
setup_quantization(VP9_COMMON * const cm,MACROBLOCKD * const xd,struct vpx_read_bit_buffer * rb)1086 static void setup_quantization(VP9_COMMON *const cm, MACROBLOCKD *const xd,
1087 struct vpx_read_bit_buffer *rb) {
1088 cm->base_qindex = vpx_rb_read_literal(rb, QINDEX_BITS);
1089 cm->y_dc_delta_q = read_delta_q(rb);
1090 cm->uv_dc_delta_q = read_delta_q(rb);
1091 cm->uv_ac_delta_q = read_delta_q(rb);
1092 cm->dequant_bit_depth = cm->bit_depth;
1093 xd->lossless = cm->base_qindex == 0 && cm->y_dc_delta_q == 0 &&
1094 cm->uv_dc_delta_q == 0 && cm->uv_ac_delta_q == 0;
1095
1096 #if CONFIG_VP9_HIGHBITDEPTH
1097 xd->bd = (int)cm->bit_depth;
1098 #endif
1099 }
1100
setup_segmentation_dequant(VP9_COMMON * const cm)1101 static void setup_segmentation_dequant(VP9_COMMON *const cm) {
1102 // Build y/uv dequant values based on segmentation.
1103 if (cm->seg.enabled) {
1104 int i;
1105 for (i = 0; i < MAX_SEGMENTS; ++i) {
1106 const int qindex = vp9_get_qindex(&cm->seg, i, cm->base_qindex);
1107 cm->y_dequant[i][0] =
1108 vp9_dc_quant(qindex, cm->y_dc_delta_q, cm->bit_depth);
1109 cm->y_dequant[i][1] = vp9_ac_quant(qindex, 0, cm->bit_depth);
1110 cm->uv_dequant[i][0] =
1111 vp9_dc_quant(qindex, cm->uv_dc_delta_q, cm->bit_depth);
1112 cm->uv_dequant[i][1] =
1113 vp9_ac_quant(qindex, cm->uv_ac_delta_q, cm->bit_depth);
1114 }
1115 } else {
1116 const int qindex = cm->base_qindex;
1117 // When segmentation is disabled, only the first value is used. The
1118 // remaining are don't cares.
1119 cm->y_dequant[0][0] = vp9_dc_quant(qindex, cm->y_dc_delta_q, cm->bit_depth);
1120 cm->y_dequant[0][1] = vp9_ac_quant(qindex, 0, cm->bit_depth);
1121 cm->uv_dequant[0][0] =
1122 vp9_dc_quant(qindex, cm->uv_dc_delta_q, cm->bit_depth);
1123 cm->uv_dequant[0][1] =
1124 vp9_ac_quant(qindex, cm->uv_ac_delta_q, cm->bit_depth);
1125 }
1126 }
1127
read_interp_filter(struct vpx_read_bit_buffer * rb)1128 static INTERP_FILTER read_interp_filter(struct vpx_read_bit_buffer *rb) {
1129 const INTERP_FILTER literal_to_filter[] = { EIGHTTAP_SMOOTH, EIGHTTAP,
1130 EIGHTTAP_SHARP, BILINEAR };
1131 return vpx_rb_read_bit(rb) ? SWITCHABLE
1132 : literal_to_filter[vpx_rb_read_literal(rb, 2)];
1133 }
1134
setup_render_size(VP9_COMMON * cm,struct vpx_read_bit_buffer * rb)1135 static void setup_render_size(VP9_COMMON *cm, struct vpx_read_bit_buffer *rb) {
1136 cm->render_width = cm->width;
1137 cm->render_height = cm->height;
1138 if (vpx_rb_read_bit(rb))
1139 vp9_read_frame_size(rb, &cm->render_width, &cm->render_height);
1140 }
1141
resize_mv_buffer(VP9_COMMON * cm)1142 static void resize_mv_buffer(VP9_COMMON *cm) {
1143 vpx_free(cm->cur_frame->mvs);
1144 cm->cur_frame->mi_rows = cm->mi_rows;
1145 cm->cur_frame->mi_cols = cm->mi_cols;
1146 CHECK_MEM_ERROR(cm, cm->cur_frame->mvs,
1147 (MV_REF *)vpx_calloc(cm->mi_rows * cm->mi_cols,
1148 sizeof(*cm->cur_frame->mvs)));
1149 }
1150
resize_context_buffers(VP9_COMMON * cm,int width,int height)1151 static void resize_context_buffers(VP9_COMMON *cm, int width, int height) {
1152 #if CONFIG_SIZE_LIMIT
1153 if (width > DECODE_WIDTH_LIMIT || height > DECODE_HEIGHT_LIMIT)
1154 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
1155 "Dimensions of %dx%d beyond allowed size of %dx%d.",
1156 width, height, DECODE_WIDTH_LIMIT, DECODE_HEIGHT_LIMIT);
1157 #endif
1158 if (cm->width != width || cm->height != height) {
1159 const int new_mi_rows =
1160 ALIGN_POWER_OF_TWO(height, MI_SIZE_LOG2) >> MI_SIZE_LOG2;
1161 const int new_mi_cols =
1162 ALIGN_POWER_OF_TWO(width, MI_SIZE_LOG2) >> MI_SIZE_LOG2;
1163
1164 // Allocations in vp9_alloc_context_buffers() depend on individual
1165 // dimensions as well as the overall size.
