2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #define UNCHECKED_BITSTREAM_READER 1
30 #include "libavutil/imgutils.h"
31 #include "libavutil/opt.h"
34 #include "cabac_functions.h"
37 #include "mpegvideo.h"
40 #include "h264_mvpred.h"
43 #include "rectangle.h"
45 #include "vdpau_internal.h"
46 #include "libavutil/avassert.h"
51 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
53 static const uint8_t rem6[QP_MAX_NUM + 1] = {
54 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
55 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
56 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
57 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
61 static const uint8_t div6[QP_MAX_NUM + 1] = {
62 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
63 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
64 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
65 10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
69 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
77 int avpriv_h264_has_num_reorder_frames(AVCodecContext *avctx)
79 H264Context *h = avctx->priv_data;
80 return h ? h->sps.num_reorder_frames : 0;
84 * Check if the top & left blocks are available if needed and
85 * change the dc mode so it only uses the available blocks.
87 int ff_h264_check_intra4x4_pred_mode(H264Context *h)
89 MpegEncContext *const s = &h->s;
90 static const int8_t top[12] = {
91 -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0
93 static const int8_t left[12] = {
94 0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED
98 if (!(h->top_samples_available & 0x8000)) {
99 for (i = 0; i < 4; i++) {
100 int status = top[h->intra4x4_pred_mode_cache[scan8[0] + i]];
102 av_log(h->s.avctx, AV_LOG_ERROR,
103 "top block unavailable for requested intra4x4 mode %d at %d %d\n",
104 status, s->mb_x, s->mb_y);
107 h->intra4x4_pred_mode_cache[scan8[0] + i] = status;
112 if ((h->left_samples_available & 0x8888) != 0x8888) {
113 static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 };
114 for (i = 0; i < 4; i++)
115 if (!(h->left_samples_available & mask[i])) {
116 int status = left[h->intra4x4_pred_mode_cache[scan8[0] + 8 * i]];
118 av_log(h->s.avctx, AV_LOG_ERROR,
119 "left block unavailable for requested intra4x4 mode %d at %d %d\n",
120 status, s->mb_x, s->mb_y);
123 h->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
129 } // FIXME cleanup like ff_h264_check_intra_pred_mode
132 * Check if the top & left blocks are available if needed and
133 * change the dc mode so it only uses the available blocks.
135 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma)
137 MpegEncContext *const s = &h->s;
138 static const int8_t top[7] = { LEFT_DC_PRED8x8, 1, -1, -1 };
139 static const int8_t left[7] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 };
142 av_log(h->s.avctx, AV_LOG_ERROR,
143 "out of range intra chroma pred mode at %d %d\n",
148 if (!(h->top_samples_available & 0x8000)) {
151 av_log(h->s.avctx, AV_LOG_ERROR,
152 "top block unavailable for requested intra mode at %d %d\n",
158 if ((h->left_samples_available & 0x8080) != 0x8080) {
160 if (is_chroma && (h->left_samples_available & 0x8080)) {
161 // mad cow disease mode, aka MBAFF + constrained_intra_pred
162 mode = ALZHEIMER_DC_L0T_PRED8x8 +
163 (!(h->left_samples_available & 0x8000)) +
164 2 * (mode == DC_128_PRED8x8);
167 av_log(h->s.avctx, AV_LOG_ERROR,
168 "left block unavailable for requested intra mode at %d %d\n",
177 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
178 int *dst_length, int *consumed, int length)
184 // src[0]&0x80; // forbidden bit
185 h->nal_ref_idc = src[0] >> 5;
186 h->nal_unit_type = src[0] & 0x1F;
191 #define STARTCODE_TEST \
192 if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
193 if (src[i + 2] != 3) { \
194 /* startcode, so we must be past the end */ \
199 #if HAVE_FAST_UNALIGNED
200 #define FIND_FIRST_ZERO \
201 if (i > 0 && !src[i]) \
206 for (i = 0; i + 1 < length; i += 9) {
207 if (!((~AV_RN64A(src + i) &
208 (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
209 0x8000800080008080ULL))
216 for (i = 0; i + 1 < length; i += 5) {
217 if (!((~AV_RN32A(src + i) &
218 (AV_RN32A(src + i) - 0x01000101U)) &
227 for (i = 0; i + 1 < length; i += 2) {
230 if (i > 0 && src[i - 1] == 0)
236 // use second escape buffer for inter data
237 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0;
239 si = h->rbsp_buffer_size[bufidx];
240 av_fast_padded_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+MAX_MBPAIR_SIZE);
241 dst = h->rbsp_buffer[bufidx];
246 if(i>=length-1){ //no escaped 0
248 *consumed= length+1; //+1 for the header
249 if(h->s.avctx->flags2 & CODEC_FLAG2_FAST){
252 memcpy(dst, src, length);
257 // printf("decoding esc\n");
260 while (si + 2 < length) {
261 // remove escapes (very rare 1:2^22)
262 if (src[si + 2] > 3) {
263 dst[di++] = src[si++];
264 dst[di++] = src[si++];
265 } else if (src[si] == 0 && src[si + 1] == 0) {
266 if (src[si + 2] == 3) { // escape
271 } else // next start code
275 dst[di++] = src[si++];
278 dst[di++] = src[si++];
281 memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
284 *consumed = si + 1; // +1 for the header
285 /* FIXME store exact number of bits in the getbitcontext
286 * (it is needed for decoding) */
291 * Identify the exact end of the bitstream
292 * @return the length of the trailing, or 0 if damaged
294 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
299 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
301 for (r = 1; r < 9; r++) {
309 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n,
310 int height, int y_offset, int list)
312 int raw_my = h->mv_cache[list][scan8[n]][1];
313 int filter_height = (raw_my & 3) ? 2 : 0;
314 int full_my = (raw_my >> 2) + y_offset;
315 int top = full_my - filter_height;
316 int bottom = full_my + filter_height + height;
318 return FFMAX(abs(top), bottom);
321 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n,
322 int height, int y_offset, int list0,
323 int list1, int *nrefs)
325 MpegEncContext *const s = &h->s;
328 y_offset += 16 * (s->mb_y >> MB_FIELD);
331 int ref_n = h->ref_cache[0][scan8[n]];
332 Picture *ref = &h->ref_list[0][ref_n];
334 // Error resilience puts the current picture in the ref list.
335 // Don't try to wait on these as it will cause a deadlock.
336 // Fields can wait on each other, though.
337 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
338 (ref->f.reference & 3) != s->picture_structure) {
339 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
340 if (refs[0][ref_n] < 0)
342 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
347 int ref_n = h->ref_cache[1][scan8[n]];
348 Picture *ref = &h->ref_list[1][ref_n];
350 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
351 (ref->f.reference & 3) != s->picture_structure) {
352 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
353 if (refs[1][ref_n] < 0)
355 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
361 * Wait until all reference frames are available for MC operations.
363 * @param h the H264 context
365 static void await_references(H264Context *h)
367 MpegEncContext *const s = &h->s;
368 const int mb_xy = h->mb_xy;
369 const int mb_type = s->current_picture.f.mb_type[mb_xy];
371 int nrefs[2] = { 0 };
374 memset(refs, -1, sizeof(refs));
376 if (IS_16X16(mb_type)) {
377 get_lowest_part_y(h, refs, 0, 16, 0,
378 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
379 } else if (IS_16X8(mb_type)) {
380 get_lowest_part_y(h, refs, 0, 8, 0,
381 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
382 get_lowest_part_y(h, refs, 8, 8, 8,
383 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
384 } else if (IS_8X16(mb_type)) {
385 get_lowest_part_y(h, refs, 0, 16, 0,
386 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
387 get_lowest_part_y(h, refs, 4, 16, 0,
388 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
392 assert(IS_8X8(mb_type));
394 for (i = 0; i < 4; i++) {
395 const int sub_mb_type = h->sub_mb_type[i];
397 int y_offset = (i & 2) << 2;
399 if (IS_SUB_8X8(sub_mb_type)) {
400 get_lowest_part_y(h, refs, n, 8, y_offset,
401 IS_DIR(sub_mb_type, 0, 0),
402 IS_DIR(sub_mb_type, 0, 1),
404 } else if (IS_SUB_8X4(sub_mb_type)) {
405 get_lowest_part_y(h, refs, n, 4, y_offset,
406 IS_DIR(sub_mb_type, 0, 0),
407 IS_DIR(sub_mb_type, 0, 1),
409 get_lowest_part_y(h, refs, n + 2, 4, y_offset + 4,
410 IS_DIR(sub_mb_type, 0, 0),
411 IS_DIR(sub_mb_type, 0, 1),
413 } else if (IS_SUB_4X8(sub_mb_type)) {
414 get_lowest_part_y(h, refs, n, 8, y_offset,
415 IS_DIR(sub_mb_type, 0, 0),
416 IS_DIR(sub_mb_type, 0, 1),
418 get_lowest_part_y(h, refs, n + 1, 8, y_offset,
419 IS_DIR(sub_mb_type, 0, 0),
420 IS_DIR(sub_mb_type, 0, 1),
424 assert(IS_SUB_4X4(sub_mb_type));
425 for (j = 0; j < 4; j++) {
426 int sub_y_offset = y_offset + 2 * (j & 2);
427 get_lowest_part_y(h, refs, n + j, 4, sub_y_offset,
428 IS_DIR(sub_mb_type, 0, 0),
429 IS_DIR(sub_mb_type, 0, 1),
436 for (list = h->list_count - 1; list >= 0; list--)
437 for (ref = 0; ref < 48 && nrefs[list]; ref++) {
438 int row = refs[list][ref];
440 Picture *ref_pic = &h->ref_list[list][ref];
441 int ref_field = ref_pic->f.reference - 1;
442 int ref_field_picture = ref_pic->field_picture;
443 int pic_height = 16 * s->mb_height >> ref_field_picture;
448 if (!FIELD_PICTURE && ref_field_picture) { // frame referencing two fields
449 ff_thread_await_progress(&ref_pic->f,
450 FFMIN((row >> 1) - !(row & 1),
453 ff_thread_await_progress(&ref_pic->f,
454 FFMIN((row >> 1), pic_height - 1),
456 } else if (FIELD_PICTURE && !ref_field_picture) { // field referencing one field of a frame
457 ff_thread_await_progress(&ref_pic->f,
458 FFMIN(row * 2 + ref_field,
461 } else if (FIELD_PICTURE) {
462 ff_thread_await_progress(&ref_pic->f,
463 FFMIN(row, pic_height - 1),
466 ff_thread_await_progress(&ref_pic->f,
467 FFMIN(row, pic_height - 1),
474 static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
475 int n, int square, int height,
477 uint8_t *dest_y, uint8_t *dest_cb,
479 int src_x_offset, int src_y_offset,
480 qpel_mc_func *qpix_op,
481 h264_chroma_mc_func chroma_op,
482 int pixel_shift, int chroma_idc)
484 MpegEncContext *const s = &h->s;
485 const int mx = h->mv_cache[list][scan8[n]][0] + src_x_offset * 8;
486 int my = h->mv_cache[list][scan8[n]][1] + src_y_offset * 8;
487 const int luma_xy = (mx & 3) + ((my & 3) << 2);
488 int offset = ((mx >> 2) << pixel_shift) + (my >> 2) * h->mb_linesize;
489 uint8_t *src_y = pic->f.data[0] + offset;
490 uint8_t *src_cb, *src_cr;
491 int extra_width = h->emu_edge_width;
492 int extra_height = h->emu_edge_height;
494 const int full_mx = mx >> 2;
495 const int full_my = my >> 2;
496 const int pic_width = 16 * s->mb_width;
497 const int pic_height = 16 * s->mb_height >> MB_FIELD;
505 if (full_mx < 0 - extra_width ||
506 full_my < 0 - extra_height ||
507 full_mx + 16 /*FIXME*/ > pic_width + extra_width ||
508 full_my + 16 /*FIXME*/ > pic_height + extra_height) {
509 s->dsp.emulated_edge_mc(s->edge_emu_buffer,
510 src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
512 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
513 full_my - 2, pic_width, pic_height);
514 src_y = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
518 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); // FIXME try variable height perhaps?
520 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
522 if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
525 if (chroma_idc == 3 /* yuv444 */) {
526 src_cb = pic->f.data[1] + offset;
528 s->dsp.emulated_edge_mc(s->edge_emu_buffer,
529 src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
531 16 + 5, 16 + 5 /*FIXME*/,
532 full_mx - 2, full_my - 2,
533 pic_width, pic_height);
534 src_cb = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
536 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); // FIXME try variable height perhaps?
538 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
540 src_cr = pic->f.data[2] + offset;
542 s->dsp.emulated_edge_mc(s->edge_emu_buffer,
543 src_cr - (2 << pixel_shift) - 2 * h->mb_linesize,
545 16 + 5, 16 + 5 /*FIXME*/,
546 full_mx - 2, full_my - 2,
547 pic_width, pic_height);
548 src_cr = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
550 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); // FIXME try variable height perhaps?
