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 AVPixelFormat 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);
259 while (si + 2 < length) {
260 // remove escapes (very rare 1:2^22)
261 if (src[si + 2] > 3) {
262 dst[di++] = src[si++];
263 dst[di++] = src[si++];
264 } else if (src[si] == 0 && src[si + 1] == 0) {
265 if (src[si + 2] == 3) { // escape
270 } else // next start code
274 dst[di++] = src[si++];
277 dst[di++] = src[si++];
280 memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
283 *consumed = si + 1; // +1 for the header
284 /* FIXME store exact number of bits in the getbitcontext
285 * (it is needed for decoding) */
290 * Identify the exact end of the bitstream
291 * @return the length of the trailing, or 0 if damaged
293 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
298 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
300 for (r = 1; r < 9; r++) {
308 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n,
309 int height, int y_offset, int list)
311 int raw_my = h->mv_cache[list][scan8[n]][1];
312 int filter_height_down = (raw_my & 3) ? 3 : 0;
313 int full_my = (raw_my >> 2) + y_offset;
314 int bottom = full_my + filter_height_down + height;
316 av_assert2(height >= 0);
318 return FFMAX(0, 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 av_assert2(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 av_assert2(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->vdsp.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->vdsp.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->vdsp.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->vdsp.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->vdsp.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 = h->bipred_scratchpad;
676 uint8_t *tmp_cr = h->bipred_scratchpad + (16 << pixel_shift);
677 uint8_t *tmp_y = h->bipred_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->vdsp.prefetch(src[0] + off, s->linesize, 4);
759 if (chroma_idc == 3 /* yuv444 */) {
760 s->vdsp.prefetch(src[1] + off, s->linesize, 4);
761 s->vdsp.prefetch(src[2] + off, s->linesize, 4);
763 off= (((mx>>1)+64)<<pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize;
764 s->vdsp.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->bipred_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 if (!h->dequant4_coeff[0])
920 init_dequant_tables(h);
930 * Mimic alloc_tables(), but for every context thread.
932 static void clone_tables(H264Context *dst, H264Context *src, int i)
934 MpegEncContext *const s = &src->s;
935 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * s->mb_stride;
936 dst->non_zero_count = src->non_zero_count;
937 dst->slice_table = src->slice_table;
938 dst->cbp_table = src->cbp_table;
939 dst->mb2b_xy = src->mb2b_xy;
940 dst->mb2br_xy = src->mb2br_xy;
941 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
942 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * s->mb_stride;
943 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * s->mb_stride;
944 dst->direct_table = src->direct_table;
945 dst->list_counts = src->list_counts;
946 dst->bipred_scratchpad = NULL;
947 ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma,
948 src->sps.chroma_format_idc);
953 * Allocate buffers which are not shared amongst multiple threads.
955 static int context_init(H264Context *h)
957 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0],
958 h->s.mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
959 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1],
960 h->s.mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
962 h->ref_cache[0][scan8[5] + 1] =
963 h->ref_cache[0][scan8[7] + 1] =
964 h->ref_cache[0][scan8[13] + 1] =
965 h->ref_cache[1][scan8[5] + 1] =
966 h->ref_cache[1][scan8[7] + 1] =
967 h->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
972 return -1; // free_tables will clean up for us
975 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
976 int parse_extradata);
978 static av_cold void common_init(H264Context *h)
980 MpegEncContext *const s = &h->s;
982 s->width = s->avctx->width;
983 s->height = s->avctx->height;
984 s->codec_id = s->avctx->codec->id;
986 s->avctx->bits_per_raw_sample = 8;
987 h->cur_chroma_format_idc = 1;
989 ff_h264dsp_init(&h->h264dsp, 8, 1);
990 ff_h264qpel_init(&h->h264qpel, 8);
991 ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
993 h->dequant_coeff_pps = -1;