1166 if (new_mi_cols > cm->mi_cols || new_mi_rows > cm->mi_rows) {
1167 if (vp9_alloc_context_buffers(cm, width, height))
1168 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
1169 "Failed to allocate context buffers");
1170 } else {
1171 vp9_set_mb_mi(cm, width, height);
1172 }
1173 vp9_init_context_buffers(cm);
1174 cm->width = width;
1175 cm->height = height;
1176 }
1177 if (cm->cur_frame->mvs == NULL || cm->mi_rows > cm->cur_frame->mi_rows ||
1178 cm->mi_cols > cm->cur_frame->mi_cols) {
1179 resize_mv_buffer(cm);
1180 }
1181 }
1182
setup_frame_size(VP9_COMMON * cm,struct vpx_read_bit_buffer * rb)1183 static void setup_frame_size(VP9_COMMON *cm, struct vpx_read_bit_buffer *rb) {
1184 int width, height;
1185 BufferPool *const pool = cm->buffer_pool;
1186 vp9_read_frame_size(rb, &width, &height);
1187 resize_context_buffers(cm, width, height);
1188 setup_render_size(cm, rb);
1189
1190 lock_buffer_pool(pool);
1191 if (vpx_realloc_frame_buffer(
1192 get_frame_new_buffer(cm), cm->width, cm->height, cm->subsampling_x,
1193 cm->subsampling_y,
1194 #if CONFIG_VP9_HIGHBITDEPTH
1195 cm->use_highbitdepth,
1196 #endif
1197 VP9_DEC_BORDER_IN_PIXELS, cm->byte_alignment,
1198 &pool->frame_bufs[cm->new_fb_idx].raw_frame_buffer, pool->get_fb_cb,
1199 pool->cb_priv)) {
1200 unlock_buffer_pool(pool);
1201 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
1202 "Failed to allocate frame buffer");
1203 }
1204 unlock_buffer_pool(pool);
1205
1206 pool->frame_bufs[cm->new_fb_idx].buf.subsampling_x = cm->subsampling_x;
1207 pool->frame_bufs[cm->new_fb_idx].buf.subsampling_y = cm->subsampling_y;
1208 pool->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth;
1209 pool->frame_bufs[cm->new_fb_idx].buf.color_space = cm->color_space;
1210 pool->frame_bufs[cm->new_fb_idx].buf.color_range = cm->color_range;
1211 pool->frame_bufs[cm->new_fb_idx].buf.render_width = cm->render_width;
1212 pool->frame_bufs[cm->new_fb_idx].buf.render_height = cm->render_height;
1213 }
1214
valid_ref_frame_img_fmt(vpx_bit_depth_t ref_bit_depth,int ref_xss,int ref_yss,vpx_bit_depth_t this_bit_depth,int this_xss,int this_yss)1215 static INLINE int valid_ref_frame_img_fmt(vpx_bit_depth_t ref_bit_depth,
1216 int ref_xss, int ref_yss,
1217 vpx_bit_depth_t this_bit_depth,
1218 int this_xss, int this_yss) {
1219 return ref_bit_depth == this_bit_depth && ref_xss == this_xss &&
1220 ref_yss == this_yss;
1221 }
1222
setup_frame_size_with_refs(VP9_COMMON * cm,struct vpx_read_bit_buffer * rb)1223 static void setup_frame_size_with_refs(VP9_COMMON *cm,
1224 struct vpx_read_bit_buffer *rb) {
1225 int width, height;
1226 int found = 0, i;
1227 int has_valid_ref_frame = 0;
1228 BufferPool *const pool = cm->buffer_pool;
1229 for (i = 0; i < REFS_PER_FRAME; ++i) {
1230 if (vpx_rb_read_bit(rb)) {
1231 if (cm->frame_refs[i].idx != INVALID_IDX) {
1232 YV12_BUFFER_CONFIG *const buf = cm->frame_refs[i].buf;
1233 width = buf->y_crop_width;
1234 height = buf->y_crop_height;
1235 found = 1;
1236 break;
1237 } else {
1238 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
1239 "Failed to decode frame size");
1240 }
1241 }
1242 }
1243
1244 if (!found) vp9_read_frame_size(rb, &width, &height);
1245
1246 if (width <= 0 || height <= 0)
1247 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
1248 "Invalid frame size");
1249
1250 // Check to make sure at least one of frames that this frame references
1251 // has valid dimensions.
1252 for (i = 0; i < REFS_PER_FRAME; ++i) {
1253 RefBuffer *const ref_frame = &cm->frame_refs[i];
1254 has_valid_ref_frame |=
1255 (ref_frame->idx != INVALID_IDX &&
1256 valid_ref_frame_size(ref_frame->buf->y_crop_width,
1257 ref_frame->buf->y_crop_height, width, height));
1258 }
1259 if (!has_valid_ref_frame)
1260 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
1261 "Referenced frame has invalid size");
1262 for (i = 0; i < REFS_PER_FRAME; ++i) {
1263 RefBuffer *const ref_frame = &cm->frame_refs[i];
1264 if (ref_frame->idx == INVALID_IDX ||
1265 !valid_ref_frame_img_fmt(ref_frame->buf->bit_depth,
1266 ref_frame->buf->subsampling_x,
1267 ref_frame->buf->subsampling_y, cm->bit_depth,
1268 cm->subsampling_x, cm->subsampling_y))
1269 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
1270 "Referenced frame has incompatible color format");
1271 }
1272
1273 resize_context_buffers(cm, width, height);
1274 setup_render_size(cm, rb);
1275
1276 lock_buffer_pool(pool);
1277 if (vpx_realloc_frame_buffer(
1278 get_frame_new_buffer(cm), cm->width, cm->height, cm->subsampling_x,
1279 cm->subsampling_y,
1280 #if CONFIG_VP9_HIGHBITDEPTH
1281 cm->use_highbitdepth,
1282 #endif
1283 VP9_DEC_BORDER_IN_PIXELS, cm->byte_alignment,
1284 &pool->frame_bufs[cm->new_fb_idx].raw_frame_buffer, pool->get_fb_cb,
1285 pool->cb_priv)) {
1286 unlock_buffer_pool(pool);
1287 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
1288 "Failed to allocate frame buffer");
1289 }
1290 unlock_buffer_pool(pool);
1291
1292 pool->frame_bufs[cm->new_fb_idx].buf.subsampling_x = cm->subsampling_x;
1293 pool->frame_bufs[cm->new_fb_idx].buf.subsampling_y = cm->subsampling_y;
1294 pool->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth;
1295 pool->frame_bufs[cm->new_fb_idx].buf.color_space = cm->color_space;
1296 pool->frame_bufs[cm->new_fb_idx].buf.color_range = cm->color_range;
1297 pool->frame_bufs[cm->new_fb_idx].buf.render_width = cm->render_width;
1298 pool->frame_bufs[cm->new_fb_idx].buf.render_height = cm->render_height;
1299 }
1300
setup_tile_info(VP9_COMMON * cm,struct vpx_read_bit_buffer * rb)1301 static void setup_tile_info(VP9_COMMON *cm, struct vpx_read_bit_buffer *rb) {
1302 int min_log2_tile_cols, max_log2_tile_cols, max_ones;
1303 vp9_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);
1304
1305 // columns
1306 max_ones = max_log2_tile_cols - min_log2_tile_cols;
1307 cm->log2_tile_cols = min_log2_tile_cols;
1308 while (max_ones-- && vpx_rb_read_bit(rb)) cm->log2_tile_cols++;
1309
1310 if (cm->log2_tile_cols > 6)
1311 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
1312 "Invalid number of tile columns");
1313
1314 // rows
1315 cm->log2_tile_rows = vpx_rb_read_bit(rb);
1316 if (cm->log2_tile_rows) cm->log2_tile_rows += vpx_rb_read_bit(rb);
1317 }
1318
1319 // Reads the next tile returning its size and adjusting '*data' accordingly
1320 // based on 'is_last'.