552 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
556 ysh = 3 - (chroma_idc == 2 /* yuv422 */);
557 if (chroma_idc == 1 /* yuv420 */ && MB_FIELD) {
558 // chroma offset when predicting from a field of opposite parity
559 my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
560 emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1);
563 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) +
564 (my >> ysh) * h->mb_uvlinesize;
565 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) +
566 (my >> ysh) * h->mb_uvlinesize;
569 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
570 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
571 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
572 src_cb = s->edge_emu_buffer;
574 chroma_op(dest_cb, src_cb, h->mb_uvlinesize,
575 height >> (chroma_idc == 1 /* yuv420 */),
576 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
579 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
580 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
581 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
582 src_cr = s->edge_emu_buffer;
584 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
585 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
588 static av_always_inline void mc_part_std(H264Context *h, int n, int square,
589 int height, int delta,
590 uint8_t *dest_y, uint8_t *dest_cb,
592 int x_offset, int y_offset,
593 qpel_mc_func *qpix_put,
594 h264_chroma_mc_func chroma_put,
595 qpel_mc_func *qpix_avg,
596 h264_chroma_mc_func chroma_avg,
597 int list0, int list1,
598 int pixel_shift, int chroma_idc)
600 MpegEncContext *const s = &h->s;
601 qpel_mc_func *qpix_op = qpix_put;
602 h264_chroma_mc_func chroma_op = chroma_put;
604 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
605 if (chroma_idc == 3 /* yuv444 */) {
606 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
607 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
608 } else if (chroma_idc == 2 /* yuv422 */) {
609 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
610 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
611 } else { /* yuv420 */
612 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
613 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
615 x_offset += 8 * s->mb_x;
616 y_offset += 8 * (s->mb_y >> MB_FIELD);
619 Picture *ref = &h->ref_list[0][h->ref_cache[0][scan8[n]]];
620 mc_dir_part(h, ref, n, square, height, delta, 0,
621 dest_y, dest_cb, dest_cr, x_offset, y_offset,
622 qpix_op, chroma_op, pixel_shift, chroma_idc);
625 chroma_op = chroma_avg;
629 Picture *ref = &h->ref_list[1][h->ref_cache[1][scan8[n]]];
630 mc_dir_part(h, ref, n, square, height, delta, 1,
631 dest_y, dest_cb, dest_cr, x_offset, y_offset,
632 qpix_op, chroma_op, pixel_shift, chroma_idc);
636 static av_always_inline void mc_part_weighted(H264Context *h, int n, int square,
637 int height, int delta,
638 uint8_t *dest_y, uint8_t *dest_cb,
640 int x_offset, int y_offset,
641 qpel_mc_func *qpix_put,
642 h264_chroma_mc_func chroma_put,
643 h264_weight_func luma_weight_op,
644 h264_weight_func chroma_weight_op,
645 h264_biweight_func luma_weight_avg,
646 h264_biweight_func chroma_weight_avg,
647 int list0, int list1,
648 int pixel_shift, int chroma_idc)
650 MpegEncContext *const s = &h->s;
653 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
654 if (chroma_idc == 3 /* yuv444 */) {
655 chroma_height = height;
656 chroma_weight_avg = luma_weight_avg;
657 chroma_weight_op = luma_weight_op;
658 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
659 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
660 } else if (chroma_idc == 2 /* yuv422 */) {
661 chroma_height = height;
662 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
663 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
664 } else { /* yuv420 */
665 chroma_height = height >> 1;
666 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
667 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
669 x_offset += 8 * s->mb_x;
670 y_offset += 8 * (s->mb_y >> MB_FIELD);
672 if (list0 && list1) {
673 /* don't optimize for luma-only case, since B-frames usually
674 * use implicit weights => chroma too. */
675 uint8_t *tmp_cb = s->obmc_scratchpad;
676 uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
677 uint8_t *tmp_y = s->obmc_scratchpad + 16 * h->mb_uvlinesize;
678 int refn0 = h->ref_cache[0][scan8[n]];
679 int refn1 = h->ref_cache[1][scan8[n]];
681 mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
682 dest_y, dest_cb, dest_cr,
683 x_offset, y_offset, qpix_put, chroma_put,
684 pixel_shift, chroma_idc);
685 mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
686 tmp_y, tmp_cb, tmp_cr,
687 x_offset, y_offset, qpix_put, chroma_put,
688 pixel_shift, chroma_idc);
690 if (h->use_weight == 2) {
691 int weight0 = h->implicit_weight[refn0][refn1][s->mb_y & 1];
692 int weight1 = 64 - weight0;
693 luma_weight_avg(dest_y, tmp_y, h->mb_linesize,
694 height, 5, weight0, weight1, 0);
695 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
696 chroma_height, 5, weight0, weight1, 0);
697 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
698 chroma_height, 5, weight0, weight1, 0);
700 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height,
701 h->luma_log2_weight_denom,
702 h->luma_weight[refn0][0][0],
703 h->luma_weight[refn1][1][0],
704 h->luma_weight[refn0][0][1] +
705 h->luma_weight[refn1][1][1]);
706 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height,
707 h->chroma_log2_weight_denom,
708 h->chroma_weight[refn0][0][0][0],
709 h->chroma_weight[refn1][1][0][0],
710 h->chroma_weight[refn0][0][0][1] +
711 h->chroma_weight[refn1][1][0][1]);
712 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height,
713 h->chroma_log2_weight_denom,
714 h->chroma_weight[refn0][0][1][0],
715 h->chroma_weight[refn1][1][1][0],
716 h->chroma_weight[refn0][0][1][1] +
717 h->chroma_weight[refn1][1][1][1]);
720 int list = list1 ? 1 : 0;
721 int refn = h->ref_cache[list][scan8[n]];
722 Picture *ref = &h->ref_list[list][refn];
723 mc_dir_part(h, ref, n, square, height, delta, list,
724 dest_y, dest_cb, dest_cr, x_offset, y_offset,
725 qpix_put, chroma_put, pixel_shift, chroma_idc);
727 luma_weight_op(dest_y, h->mb_linesize, height,
728 h->luma_log2_weight_denom,
729 h->luma_weight[refn][list][0],
730 h->luma_weight[refn][list][1]);
731 if (h->use_weight_chroma) {
732 chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height,
733 h->chroma_log2_weight_denom,
734 h->chroma_weight[refn][list][0][0],
735 h->chroma_weight[refn][list][0][1]);
736 chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height,
737 h->chroma_log2_weight_denom,
738 h->chroma_weight[refn][list][1][0],
739 h->chroma_weight[refn][list][1][1]);
744 static av_always_inline void prefetch_motion(H264Context *h, int list,
745 int pixel_shift, int chroma_idc)
747 /* fetch pixels for estimated mv 4 macroblocks ahead
748 * optimized for 64byte cache lines */
749 MpegEncContext *const s = &h->s;
750 const int refn = h->ref_cache[list][scan8[0]];
752 const int mx = (h->mv_cache[list][scan8[0]][0] >> 2) + 16 * s->mb_x + 8;
753 const int my = (h->mv_cache[list][scan8[0]][1] >> 2) + 16 * s->mb_y;
754 uint8_t **src = h->ref_list[list][refn].f.data;
755 int off = (mx << pixel_shift) +
756 (my + (s->mb_x & 3) * 4) * h->mb_linesize +
758 s->dsp.prefetch(src[0] + off, s->linesize, 4);
759 if (chroma_idc == 3 /* yuv444 */) {
760 s->dsp.prefetch(src[1] + off, s->linesize, 4);
761 s->dsp.prefetch(src[2] + off, s->linesize, 4);
763 off= (((mx>>1)+64)<<pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize;
764 s->dsp.prefetch(src[1] + off, src[2] - src[1], 2);
769 static void free_tables(H264Context *h, int free_rbsp)
774 av_freep(&h->intra4x4_pred_mode);
775 av_freep(&h->chroma_pred_mode_table);
776 av_freep(&h->cbp_table);
777 av_freep(&h->mvd_table[0]);
778 av_freep(&h->mvd_table[1]);
779 av_freep(&h->direct_table);
780 av_freep(&h->non_zero_count);
781 av_freep(&h->slice_table_base);
782 h->slice_table = NULL;
783 av_freep(&h->list_counts);
785 av_freep(&h->mb2b_xy);
786 av_freep(&h->mb2br_xy);
788 for (i = 0; i < MAX_THREADS; i++) {
789 hx = h->thread_context[i];
792 av_freep(&hx->top_borders[1]);
793 av_freep(&hx->top_borders[0]);
794 av_freep(&hx->s.obmc_scratchpad);
796 av_freep(&hx->rbsp_buffer[1]);
797 av_freep(&hx->rbsp_buffer[0]);
798 hx->rbsp_buffer_size[0] = 0;
799 hx->rbsp_buffer_size[1] = 0;
802 av_freep(&h->thread_context[i]);
806 static void init_dequant8_coeff_table(H264Context *h)
809 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
811 for (i = 0; i < 6; i++) {
812 h->dequant8_coeff[i] = h->dequant8_buffer[i];
813 for (j = 0; j < i; j++)
814 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
815 64 * sizeof(uint8_t))) {
816 h->dequant8_coeff[i] = h->dequant8_buffer[j];
822 for (q = 0; q < max_qp + 1; q++) {
825 for (x = 0; x < 64; x++)
826 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
827 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
828 h->pps.scaling_matrix8[i][x]) << shift;
833 static void init_dequant4_coeff_table(H264Context *h)
836 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
837 for (i = 0; i < 6; i++) {
838 h->dequant4_coeff[i] = h->dequant4_buffer[i];
839 for (j = 0; j < i; j++)
840 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
841 16 * sizeof(uint8_t))) {
842 h->dequant4_coeff[i] = h->dequant4_buffer[j];
848 for (q = 0; q < max_qp + 1; q++) {
849 int shift = div6[q] + 2;
851 for (x = 0; x < 16; x++)
852 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
853 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
854 h->pps.scaling_matrix4[i][x]) << shift;
859 static void init_dequant_tables(H264Context *h)
862 init_dequant4_coeff_table(h);
863 if (h->pps.transform_8x8_mode)
864 init_dequant8_coeff_table(h);
865 if (h->sps.transform_bypass) {
866 for (i = 0; i < 6; i++)
867 for (x = 0; x < 16; x++)
868 h->dequant4_coeff[i][0][x] = 1 << 6;
869 if (h->pps.transform_8x8_mode)
870 for (i = 0; i < 6; i++)
871 for (x = 0; x < 64; x++)
872 h->dequant8_coeff[i][0][x] = 1 << 6;
876 int ff_h264_alloc_tables(H264Context *h)
878 MpegEncContext *const s = &h->s;
879 const int big_mb_num = s->mb_stride * (s->mb_height + 1);
880 const int row_mb_num = 2*s->mb_stride*FFMAX(s->avctx->thread_count, 1);
883 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode,
884 row_mb_num * 8 * sizeof(uint8_t), fail)
885 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count,
886 big_mb_num * 48 * sizeof(uint8_t), fail)
887 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base,
888 (big_mb_num + s->mb_stride) * sizeof(*h->slice_table_base), fail)
889 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table,
890 big_mb_num * sizeof(uint16_t), fail)
891 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table,
892 big_mb_num * sizeof(uint8_t), fail)
893 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0],
894 16 * row_mb_num * sizeof(uint8_t), fail);
895 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1],
896 16 * row_mb_num * sizeof(uint8_t), fail);
897 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table,
898 4 * big_mb_num * sizeof(uint8_t), fail);
899 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts,
900 big_mb_num * sizeof(uint8_t), fail)
902 memset(h->slice_table_base, -1,
903 (big_mb_num + s->mb_stride) * sizeof(*h->slice_table_base));
904 h->slice_table = h->slice_table_base + s->mb_stride * 2 + 1;
906 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy,
907 big_mb_num * sizeof(uint32_t), fail);
908 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy,
909 big_mb_num * sizeof(uint32_t), fail);
910 for (y = 0; y < s->mb_height; y++)
911 for (x = 0; x < s->mb_width; x++) {
912 const int mb_xy = x + y * s->mb_stride;
913 const int b_xy = 4 * x + 4 * y * h->b_stride;
915 h->mb2b_xy[mb_xy] = b_xy;
916 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * s->mb_stride)));
919 s->obmc_scratchpad = NULL;
921 if (!h->dequant4_coeff[0])
922 init_dequant_tables(h);
932 * Mimic alloc_tables(), but for every context thread.
934 static void clone_tables(H264Context *dst, H264Context *src, int i)
936 MpegEncContext *const s = &src->s;
937 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * s->mb_stride;
938 dst->non_zero_count = src->non_zero_count;
939 dst->slice_table = src->slice_table;
940 dst->cbp_table = src->cbp_table;
941 dst->mb2b_xy = src->mb2b_xy;
942 dst->mb2br_xy = src->mb2br_xy;
943 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
944 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * s->mb_stride;
945 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * s->mb_stride;
946 dst->direct_table = src->direct_table;
947 dst->list_counts = src->list_counts;
948 dst->s.obmc_scratchpad = NULL;
949 ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma,
950 src->sps.chroma_format_idc);
955 * Allocate buffers which are not shared amongst multiple threads.
957 static int context_init(H264Context *h)
959 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0],
960 h->s.mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
961 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1],
962 h->s.mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
964 h->ref_cache[0][scan8[5] + 1] =
965 h->ref_cache[0][scan8[7] + 1] =
966 h->ref_cache[0][scan8[13] + 1] =
967 h->ref_cache[1][scan8[5] + 1] =
968 h->ref_cache[1][scan8[7] + 1] =
969 h->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
974 return -1; // free_tables will clean up for us
977 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
979 static av_cold void common_init(H264Context *h)
981 MpegEncContext *const s = &h->s;
983 s->width = s->avctx->width;
984 s->height = s->avctx->height;
985 s->codec_id = s->avctx->codec->id;
987 s->avctx->bits_per_raw_sample = 8;
988 h->cur_chroma_format_idc = 1;
990 ff_h264dsp_init(&h->h264dsp,
991 s->avctx->bits_per_raw_sample, h->cur_chroma_format_idc);
992 ff_h264_pred_init(&h->hpc, s->codec_id,
993 s->avctx->bits_per_raw_sample, h->cur_chroma_format_idc);
995 h->dequant_coeff_pps = -1;
996 s->unrestricted_mv = 1;
998 s->dsp.dct_bits = 16;
999 /* needed so that IDCT permutation is known early */
1000 ff_dsputil_init(&s->dsp, s->avctx);
1002 memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
1003 memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
1006 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size)
1008 AVCodecContext *avctx = h->s.avctx;
1010 if (!buf || size <= 0)
1014 int i, cnt, nalsize;
1015 const unsigned char *p = buf;
1020 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1023 /* sps and pps in the avcC always have length coded with 2 bytes,
1024 * so put a fake nal_length_size = 2 while parsing them */
1025 h->nal_length_size = 2;
1026 // Decode sps from avcC
1027 cnt = *(p + 5) & 0x1f; // Number of sps
1029 for (i = 0; i < cnt; i++) {
1030 nalsize = AV_RB16(p) + 2;
1031 if(nalsize > size - (p-buf))
1033 if (decode_nal_units(h, p, nalsize) < 0) {
1034 av_log(avctx, AV_LOG_ERROR,
1035 "Decoding sps %d from avcC failed\n", i);
1040 // Decode pps from avcC
1041 cnt = *(p++); // Number of pps
1042 for (i = 0; i < cnt; i++) {
1043 nalsize = AV_RB16(p) + 2;
1044 if(nalsize > size - (p-buf))
1046 if (decode_nal_units(h, p, nalsize) < 0) {
1047 av_log(avctx, AV_LOG_ERROR,
1048 "Decoding pps %d from avcC failed\n", i);
1053 // Now store right nal length size, that will be used to parse all other nals
1054 h->nal_length_size = (buf[4] & 0x03) + 1;
1057 if (decode_nal_units(h, buf, size) < 0)
1063 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
1065 H264Context *h = avctx->priv_data;
1066 MpegEncContext *const s = &h->s;
1069 ff_MPV_decode_defaults(s);
1074 s->out_format = FMT_H264;
1075 s->workaround_bugs = avctx->workaround_bugs;
1078 // s->decode_mb = ff_h263_decode_mb;
1079 s->quarter_sample = 1;
1080 if (!avctx->has_b_frames)
1083 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1085 ff_h264_decode_init_vlc();
1088 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1090 h->thread_context[0] = h;
1091 h->outputed_poc = h->next_outputed_poc = INT_MIN;
1092 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1093 h->last_pocs[i] = INT_MIN;
1094 h->prev_poc_msb = 1 << 16;
1095 h->prev_frame_num = -1;
1097 ff_h264_reset_sei(h);
1098 if (avctx->codec_id == AV_CODEC_ID_H264) {
1099 if (avctx->ticks_per_frame == 1)
1100 s->avctx->time_base.den *= 2;
1101 avctx->ticks_per_frame = 2;
1104 if (avctx->extradata_size > 0 && avctx->extradata &&
1105 ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size) < 0) {
1106 ff_h264_free_context(h);
1110 if (h->sps.bitstream_restriction_flag &&
1111 s->avctx->has_b_frames < h->sps.num_reorder_frames) {
1112 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1119 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
1121 static void copy_picture_range(Picture **to, Picture **from, int count,
1122 MpegEncContext *new_base,
1123 MpegEncContext *old_base)
1127 for (i = 0; i < count; i++) {
1128 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1129 IN_RANGE(from[i], old_base->picture,
1130 sizeof(Picture) * old_base->picture_count) ||
1132 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1136 static void copy_parameter_set(void **to, void **from, int count, int size)
1140 for (i = 0; i < count; i++) {
1141 if (to[i] && !from[i])
1143 else if (from[i] && !to[i])
1144 to[i] = av_malloc(size);
1147 memcpy(to[i], from[i], size);
1151 static int decode_init_thread_copy(AVCodecContext *avctx)
1153 H264Context *h = avctx->priv_data;
1155 if (!avctx->internal->is_copy)
1157 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1158 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1163 #define copy_fields(to, from, start_field, end_field) \
1164 memcpy(&to->start_field, &from->start_field, \
1165 (char *)&to->end_field - (char *)&to->start_field)
1167 static int decode_update_thread_context(AVCodecContext *dst,
1168 const AVCodecContext *src)
1170 H264Context *h = dst->priv_data, *h1 = src->priv_data;
1171 MpegEncContext *const s = &h->s, *const s1 = &h1->s;
1172 int inited = s->context_initialized, err;
1178 err = ff_mpeg_update_thread_context(dst, src);
1182 // FIXME handle width/height changing
1184 for (i = 0; i < MAX_SPS_COUNT; i++)
1185 av_freep(h->sps_buffers + i);
1187 for (i = 0; i < MAX_PPS_COUNT; i++)
1188 av_freep(h->pps_buffers + i);
1190 // copy all fields after MpegEnc
1191 memcpy(&h->s + 1, &h1->s + 1,
1192 sizeof(H264Context) - sizeof(MpegEncContext));
1193 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1194 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1196 if (s1->context_initialized) {
1197 if (ff_h264_alloc_tables(h) < 0) {
1198 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1199 return AVERROR(ENOMEM);
1203 /* frame_start may not be called for the next thread (if it's decoding
1204 * a bottom field) so this has to be allocated here */
1205 h->s.obmc_scratchpad = av_malloc(16 * 6 * s->linesize);
1208 for (i = 0; i < 2; i++) {
1209 h->rbsp_buffer[i] = NULL;
1210 h->rbsp_buffer_size[i] = 0;
1213 h->thread_context[0] = h;
1215 s->dsp.clear_blocks(h->mb);
1216 s->dsp.clear_blocks(h->mb + (24 * 16 << h->pixel_shift));
1219 // extradata/NAL handling
1220 h->is_avc = h1->is_avc;
1223 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1224 MAX_SPS_COUNT, sizeof(SPS));
1226 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1227 MAX_PPS_COUNT, sizeof(PPS));
1230 // Dequantization matrices
1231 // FIXME these are big - can they be only copied when PPS changes?