994 s->unrestricted_mv = 1;
996 s->dsp.dct_bits = 16;
997 /* needed so that IDCT permutation is known early */
998 ff_dsputil_init(&s->dsp, s->avctx);
999 ff_videodsp_init(&s->vdsp, 8);
1001 memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
1002 memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
1005 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size)
1007 AVCodecContext *avctx = h->s.avctx;
1009 if (!buf || size <= 0)
1013 int i, cnt, nalsize;
1014 const unsigned char *p = buf;
1019 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1022 /* sps and pps in the avcC always have length coded with 2 bytes,
1023 * so put a fake nal_length_size = 2 while parsing them */
1024 h->nal_length_size = 2;
1025 // Decode sps from avcC
1026 cnt = *(p + 5) & 0x1f; // Number of sps
1028 for (i = 0; i < cnt; i++) {
1029 nalsize = AV_RB16(p) + 2;
1030 if(nalsize > size - (p-buf))
1032 if (decode_nal_units(h, p, nalsize, 1) < 0) {
1033 av_log(avctx, AV_LOG_ERROR,
1034 "Decoding sps %d from avcC failed\n", i);
1039 // Decode pps from avcC
1040 cnt = *(p++); // Number of pps
1041 for (i = 0; i < cnt; i++) {
1042 nalsize = AV_RB16(p) + 2;
1043 if(nalsize > size - (p-buf))
1045 if (decode_nal_units(h, p, nalsize, 1) < 0) {
1046 av_log(avctx, AV_LOG_ERROR,
1047 "Decoding pps %d from avcC failed\n", i);
1052 // Now store right nal length size, that will be used to parse all other nals
1053 h->nal_length_size = (buf[4] & 0x03) + 1;
1056 if (decode_nal_units(h, buf, size, 1) < 0)
1062 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
1064 H264Context *h = avctx->priv_data;
1065 MpegEncContext *const s = &h->s;
1068 ff_MPV_decode_defaults(s);
1073 s->out_format = FMT_H264;
1074 s->workaround_bugs = avctx->workaround_bugs;
1077 // s->decode_mb = ff_h263_decode_mb;
1078 s->quarter_sample = 1;
1079 if (!avctx->has_b_frames)
1082 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1084 ff_h264_decode_init_vlc();
1087 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1089 h->thread_context[0] = h;
1090 h->outputed_poc = h->next_outputed_poc = INT_MIN;
1091 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1092 h->last_pocs[i] = INT_MIN;
1093 h->prev_poc_msb = 1 << 16;
1094 h->prev_frame_num = -1;
1096 ff_h264_reset_sei(h);
1097 if (avctx->codec_id == AV_CODEC_ID_H264) {
1098 if (avctx->ticks_per_frame == 1) {
1099 if(s->avctx->time_base.den < INT_MAX/2) {
1100 s->avctx->time_base.den *= 2;
1102 s->avctx->time_base.num /= 2;
1104 avctx->ticks_per_frame = 2;
1107 if (avctx->extradata_size > 0 && avctx->extradata &&
1108 ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size) < 0) {
1109 ff_h264_free_context(h);
1113 if (h->sps.bitstream_restriction_flag &&
1114 s->avctx->has_b_frames < h->sps.num_reorder_frames) {
1115 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1119 ff_init_cabac_states();
1124 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
1126 static void copy_picture_range(Picture **to, Picture **from, int count,
1127 MpegEncContext *new_base,
1128 MpegEncContext *old_base)
1132 for (i = 0; i < count; i++) {
1133 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1134 IN_RANGE(from[i], old_base->picture,
1135 sizeof(Picture) * old_base->picture_count) ||
1137 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1141 static void copy_parameter_set(void **to, void **from, int count, int size)
1145 for (i = 0; i < count; i++) {
1146 if (to[i] && !from[i])
1148 else if (from[i] && !to[i])
1149 to[i] = av_malloc(size);
1152 memcpy(to[i], from[i], size);
1156 static int decode_init_thread_copy(AVCodecContext *avctx)
1158 H264Context *h = avctx->priv_data;
1160 if (!avctx->internal->is_copy)
1162 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1163 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1165 h->s.context_initialized = 0;
1170 #define copy_fields(to, from, start_field, end_field) \
1171 memcpy(&to->start_field, &from->start_field, \
1172 (char *)&to->end_field - (char *)&to->start_field)
1174 static int h264_slice_header_init(H264Context *, int);
1176 static int h264_set_parameter_from_sps(H264Context *h);
1178 static int decode_update_thread_context(AVCodecContext *dst,
1179 const AVCodecContext *src)
1181 H264Context *h = dst->priv_data, *h1 = src->priv_data;
1182 MpegEncContext *const s = &h->s, *const s1 = &h1->s;
1183 int inited = s->context_initialized, err;