get_tile_buffer(const uint8_t * const data_end,int is_last,struct vpx_internal_error_info * error_info,const uint8_t ** data,vpx_decrypt_cb decrypt_cb,void * decrypt_state,TileBuffer * buf)1321 static void get_tile_buffer(const uint8_t *const data_end, int is_last,
1322 struct vpx_internal_error_info *error_info,
1323 const uint8_t **data, vpx_decrypt_cb decrypt_cb,
1324 void *decrypt_state, TileBuffer *buf) {
1325 size_t size;
1326
1327 if (!is_last) {
1328 if (!read_is_valid(*data, 4, data_end))
1329 vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME,
1330 "Truncated packet or corrupt tile length");
1331
1332 if (decrypt_cb) {
1333 uint8_t be_data[4];
1334 decrypt_cb(decrypt_state, *data, be_data, 4);
1335 size = mem_get_be32(be_data);
1336 } else {
1337 size = mem_get_be32(*data);
1338 }
1339 *data += 4;
1340
1341 if (size > (size_t)(data_end - *data))
1342 vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME,
1343 "Truncated packet or corrupt tile size");
1344 } else {
1345 size = data_end - *data;
1346 }
1347
1348 buf->data = *data;
1349 buf->size = size;
1350
1351 *data += size;
1352 }
1353
get_tile_buffers(VP9Decoder * pbi,const uint8_t * data,const uint8_t * data_end,int tile_cols,int tile_rows,TileBuffer (* tile_buffers)[1<<6])1354 static void get_tile_buffers(VP9Decoder *pbi, const uint8_t *data,
1355 const uint8_t *data_end, int tile_cols,
1356 int tile_rows,
1357 TileBuffer (*tile_buffers)[1 << 6]) {
1358 int r, c;
1359
1360 for (r = 0; r < tile_rows; ++r) {
1361 for (c = 0; c < tile_cols; ++c) {
1362 const int is_last = (r == tile_rows - 1) && (c == tile_cols - 1);
1363 TileBuffer *const buf = &tile_buffers[r][c];
1364 buf->col = c;
1365 get_tile_buffer(data_end, is_last, &pbi->common.error, &data,
1366 pbi->decrypt_cb, pbi->decrypt_state, buf);
1367 }
1368 }
1369 }
1370
decode_tiles(VP9Decoder * pbi,const uint8_t * data,const uint8_t * data_end)1371 static const uint8_t *decode_tiles(VP9Decoder *pbi, const uint8_t *data,
1372 const uint8_t *data_end) {
1373 VP9_COMMON *const cm = &pbi->common;
1374 const VPxWorkerInterface *const winterface = vpx_get_worker_interface();
1375 const int aligned_cols = mi_cols_aligned_to_sb(cm->mi_cols);
1376 const int tile_cols = 1 << cm->log2_tile_cols;
1377 const int tile_rows = 1 << cm->log2_tile_rows;
1378 TileBuffer tile_buffers[4][1 << 6];
1379 int tile_row, tile_col;
1380 int mi_row, mi_col;
1381 TileWorkerData *tile_data = NULL;
1382
1383 if (cm->lf.filter_level && !cm->skip_loop_filter &&
1384 pbi->lf_worker.data1 == NULL) {
1385 CHECK_MEM_ERROR(cm, pbi->lf_worker.data1,
1386 vpx_memalign(32, sizeof(LFWorkerData)));
1387 pbi->lf_worker.hook = (VPxWorkerHook)vp9_loop_filter_worker;
1388 if (pbi->max_threads > 1 && !winterface->reset(&pbi->lf_worker)) {
1389 vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
1390 "Loop filter thread creation failed");
1391 }
1392 }
1393
1394 if (cm->lf.filter_level && !cm->skip_loop_filter) {
1395 LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1;
1396 // Be sure to sync as we might be resuming after a failed frame decode.
1397 winterface->sync(&pbi->lf_worker);
1398 vp9_loop_filter_data_reset(lf_data, get_frame_new_buffer(cm), cm,
1399 pbi->mb.plane);
1400 }
1401
1402 assert(tile_rows <= 4);
1403 assert(tile_cols <= (1 << 6));
1404
1405 // Note: this memset assumes above_context[0], [1] and [2]
1406 // are allocated as part of the same buffer.
1407 memset(cm->above_context, 0,
1408 sizeof(*cm->above_context) * MAX_MB_PLANE * 2 * aligned_cols);
1409
1410 memset(cm->above_seg_context, 0,
1411 sizeof(*cm->above_seg_context) * aligned_cols);
1412
1413 vp9_reset_lfm(cm);
1414
1415 get_tile_buffers(pbi, data, data_end, tile_cols, tile_rows, tile_buffers);
1416
1417 // Load all tile information into tile_data.
1418 for (tile_row = 0; tile_row < tile_rows; ++tile_row) {
1419 for (tile_col = 0; tile_col < tile_cols; ++tile_col) {
1420 const TileBuffer *const buf = &tile_buffers[tile_row][tile_col];
1421 tile_data = pbi->tile_worker_data + tile_cols * tile_row + tile_col;
1422 tile_data->xd = pbi->mb;
1423 tile_data->xd.corrupted = 0;
1424 tile_data->xd.counts =
1425 cm->frame_parallel_decoding_mode ? NULL : &cm->counts;
1426 vp9_zero(tile_data->dqcoeff);
1427 vp9_tile_init(&tile_data->xd.tile, cm, tile_row, tile_col);
1428 setup_token_decoder(buf->data, data_end, buf->size, &cm->error,
1429 &tile_data->bit_reader, pbi->decrypt_cb,
1430 pbi->decrypt_state);
1431 vp9_init_macroblockd(cm, &tile_data->xd, tile_data->dqcoeff);
1432 }
1433 }
1434
1435 for (tile_row = 0; tile_row < tile_rows; ++tile_row) {
1436 TileInfo tile;
1437 vp9_tile_set_row(&tile, cm, tile_row);
1438 for (mi_row = tile.mi_row_start; mi_row < tile.mi_row_end;
1439 mi_row += MI_BLOCK_SIZE) {
1440 for (tile_col = 0; tile_col < tile_cols; ++tile_col) {
1441 const int col =
1442 pbi->inv_tile_order ? tile_cols - tile_col - 1 : tile_col;
1443 tile_data = pbi->tile_worker_data + tile_cols * tile_row + col;
1444 vp9_tile_set_col(&tile, cm, col);
1445 vp9_zero(tile_data->xd.left_context);
1446 vp9_zero(tile_data->xd.left_seg_context);
1447 for (mi_col = tile.mi_col_start; mi_col < tile.mi_col_end;
1448 mi_col += MI_BLOCK_SIZE) {
1449 decode_partition(tile_data, pbi, mi_row, mi_col, BLOCK_64X64, 4);
1450 }
1451 pbi->mb.corrupted |= tile_data->xd.corrupted;
1452 if (pbi->mb.corrupted)
1453 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
1454 "Failed to decode tile data");
1455 }
1456 // Loopfilter one row.