1232 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1234 for (i = 0; i < 6; i++)
1235 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
1236 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1238 for (i = 0; i < 6; i++)
1239 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
1240 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1242 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1245 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1248 copy_fields(h, h1, ref_count, list_count);
1249 copy_fields(h, h1, ref_list, intra_gb);
1250 copy_fields(h, h1, short_ref, cabac_init_idc);
1252 copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1253 copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1254 copy_picture_range(h->delayed_pic, h1->delayed_pic,
1255 MAX_DELAYED_PIC_COUNT + 2, s, s1);
1257 h->last_slice_type = h1->last_slice_type;
1260 if (!s->current_picture_ptr)
1264 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1265 h->prev_poc_msb = h->poc_msb;
1266 h->prev_poc_lsb = h->poc_lsb;
1268 h->prev_frame_num_offset = h->frame_num_offset;
1269 h->prev_frame_num = h->frame_num;
1270 h->outputed_poc = h->next_outputed_poc;
1275 int ff_h264_frame_start(H264Context *h)
1277 MpegEncContext *const s = &h->s;
1279 const int pixel_shift = h->pixel_shift;
1281 if (ff_MPV_frame_start(s, s->avctx) < 0)
1283 ff_er_frame_start(s);
1285 * ff_MPV_frame_start uses pict_type to derive key_frame.
1286 * This is incorrect for H.264; IDR markings must be used.
1287 * Zero here; IDR markings per slice in frame or fields are ORed in later.
1288 * See decode_nal_units().
1290 s->current_picture_ptr->f.key_frame = 0;
1291 s->current_picture_ptr->sync = 0;
1292 s->current_picture_ptr->mmco_reset = 0;
1294 assert(s->linesize && s->uvlinesize);
1296 for (i = 0; i < 16; i++) {
1297 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
1298 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
1300 for (i = 0; i < 16; i++) {
1301 h->block_offset[16 + i] =
1302 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1303 h->block_offset[48 + 16 + i] =
1304 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1307 /* can't be in alloc_tables because linesize isn't known there.
1308 * FIXME: redo bipred weight to not require extra buffer? */
1309 for (i = 0; i < s->slice_context_count; i++)
1310 if (h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1311 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16 * 6 * s->linesize);
1313 /* Some macroblocks can be accessed before they're available in case
1314 * of lost slices, MBAFF or threading. */
1315 memset(h->slice_table, -1,
1316 (s->mb_height * s->mb_stride - 1) * sizeof(*h->slice_table));
1318 // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding ||
1319 // s->current_picture.f.reference /* || h->contains_intra */ || 1;
1321 /* We mark the current picture as non-reference after allocating it, so
1322 * that if we break out due to an error it can be released automatically
1323 * in the next ff_MPV_frame_start().
1324 * SVQ3 as well as most other codecs have only last/next/current and thus
1325 * get released even with set reference, besides SVQ3 and others do not
1326 * mark frames as reference later "naturally". */
1327 if (s->codec_id != AV_CODEC_ID_SVQ3)
1328 s->current_picture_ptr->f.reference = 0;
1330 s->current_picture_ptr->field_poc[0] =
1331 s->current_picture_ptr->field_poc[1] = INT_MAX;
1333 h->next_output_pic = NULL;
1335 assert(s->current_picture_ptr->long_ref == 0);
1341 * Run setup operations that must be run after slice header decoding.
1342 * This includes finding the next displayed frame.
1344 * @param h h264 master context
1345 * @param setup_finished enough NALs have been read that we can call
1346 * ff_thread_finish_setup()
1348 static void decode_postinit(H264Context *h, int setup_finished)
1350 MpegEncContext *const s = &h->s;
1351 Picture *out = s->current_picture_ptr;
1352 Picture *cur = s->current_picture_ptr;
1353 int i, pics, out_of_order, out_idx;
1355 s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1356 s->current_picture_ptr->f.pict_type = s->pict_type;
1358 if (h->next_output_pic)
1361 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
1362 /* FIXME: if we have two PAFF fields in one packet, we can't start
1363 * the next thread here. If we have one field per packet, we can.
1364 * The check in decode_nal_units() is not good enough to find this
1365 * yet, so we assume the worst for now. */
1366 // if (setup_finished)
1367 // ff_thread_finish_setup(s->avctx);
1371 cur->f.interlaced_frame = 0;
1372 cur->f.repeat_pict = 0;
1374 /* Signal interlacing information externally. */
1375 /* Prioritize picture timing SEI information over used
1376 * decoding process if it exists. */
1378 if (h->sps.pic_struct_present_flag) {
1379 switch (h->sei_pic_struct) {
1380 case SEI_PIC_STRUCT_FRAME:
1382 case SEI_PIC_STRUCT_TOP_FIELD:
1383 case SEI_PIC_STRUCT_BOTTOM_FIELD:
1384 cur->f.interlaced_frame = 1;
1386 case SEI_PIC_STRUCT_TOP_BOTTOM:
1387 case SEI_PIC_STRUCT_BOTTOM_TOP:
1388 if (FIELD_OR_MBAFF_PICTURE)
1389 cur->f.interlaced_frame = 1;
1391 // try to flag soft telecine progressive
1392 cur->f.interlaced_frame = h->prev_interlaced_frame;
1394 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1395 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1396 /* Signal the possibility of telecined film externally
1397 * (pic_struct 5,6). From these hints, let the applications
1398 * decide if they apply deinterlacing. */
1399 cur->f.repeat_pict = 1;
1401 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1402 // Force progressive here, doubling interlaced frame is a bad idea.
1403 cur->f.repeat_pict = 2;
1405 case SEI_PIC_STRUCT_FRAME_TRIPLING:
1406 cur->f.repeat_pict = 4;
1410 if ((h->sei_ct_type & 3) &&
1411 h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1412 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1414 /* Derive interlacing flag from used decoding process. */
1415 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1417 h->prev_interlaced_frame = cur->f.interlaced_frame;
1419 if (cur->field_poc[0] != cur->field_poc[1]) {
1420 /* Derive top_field_first from field pocs. */
1421 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1423 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1424 /* Use picture timing SEI information. Even if it is a
1425 * information of a past frame, better than nothing. */
1426 if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
1427 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1428 cur->f.top_field_first = 1;
1430 cur->f.top_field_first = 0;
1432 /* Most likely progressive */
1433 cur->f.top_field_first = 0;
1437 cur->mmco_reset = h->mmco_reset;
1439 // FIXME do something with unavailable reference frames
1441 /* Sort B-frames into display order */
1443 if (h->sps.bitstream_restriction_flag &&
1444 s->avctx->has_b_frames < h->sps.num_reorder_frames) {
1445 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1449 if (s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
1450 !h->sps.bitstream_restriction_flag) {
1451 s->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
1455 for (i = 0; 1; i++) {
1456 if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
1458 h->last_pocs[i-1] = cur->poc;
1461 h->last_pocs[i-1]= h->last_pocs[i];
1464 out_of_order = MAX_DELAYED_PIC_COUNT - i;
1465 if( cur->f.pict_type == AV_PICTURE_TYPE_B
1466 || (h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > INT_MIN && h->last_pocs[MAX_DELAYED_PIC_COUNT-1] - h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > 2))
1467 out_of_order = FFMAX(out_of_order, 1);
1468 if(s->avctx->has_b_frames < out_of_order && !h->sps.bitstream_restriction_flag){
1469 av_log(s->avctx, AV_LOG_VERBOSE, "Increasing reorder buffer to %d\n", out_of_order);
1470 s->avctx->has_b_frames = out_of_order;
1475 while (h->delayed_pic[pics])
1478 av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
1480 h->delayed_pic[pics++] = cur;
1481 if (cur->f.reference == 0)
1482 cur->f.reference = DELAYED_PIC_REF;
1484 out = h->delayed_pic[0];
1486 for (i = 1; h->delayed_pic[i] &&
1487 !h->delayed_pic[i]->f.key_frame &&
1488 !h->delayed_pic[i]->mmco_reset;
1490 if (h->delayed_pic[i]->poc < out->poc) {
1491 out = h->delayed_pic[i];
1494 if (s->avctx->has_b_frames == 0 &&
1495 (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
1496 h->next_outputed_poc = INT_MIN;
1497 out_of_order = out->poc < h->next_outputed_poc;
1499 if (out_of_order || pics > s->avctx->has_b_frames) {
1500 out->f.reference &= ~DELAYED_PIC_REF;
1501 // for frame threading, the owner must be the second field's thread or
1502 // else the first thread can release the picture and reuse it unsafely
1504 for (i = out_idx; h->delayed_pic[i]; i++)
1505 h->delayed_pic[i] = h->delayed_pic[i + 1];
1507 if (!out_of_order && pics > s->avctx->has_b_frames) {
1508 h->next_output_pic = out;
1509 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
1510 h->next_outputed_poc = INT_MIN;
1512 h->next_outputed_poc = out->poc;
1514 av_log(s->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
1517 if (h->next_output_pic && h->next_output_pic->sync) {
1522 ff_thread_finish_setup(s->avctx);
1525 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1526 uint8_t *src_cb, uint8_t *src_cr,
1527 int linesize, int uvlinesize,
1530 MpegEncContext *const s = &h->s;
1531 uint8_t *top_border;
1533 const int pixel_shift = h->pixel_shift;
1534 int chroma444 = CHROMA444;
1535 int chroma422 = CHROMA422;
1538 src_cb -= uvlinesize;
1539 src_cr -= uvlinesize;
1541 if (!simple && FRAME_MBAFF) {
1544 top_border = h->top_borders[0][s->mb_x];
1545 AV_COPY128(top_border, src_y + 15 * linesize);
1547 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
1548 if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1551 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
1552 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
1553 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
1554 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
1556 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
1557 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
1559 } else if (chroma422) {
1561 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
1562 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
1564 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
1565 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
1569 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
1570 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
1572 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
1573 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
1578 } else if (MB_MBAFF) {
1584 top_border = h->top_borders[top_idx][s->mb_x];
1585 /* There are two lines saved, the line above the top macroblock
1586 * of a pair, and the line above the bottom macroblock. */
1587 AV_COPY128(top_border, src_y + 16 * linesize);
1589 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
1591 if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1594 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
1595 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
1596 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
1597 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
1599 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
1600 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
1602 } else if (chroma422) {
1604 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
1605 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
1607 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
1608 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
1612 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
1613 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
1615 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
1616 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
1622 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1623 uint8_t *src_cb, uint8_t *src_cr,
1624 int linesize, int uvlinesize,
1625 int xchg, int chroma444,
1626 int simple, int pixel_shift)
1628 MpegEncContext *const s = &h->s;
1629 int deblock_topleft;
1632 uint8_t *top_border_m1;
1633 uint8_t *top_border;
1635 if (!simple && FRAME_MBAFF) {
1640 top_idx = MB_MBAFF ? 0 : 1;
1644 if (h->deblocking_filter == 2) {
1645 deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1646 deblock_top = h->top_type;
1648 deblock_topleft = (s->mb_x > 0);
1649 deblock_top = (s->mb_y > !!MB_FIELD);
1652 src_y -= linesize + 1 + pixel_shift;
1653 src_cb -= uvlinesize + 1 + pixel_shift;
1654 src_cr -= uvlinesize + 1 + pixel_shift;
1656 top_border_m1 = h->top_borders[top_idx][s->mb_x - 1];
1657 top_border = h->top_borders[top_idx][s->mb_x];
1659 #define XCHG(a, b, xchg) \
1660 if (pixel_shift) { \
1662 AV_SWAP64(b + 0, a + 0); \
1663 AV_SWAP64(b + 8, a + 8); \
1673 if (deblock_topleft) {
1674 XCHG(top_border_m1 + (8 << pixel_shift),
1675 src_y - (7 << pixel_shift), 1);
1677 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1678 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1679 if (s->mb_x + 1 < s->mb_width) {
1680 XCHG(h->top_borders[top_idx][s->mb_x + 1],
1681 src_y + (17 << pixel_shift), 1);
1684 if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1686 if (deblock_topleft) {
1687 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1688 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1690 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1691 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1692 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1693 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1694 if (s->mb_x + 1 < s->mb_width) {
1695 XCHG(h->top_borders[top_idx][s->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1696 XCHG(h->top_borders[top_idx][s->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1700 if (deblock_topleft) {
1701 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1702 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1704 XCHG(top_border + (16 << pixel_shift), src_cb + 1 + pixel_shift, 1);
1705 XCHG(top_border + (24 << pixel_shift), src_cr + 1 + pixel_shift, 1);
1711 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth,
1714 if (high_bit_depth) {
1715 return AV_RN32A(((int32_t *)mb) + index);
1717 return AV_RN16A(mb + index);
1720 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth,
1721 int index, int value)
1723 if (high_bit_depth) {
1724 AV_WN32A(((int32_t *)mb) + index, value);
1726 AV_WN16A(mb + index, value);
1729 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h,
1730 int mb_type, int is_h264,
1732 int transform_bypass,
1736 uint8_t *dest_y, int p)
1738 MpegEncContext *const s = &h->s;
1739 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1740 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1742 int qscale = p == 0 ? s->qscale : h->chroma_qp[p - 1];