1190 (s->width != s1->width ||
1191 s->height != s1->height ||
1192 s->mb_width != s1->mb_width ||
1193 s->mb_height != s1->mb_height ||
1194 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
1195 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
1196 h->sps.colorspace != h1->sps.colorspace)) {
1198 av_freep(&h->bipred_scratchpad);
1200 s->width = s1->width;
1201 s->height = s1->height;
1202 s->mb_height = s1->mb_height;
1203 h->b_stride = h1->b_stride;
1205 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1206 MAX_SPS_COUNT, sizeof(SPS));
1208 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1209 MAX_PPS_COUNT, sizeof(PPS));
1212 if ((err = h264_slice_header_init(h, 1)) < 0) {
1213 av_log(h->s.avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
1216 h->context_reinitialized = 1;
1218 h264_set_parameter_from_sps(h);
1219 //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
1220 h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
1222 /* update linesize on resize for h264. The h264 decoder doesn't
1223 * necessarily call ff_MPV_frame_start in the new thread */
1224 s->linesize = s1->linesize;
1225 s->uvlinesize = s1->uvlinesize;
1227 /* copy block_offset since frame_start may not be called */
1228 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
1230 err = ff_mpeg_update_thread_context(dst, src);
1235 for (i = 0; i < MAX_SPS_COUNT; i++)
1236 av_freep(h->sps_buffers + i);
1238 for (i = 0; i < MAX_PPS_COUNT; i++)
1239 av_freep(h->pps_buffers + i);
1241 // copy all fields after MpegEnc
1242 memcpy(&h->s + 1, &h1->s + 1,
1243 sizeof(H264Context) - sizeof(MpegEncContext));
1244 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1245 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1247 if (s1->context_initialized) {
1248 if (ff_h264_alloc_tables(h) < 0) {
1249 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1250 return AVERROR(ENOMEM);
1256 for (i = 0; i < 2; i++) {
1257 h->rbsp_buffer[i] = NULL;
1258 h->rbsp_buffer_size[i] = 0;
1260 h->bipred_scratchpad = NULL;
1262 h->thread_context[0] = h;
1264 s->dsp.clear_blocks(h->mb);
1265 s->dsp.clear_blocks(h->mb + (24 * 16 << h->pixel_shift));
1268 /* frame_start may not be called for the next thread (if it's decoding
1269 * a bottom field) so this has to be allocated here */
1270 if (!h->bipred_scratchpad && s->linesize)
1271 h->bipred_scratchpad = av_malloc(16 * 6 * s->linesize);
1273 // extradata/NAL handling
1274 h->is_avc = h1->is_avc;
1277 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1278 MAX_SPS_COUNT, sizeof(SPS));
1280 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1281 MAX_PPS_COUNT, sizeof(PPS));
1284 // Dequantization matrices
1285 // FIXME these are big - can they be only copied when PPS changes?
1286 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1288 for (i = 0; i < 6; i++)
1289 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
1290 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1292 for (i = 0; i < 6; i++)
1293 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
1294 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1296 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1299 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1302 copy_fields(h, h1, ref_count, list_count);
1303 copy_fields(h, h1, ref2frm, intra_gb);
1304 copy_fields(h, h1, short_ref, cabac_init_idc);
1306 copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1307 copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1308 copy_picture_range(h->delayed_pic, h1->delayed_pic,
1309 MAX_DELAYED_PIC_COUNT + 2, s, s1);
1311 h->last_slice_type = h1->last_slice_type;
1314 if (!s->current_picture_ptr)
1317 if (!s->droppable) {
1318 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1319 h->prev_poc_msb = h->poc_msb;
1320 h->prev_poc_lsb = h->poc_lsb;
1322 h->prev_frame_num_offset = h->frame_num_offset;
1323 h->prev_frame_num = h->frame_num;
1324 h->outputed_poc = h->next_outputed_poc;
1329 int ff_h264_frame_start(H264Context *h)
1331 MpegEncContext *const s = &h->s;
1333 const int pixel_shift = h->pixel_shift;
1335 if (ff_MPV_frame_start(s, s->avctx) < 0)
1337 ff_er_frame_start(s);
1339 * ff_MPV_frame_start uses pict_type to derive key_frame.
1340 * This is incorrect for H.264; IDR markings must be used.
1341 * Zero here; IDR markings per slice in frame or fields are ORed in later.
1342 * See decode_nal_units().