1457 if (cm->lf.filter_level && !cm->skip_loop_filter) {
1458 const int lf_start = mi_row - MI_BLOCK_SIZE;
1459 LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1;
1460
1461 // delay the loopfilter by 1 macroblock row.
1462 if (lf_start < 0) continue;
1463
1464 // decoding has completed: finish up the loop filter in this thread.
1465 if (mi_row + MI_BLOCK_SIZE >= cm->mi_rows) continue;
1466
1467 winterface->sync(&pbi->lf_worker);
1468 lf_data->start = lf_start;
1469 lf_data->stop = mi_row;
1470 if (pbi->max_threads > 1) {
1471 winterface->launch(&pbi->lf_worker);
1472 } else {
1473 winterface->execute(&pbi->lf_worker);
1474 }
1475 }
1476 // After loopfiltering, the last 7 row pixels in each superblock row may
1477 // still be changed by the longest loopfilter of the next superblock
1478 // row.
1479 if (pbi->frame_parallel_decode)
1480 vp9_frameworker_broadcast(pbi->cur_buf, mi_row << MI_BLOCK_SIZE_LOG2);
1481 }
1482 }
1483
1484 // Loopfilter remaining rows in the frame.
1485 if (cm->lf.filter_level && !cm->skip_loop_filter) {
1486 LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1;
1487 winterface->sync(&pbi->lf_worker);
1488 lf_data->start = lf_data->stop;
1489 lf_data->stop = cm->mi_rows;
1490 winterface->execute(&pbi->lf_worker);
1491 }
1492
1493 // Get last tile data.
1494 tile_data = pbi->tile_worker_data + tile_cols * tile_rows - 1;
1495
1496 if (pbi->frame_parallel_decode)
1497 vp9_frameworker_broadcast(pbi->cur_buf, INT_MAX);
1498 return vpx_reader_find_end(&tile_data->bit_reader);
1499 }
1500
1501 // On entry 'tile_data->data_end' points to the end of the input frame, on exit
1502 // it is updated to reflect the bitreader position of the final tile column if
1503 // present in the tile buffer group or NULL otherwise.
tile_worker_hook(TileWorkerData * const tile_data,VP9Decoder * const pbi)1504 static int tile_worker_hook(TileWorkerData *const tile_data,
1505 VP9Decoder *const pbi) {
1506 TileInfo *volatile tile = &tile_data->xd.tile;
1507 const int final_col = (1 << pbi->common.log2_tile_cols) - 1;
1508 const uint8_t *volatile bit_reader_end = NULL;
1509 volatile int n = tile_data->buf_start;
1510 tile_data->error_info.setjmp = 1;
1511
1512 if (setjmp(tile_data->error_info.jmp)) {
1513 tile_data->error_info.setjmp = 0;
1514 tile_data->xd.corrupted = 1;
1515 tile_data->data_end = NULL;
1516 return 0;
1517 }
1518
1519 tile_data->xd.corrupted = 0;
1520
1521 do {
1522 int mi_row, mi_col;
1523 const TileBuffer *const buf = pbi->tile_buffers + n;
1524 vp9_zero(tile_data->dqcoeff);
1525 vp9_tile_init(tile, &pbi->common, 0, buf->col);
1526 setup_token_decoder(buf->data, tile_data->data_end, buf->size,
1527 &tile_data->error_info, &tile_data->bit_reader,
1528 pbi->decrypt_cb, pbi->decrypt_state);
1529 vp9_init_macroblockd(&pbi->common, &tile_data->xd, tile_data->dqcoeff);
1530 // init resets xd.error_info
1531 tile_data->xd.error_info = &tile_data->error_info;
1532
1533 for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end;
1534 mi_row += MI_BLOCK_SIZE) {
1535 vp9_zero(tile_data->xd.left_context);
1536 vp9_zero(tile_data->xd.left_seg_context);
1537 for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
1538 mi_col += MI_BLOCK_SIZE) {
1539 decode_partition(tile_data, pbi, mi_row, mi_col, BLOCK_64X64, 4);
1540 }
1541 }
1542
1543 if (buf->col == final_col) {
1544 bit_reader_end = vpx_reader_find_end(&tile_data->bit_reader);
1545 }
1546 } while (!tile_data->xd.corrupted && ++n <= tile_data->buf_end);
1547
1548 tile_data->data_end = bit_reader_end;
1549 return !tile_data->xd.corrupted;
1550 }
1551
1552 // sorts in descending order
compare_tile_buffers(const void * a,const void * b)1553 static int compare_tile_buffers(const void *a, const void *b) {
1554 const TileBuffer *const buf1 = (const TileBuffer *)a;
1555 const TileBuffer *const buf2 = (const TileBuffer *)b;
1556 return (int)(buf2->size - buf1->size);
1557 }
1558
decode_tiles_mt(VP9Decoder * pbi,const uint8_t * data,const uint8_t * data_end)1559 static const uint8_t *decode_tiles_mt(VP9Decoder *pbi, const uint8_t *data,
1560 const uint8_t *data_end) {
1561 VP9_COMMON *const cm = &pbi->common;
1562 const VPxWorkerInterface *const winterface = vpx_get_worker_interface();
1563 const uint8_t *bit_reader_end = NULL;
1564 const int aligned_mi_cols = mi_cols_aligned_to_sb(cm->mi_cols);
1565 const int tile_cols = 1 << cm->log2_tile_cols;
1566 const int tile_rows = 1 << cm->log2_tile_rows;
1567 const int num_workers = VPXMIN(pbi->max_threads, tile_cols);
1568 int n;
1569
1570 assert(tile_cols <= (1 << 6));
1571 assert(tile_rows == 1);
1572 (void)tile_rows;
1573
1574 if (pbi->num_tile_workers == 0) {
1575 const int num_threads = pbi->max_threads;
1576 CHECK_MEM_ERROR(cm, pbi->tile_workers,
1577 vpx_malloc(num_threads * sizeof(*pbi->tile_workers)));
1578 for (n = 0; n < num_threads; ++n) {
1579 VPxWorker *const worker = &pbi->tile_workers[n];
1580 ++pbi->num_tile_workers;
1581
1582 winterface->init(worker);
1583 if (n < num_threads - 1 && !winterface->reset(worker)) {
1584 vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
1585 "Tile decoder thread creation failed");
1586 }
1587 }
1588 }
1589
1590 // Reset tile decoding hook
1591 for (n = 0; n < num_workers; ++n) {
1592 VPxWorker *const worker = &pbi->tile_workers[n];
1593 TileWorkerData *const tile_data =
1594 &pbi->tile_worker_data[n + pbi->total_tiles];
1595 winterface->sync(worker);
1596 tile_data->xd = pbi->mb;
1597 tile_data->xd.counts =
1598 cm->frame_parallel_decoding_mode ? NULL : &tile_data->counts;
1599 worker->hook = (VPxWorkerHook)tile_worker_hook;
1600 worker->data1 = tile_data;
1601 worker->data2 = pbi;
1602 }
1603
1604 // Note: this memset assumes above_context[0], [1] and [2]
1605 // are allocated as part of the same buffer.