1743 block_offset += 16 * p;
1744 if (IS_INTRA4x4(mb_type)) {
1745 if (simple || !s->encoding) {
1746 if (IS_8x8DCT(mb_type)) {
1747 if (transform_bypass) {
1749 idct_add = s->dsp.add_pixels8;
1751 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1752 idct_add = h->h264dsp.h264_idct8_add;
1754 for (i = 0; i < 16; i += 4) {
1755 uint8_t *const ptr = dest_y + block_offset[i];
1756 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
1757 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1758 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1760 const int nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1761 h->hpc.pred8x8l[dir](ptr, (h->topleft_samples_available << i) & 0x8000,
1762 (h->topright_samples_available << i) & 0x4000, linesize);
1764 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1765 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1767 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1772 if (transform_bypass) {
1774 idct_add = s->dsp.add_pixels4;
1776 idct_dc_add = h->h264dsp.h264_idct_dc_add;
1777 idct_add = h->h264dsp.h264_idct_add;
1779 for (i = 0; i < 16; i++) {
1780 uint8_t *const ptr = dest_y + block_offset[i];
1781 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
1783 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1784 h->hpc.pred4x4_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1789 if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
1790 const int topright_avail = (h->topright_samples_available << i) & 0x8000;
1791 assert(s->mb_y || linesize <= block_offset[i]);
1792 if (!topright_avail) {
1794 tr_high = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;
1795 topright = (uint8_t *)&tr_high;
1797 tr = ptr[3 - linesize] * 0x01010101u;
1798 topright = (uint8_t *)&tr;
1801 topright = ptr + (4 << pixel_shift) - linesize;
1805 h->hpc.pred4x4[dir](ptr, topright, linesize);
1806 nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1809 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1810 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1812 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1813 } else if (CONFIG_SVQ3_DECODER)
1814 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, qscale, 0);
1821 h->hpc.pred16x16[h->intra16x16_pred_mode](dest_y, linesize);
1823 if (h->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) {
1824 if (!transform_bypass)
1825 h->h264dsp.h264_luma_dc_dequant_idct(h->mb + (p * 256 << pixel_shift),
1827 h->dequant4_coeff[p][qscale][0]);
1829 static const uint8_t dc_mapping[16] = {
1830 0 * 16, 1 * 16, 4 * 16, 5 * 16,
1831 2 * 16, 3 * 16, 6 * 16, 7 * 16,
1832 8 * 16, 9 * 16, 12 * 16, 13 * 16,
1833 10 * 16, 11 * 16, 14 * 16, 15 * 16 };
1834 for (i = 0; i < 16; i++)
1835 dctcoef_set(h->mb + (p * 256 << pixel_shift),
1836 pixel_shift, dc_mapping[i],
1837 dctcoef_get(h->mb_luma_dc[p],
1841 } else if (CONFIG_SVQ3_DECODER)
1842 ff_svq3_luma_dc_dequant_idct_c(h->mb + p * 256,
1843 h->mb_luma_dc[p], qscale);
1847 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type,
1848 int is_h264, int simple,
1849 int transform_bypass,
1853 uint8_t *dest_y, int p)
1855 MpegEncContext *const s = &h->s;
1856 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1858 block_offset += 16 * p;
1859 if (!IS_INTRA4x4(mb_type)) {
1861 if (IS_INTRA16x16(mb_type)) {
1862 if (transform_bypass) {
1863 if (h->sps.profile_idc == 244 &&
1864 (h->intra16x16_pred_mode == VERT_PRED8x8 ||
1865 h->intra16x16_pred_mode == HOR_PRED8x8)) {
1866 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset,
1867 h->mb + (p * 256 << pixel_shift),
1870 for (i = 0; i < 16; i++)
1871 if (h->non_zero_count_cache[scan8[i + p * 16]] ||
1872 dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1873 s->dsp.add_pixels4(dest_y + block_offset[i],
1874 h->mb + (i * 16 + p * 256 << pixel_shift),
1878 h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
1879 h->mb + (p * 256 << pixel_shift),
1881 h->non_zero_count_cache + p * 5 * 8);
1883 } else if (h->cbp & 15) {
1884 if (transform_bypass) {
1885 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1886 idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8
1887 : s->dsp.add_pixels4;
1888 for (i = 0; i < 16; i += di)
1889 if (h->non_zero_count_cache[scan8[i + p * 16]])
1890 idct_add(dest_y + block_offset[i],
1891 h->mb + (i * 16 + p * 256 << pixel_shift),
1894 if (IS_8x8DCT(mb_type))
1895 h->h264dsp.h264_idct8_add4(dest_y, block_offset,
1896 h->mb + (p * 256 << pixel_shift),
1898 h->non_zero_count_cache + p * 5 * 8);
1900 h->h264dsp.h264_idct_add16(dest_y, block_offset,
1901 h->mb + (p * 256 << pixel_shift),
1903 h->non_zero_count_cache + p * 5 * 8);
1906 } else if (CONFIG_SVQ3_DECODER) {
1907 for (i = 0; i < 16; i++)
1908 if (h->non_zero_count_cache[scan8[i + p * 16]] || h->mb[i * 16 + p * 256]) {
1909 // FIXME benchmark weird rule, & below
1910 uint8_t *const ptr = dest_y + block_offset[i];
1911 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize,
1912 s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1920 #include "h264_mb_template.c"
1924 #include "h264_mb_template.c"
1928 #include "h264_mb_template.c"
1930 void ff_h264_hl_decode_mb(H264Context *h)
1932 MpegEncContext *const s = &h->s;
1933 const int mb_xy = h->mb_xy;
1934 const int mb_type = s->current_picture.f.mb_type[mb_xy];
1935 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1938 if (is_complex || h->pixel_shift)
1939 hl_decode_mb_444_complex(h);
1941 hl_decode_mb_444_simple_8(h);
1942 } else if (is_complex) {
1943 hl_decode_mb_complex(h);
1944 } else if (h->pixel_shift) {
1945 hl_decode_mb_simple_16(h);
1947 hl_decode_mb_simple_8(h);
1950 static int pred_weight_table(H264Context *h)
1952 MpegEncContext *const s = &h->s;
1954 int luma_def, chroma_def;
1957 h->use_weight_chroma = 0;
1958 h->luma_log2_weight_denom = get_ue_golomb(&s->gb);
1959 if (h->sps.chroma_format_idc)
1960 h->chroma_log2_weight_denom = get_ue_golomb(&s->gb);
1961 luma_def = 1 << h->luma_log2_weight_denom;
1962 chroma_def = 1 << h->chroma_log2_weight_denom;
1964 for (list = 0; list < 2; list++) {
1965 h->luma_weight_flag[list] = 0;
1966 h->chroma_weight_flag[list] = 0;
1967 for (i = 0; i < h->ref_count[list]; i++) {
1968 int luma_weight_flag, chroma_weight_flag;
1970 luma_weight_flag = get_bits1(&s->gb);
1971 if (luma_weight_flag) {
1972 h->luma_weight[i][list][0] = get_se_golomb(&s->gb);
1973 h->luma_weight[i][list][1] = get_se_golomb(&s->gb);
1974 if (h->luma_weight[i][list][0] != luma_def ||
1975 h->luma_weight[i][list][1] != 0) {
1977 h->luma_weight_flag[list] = 1;
1980 h->luma_weight[i][list][0] = luma_def;
1981 h->luma_weight[i][list][1] = 0;
1984 if (h->sps.chroma_format_idc) {
1985 chroma_weight_flag = get_bits1(&s->gb);
1986 if (chroma_weight_flag) {
1988 for (j = 0; j < 2; j++) {
1989 h->chroma_weight[i][list][j][0] = get_se_golomb(&s->gb);
1990 h->chroma_weight[i][list][j][1] = get_se_golomb(&s->gb);
1991 if (h->chroma_weight[i][list][j][0] != chroma_def ||
1992 h->chroma_weight[i][list][j][1] != 0) {
1993 h->use_weight_chroma = 1;
1994 h->chroma_weight_flag[list] = 1;
1999 for (j = 0; j < 2; j++) {
2000 h->chroma_weight[i][list][j][0] = chroma_def;
2001 h->chroma_weight[i][list][j][1] = 0;
2006 if (h->slice_type_nos != AV_PICTURE_TYPE_B)
2009 h->use_weight = h->use_weight || h->use_weight_chroma;
2014 * Initialize implicit_weight table.
2015 * @param field 0/1 initialize the weight for interlaced MBAFF
2016 * -1 initializes the rest
2018 static void implicit_weight_table(H264Context *h, int field)
2020 MpegEncContext *const s = &h->s;
2021 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2023 for (i = 0; i < 2; i++) {
2024 h->luma_weight_flag[i] = 0;
2025 h->chroma_weight_flag[i] = 0;
2029 if (s->picture_structure == PICT_FRAME) {
2030 cur_poc = s->current_picture_ptr->poc;
2032 cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2034 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF &&
2035 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
2037 h->use_weight_chroma = 0;
2041 ref_count0 = h->ref_count[0];
2042 ref_count1 = h->ref_count[1];
2044 cur_poc = s->current_picture_ptr->field_poc[field];
2046 ref_count0 = 16 + 2 * h->ref_count[0];
2047 ref_count1 = 16 + 2 * h->ref_count[1];
2051 h->use_weight_chroma = 2;
2052 h->luma_log2_weight_denom = 5;
2053 h->chroma_log2_weight_denom = 5;
2055 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
2056 int poc0 = h->ref_list[0][ref0].poc;
2057 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
2059 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2060 int poc1 = h->ref_list[1][ref1].poc;
2061 int td = av_clip(poc1 - poc0, -128, 127);
2063 int tb = av_clip(cur_poc - poc0, -128, 127);
2064 int tx = (16384 + (FFABS(td) >> 1)) / td;
2065 int dist_scale_factor = (tb * tx + 32) >> 8;
2066 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
2067 w = 64 - dist_scale_factor;
2071 h->implicit_weight[ref0][ref1][0] =
2072 h->implicit_weight[ref0][ref1][1] = w;
2074 h->implicit_weight[ref0][ref1][field] = w;
2081 * instantaneous decoder refresh.
2083 static void idr(H264Context *h)
2086 ff_h264_remove_all_refs(h);
2087 h->prev_frame_num = 0;
2088 h->prev_frame_num_offset = 0;
2089 h->prev_poc_msb = 1<<16;
2090 h->prev_poc_lsb = 0;
2091 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2092 h->last_pocs[i] = INT_MIN;
2095 /* forget old pics after a seek */
2096 static void flush_dpb(AVCodecContext *avctx)
2098 H264Context *h = avctx->priv_data;
2100 for (i=0; i<=MAX_DELAYED_PIC_COUNT; i++) {
2101 if (h->delayed_pic[i])
2102 h->delayed_pic[i]->f.reference = 0;
2103 h->delayed_pic[i] = NULL;
2105 h->outputed_poc = h->next_outputed_poc = INT_MIN;
2106 h->prev_interlaced_frame = 1;
2108 h->prev_frame_num = -1;
2109 if (h->s.current_picture_ptr)
2110 h->s.current_picture_ptr->f.reference = 0;
2111 h->s.first_field = 0;
2112 ff_h264_reset_sei(h);
2113 ff_mpeg_flush(avctx);
2114 h->recovery_frame= -1;
2118 static int init_poc(H264Context *h)
2120 MpegEncContext *const s = &h->s;
2121 const int max_frame_num = 1 << h->sps.log2_max_frame_num;
2123 Picture *cur = s->current_picture_ptr;
2125 h->frame_num_offset = h->prev_frame_num_offset;
2126 if (h->frame_num < h->prev_frame_num)
2127 h->frame_num_offset += max_frame_num;
2129 if (h->sps.poc_type == 0) {
2130 const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
2132 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
2133 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2134 else if (h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
2135 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2137 h->poc_msb = h->prev_poc_msb;
2138 // printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2140 field_poc[1] = h->poc_msb + h->poc_lsb;
2141 if (s->picture_structure == PICT_FRAME)
2142 field_poc[1] += h->delta_poc_bottom;
2143 } else if (h->sps.poc_type == 1) {
2144 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2147 if (h->sps.poc_cycle_length != 0)
2148 abs_frame_num = h->frame_num_offset + h->frame_num;
2152 if (h->nal_ref_idc == 0 && abs_frame_num > 0)
2155 expected_delta_per_poc_cycle = 0;
2156 for (i = 0; i < h->sps.poc_cycle_length; i++)
2157 // FIXME integrate during sps parse
2158 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
2160 if (abs_frame_num > 0) {
2161 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2162 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2164 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2165 for (i = 0; i <= frame_num_in_poc_cycle; i++)
2166 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
2170 if (h->nal_ref_idc == 0)
2171 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2173 field_poc[0] = expectedpoc + h->delta_poc[0];
2174 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2176 if (s->picture_structure == PICT_FRAME)
2177 field_poc[1] += h->delta_poc[1];
2179 int poc = 2 * (h->frame_num_offset + h->frame_num);
2181 if (!h->nal_ref_idc)
2188 if (s->picture_structure != PICT_BOTTOM_FIELD)
2189 s->current_picture_ptr->field_poc[0] = field_poc[0];
2190 if (s->picture_structure != PICT_TOP_FIELD)
2191 s->current_picture_ptr->field_poc[1] = field_poc[1];
2192 cur->poc = FFMIN(cur->field_poc[0], cur->field_poc[1]);
2198 * initialize scan tables
2200 static void init_scan_tables(H264Context *h)
2203 for (i = 0; i < 16; i++) {
2204 #define T(x) (x >> 2) | ((x << 2) & 0xF)
2205 h->zigzag_scan[i] = T(zigzag_scan[i]);
2206 h->field_scan[i] = T(field_scan[i]);
2209 for (i = 0; i < 64; i++) {
2210 #define T(x) (x >> 3) | ((x & 7) << 3)
2211 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2212 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2213 h->field_scan8x8[i] = T(field_scan8x8[i]);
2214 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
2217 if (h->sps.transform_bypass) { // FIXME same ugly
2218 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2219 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
2220 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2221 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
2222 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2223 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2225 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2226 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
2227 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2228 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
2229 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2230 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2234 static int field_end(H264Context *h, int in_setup)
2236 MpegEncContext *const s = &h->s;
2237 AVCodecContext *const avctx = s->avctx;
2241 if (!in_setup && !s->dropable)
2242 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
2243 s->picture_structure == PICT_BOTTOM_FIELD);
2245 if (CONFIG_H264_VDPAU_DECODER &&
2246 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2247 ff_vdpau_h264_set_reference_frames(s);
2249 if (in_setup || !(avctx->active_thread_type & FF_THREAD_FRAME)) {
2251 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2252 h->prev_poc_msb = h->poc_msb;
2253 h->prev_poc_lsb = h->poc_lsb;
2255 h->prev_frame_num_offset = h->frame_num_offset;
2256 h->prev_frame_num = h->frame_num;
2257 h->outputed_poc = h->next_outputed_poc;
2260 if (avctx->hwaccel) {
2261 if (avctx->hwaccel->end_frame(avctx) < 0)
2262 av_log(avctx, AV_LOG_ERROR,
2263 "hardware accelerator failed to decode picture\n");
2266 if (CONFIG_H264_VDPAU_DECODER &&
2267 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2268 ff_vdpau_h264_picture_complete(s);
2271 * FIXME: Error handling code does not seem to support interlaced
2272 * when slices span multiple rows
2273 * The ff_er_add_slice calls don't work right for bottom
2274 * fields; they cause massive erroneous error concealing
2275 * Error marking covers both fields (top and bottom).
2276 * This causes a mismatched s->error_count
2277 * and a bad error table. Further, the error count goes to
2278 * INT_MAX when called for bottom field, because mb_y is
2279 * past end by one (callers fault) and resync_mb_y != 0
2280 * causes problems for the first MB line, too.
2285 ff_MPV_frame_end(s);
2287 h->current_slice = 0;
2293 * Replicate H264 "master" context to thread contexts.
2295 static void clone_slice(H264Context *dst, H264Context *src)
2297 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2298 dst->s.current_picture_ptr = src->s.current_picture_ptr;
2299 dst->s.current_picture = src->s.current_picture;
2300 dst->s.linesize = src->s.linesize;
2301 dst->s.uvlinesize = src->s.uvlinesize;
2302 dst->s.first_field = src->s.first_field;
2304 dst->prev_poc_msb = src->prev_poc_msb;
2305 dst->prev_poc_lsb = src->prev_poc_lsb;
2306 dst->prev_frame_num_offset = src->prev_frame_num_offset;
2307 dst->prev_frame_num = src->prev_frame_num;
2308 dst->short_ref_count = src->short_ref_count;
2310 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2311 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2312 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2313 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
2315 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2316 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
2320 * Compute profile from profile_idc and constraint_set?_flags.
2324 * @return profile as defined by FF_PROFILE_H264_*
2326 int ff_h264_get_profile(SPS *sps)
2328 int profile = sps->profile_idc;
2330 switch (sps->profile_idc) {
2331 case FF_PROFILE_H264_BASELINE:
2332 // constraint_set1_flag set to 1
2333 profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2335 case FF_PROFILE_H264_HIGH_10:
2336 case FF_PROFILE_H264_HIGH_422:
2337 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2338 // constraint_set3_flag set to 1
2339 profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
2347 * Decode a slice header.
2348 * This will also call ff_MPV_common_init() and frame_start() as needed.
2350 * @param h h264context
2351 * @param h0 h264 master context (differs from 'h' when doing sliced based
2352 * parallel decoding)
2354 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2356 static int decode_slice_header(H264Context *h, H264Context *h0)
2358 MpegEncContext *const s = &h->s;
2359 MpegEncContext *const s0 = &h0->s;
2360 unsigned int first_mb_in_slice;
2361 unsigned int pps_id;
2362 int num_ref_idx_active_override_flag;
2363 unsigned int slice_type, tmp, i, j;
2364 int default_ref_list_done = 0;
2365 int last_pic_structure, last_pic_dropable;
2368 /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2369 if ((s->avctx->flags2 & CODEC_FLAG2_FAST) &&
2370 !h->nal_ref_idc && !h->pixel_shift) {
2371 s->me.qpel_put = s->dsp.put_2tap_qpel_pixels_tab;
2372 s->me.qpel_avg = s->dsp.avg_2tap_qpel_pixels_tab;
2374 s->me.qpel_put = s->dsp.put_h264_qpel_pixels_tab;
2375 s->me.qpel_avg = s->dsp.avg_h264_qpel_pixels_tab;
2378 first_mb_in_slice = get_ue_golomb_long(&s->gb);
2380 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
2381 if (h0->current_slice && FIELD_PICTURE) {
2385 h0->current_slice = 0;
2386 if (!s0->first_field) {
2387 if (s->current_picture_ptr && !s->dropable &&
2388 s->current_picture_ptr->owner2 == s) {
2389 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
2390 s->picture_structure == PICT_BOTTOM_FIELD);
2392 s->current_picture_ptr = NULL;
2396 slice_type = get_ue_golomb_31(&s->gb);
2397 if (slice_type > 9) {
2398 av_log(h->s.avctx, AV_LOG_ERROR,
2399 "slice type too large (%d) at %d %d\n",
2400 h->slice_type, s->mb_x, s->mb_y);
2403 if (slice_type > 4) {
2405 h->slice_type_fixed = 1;
2407 h->slice_type_fixed = 0;
2409 slice_type = golomb_to_pict_type[slice_type];
2410 if (slice_type == AV_PICTURE_TYPE_I ||
2411 (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
2412 default_ref_list_done = 1;
2414 h->slice_type = slice_type;
2415 h->slice_type_nos = slice_type & 3;
2417 // to make a few old functions happy, it's wrong though
2418 s->pict_type = h->slice_type;
2420 pps_id = get_ue_golomb(&s->gb);
2421 if (pps_id >= MAX_PPS_COUNT) {
2422 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id %d out of range\n", pps_id);
2425 if (!h0->pps_buffers[pps_id]) {
2426 av_log(h->s.avctx, AV_LOG_ERROR,
2427 "non-existing PPS %u referenced\n",
2431 h->pps = *h0->pps_buffers[pps_id];
2433 if (!h0->sps_buffers[h->pps.sps_id]) {
2434 av_log(h->s.avctx, AV_LOG_ERROR,
2435 "non-existing SPS %u referenced\n",
2439 h->sps = *h0->sps_buffers[h->pps.sps_id];
2441 s->avctx->profile = ff_h264_get_profile(&h->sps);
2442 s->avctx->level = h->sps.level_idc;
2443 s->avctx->refs = h->sps.ref_frame_count;
2445 must_reinit = (s->context_initialized &&
2446 ( 16*h->sps.mb_width != s->avctx->coded_width
2447 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != s->avctx->coded_height
2448 || s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
2449 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
2450 || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio)));
2452 if(must_reinit && (h != h0 || (s->avctx->active_thread_type & FF_THREAD_FRAME))) {
2453 av_log_missing_feature(s->avctx,
2454 "Width/height/bit depth/chroma idc changing with threads is", 0);
2455 return AVERROR_PATCHWELCOME; // width / height changed during parallelized decoding
2458 s->mb_width = h->sps.mb_width;
2459 s->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2461 h->b_stride = s->mb_width * 4;
2463 s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2465 s->width = 16 * s->mb_width;
2466 s->height = 16 * s->mb_height;
2470 flush_dpb(s->avctx);
2471 ff_MPV_common_end(s);
2473 h->current_slice = 0;
2475 if (!s->context_initialized) {
2477 av_log(h->s.avctx, AV_LOG_ERROR,
2478 "Cannot (re-)initialize context during parallel decoding.\n");
2481 if( FFALIGN(s->avctx->width , 16 ) == s->width
2482 && FFALIGN(s->avctx->height, 16*(2 - h->sps.frame_mbs_only_flag)) == s->height
2483 && !h->sps.crop_right && !h->sps.crop_bottom
2484 && (s->avctx->width != s->width || s->avctx->height && s->height)
2486 av_log(h->s.avctx, AV_LOG_DEBUG, "Using externally provided dimensions\n");
2487 s->avctx->coded_width = s->width;
2488 s->avctx->coded_height = s->height;
2490 avcodec_set_dimensions(s->avctx, s->width, s->height);
2491 s->avctx->width -= (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2492 s->avctx->height -= (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1) * (2 - h->sps.frame_mbs_only_flag);
2494 s->avctx->sample_aspect_ratio = h->sps.sar;
2495 av_assert0(s->avctx->sample_aspect_ratio.den);
2497 if (s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU
2498 && (h->sps.bit_depth_luma != 8 ||
2499 h->sps.chroma_format_idc > 1)) {
2500 av_log(s->avctx, AV_LOG_ERROR,
2501 "VDPAU decoding does not support video "
2506 if (s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
2507 h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
2508 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 14 && h->sps.bit_depth_luma != 11 && h->sps.bit_depth_luma != 13 &&
2509 (h->sps.bit_depth_luma != 9 || !CHROMA422)) {
2510 s->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
2511 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
2512 h->pixel_shift = h->sps.bit_depth_luma > 8;
2514 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
2515 ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
2516 s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
2517 ff_dsputil_init(&s->dsp, s->avctx);
2519 av_log(s->avctx, AV_LOG_ERROR, "Unsupported bit depth: %d chroma_idc: %d\n",
2520 h->sps.bit_depth_luma, h->sps.chroma_format_idc);
2525 if (h->sps.video_signal_type_present_flag) {
2526 s->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
2528 if (h->sps.colour_description_present_flag) {
2529 s->avctx->color_primaries = h->sps.color_primaries;
2530 s->avctx->color_trc = h->sps.color_trc;
2531 s->avctx->colorspace = h->sps.colorspace;
2535 if (h->sps.timing_info_present_flag) {
2536 int64_t den = h->sps.time_scale;
2537 if (h->x264_build < 44U)
2539 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2540 h->sps.num_units_in_tick, den, 1 << 30);
2543 switch (h->sps.bit_depth_luma) {
2546 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2547 s->avctx->pix_fmt = PIX_FMT_GBRP9;
2549 s->avctx->pix_fmt = PIX_FMT_YUV444P9;
2550 } else if (CHROMA422)
2551 s->avctx->pix_fmt = PIX_FMT_YUV422P9;
2553 s->avctx->pix_fmt = PIX_FMT_YUV420P9;
2557 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2558 s->avctx->pix_fmt = PIX_FMT_GBRP10;
2560 s->avctx->pix_fmt = PIX_FMT_YUV444P10;
2561 } else if (CHROMA422)
2562 s->avctx->pix_fmt = PIX_FMT_YUV422P10;
2564 s->avctx->pix_fmt = PIX_FMT_YUV420P10;
2568 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2569 s->avctx->pix_fmt = PIX_FMT_GBRP12;
2571 s->avctx->pix_fmt = PIX_FMT_YUV444P12;
2572 } else if (CHROMA422)
2573 s->avctx->pix_fmt = PIX_FMT_YUV422P12;
2575 s->avctx->pix_fmt = PIX_FMT_YUV420P12;
2579 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2580 s->avctx->pix_fmt = PIX_FMT_GBRP14;
2582 s->avctx->pix_fmt = PIX_FMT_YUV444P14;
2583 } else if (CHROMA422)
2584 s->avctx->pix_fmt = PIX_FMT_YUV422P14;
2586 s->avctx->pix_fmt = PIX_FMT_YUV420P14;
2590 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P
2592 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2593 s->avctx->pix_fmt = PIX_FMT_GBR24P;
2594 av_log(h->s.avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
2595 } else if (s->avctx->colorspace == AVCOL_SPC_YCGCO) {
2596 av_log(h->s.avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
2598 } else if (CHROMA422) {
2599 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ422P
2602 s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2603 s->avctx->codec->pix_fmts ?
2604 s->avctx->codec->pix_fmts :
2605 s->avctx->color_range == AVCOL_RANGE_JPEG ?
2606 hwaccel_pixfmt_list_h264_jpeg_420 :
2607 ff_hwaccel_pixfmt_list_420);
2611 av_log(s->avctx, AV_LOG_ERROR,
2612 "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
2613 return AVERROR_INVALIDDATA;
2616 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id,
2619 if (ff_MPV_common_init(s) < 0) {
2620 av_log(h->s.avctx, AV_LOG_ERROR, "ff_MPV_common_init() failed.\n");
2624 h->prev_interlaced_frame = 1;
2626 init_scan_tables(h);
2627 if (ff_h264_alloc_tables(h) < 0) {
2628 av_log(h->s.avctx, AV_LOG_ERROR,
2629 "Could not allocate memory for h264\n");
2630 return AVERROR(ENOMEM);
2633 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_SLICE)) {
2634 if (context_init(h) < 0) {
2635 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2639 for (i = 1; i < s->slice_context_count; i++) {
2641 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2642 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2643 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2644 c->h264dsp = h->h264dsp;
2647 c->pixel_shift = h->pixel_shift;
2648 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
2649 init_scan_tables(c);
2650 clone_tables(c, h, i);
2653 for (i = 0; i < s->slice_context_count; i++)
2654 if (context_init(h->thread_context[i]) < 0) {
2655 av_log(h->s.avctx, AV_LOG_ERROR,
2656 "context_init() failed.\n");
2662 if (h == h0 && h->dequant_coeff_pps != pps_id) {
2663 h->dequant_coeff_pps = pps_id;
2664 init_dequant_tables(h);
2667 h->frame_num = get_bits(&s->gb, h->sps.log2_max_frame_num);
2670 h->mb_aff_frame = 0;
2671 last_pic_structure = s0->picture_structure;
2672 last_pic_dropable = s->dropable;
2673 s->dropable = h->nal_ref_idc == 0;
2674 if (h->sps.frame_mbs_only_flag) {
2675 s->picture_structure = PICT_FRAME;
2677 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
2678 av_log(h->s.avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
2681 if (get_bits1(&s->gb)) { // field_pic_flag
2682 s->picture_structure = PICT_TOP_FIELD + get_bits1(&s->gb); // bottom_field_flag
2684 s->picture_structure = PICT_FRAME;
2685 h->mb_aff_frame = h->sps.mb_aff;
2688 h->mb_field_decoding_flag = s->picture_structure != PICT_FRAME;
2690 if (h0->current_slice != 0) {
2691 if (last_pic_structure != s->picture_structure ||
2692 last_pic_dropable != s->dropable) {
2693 av_log(h->s.avctx, AV_LOG_ERROR,
2694 "Changing field mode (%d -> %d) between slices is not allowed\n",
2695 last_pic_structure, s->picture_structure);
2696 s->picture_structure = last_pic_structure;
2697 s->dropable = last_pic_dropable;
2698 return AVERROR_INVALIDDATA;
2701 /* Shorten frame num gaps so we don't have to allocate reference
2702 * frames just to throw them away */
2703 if (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0) {
2704 int unwrap_prev_frame_num = h->prev_frame_num;
2705 int max_frame_num = 1 << h->sps.log2_max_frame_num;
2707 if (unwrap_prev_frame_num > h->frame_num)
2708 unwrap_prev_frame_num -= max_frame_num;
2710 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2711 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2712 if (unwrap_prev_frame_num < 0)
2713 unwrap_prev_frame_num += max_frame_num;
2715 h->prev_frame_num = unwrap_prev_frame_num;
2719 /* See if we have a decoded first field looking for a pair...
2720 * Here, we're using that to see if we should mark previously
2721 * decode frames as "finished".