1606 memset(cm->above_context, 0,
1607 sizeof(*cm->above_context) * MAX_MB_PLANE * 2 * aligned_mi_cols);
1608 memset(cm->above_seg_context, 0,
1609 sizeof(*cm->above_seg_context) * aligned_mi_cols);
1610
1611 vp9_reset_lfm(cm);
1612
1613 // Load tile data into tile_buffers
1614 get_tile_buffers(pbi, data, data_end, tile_cols, tile_rows,
1615 &pbi->tile_buffers);
1616
1617 // Sort the buffers based on size in descending order.
1618 qsort(pbi->tile_buffers, tile_cols, sizeof(pbi->tile_buffers[0]),
1619 compare_tile_buffers);
1620
1621 if (num_workers == tile_cols) {
1622 // Rearrange the tile buffers such that the largest, and
1623 // presumably the most difficult, tile will be decoded in the main thread.
1624 // This should help minimize the number of instances where the main thread
1625 // is waiting for a worker to complete.
1626 const TileBuffer largest = pbi->tile_buffers[0];
1627 memmove(pbi->tile_buffers, pbi->tile_buffers + 1,
1628 (tile_cols - 1) * sizeof(pbi->tile_buffers[0]));
1629 pbi->tile_buffers[tile_cols - 1] = largest;
1630 } else {
1631 int start = 0, end = tile_cols - 2;
1632 TileBuffer tmp;
1633
1634 // Interleave the tiles to distribute the load between threads, assuming a
1635 // larger tile implies it is more difficult to decode.
1636 while (start < end) {
1637 tmp = pbi->tile_buffers[start];
1638 pbi->tile_buffers[start] = pbi->tile_buffers[end];
1639 pbi->tile_buffers[end] = tmp;
1640 start += 2;
1641 end -= 2;
1642 }
1643 }
1644
1645 // Initialize thread frame counts.
1646 if (!cm->frame_parallel_decoding_mode) {
1647 for (n = 0; n < num_workers; ++n) {
1648 TileWorkerData *const tile_data =
1649 (TileWorkerData *)pbi->tile_workers[n].data1;
1650 vp9_zero(tile_data->counts);
1651 }
1652 }
1653
1654 {
1655 const int base = tile_cols / num_workers;
1656 const int remain = tile_cols % num_workers;
1657 int buf_start = 0;
1658
1659 for (n = 0; n < num_workers; ++n) {
1660 const int count = base + (remain + n) / num_workers;
1661 VPxWorker *const worker = &pbi->tile_workers[n];
1662 TileWorkerData *const tile_data = (TileWorkerData *)worker->data1;
1663
1664 tile_data->buf_start = buf_start;
1665 tile_data->buf_end = buf_start + count - 1;
1666 tile_data->data_end = data_end;
1667 buf_start += count;
1668
1669 worker->had_error = 0;
1670 if (n == num_workers - 1) {
1671 assert(tile_data->buf_end == tile_cols - 1);
1672 winterface->execute(worker);
1673 } else {
1674 winterface->launch(worker);
1675 }
1676 }
1677
1678 for (; n > 0; --n) {
1679 VPxWorker *const worker = &pbi->tile_workers[n - 1];
1680 TileWorkerData *const tile_data = (TileWorkerData *)worker->data1;
1681 // TODO(jzern): The tile may have specific error data associated with
1682 // its vpx_internal_error_info which could be propagated to the main info
1683 // in cm. Additionally once the threads have been synced and an error is
1684 // detected, there's no point in continuing to decode tiles.
1685 pbi->mb.corrupted |= !winterface->sync(worker);
1686 if (!bit_reader_end) bit_reader_end = tile_data->data_end;
1687 }
1688 }
1689
1690 // Accumulate thread frame counts.
1691 if (!cm->frame_parallel_decoding_mode) {
1692 for (n = 0; n < num_workers; ++n) {
1693 TileWorkerData *const tile_data =
1694 (TileWorkerData *)pbi->tile_workers[n].data1;
1695 vp9_accumulate_frame_counts(&cm->counts, &tile_data->counts, 1);
1696 }
1697 }
1698
1699 assert(bit_reader_end || pbi->mb.corrupted);
1700 return bit_reader_end;
1701 }
1702
error_handler(void * data)1703 static void error_handler(void *data) {
1704 VP9_COMMON *const cm = (VP9_COMMON *)data;
1705 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, "Truncated packet");
1706 }
1707
read_bitdepth_colorspace_sampling(VP9_COMMON * cm,struct vpx_read_bit_buffer * rb)1708 static void read_bitdepth_colorspace_sampling(VP9_COMMON *cm,
1709 struct vpx_read_bit_buffer *rb) {
1710 if (cm->profile >= PROFILE_2) {
1711 cm->bit_depth = vpx_rb_read_bit(rb) ? VPX_BITS_12 : VPX_BITS_10;
1712 #if CONFIG_VP9_HIGHBITDEPTH
1713 cm->use_highbitdepth = 1;
1714 #endif
1715 } else {
1716 cm->bit_depth = VPX_BITS_8;
1717 #if CONFIG_VP9_HIGHBITDEPTH
1718 cm->use_highbitdepth = 0;
1719 #endif
1720 }
1721 cm->color_space = vpx_rb_read_literal(rb, 3);
1722 if (cm->color_space != VPX_CS_SRGB) {
1723 cm->color_range = (vpx_color_range_t)vpx_rb_read_bit(rb);
1724 if (cm->profile == PROFILE_1 || cm->profile == PROFILE_3) {
1725 cm->subsampling_x = vpx_rb_read_bit(rb);
1726 cm->subsampling_y = vpx_rb_read_bit(rb);
1727 if (cm->subsampling_x == 1 && cm->subsampling_y == 1)
1728 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
1729 "4:2:0 color not supported in profile 1 or 3");
1730 if (vpx_rb_read_bit(rb))
1731 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
1732 "Reserved bit set");
1733 } else {
1734 cm->subsampling_y = cm->subsampling_x = 1;
1735 }
1736 } else {
1737 cm->color_range = VPX_CR_FULL_RANGE;
1738 if (cm->profile == PROFILE_1 || cm->profile == PROFILE_3) {
1739 // Note if colorspace is SRGB then 4:4:4 chroma sampling is assumed.