2722 * We have to do that before the "dummy" in-between frame allocation,
2723 * since that can modify s->current_picture_ptr. */
2724 if (s0->first_field) {
2725 assert(s0->current_picture_ptr);
2726 assert(s0->current_picture_ptr->f.data[0]);
2727 assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2729 /* Mark old field/frame as completed */
2730 if (!last_pic_dropable && s0->current_picture_ptr->owner2 == s0) {
2731 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2732 last_pic_structure == PICT_BOTTOM_FIELD);
2735 /* figure out if we have a complementary field pair */
2736 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2737 /* Previous field is unmatched. Don't display it, but let it
2738 * remain for reference if marked as such. */
2739 if (!last_pic_dropable && last_pic_structure != PICT_FRAME) {
2740 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2741 last_pic_structure == PICT_TOP_FIELD);
2744 if (s0->current_picture_ptr->frame_num != h->frame_num) {
2745 /* This and previous field were reference, but had
2746 * different frame_nums. Consider this field first in
2747 * pair. Throw away previous field except for reference
2749 if (!last_pic_dropable && last_pic_structure != PICT_FRAME) {
2750 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2751 last_pic_structure == PICT_TOP_FIELD);
2754 /* Second field in complementary pair */
2755 if (!((last_pic_structure == PICT_TOP_FIELD &&
2756 s->picture_structure == PICT_BOTTOM_FIELD) ||
2757 (last_pic_structure == PICT_BOTTOM_FIELD &&
2758 s->picture_structure == PICT_TOP_FIELD))) {
2759 av_log(s->avctx, AV_LOG_ERROR,
2760 "Invalid field mode combination %d/%d\n",
2761 last_pic_structure, s->picture_structure);
2762 s->picture_structure = last_pic_structure;
2763 s->dropable = last_pic_dropable;
2764 return AVERROR_INVALIDDATA;
2765 } else if (last_pic_dropable != s->dropable) {
2766 av_log(s->avctx, AV_LOG_ERROR,
2767 "Cannot combine reference and non-reference fields in the same frame\n");
2768 av_log_ask_for_sample(s->avctx, NULL);
2769 s->picture_structure = last_pic_structure;
2770 s->dropable = last_pic_dropable;
2771 return AVERROR_INVALIDDATA;
2774 /* Take ownership of this buffer. Note that if another thread owned
2775 * the first field of this buffer, we're not operating on that pointer,
2776 * so the original thread is still responsible for reporting progress
2777 * on that first field (or if that was us, we just did that above).
2778 * By taking ownership, we assign responsibility to ourselves to
2779 * report progress on the second field. */
2780 s0->current_picture_ptr->owner2 = s0;
2785 while (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0 &&
2786 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
2787 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2788 av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
2789 h->frame_num, h->prev_frame_num);
2790 if (ff_h264_frame_start(h) < 0)
2792 h->prev_frame_num++;
2793 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
2794 s->current_picture_ptr->frame_num = h->prev_frame_num;
2795 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
2796 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 1);
2797 ff_generate_sliding_window_mmcos(h);
2798 if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2799 (s->avctx->err_recognition & AV_EF_EXPLODE))
2800 return AVERROR_INVALIDDATA;
2801 /* Error concealment: if a ref is missing, copy the previous ref in its place.
2802 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2803 * about there being no actual duplicates.
2804 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
2805 * concealing a lost frame, this probably isn't noticeable by comparison, but it should
2807 if (h->short_ref_count) {
2809 av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2810 (const uint8_t **)prev->f.data, prev->f.linesize,
2811 s->avctx->pix_fmt, s->mb_width * 16, s->mb_height * 16);
2812 h->short_ref[0]->poc = prev->poc + 2;
2814 h->short_ref[0]->frame_num = h->prev_frame_num;
2818 /* See if we have a decoded first field looking for a pair...
2819 * We're using that to see whether to continue decoding in that
2820 * frame, or to allocate a new one. */
2821 if (s0->first_field) {
2822 assert(s0->current_picture_ptr);
2823 assert(s0->current_picture_ptr->f.data[0]);
2824 assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2826 /* figure out if we have a complementary field pair */
2827 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2828 /* Previous field is unmatched. Don't display it, but let it
2829 * remain for reference if marked as such. */
2830 s0->current_picture_ptr = NULL;
2831 s0->first_field = FIELD_PICTURE;
2833 if (s0->current_picture_ptr->frame_num != h->frame_num) {
2834 ff_thread_report_progress((AVFrame*)s0->current_picture_ptr, INT_MAX,
2835 s0->picture_structure==PICT_BOTTOM_FIELD);
2836 /* This and the previous field had different frame_nums.
2837 * Consider this field first in pair. Throw away previous
2838 * one except for reference purposes. */
2839 s0->first_field = 1;
2840 s0->current_picture_ptr = NULL;
2842 /* Second field in complementary pair */
2843 s0->first_field = 0;
2847 /* Frame or first field in a potentially complementary pair */
2848 s0->first_field = FIELD_PICTURE;
2851 if (!FIELD_PICTURE || s0->first_field) {
2852 if (ff_h264_frame_start(h) < 0) {
2853 s0->first_field = 0;
2857 ff_release_unused_pictures(s, 0);
2863 s->current_picture_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
2865 assert(s->mb_num == s->mb_width * s->mb_height);
2866 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2867 first_mb_in_slice >= s->mb_num) {
2868 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2871 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2872 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2873 if (s->picture_structure == PICT_BOTTOM_FIELD)
2874 s->resync_mb_y = s->mb_y = s->mb_y + 1;
2875 assert(s->mb_y < s->mb_height);
2877 if (s->picture_structure == PICT_FRAME) {
2878 h->curr_pic_num = h->frame_num;
2879 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
2881 h->curr_pic_num = 2 * h->frame_num + 1;
2882 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
2885 if (h->nal_unit_type == NAL_IDR_SLICE)
2886 get_ue_golomb(&s->gb); /* idr_pic_id */
2888 if (h->sps.poc_type == 0) {
2889 h->poc_lsb = get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2891 if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
2892 h->delta_poc_bottom = get_se_golomb(&s->gb);
2895 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
2896 h->delta_poc[0] = get_se_golomb(&s->gb);
2898 if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
2899 h->delta_poc[1] = get_se_golomb(&s->gb);
2904 if (h->pps.redundant_pic_cnt_present)
2905 h->redundant_pic_count = get_ue_golomb(&s->gb);
2907 // set defaults, might be overridden a few lines later
2908 h->ref_count[0] = h->pps.ref_count[0];
2909 h->ref_count[1] = h->pps.ref_count[1];
2911 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
2913 max[0] = max[1] = s->picture_structure == PICT_FRAME ? 15 : 31;
2915 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
2916 h->direct_spatial_mv_pred = get_bits1(&s->gb);
2917 num_ref_idx_active_override_flag = get_bits1(&s->gb);
2919 if (num_ref_idx_active_override_flag) {
2920 h->ref_count[0] = get_ue_golomb(&s->gb) + 1;
2921 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
2922 h->ref_count[1] = get_ue_golomb(&s->gb) + 1;
2924 // full range is spec-ok in this case, even for frames
2928 if (h->ref_count[0]-1 > max[0] || h->ref_count[1]-1 > max[1]){
2929 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow %u > %u or %u > %u\n", h->ref_count[0]-1, max[0], h->ref_count[1]-1, max[1]);
2930 h->ref_count[0] = h->ref_count[1] = 1;
2931 return AVERROR_INVALIDDATA;
2934 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
2939 h->ref_count[1]= h->ref_count[0]= h->list_count= 0;
2941 if (!default_ref_list_done)
2942 ff_h264_fill_default_ref_list(h);
2944 if (h->slice_type_nos != AV_PICTURE_TYPE_I &&
2945 ff_h264_decode_ref_pic_list_reordering(h) < 0) {
2946 h->ref_count[1] = h->ref_count[0] = 0;
2950 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
2951 s->last_picture_ptr = &h->ref_list[0][0];
2952 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2954 if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
2955 s->next_picture_ptr = &h->ref_list[1][0];
2956 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2959 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
2960 (h->pps.weighted_bipred_idc == 1 &&
2961 h->slice_type_nos == AV_PICTURE_TYPE_B))
2962 pred_weight_table(h);
2963 else if (h->pps.weighted_bipred_idc == 2 &&
2964 h->slice_type_nos == AV_PICTURE_TYPE_B) {
2965 implicit_weight_table(h, -1);
2968 for (i = 0; i < 2; i++) {
2969 h->luma_weight_flag[i] = 0;
2970 h->chroma_weight_flag[i] = 0;
2974 if (h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
2975 (s->avctx->err_recognition & AV_EF_EXPLODE))
2976 return AVERROR_INVALIDDATA;
2979 ff_h264_fill_mbaff_ref_list(h);
2981 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
2982 implicit_weight_table(h, 0);
2983 implicit_weight_table(h, 1);
2987 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
2988 ff_h264_direct_dist_scale_factor(h);
2989 ff_h264_direct_ref_list_init(h);
2991 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
2992 tmp = get_ue_golomb_31(&s->gb);
2994 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2997 h->cabac_init_idc = tmp;
3000 h->last_qscale_diff = 0;
3001 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3002 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
3003 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3007 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3008 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3009 // FIXME qscale / qp ... stuff
3010 if (h->slice_type == AV_PICTURE_TYPE_SP)
3011 get_bits1(&s->gb); /* sp_for_switch_flag */
3012 if (h->slice_type == AV_PICTURE_TYPE_SP ||
3013 h->slice_type == AV_PICTURE_TYPE_SI)
3014 get_se_golomb(&s->gb); /* slice_qs_delta */
3016 h->deblocking_filter = 1;
3017 h->slice_alpha_c0_offset = 52;
3018 h->slice_beta_offset = 52;
3019 if (h->pps.deblocking_filter_parameters_present) {
3020 tmp = get_ue_golomb_31(&s->gb);
3022 av_log(s->avctx, AV_LOG_ERROR,
3023 "deblocking_filter_idc %u out of range\n", tmp);
3026 h->deblocking_filter = tmp;
3027 if (h->deblocking_filter < 2)
3028 h->deblocking_filter ^= 1; // 1<->0
3030 if (h->deblocking_filter) {
3031 h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
3032 h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
3033 if (h->slice_alpha_c0_offset > 104U ||
3034 h->slice_beta_offset > 104U) {
3035 av_log(s->avctx, AV_LOG_ERROR,
3036 "deblocking filter parameters %d %d out of range\n",
3037 h->slice_alpha_c0_offset, h->slice_beta_offset);
3043 if (s->avctx->skip_loop_filter >= AVDISCARD_ALL ||
3044 (s->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
3045 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
3046 (s->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
3047 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
3048 (s->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
3049 h->nal_ref_idc == 0))
3050 h->deblocking_filter = 0;
3052 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
3053 if (s->avctx->flags2 & CODEC_FLAG2_FAST) {
3054 /* Cheat slightly for speed:
3055 * Do not bother to deblock across slices. */
3056 h->deblocking_filter = 2;
3058 h0->max_contexts = 1;
3059 if (!h0->single_decode_warning) {
3060 av_log(s->avctx, AV_LOG_INFO,
3061 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3062 h0->single_decode_warning = 1;
3065 av_log(h->s.avctx, AV_LOG_ERROR,
3066 "Deblocking switched inside frame.\n");
3071 h->qp_thresh = 15 + 52 -
3072 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
3074 h->pps.chroma_qp_index_offset[0],
3075 h->pps.chroma_qp_index_offset[1]) +
3076 6 * (h->sps.bit_depth_luma - 8);
3078 h0->last_slice_type = slice_type;
3079 h->slice_num = ++h0->current_slice;
3082 h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= s->resync_mb_y;
3083 if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= s->resync_mb_y
3084 && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= s->resync_mb_y
3085 && h->slice_num >= MAX_SLICES) {
3086 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
3087 av_log(s->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", h->slice_num, MAX_SLICES);
3090 for (j = 0; j < 2; j++) {
3092 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
3093 for (i = 0; i < 16; i++) {
3095 if (h->ref_list[j][i].f.data[0]) {
3097 uint8_t *base = h->ref_list[j][i].f.base[0];
3098 for (k = 0; k < h->short_ref_count; k++)
3099 if (h->short_ref[k]->f.base[0] == base) {
3103 for (k = 0; k < h->long_ref_count; k++)
3104 if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
3105 id_list[i] = h->short_ref_count + k;
3113 for (i = 0; i < 16; i++)
3114 ref2frm[i + 2] = 4 * id_list[i] +
3115 (h->ref_list[j][i].f.reference & 3);
3117 ref2frm[18 + 1] = -1;
3118 for (i = 16; i < 48; i++)
3119 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
3120 (h->ref_list[j][i].f.reference & 3);
3123 // FIXME: fix draw_edges + PAFF + frame threads
3124 h->emu_edge_width = (s->flags & CODEC_FLAG_EMU_EDGE ||
3125 (!h->sps.frame_mbs_only_flag &&
3126 s->avctx->active_thread_type))
3128 h->emu_edge_height = (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3130 if (s->avctx->debug & FF_DEBUG_PICT_INFO) {
3131 av_log(h->s.avctx, AV_LOG_DEBUG,
3132 "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
3134 (s->picture_structure == PICT_FRAME ? "F" : s->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
3136 av_get_picture_type_char(h->slice_type),
3137 h->slice_type_fixed ? " fix" : "",
3138 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3139 pps_id, h->frame_num,
3140 s->current_picture_ptr->field_poc[0],
3141 s->current_picture_ptr->field_poc[1],
3142 h->ref_count[0], h->ref_count[1],
3144 h->deblocking_filter,
3145 h->slice_alpha_c0_offset / 2 - 26, h->slice_beta_offset / 2 - 26,
3147 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
3148 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
3154 int ff_h264_get_slice_type(const H264Context *h)
3156 switch (h->slice_type) {
3157 case AV_PICTURE_TYPE_P:
3159 case AV_PICTURE_TYPE_B:
3161 case AV_PICTURE_TYPE_I:
3163 case AV_PICTURE_TYPE_SP:
3165 case AV_PICTURE_TYPE_SI:
3172 static av_always_inline void fill_filter_caches_inter(H264Context *h,
3173 MpegEncContext *const s,
3174 int mb_type, int top_xy,
3175 int left_xy[LEFT_MBS],
3177 int left_type[LEFT_MBS],
3178 int mb_xy, int list)
3180 int b_stride = h->b_stride;
3181 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3182 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3183 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
3184 if (USES_LIST(top_type, list)) {
3185 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
3186 const int b8_xy = 4 * top_xy + 2;
3187 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3188 AV_COPY128(mv_dst - 1 * 8, s->current_picture.f.motion_val[list][b_xy + 0]);
3189 ref_cache[0 - 1 * 8] =
3190 ref_cache[1 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3191 ref_cache[2 - 1 * 8] =
3192 ref_cache[3 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3194 AV_ZERO128(mv_dst - 1 * 8);
3195 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3198 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
3199 if (USES_LIST(left_type[LTOP], list)) {
3200 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
3201 const int b8_xy = 4 * left_xy[LTOP] + 1;
3202 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3203 AV_COPY32(mv_dst - 1 + 0, s->current_picture.f.motion_val[list][b_xy + b_stride * 0]);
3204 AV_COPY32(mv_dst - 1 + 8, s->current_picture.f.motion_val[list][b_xy + b_stride * 1]);
3205 AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride * 2]);
3206 AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride * 3]);
3208 ref_cache[-1 + 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 0]];
3209 ref_cache[-1 + 16] =
3210 ref_cache[-1 + 24] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 1]];
3212 AV_ZERO32(mv_dst - 1 + 0);
3213 AV_ZERO32(mv_dst - 1 + 8);
3214 AV_ZERO32(mv_dst - 1 + 16);
3215 AV_ZERO32(mv_dst - 1 + 24);
3218 ref_cache[-1 + 16] =
3219 ref_cache[-1 + 24] = LIST_NOT_USED;
3224 if (!USES_LIST(mb_type, list)) {
3225 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
3226 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3227 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3228 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3229 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3234 int8_t *ref = &s->current_picture.f.ref_index[list][4 * mb_xy];
3235 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3236 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
3237 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
3238 AV_WN32A(&ref_cache[0 * 8], ref01);
3239 AV_WN32A(&ref_cache[1 * 8], ref01);
3240 AV_WN32A(&ref_cache[2 * 8], ref23);
3241 AV_WN32A(&ref_cache[3 * 8], ref23);
3245 int16_t(*mv_src)[2] = &s->current_picture.f.motion_val[list][4 * s->mb_x + 4 * s->mb_y * b_stride];
3246 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
3247 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
3248 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
3249 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
3255 * @return non zero if the loop filter can be skipped
3257 static int fill_filter_caches(H264Context *h, int mb_type)
3259 MpegEncContext *const s = &h->s;
3260 const int mb_xy = h->mb_xy;
3261 int top_xy, left_xy[LEFT_MBS];
3262 int top_type, left_type[LEFT_MBS];
3266 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
3268 /* Wow, what a mess, why didn't they simplify the interlacing & intra
3269 * stuff, I can't imagine that these complex rules are worth it. */
3271 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
3273 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
3274 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
3276 if (left_mb_field_flag != curr_mb_field_flag)
3277 left_xy[LTOP] -= s->mb_stride;
3279 if (curr_mb_field_flag)
3280 top_xy += s->mb_stride &
3281 (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
3282 if (left_mb_field_flag != curr_mb_field_flag)
3283 left_xy[LBOT] += s->mb_stride;
3287 h->top_mb_xy = top_xy;
3288 h->left_mb_xy[LTOP] = left_xy[LTOP];
3289 h->left_mb_xy[LBOT] = left_xy[LBOT];
3291 /* For sufficiently low qp, filtering wouldn't do anything.
3292 * This is a conservative estimate: could also check beta_offset
3293 * and more accurate chroma_qp. */
3294 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
3295 int qp = s->current_picture.f.qscale_table[mb_xy];
3296 if (qp <= qp_thresh &&
3297 (left_xy[LTOP] < 0 ||
3298 ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
3300 ((qp + s->current_picture.f.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
3303 if ((left_xy[LTOP] < 0 ||
3304 ((qp + s->current_picture.f.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
3305 (top_xy < s->mb_stride ||
3306 ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
3311 top_type = s->current_picture.f.mb_type[top_xy];
3312 left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
3313 left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
3314 if (h->deblocking_filter == 2) {
3315 if (h->slice_table[top_xy] != h->slice_num)
3317 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
3318 left_type[LTOP] = left_type[LBOT] = 0;
3320 if (h->slice_table[top_xy] == 0xFFFF)
3322 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
3323 left_type[LTOP] = left_type[LBOT] = 0;
3325 h->top_type = top_type;
3326 h->left_type[LTOP] = left_type[LTOP];
3327 h->left_type[LBOT] = left_type[LBOT];
3329 if (IS_INTRA(mb_type))
3332 fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy,
3333 top_type, left_type, mb_xy, 0);
3334 if (h->list_count == 2)
3335 fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy,
3336 top_type, left_type, mb_xy, 1);
3338 nnz = h->non_zero_count[mb_xy];
3339 nnz_cache = h->non_zero_count_cache;
3340 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
3341 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
3342 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
3343 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
3344 h->cbp = h->cbp_table[mb_xy];
3347 nnz = h->non_zero_count[top_xy];
3348 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
3351 if (left_type[LTOP]) {
3352 nnz = h->non_zero_count[left_xy[LTOP]];
3353 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
3354 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
3355 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
3356 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
3359 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
3360 * from what the loop filter needs */
3361 if (!CABAC && h->pps.transform_8x8_mode) {
3362 if (IS_8x8DCT(top_type)) {
3363 nnz_cache[4 + 8 * 0] =
3364 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
3365 nnz_cache[6 + 8 * 0] =
3366 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
3368 if (IS_8x8DCT(left_type[LTOP])) {
3369 nnz_cache[3 + 8 * 1] =
3370 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
3372 if (IS_8x8DCT(left_type[LBOT])) {
3373 nnz_cache[3 + 8 * 3] =
3374 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
3377 if (IS_8x8DCT(mb_type)) {
3378 nnz_cache[scan8[0]] =
3379 nnz_cache[scan8[1]] =
3380 nnz_cache[scan8[2]] =
3381 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
3383 nnz_cache[scan8[0 + 4]] =
3384 nnz_cache[scan8[1 + 4]] =
3385 nnz_cache[scan8[2 + 4]] =
3386 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
3388 nnz_cache[scan8[0 + 8]] =
3389 nnz_cache[scan8[1 + 8]] =
3390 nnz_cache[scan8[2 + 8]] =
3391 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
3393 nnz_cache[scan8[0 + 12]] =
3394 nnz_cache[scan8[1 + 12]] =
3395 nnz_cache[scan8[2 + 12]] =
3396 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
3403 static void loop_filter(H264Context *h, int start_x, int end_x)
3405 MpegEncContext *const s = &h->s;
3406 uint8_t *dest_y, *dest_cb, *dest_cr;
3407 int linesize, uvlinesize, mb_x, mb_y;
3408 const int end_mb_y = s->mb_y + FRAME_MBAFF;
3409 const int old_slice_type = h->slice_type;
3410 const int pixel_shift = h->pixel_shift;
3411 const int block_h = 16 >> s->chroma_y_shift;
3413 if (h->deblocking_filter) {
3414 for (mb_x = start_x; mb_x < end_x; mb_x++)
3415 for (mb_y = end_mb_y - FRAME_MBAFF; mb_y <= end_mb_y; mb_y++) {
3417 mb_xy = h->mb_xy = mb_x + mb_y * s->mb_stride;
3418 h->slice_num = h->slice_table[mb_xy];
3419 mb_type = s->current_picture.f.mb_type[mb_xy];
3420 h->list_count = h->list_counts[mb_xy];
3424 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3428 dest_y = s->current_picture.f.data[0] +
3429 ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
3430 dest_cb = s->current_picture.f.data[1] +
3431 (mb_x << pixel_shift) * (8 << CHROMA444) +
3432 mb_y * s->uvlinesize * block_h;
3433 dest_cr = s->current_picture.f.data[2] +
3434 (mb_x << pixel_shift) * (8 << CHROMA444) +
3435 mb_y * s->uvlinesize * block_h;
3436 // FIXME simplify above
3439 linesize = h->mb_linesize = s->linesize * 2;
3440 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3441 if (mb_y & 1) { // FIXME move out of this function?
3442 dest_y -= s->linesize * 15;
3443 dest_cb -= s->uvlinesize * (block_h - 1);
3444 dest_cr -= s->uvlinesize * (block_h - 1);
3447 linesize = h->mb_linesize = s->linesize;
3448 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3450 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
3452 if (fill_filter_caches(h, mb_type))
3454 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
3455 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
3458 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
3459 linesize, uvlinesize);
3461 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
3462 dest_cr, linesize, uvlinesize);
3466 h->slice_type = old_slice_type;
3468 s->mb_y = end_mb_y - FRAME_MBAFF;
3469 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3470 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3473 static void predict_field_decoding_flag(H264Context *h)
3475 MpegEncContext *const s = &h->s;
3476 const int mb_xy = s->mb_x + s->mb_y * s->mb_stride;
3477 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
3478 s->current_picture.f.mb_type[mb_xy - 1] :
3479 (h->slice_table[mb_xy - s->mb_stride] == h->slice_num) ?
3480 s->current_picture.f.mb_type[mb_xy - s->mb_stride] : 0;
3481 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3485 * Draw edges and report progress for the last MB row.
3487 static void decode_finish_row(H264Context *h)
3489 MpegEncContext *const s = &h->s;
3490 int top = 16 * (s->mb_y >> FIELD_PICTURE);
3491 int pic_height = 16 * s->mb_height >> FIELD_PICTURE;
3492 int height = 16 << FRAME_MBAFF;
3493 int deblock_border = (16 + 4) << FRAME_MBAFF;
3495 if (h->deblocking_filter) {
3496 if ((top + height) >= pic_height)
3497 height += deblock_border;
3498 top -= deblock_border;
3501 if (top >= pic_height || (top + height) < h->emu_edge_height)
3504 height = FFMIN(height, pic_height - top);
3505 if (top < h->emu_edge_height) {
3506 height = top + height;
3510 ff_draw_horiz_band(s, top, height);
3515 ff_thread_report_progress(&s->current_picture_ptr->f, top + height - 1,
3516 s->picture_structure == PICT_BOTTOM_FIELD);
3519 static int decode_slice(struct AVCodecContext *avctx, void *arg)
3521 H264Context *h = *(void **)arg;
3522 MpegEncContext *const s = &h->s;
3523 const int part_mask = s->partitioned_frame ? (ER_AC_END | ER_AC_ERROR)
3525 int lf_x_start = s->mb_x;
3527 s->mb_skip_run = -1;
3529 h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME ||
3530 s->codec_id != AV_CODEC_ID_H264 ||
3531 (CONFIG_GRAY && (s->flags & CODEC_FLAG_GRAY));
3535 align_get_bits(&s->gb);
3538 ff_init_cabac_states();
3539 ff_init_cabac_decoder(&h->cabac,
3540 s->gb.buffer + get_bits_count(&s->gb) / 8,
3541 (get_bits_left(&s->gb) + 7) / 8);
3543 ff_h264_init_cabac_states(h);
3547 int ret = ff_h264_decode_mb_cabac(h);
3549 // STOP_TIMER("decode_mb_cabac")
3552 ff_h264_hl_decode_mb(h);
3554 // FIXME optimal? or let mb_decode decode 16x32 ?
3555 if (ret >= 0 && FRAME_MBAFF) {
3558 ret = ff_h264_decode_mb_cabac(h);
3561 ff_h264_hl_decode_mb(h);
3564 eos = get_cabac_terminate(&h->cabac);
3566 if ((s->workaround_bugs & FF_BUG_TRUNCATED) &&
3567 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3568 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x - 1,
3569 s->mb_y, ER_MB_END & part_mask);
3570 if (s->mb_x >= lf_x_start)
3571 loop_filter(h, lf_x_start, s->mb_x + 1);
3574 if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
3575 av_log(h->s.avctx, AV_LOG_DEBUG, "bytestream overread %td\n", h->cabac.bytestream_end - h->cabac.bytestream);
3576 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
3577 av_log(h->s.avctx, AV_LOG_ERROR,
3578 "error while decoding MB %d %d, bytestream (%td)\n",
3580 h->cabac.bytestream_end - h->cabac.bytestream);
3581 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3582 s->mb_y, ER_MB_ERROR & part_mask);
3586 if (++s->mb_x >= s->mb_width) {
3587 loop_filter(h, lf_x_start, s->mb_x);
3588 s->mb_x = lf_x_start = 0;
3589 decode_finish_row(h);
3591 if (FIELD_OR_MBAFF_PICTURE) {
3593 if (FRAME_MBAFF && s->mb_y < s->mb_height)
3594 predict_field_decoding_flag(h);
3598 if (eos || s->mb_y >= s->mb_height) {
3599 tprintf(s->avctx, "slice end %d %d\n",
3600 get_bits_count(&s->gb), s->gb.size_in_bits);
3601 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x - 1,
3602 s->mb_y, ER_MB_END & part_mask);
3603 if (s->mb_x > lf_x_start)
3604 loop_filter(h, lf_x_start, s->mb_x);
3610 int ret = ff_h264_decode_mb_cavlc(h);
3613 ff_h264_hl_decode_mb(h);
3615 // FIXME optimal? or let mb_decode decode 16x32 ?
3616 if (ret >= 0 && FRAME_MBAFF) {
3618 ret = ff_h264_decode_mb_cavlc(h);
3621 ff_h264_hl_decode_mb(h);
3626 av_log(h->s.avctx, AV_LOG_ERROR,
3627 "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3628 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3629 s->mb_y, ER_MB_ERROR & part_mask);
3633 if (++s->mb_x >= s->mb_width) {
3634 loop_filter(h, lf_x_start, s->mb_x);
3635 s->mb_x = lf_x_start = 0;
3636 decode_finish_row(h);
3638 if (FIELD_OR_MBAFF_PICTURE) {
3640 if (FRAME_MBAFF && s->mb_y < s->mb_height)
3641 predict_field_decoding_flag(h);
3643 if (s->mb_y >= s->mb_height) {
3644 tprintf(s->avctx, "slice end %d %d\n",
3645 get_bits_count(&s->gb), s->gb.size_in_bits);
3647 if ( get_bits_left(&s->gb) == 0
3648 || get_bits_left(&s->gb) > 0 && !(s->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
3649 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3650 s->mb_x - 1, s->mb_y,
3651 ER_MB_END & part_mask);
3655 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3657 ER_MB_END & part_mask);
3664 if (get_bits_left(&s->gb) <= 0 && s->mb_skip_run <= 0) {
3665 tprintf(s->avctx, "slice end %d %d\n",
3666 get_bits_count(&s->gb), s->gb.size_in_bits);
3667 if (get_bits_left(&s->gb) == 0) {
3668 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3669 s->mb_x - 1, s->mb_y,
3670 ER_MB_END & part_mask);
3671 if (s->mb_x > lf_x_start)
3672 loop_filter(h, lf_x_start, s->mb_x);
3676 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3677 s->mb_y, ER_MB_ERROR & part_mask);
3687 * Call decode_slice() for each context.
3689 * @param h h264 master context
3690 * @param context_count number of contexts to execute
3692 static int execute_decode_slices(H264Context *h, int context_count)
3694 MpegEncContext *const s = &h->s;
3695 AVCodecContext *const avctx = s->avctx;
3699 if (s->avctx->hwaccel ||
3700 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
3702 if (context_count == 1) {
3703 return decode_slice(avctx, &h);
3705 for (i = 1; i < context_count; i++) {
3706 hx = h->thread_context[i];
3707 hx->s.err_recognition = avctx->err_recognition;
3708 hx->s.error_count = 0;
3709 hx->x264_build = h->x264_build;
3712 avctx->execute(avctx, decode_slice, h->thread_context,
3713 NULL, context_count, sizeof(void *));
3715 /* pull back stuff from slices to master context */
3716 hx = h->thread_context[context_count - 1];
3717 s->mb_x = hx->s.mb_x;
3718 s->mb_y = hx->s.mb_y;
3719 s->dropable = hx->s.dropable;
3720 s->picture_structure = hx->s.picture_structure;
3721 for (i = 1; i < context_count; i++)
3722 h->s.error_count += h->thread_context[i]->s.error_count;
3728 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size)
3730 MpegEncContext *const s = &h->s;
3731 AVCodecContext *const avctx = s->avctx;
3732 H264Context *hx; ///< thread context
3736 int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3737 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
3740 h->nal_unit_type= 0;
3742 if(!s->slice_context_count)
3743 s->slice_context_count= 1;
3744 h->max_contexts = s->slice_context_count;
3745 if (!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3746 h->current_slice = 0;
3747 if (!s->first_field)
3748 s->current_picture_ptr = NULL;
3749 ff_h264_reset_sei(h);
3752 for (; pass <= 1; pass++) {
3755 next_avc = h->is_avc ? 0 : buf_size;
3765 if (buf_index >= next_avc) {
3766 if (buf_index >= buf_size - h->nal_length_size)
3769 for (i = 0; i < h->nal_length_size; i++)
3770 nalsize = (nalsize << 8) | buf[buf_index++];
3771 if (nalsize <= 0 || nalsize > buf_size - buf_index) {
3772 av_log(h->s.avctx, AV_LOG_ERROR,
3773 "AVC: nal size %d\n", nalsize);
3776 next_avc = buf_index + nalsize;
3778 // start code prefix search
3779 for (; buf_index + 3 < next_avc; buf_index++)
3780 // This should always succeed in the first iteration.