1740 // 4:2:2 or 4:4:0 chroma sampling is not allowed.
1741 cm->subsampling_y = cm->subsampling_x = 0;
1742 if (vpx_rb_read_bit(rb))
1743 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
1744 "Reserved bit set");
1745 } else {
1746 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
1747 "4:4:4 color not supported in profile 0 or 2");
1748 }
1749 }
1750 }
1751
read_uncompressed_header(VP9Decoder * pbi,struct vpx_read_bit_buffer * rb)1752 static size_t read_uncompressed_header(VP9Decoder *pbi,
1753 struct vpx_read_bit_buffer *rb) {
1754 VP9_COMMON *const cm = &pbi->common;
1755 BufferPool *const pool = cm->buffer_pool;
1756 RefCntBuffer *const frame_bufs = pool->frame_bufs;
1757 int i, mask, ref_index = 0;
1758 size_t sz;
1759
1760 cm->last_frame_type = cm->frame_type;
1761 cm->last_intra_only = cm->intra_only;
1762
1763 if (vpx_rb_read_literal(rb, 2) != VP9_FRAME_MARKER)
1764 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
1765 "Invalid frame marker");
1766
1767 cm->profile = vp9_read_profile(rb);
1768 #if CONFIG_VP9_HIGHBITDEPTH
1769 if (cm->profile >= MAX_PROFILES)
1770 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
1771 "Unsupported bitstream profile");
1772 #else
1773 if (cm->profile >= PROFILE_2)
1774 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
1775 "Unsupported bitstream profile");
1776 #endif
1777
1778 cm->show_existing_frame = vpx_rb_read_bit(rb);
1779 if (cm->show_existing_frame) {
1780 // Show an existing frame directly.
1781 const int frame_to_show = cm->ref_frame_map[vpx_rb_read_literal(rb, 3)];
1782 lock_buffer_pool(pool);
1783 if (frame_to_show < 0 || frame_bufs[frame_to_show].ref_count < 1) {
1784 unlock_buffer_pool(pool);
1785 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
1786 "Buffer %d does not contain a decoded frame",
1787 frame_to_show);
1788 }
1789
1790 ref_cnt_fb(frame_bufs, &cm->new_fb_idx, frame_to_show);
1791 unlock_buffer_pool(pool);
1792 pbi->refresh_frame_flags = 0;
1793 cm->lf.filter_level = 0;
1794 cm->show_frame = 1;
1795
1796 if (pbi->frame_parallel_decode) {
1797 for (i = 0; i < REF_FRAMES; ++i)
1798 cm->next_ref_frame_map[i] = cm->ref_frame_map[i];
1799 }
1800 return 0;
1801 }
1802
1803 cm->frame_type = (FRAME_TYPE)vpx_rb_read_bit(rb);
1804 cm->show_frame = vpx_rb_read_bit(rb);
1805 cm->error_resilient_mode = vpx_rb_read_bit(rb);
1806
1807 if (cm->frame_type == KEY_FRAME) {
1808 if (!vp9_read_sync_code(rb))
1809 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
1810 "Invalid frame sync code");
1811
1812 read_bitdepth_colorspace_sampling(cm, rb);
1813 pbi->refresh_frame_flags = (1 << REF_FRAMES) - 1;
1814
1815 for (i = 0; i < REFS_PER_FRAME; ++i) {
1816 cm->frame_refs[i].idx = INVALID_IDX;
1817 cm->frame_refs[i].buf = NULL;
1818 }
1819
1820 setup_frame_size(cm, rb);
1821 if (pbi->need_resync) {
1822 memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
1823 pbi->need_resync = 0;
1824 }
1825 } else {
1826 cm->intra_only = cm->show_frame ? 0 : vpx_rb_read_bit(rb);
1827
1828 cm->reset_frame_context =
1829 cm->error_resilient_mode ? 0 : vpx_rb_read_literal(rb, 2);
1830
1831 if (cm->intra_only) {
1832 if (!vp9_read_sync_code(rb))
1833 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
1834 "Invalid frame sync code");
1835 if (cm->profile > PROFILE_0) {
1836 read_bitdepth_colorspace_sampling(cm, rb);
1837 } else {
1838 // NOTE: The intra-only frame header does not include the specification
1839 // of either the color format or color sub-sampling in profile 0. VP9
1840 // specifies that the default color format should be YUV 4:2:0 in this
1841 // case (normative).
1842 cm->color_space = VPX_CS_BT_601;
1843 cm->color_range = VPX_CR_STUDIO_RANGE;
1844 cm->subsampling_y = cm->subsampling_x = 1;
1845 cm->bit_depth = VPX_BITS_8;
1846 #if CONFIG_VP9_HIGHBITDEPTH
1847 cm->use_highbitdepth = 0;
1848 #endif
1849 }
1850
1851 pbi->refresh_frame_flags = vpx_rb_read_literal(rb, REF_FRAMES);
1852 setup_frame_size(cm, rb);
1853 if (pbi->need_resync) {
1854 memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
1855 pbi->need_resync = 0;
1856 }
1857 } else if (pbi->need_resync != 1) { /* Skip if need resync */
1858 pbi->refresh_frame_flags = vpx_rb_read_literal(rb, REF_FRAMES);
1859 for (i = 0; i < REFS_PER_FRAME; ++i) {
1860 const int ref = vpx_rb_read_literal(rb, REF_FRAMES_LOG2);
1861 const int idx = cm->ref_frame_map[ref];
1862 RefBuffer *const ref_frame = &cm->frame_refs[i];
1863 ref_frame->idx = idx;
1864 ref_frame->buf = &frame_bufs[idx].buf;
1865 cm->ref_frame_sign_bias[LAST_FRAME + i] = vpx_rb_read_bit(rb);
1866 }
1867
1868 setup_frame_size_with_refs(cm, rb);
1869
1870 cm->allow_high_precision_mv = vpx_rb_read_bit(rb);
1871 cm->interp_filter = read_interp_filter(rb);
1872
1873 for (i = 0; i < REFS_PER_FRAME; ++i) {
1874 RefBuffer *const ref_buf = &cm->frame_refs[i];
1875 #if CONFIG_VP9_HIGHBITDEPTH
1876 vp9_setup_scale_factors_for_frame(
1877 &ref_buf->sf, ref_buf->buf->y_crop_width,
1878 ref_buf->buf->y_crop_height, cm->width, cm->height,
1879 cm->use_highbitdepth);
1880 #else
1881 vp9_setup_scale_factors_for_frame(
1882 &ref_buf->sf, ref_buf->buf->y_crop_width,
1883 ref_buf->buf->y_crop_height, cm->width, cm->height);
1884 #endif
1885 }
1886 }
1887 }
1888 #if CONFIG_VP9_HIGHBITDEPTH
1889 get_frame_new_buffer(cm)->bit_depth = cm->bit_depth;
1890 #endif
1891 get_frame_new_buffer(cm)->color_space = cm->color_space;
1892 get_frame_new_buffer(cm)->color_range = cm->color_range;
1893 get_frame_new_buffer(cm)->render_width = cm->render_width;
1894 get_frame_new_buffer(cm)->render_height = cm->render_height;
1895
1896 if (pbi->need_resync) {
1897 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
1898 "Keyframe / intra-only frame required to reset decoder"
1899 " state");
1900 }
1901
1902 if (!cm->error_resilient_mode) {
1903 cm->refresh_frame_context = vpx_rb_read_bit(rb);
1904 cm->frame_parallel_decoding_mode = vpx_rb_read_bit(rb);
1905 if (!cm->frame_parallel_decoding_mode) vp9_zero(cm->counts);
1906 } else {
1907 cm->refresh_frame_context = 0;
1908 cm->frame_parallel_decoding_mode = 1;
1909 }
1910
1911 // This flag will be overridden by the call to vp9_setup_past_independence
1912 // below, forcing the use of context 0 for those frame types.