3781 if (buf[buf_index] == 0 &&
3782 buf[buf_index + 1] == 0 &&
3783 buf[buf_index + 2] == 1)
3786 if (buf_index + 3 >= buf_size)
3790 if (buf_index >= next_avc)
3794 hx = h->thread_context[context_count];
3796 ptr = ff_h264_decode_nal(hx, buf + buf_index, &dst_length,
3797 &consumed, next_avc - buf_index);
3798 if (ptr == NULL || dst_length < 0) {
3802 i = buf_index + consumed;
3803 if ((s->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
3804 buf[i] == 0x00 && buf[i + 1] == 0x00 &&
3805 buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
3806 s->workaround_bugs |= FF_BUG_TRUNCATED;
3808 if (!(s->workaround_bugs & FF_BUG_TRUNCATED))
3809 while(dst_length > 0 && ptr[dst_length - 1] == 0)
3811 bit_length = !dst_length ? 0
3813 decode_rbsp_trailing(h, ptr + dst_length - 1));
3815 if (s->avctx->debug & FF_DEBUG_STARTCODE)
3816 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d/%d at %d/%d length %d pass %d\n", hx->nal_unit_type, hx->nal_ref_idc, buf_index, buf_size, dst_length, pass);
3818 if (h->is_avc && (nalsize != consumed) && nalsize)
3819 av_log(h->s.avctx, AV_LOG_DEBUG,
3820 "AVC: Consumed only %d bytes instead of %d\n",
3823 buf_index += consumed;
3827 /* packets can sometimes contain multiple PPS/SPS,
3828 * e.g. two PAFF field pictures in one packet, or a demuxer
3829 * which splits NALs strangely if so, when frame threading we
3830 * can't start the next thread until we've read all of them */
3831 switch (hx->nal_unit_type) {
3834 nals_needed = nal_index;
3838 init_get_bits(&hx->s.gb, ptr, bit_length);
3839 if (!get_ue_golomb(&hx->s.gb))
3840 nals_needed = nal_index;
3845 // FIXME do not discard SEI id
3846 if (avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
3851 switch (hx->nal_unit_type) {
3853 if (h->nal_unit_type != NAL_IDR_SLICE) {
3854 av_log(h->s.avctx, AV_LOG_ERROR,
3855 "Invalid mix of idr and non-idr slices\n");
3859 idr(h); // FIXME ensure we don't lose some frames if there is reordering
3861 init_get_bits(&hx->s.gb, ptr, bit_length);
3863 hx->inter_gb_ptr = &hx->s.gb;
3864 hx->s.data_partitioning = 0;
3866 if ((err = decode_slice_header(hx, h)))
3869 if (h->sei_recovery_frame_cnt >= 0 && (h->frame_num != h->sei_recovery_frame_cnt || hx->slice_type_nos != AV_PICTURE_TYPE_I))
3870 h->valid_recovery_point = 1;
3872 if ( h->sei_recovery_frame_cnt >= 0
3873 && ( h->recovery_frame<0
3874 || ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt)) {
3875 h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) %
3876 (1 << h->sps.log2_max_frame_num);
3878 if (!h->valid_recovery_point)
3879 h->recovery_frame = h->frame_num;
3882 s->current_picture_ptr->f.key_frame |=
3883 (hx->nal_unit_type == NAL_IDR_SLICE);
3885 if (h->recovery_frame == h->frame_num) {
3886 s->current_picture_ptr->sync |= 1;
3887 h->recovery_frame = -1;
3890 h->sync |= !!s->current_picture_ptr->f.key_frame;
3891 h->sync |= 3*!!(s->flags2 & CODEC_FLAG2_SHOW_ALL);
3892 s->current_picture_ptr->sync |= h->sync;
3894 if (h->current_slice == 1) {
3895 if (!(s->flags2 & CODEC_FLAG2_CHUNKS))
3896 decode_postinit(h, nal_index >= nals_needed);
3898 if (s->avctx->hwaccel &&
3899 s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
3901 if (CONFIG_H264_VDPAU_DECODER &&
3902 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
3903 ff_vdpau_h264_picture_start(s);
3906 if (hx->redundant_pic_count == 0 &&
3907 (avctx->skip_frame < AVDISCARD_NONREF ||
3909 (avctx->skip_frame < AVDISCARD_BIDIR ||
3910 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
3911 (avctx->skip_frame < AVDISCARD_NONKEY ||
3912 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
3913 avctx->skip_frame < AVDISCARD_ALL) {
3914 if (avctx->hwaccel) {
3915 if (avctx->hwaccel->decode_slice(avctx,
3916 &buf[buf_index - consumed],
3919 } else if (CONFIG_H264_VDPAU_DECODER &&
3920 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) {
3921 static const uint8_t start_code[] = {
3923 ff_vdpau_add_data_chunk(s, start_code,
3924 sizeof(start_code));
3925 ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed],
3932 init_get_bits(&hx->s.gb, ptr, bit_length);
3934 hx->inter_gb_ptr = NULL;
3936 if ((err = decode_slice_header(hx, h)) < 0)
3939 hx->s.data_partitioning = 1;
3942 init_get_bits(&hx->intra_gb, ptr, bit_length);
3943 hx->intra_gb_ptr = &hx->intra_gb;
3946 init_get_bits(&hx->inter_gb, ptr, bit_length);
3947 hx->inter_gb_ptr = &hx->inter_gb;
3949 av_log(h->s.avctx, AV_LOG_ERROR, "Partitioned H.264 support is incomplete\n");
3950 return AVERROR_PATCHWELCOME;
3952 if (hx->redundant_pic_count == 0 &&
3954 hx->s.data_partitioning &&
3955 s->context_initialized &&
3956 (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc) &&
3957 (avctx->skip_frame < AVDISCARD_BIDIR ||
3958 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
3959 (avctx->skip_frame < AVDISCARD_NONKEY ||
3960 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
3961 avctx->skip_frame < AVDISCARD_ALL)
3965 init_get_bits(&s->gb, ptr, bit_length);
3966 ff_h264_decode_sei(h);
3969 init_get_bits(&s->gb, ptr, bit_length);
3970 if (ff_h264_decode_seq_parameter_set(h) < 0 && (h->is_avc ? (nalsize != consumed) && nalsize : 1)) {
3971 av_log(h->s.avctx, AV_LOG_DEBUG,
3972 "SPS decoding failure, trying again with the complete NAL\n");
3974 av_assert0(next_avc - buf_index + consumed == nalsize);
3975 init_get_bits(&s->gb, &buf[buf_index + 1 - consumed],
3976 8*(next_avc - buf_index + consumed - 1));
3977 ff_h264_decode_seq_parameter_set(h);
3980 if (s->flags & CODEC_FLAG_LOW_DELAY ||
3981 (h->sps.bitstream_restriction_flag &&
3982 !h->sps.num_reorder_frames))
3984 if (avctx->has_b_frames < 2)
3985 avctx->has_b_frames = !s->low_delay;
3988 init_get_bits(&s->gb, ptr, bit_length);
3989 ff_h264_decode_picture_parameter_set(h, bit_length);
3992 case NAL_END_SEQUENCE:
3993 case NAL_END_STREAM:
3994 case NAL_FILLER_DATA:
3996 case NAL_AUXILIARY_SLICE:
3999 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
4000 hx->nal_unit_type, bit_length);
4003 if (context_count == h->max_contexts) {
4004 execute_decode_slices(h, context_count);
4009 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
4010 else if (err == 1) {
4011 /* Slice could not be decoded in parallel mode, copy down
4012 * NAL unit stuff to context 0 and restart. Note that
4013 * rbsp_buffer is not transferred, but since we no longer
4014 * run in parallel mode this should not be an issue. */
4015 h->nal_unit_type = hx->nal_unit_type;
4016 h->nal_ref_idc = hx->nal_ref_idc;
4023 execute_decode_slices(h, context_count);
4027 if (s->current_picture_ptr && s->current_picture_ptr->owner2 == s &&
4029 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
4030 s->picture_structure == PICT_BOTTOM_FIELD);
4037 * Return the number of bytes consumed for building the current frame.
4039 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size)
4042 pos = 1; // avoid infinite loops (i doubt that is needed but ...)
4043 if (pos + 10 > buf_size)
4044 pos = buf_size; // oops ;)
4049 static int decode_frame(AVCodecContext *avctx, void *data,
4050 int *data_size, AVPacket *avpkt)
4052 const uint8_t *buf = avpkt->data;
4053 int buf_size = avpkt->size;
4054 H264Context *h = avctx->priv_data;
4055 MpegEncContext *s = &h->s;
4056 AVFrame *pict = data;
4061 s->flags = avctx->flags;
4062 s->flags2 = avctx->flags2;
4064 /* end of stream, output what is still in the buffers */
4065 if (buf_size == 0) {
4068 s->current_picture_ptr = NULL;
4070 // FIXME factorize this with the output code below
4071 out = h->delayed_pic[0];
4074 h->delayed_pic[i] &&
4075 !h->delayed_pic[i]->f.key_frame &&
4076 !h->delayed_pic[i]->mmco_reset;
4078 if (h->delayed_pic[i]->poc < out->poc) {
4079 out = h->delayed_pic[i];
4083 for (i = out_idx; h->delayed_pic[i]; i++)
4084 h->delayed_pic[i] = h->delayed_pic[i + 1];
4087 *data_size = sizeof(AVFrame);
4093 if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
4094 int cnt= buf[5]&0x1f;
4095 const uint8_t *p= buf+6;
4097 int nalsize= AV_RB16(p) + 2;
4098 if(nalsize > buf_size - (p-buf) || p[2]!=0x67)
4106 int nalsize= AV_RB16(p) + 2;
4107 if(nalsize > buf_size - (p-buf) || p[2]!=0x68)
4112 return ff_h264_decode_extradata(h, buf, buf_size);
4116 buf_index = decode_nal_units(h, buf, buf_size);
4120 if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
4121 av_assert0(buf_index <= buf_size);
4125 if (!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr) {
4126 if (avctx->skip_frame >= AVDISCARD_NONREF ||
4127 buf_size >= 4 && !memcmp("Q264", buf, 4))
4129 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
4133 if (!(s->flags2 & CODEC_FLAG2_CHUNKS) ||
4134 (s->mb_y >= s->mb_height && s->mb_height)) {
4135 if (s->flags2 & CODEC_FLAG2_CHUNKS)
4136 decode_postinit(h, 1);
4140 /* Wait for second field. */
4142 if (h->next_output_pic && (h->next_output_pic->sync || h->sync>1)) {
4143 *data_size = sizeof(AVFrame);
4144 *pict = h->next_output_pic->f;
4148 assert(pict->data[0] || !*data_size);
4149 ff_print_debug_info(s, pict);
4150 // printf("out %d\n", (int)pict->data[0]);
4152 return get_consumed_bytes(s, buf_index, buf_size);
4155 av_cold void ff_h264_free_context(H264Context *h)
4159 free_tables(h, 1); // FIXME cleanup init stuff perhaps
4161 for (i = 0; i < MAX_SPS_COUNT; i++)
4162 av_freep(h->sps_buffers + i);
4164 for (i = 0; i < MAX_PPS_COUNT; i++)
4165 av_freep(h->pps_buffers + i);
4168 static av_cold int h264_decode_end(AVCodecContext *avctx)
4170 H264Context *h = avctx->priv_data;
4171 MpegEncContext *s = &h->s;
4173 ff_h264_remove_all_refs(h);
4174 ff_h264_free_context(h);
4176 ff_MPV_common_end(s);
4178 // memset(h, 0, sizeof(H264Context));
4183 static const AVProfile profiles[] = {
4184 { FF_PROFILE_H264_BASELINE, "Baseline" },
4185 { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
4186 { FF_PROFILE_H264_MAIN, "Main" },
4187 { FF_PROFILE_H264_EXTENDED, "Extended" },
4188 { FF_PROFILE_H264_HIGH, "High" },
4189 { FF_PROFILE_H264_HIGH_10, "High 10" },
4190 { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
4191 { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
4192 { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
4193 { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
4194 { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
4195 { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
4196 { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
4197 { FF_PROFILE_UNKNOWN },
4200 static const AVOption h264_options[] = {
4201 {"is_avc", "is avc", offsetof(H264Context, is_avc), FF_OPT_TYPE_INT, {.dbl = 0}, 0, 1, 0},
4202 {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), FF_OPT_TYPE_INT, {.dbl = 0}, 0, 4, 0},
4206 static const AVClass h264_class = {
4208 av_default_item_name,
4210 LIBAVUTIL_VERSION_INT,
4213 static const AVClass h264_vdpau_class = {
4214 "H264 VDPAU Decoder",
4215 av_default_item_name,
4217 LIBAVUTIL_VERSION_INT,
4220 AVCodec ff_h264_decoder = {
4222 .type = AVMEDIA_TYPE_VIDEO,
4223 .id = AV_CODEC_ID_H264,
4224 .priv_data_size = sizeof(H264Context),
4225 .init = ff_h264_decode_init,
4226 .close = h264_decode_end,
4227 .decode = decode_frame,
4228 .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 |
4229 CODEC_CAP_DELAY | CODEC_CAP_SLICE_THREADS |
4230 CODEC_CAP_FRAME_THREADS,
4232 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
4233 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
4234 .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
4235 .profiles = NULL_IF_CONFIG_SMALL(profiles),
4236 .priv_class = &h264_class,
4239 #if CONFIG_H264_VDPAU_DECODER
4240 AVCodec ff_h264_vdpau_decoder = {
4241 .name = "h264_vdpau",
4242 .type = AVMEDIA_TYPE_VIDEO,
4243 .id = AV_CODEC_ID_H264,
4244 .priv_data_size = sizeof(H264Context),
4245 .init = ff_h264_decode_init,
4246 .close = h264_decode_end,
4247 .decode = decode_frame,
4248 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
4250 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
4251 .pix_fmts = (const enum PixelFormat[]) { PIX_FMT_VDPAU_H264,
4253 .profiles = NULL_IF_CONFIG_SMALL(profiles),
4254 .priv_class = &h264_vdpau_class,