1913 cm->frame_context_idx = vpx_rb_read_literal(rb, FRAME_CONTEXTS_LOG2);
1914
1915 // Generate next_ref_frame_map.
1916 lock_buffer_pool(pool);
1917 for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
1918 if (mask & 1) {
1919 cm->next_ref_frame_map[ref_index] = cm->new_fb_idx;
1920 ++frame_bufs[cm->new_fb_idx].ref_count;
1921 } else {
1922 cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index];
1923 }
1924 // Current thread holds the reference frame.
1925 if (cm->ref_frame_map[ref_index] >= 0)
1926 ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count;
1927 ++ref_index;
1928 }
1929
1930 for (; ref_index < REF_FRAMES; ++ref_index) {
1931 cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index];
1932 // Current thread holds the reference frame.
1933 if (cm->ref_frame_map[ref_index] >= 0)
1934 ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count;
1935 }
1936 unlock_buffer_pool(pool);
1937 pbi->hold_ref_buf = 1;
1938
1939 if (frame_is_intra_only(cm) || cm->error_resilient_mode)
1940 vp9_setup_past_independence(cm);
1941
1942 setup_loopfilter(&cm->lf, rb);
1943 setup_quantization(cm, &pbi->mb, rb);
1944 setup_segmentation(&cm->seg, rb);
1945 setup_segmentation_dequant(cm);
1946
1947 setup_tile_info(cm, rb);
1948 sz = vpx_rb_read_literal(rb, 16);
1949
1950 if (sz == 0)
1951 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
1952 "Invalid header size");
1953
1954 return sz;
1955 }
1956
read_compressed_header(VP9Decoder * pbi,const uint8_t * data,size_t partition_size)1957 static int read_compressed_header(VP9Decoder *pbi, const uint8_t *data,
1958 size_t partition_size) {
1959 VP9_COMMON *const cm = &pbi->common;
1960 MACROBLOCKD *const xd = &pbi->mb;
1961 FRAME_CONTEXT *const fc = cm->fc;
1962 vpx_reader r;
1963 int k;
1964
1965 if (vpx_reader_init(&r, data, partition_size, pbi->decrypt_cb,
1966 pbi->decrypt_state))
1967 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
1968 "Failed to allocate bool decoder 0");
1969
1970 cm->tx_mode = xd->lossless ? ONLY_4X4 : read_tx_mode(&r);
1971 if (cm->tx_mode == TX_MODE_SELECT) read_tx_mode_probs(&fc->tx_probs, &r);
1972 read_coef_probs(fc, cm->tx_mode, &r);
1973
1974 for (k = 0; k < SKIP_CONTEXTS; ++k)
1975 vp9_diff_update_prob(&r, &fc->skip_probs[k]);
1976
1977 if (!frame_is_intra_only(cm)) {
1978 nmv_context *const nmvc = &fc->nmvc;
1979 int i, j;
1980
1981 read_inter_mode_probs(fc, &r);
1982
1983 if (cm->interp_filter == SWITCHABLE) read_switchable_interp_probs(fc, &r);
1984
1985 for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
1986 vp9_diff_update_prob(&r, &fc->intra_inter_prob[i]);
1987
1988 cm->reference_mode = read_frame_reference_mode(cm, &r);
1989 if (cm->reference_mode != SINGLE_REFERENCE)
1990 setup_compound_reference_mode(cm);
1991 read_frame_reference_mode_probs(cm, &r);
1992
1993 for (j = 0; j < BLOCK_SIZE_GROUPS; j++)
1994 for (i = 0; i < INTRA_MODES - 1; ++i)
1995 vp9_diff_update_prob(&r, &fc->y_mode_prob[j][i]);
1996
1997 for (j = 0; j < PARTITION_CONTEXTS; ++j)
1998 for (i = 0; i < PARTITION_TYPES - 1; ++i)
1999 vp9_diff_update_prob(&r, &fc->partition_prob[j][i]);
2000
2001 read_mv_probs(nmvc, cm->allow_high_precision_mv, &r);
2002 }
2003
2004 return vpx_reader_has_error(&r);
2005 }
2006
init_read_bit_buffer(VP9Decoder * pbi,struct vpx_read_bit_buffer * rb,const uint8_t * data,const uint8_t * data_end,uint8_t clear_data[MAX_VP9_HEADER_SIZE])2007 static struct vpx_read_bit_buffer *init_read_bit_buffer(
2008 VP9Decoder *pbi, struct vpx_read_bit_buffer *rb, const uint8_t *data,
2009 const uint8_t *data_end, uint8_t clear_data[MAX_VP9_HEADER_SIZE]) {
2010 rb->bit_offset = 0;
2011 rb->error_handler = error_handler;
2012 rb->error_handler_data = &pbi->common;
2013 if (pbi->decrypt_cb) {
2014 const int n = (int)VPXMIN(MAX_VP9_HEADER_SIZE, data_end - data);
2015 pbi->decrypt_cb(pbi->decrypt_state, data, clear_data, n);
2016 rb->bit_buffer = clear_data;
2017 rb->bit_buffer_end = clear_data + n;
2018 } else {
2019 rb->bit_buffer = data;
2020 rb->bit_buffer_end = data_end;
2021 }
2022 return rb;
2023 }
2024
2025 //------------------------------------------------------------------------------
2026
vp9_read_sync_code(struct vpx_read_bit_buffer * const rb)2027 int vp9_read_sync_code(struct vpx_read_bit_buffer *const rb) {
2028 return vpx_rb_read_literal(rb, 8) == VP9_SYNC_CODE_0 &&
2029 vpx_rb_read_literal(rb, 8) == VP9_SYNC_CODE_1 &&
2030 vpx_rb_read_literal(rb, 8) == VP9_SYNC_CODE_2;
2031 }
2032
vp9_read_frame_size(struct vpx_read_bit_buffer * rb,int * width,int * height)2033 void vp9_read_frame_size(struct vpx_read_bit_buffer *rb, int *width,
2034 int *height) {
2035 *width = vpx_rb_read_literal(rb, 16) + 1;
2036 *height = vpx_rb_read_literal(rb, 16) + 1;
2037 }
2038
vp9_read_profile(struct vpx_read_bit_buffer * rb)2039 BITSTREAM_PROFILE vp9_read_profile(struct vpx_read_bit_buffer *rb) {
2040 int profile = vpx_rb_read_bit(rb);
2041 profile |= vpx_rb_read_bit(rb) << 1;
2042 if (profile > 2) profile += vpx_rb_read_bit(rb);
2043 return (BITSTREAM_PROFILE)profile;
2044 }
2045
vp9_decode_frame(VP9Decoder * pbi,const uint8_t * data,const uint8_t * data_end,const uint8_t ** p_data_end)2046 void vp9_decode_frame(VP9Decoder *pbi, const uint8_t *data,
2047 const uint8_t *data_end, const uint8_t **p_data_end) {
2048 VP9_COMMON *const cm = &pbi->common;
2049 MACROBLOCKD *const xd = &pbi->mb;
2050 struct vpx_read_bit_buffer rb;
2051 int context_updated = 0;
2052 uint8_t clear_data[MAX_VP9_HEADER_SIZE];
2053 const size_t first_partition_size = read_uncompressed_header(
2054 pbi, init_read_bit_buffer(pbi, &rb, data, data_end, clear_data));
2055 const int tile_rows = 1 << cm->log2_tile_rows;
2056 const int tile_cols = 1 << cm->log2_tile_cols;
2057 YV12_BUFFER_CONFIG *const new_fb = get_frame_new_buffer(cm);
2058 xd->cur_buf = new_fb;
2059
2060 if (!first_partition_size) {
2061 // showing a frame directly
2062 *p_data_end = data + (cm->profile <= PROFILE_2 ? 1 : 2);
2063 return;
2064 }
2065
2066 data += vpx_rb_bytes_read(&rb);
2067 if (!read_is_valid(data, first_partition_size, data_end))
2068 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
2069 "Truncated packet or corrupt header length");
2070
2071 cm->use_prev_frame_mvs =
2072 !cm->error_resilient_mode && cm->width == cm->last_width &&
2073 cm->height == cm->last_height && !cm->last_intra_only &&
2074 cm->last_show_frame && (cm->last_frame_type != KEY_FRAME);
2075
2076 vp9_setup_block_planes(xd, cm->subsampling_x, cm->subsampling_y);
2077
2078 *cm->fc = cm->frame_contexts[cm->frame_context_idx];
2079 if (!cm->fc->initialized)
2080 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
2081 "Uninitialized entropy context.");
2082
2083 xd->corrupted = 0;
2084 new_fb->corrupted = read_compressed_header(pbi, data, first_partition_size);
2085 if (new_fb->corrupted)
2086 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
2087 "Decode failed. Frame data header is corrupted.");
2088
2089 if (cm->lf.filter_level && !cm->skip_loop_filter) {
2090 vp9_loop_filter_frame_init(cm, cm->lf.filter_level);
2091 }
2092
2093 // If encoded in frame parallel mode, frame context is ready after decoding
2094 // the frame header.
2095 if (pbi->frame_parallel_decode && cm->frame_parallel_decoding_mode) {
2096 VPxWorker *const worker = pbi->frame_worker_owner;
2097 FrameWorkerData *const frame_worker_data = worker->data1;
2098 if (cm->refresh_frame_context) {
2099 context_updated = 1;
2100 cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
2101 }
2102 vp9_frameworker_lock_stats(worker);
2103 pbi->cur_buf->row = -1;
2104 pbi->cur_buf->col = -1;
2105 frame_worker_data->frame_context_ready = 1;
2106 // Signal the main thread that context is ready.
2107 vp9_frameworker_signal_stats(worker);
2108 vp9_frameworker_unlock_stats(worker);
2109 }
2110
2111 if (pbi->tile_worker_data == NULL ||
2112 (tile_cols * tile_rows) != pbi->total_tiles) {
2113 const int num_tile_workers =
2114 tile_cols * tile_rows + ((pbi->max_threads > 1) ? pbi->max_threads : 0);
2115 const size_t twd_size = num_tile_workers * sizeof(*pbi->tile_worker_data);
2116 // Ensure tile data offsets will be properly aligned. This may fail on
2117 // platforms without DECLARE_ALIGNED().
2118 assert((sizeof(*pbi->tile_worker_data) % 16) == 0);
2119 vpx_free(pbi->tile_worker_data);
2120 CHECK_MEM_ERROR(cm, pbi->tile_worker_data, vpx_memalign(32, twd_size));
2121 pbi->total_tiles = tile_rows * tile_cols;
2122 }
2123
2124 if (pbi->max_threads > 1 && tile_rows == 1 && tile_cols > 1) {
2125 // Multi-threaded tile decoder
2126 *p_data_end = decode_tiles_mt(pbi, data + first_partition_size, data_end);
2127 if (!xd->corrupted) {
2128 if (!cm->skip_loop_filter) {
2129 // If multiple threads are used to decode tiles, then we use those
2130 // threads to do parallel loopfiltering.
2131 vp9_loop_filter_frame_mt(new_fb, cm, pbi->mb.plane, cm->lf.filter_level,
2132 0, 0, pbi->tile_workers, pbi->num_tile_workers,
2133 &pbi->lf_row_sync);
2134 }
2135 } else {
2136 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
2137 "Decode failed. Frame data is corrupted.");
2138 }
2139 } else {
2140 *p_data_end = decode_tiles(pbi, data + first_partition_size, data_end);
2141 }
2142
2143 if (!xd->corrupted) {
2144 if (!cm->error_resilient_mode && !cm->frame_parallel_decoding_mode) {
2145 vp9_adapt_coef_probs(cm);
2146
2147 if (!frame_is_intra_only(cm)) {
2148 vp9_adapt_mode_probs(cm);
2149 vp9_adapt_mv_probs(cm, cm->allow_high_precision_mv);
2150 }
2151 }
2152 } else {
2153 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
2154 "Decode failed. Frame data is corrupted.");
2155 }
2156
2157 // Non frame parallel update frame context here.
2158 if (cm->refresh_frame_context && !context_updated)
2159 cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
2160 }
2161