1344 s->current_picture_ptr->f.key_frame = 0;
1345 s->current_picture_ptr->sync = 0;
1346 s->current_picture_ptr->mmco_reset = 0;
1348 assert(s->linesize && s->uvlinesize);
1350 for (i = 0; i < 16; i++) {
1351 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
1352 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
1354 for (i = 0; i < 16; i++) {
1355 h->block_offset[16 + i] =
1356 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1357 h->block_offset[48 + 16 + i] =
1358 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1361 /* can't be in alloc_tables because linesize isn't known there.
1362 * FIXME: redo bipred weight to not require extra buffer? */
1363 for (i = 0; i < s->slice_context_count; i++)
1364 if (h->thread_context[i] && !h->thread_context[i]->bipred_scratchpad)
1365 h->thread_context[i]->bipred_scratchpad = av_malloc(16 * 6 * s->linesize);
1367 /* Some macroblocks can be accessed before they're available in case
1368 * of lost slices, MBAFF or threading. */
1369 memset(h->slice_table, -1,
1370 (s->mb_height * s->mb_stride - 1) * sizeof(*h->slice_table));
1372 // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding ||
1373 // s->current_picture.f.reference /* || h->contains_intra */ || 1;
1375 /* We mark the current picture as non-reference after allocating it, so
1376 * that if we break out due to an error it can be released automatically
1377 * in the next ff_MPV_frame_start().
1378 * SVQ3 as well as most other codecs have only last/next/current and thus
1379 * get released even with set reference, besides SVQ3 and others do not
1380 * mark frames as reference later "naturally". */
1381 if (s->codec_id != AV_CODEC_ID_SVQ3)
1382 s->current_picture_ptr->f.reference = 0;
1384 s->current_picture_ptr->field_poc[0] =
1385 s->current_picture_ptr->field_poc[1] = INT_MAX;
1387 h->next_output_pic = NULL;
1389 assert(s->current_picture_ptr->long_ref == 0);
1395 * Run setup operations that must be run after slice header decoding.
1396 * This includes finding the next displayed frame.
1398 * @param h h264 master context
1399 * @param setup_finished enough NALs have been read that we can call
1400 * ff_thread_finish_setup()
1402 static void decode_postinit(H264Context *h, int setup_finished)
1404 MpegEncContext *const s = &h->s;
1405 Picture *out = s->current_picture_ptr;
1406 Picture *cur = s->current_picture_ptr;
1407 int i, pics, out_of_order, out_idx;
1409 s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1410 s->current_picture_ptr->f.pict_type = s->pict_type;
1412 if (h->next_output_pic)
1415 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
1416 /* FIXME: if we have two PAFF fields in one packet, we can't start
1417 * the next thread here. If we have one field per packet, we can.
1418 * The check in decode_nal_units() is not good enough to find this
1419 * yet, so we assume the worst for now. */
1420 // if (setup_finished)
1421 // ff_thread_finish_setup(s->avctx);
1425 cur->f.interlaced_frame = 0;
1426 cur->f.repeat_pict = 0;
1428 /* Signal interlacing information externally. */
1429 /* Prioritize picture timing SEI information over used
1430 * decoding process if it exists. */
1432 if (h->sps.pic_struct_present_flag) {
1433 switch (h->sei_pic_struct) {
1434 case SEI_PIC_STRUCT_FRAME:
1436 case SEI_PIC_STRUCT_TOP_FIELD:
1437 case SEI_PIC_STRUCT_BOTTOM_FIELD:
1438 cur->f.interlaced_frame = 1;
1440 case SEI_PIC_STRUCT_TOP_BOTTOM:
1441 case SEI_PIC_STRUCT_BOTTOM_TOP:
1442 if (FIELD_OR_MBAFF_PICTURE)
1443 cur->f.interlaced_frame = 1;
1445 // try to flag soft telecine progressive
1446 cur->f.interlaced_frame = h->prev_interlaced_frame;
1448 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1449 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1450 /* Signal the possibility of telecined film externally
1451 * (pic_struct 5,6). From these hints, let the applications
1452 * decide if they apply deinterlacing. */
1453 cur->f.repeat_pict = 1;
1455 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1456 // Force progressive here, doubling interlaced frame is a bad idea.
1457 cur->f.repeat_pict = 2;
1459 case SEI_PIC_STRUCT_FRAME_TRIPLING:
1460 cur->f.repeat_pict = 4;
1464 if ((h->sei_ct_type & 3) &&
1465 h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1466 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1468 /* Derive interlacing flag from used decoding process. */
1469 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1471 h->prev_interlaced_frame = cur->f.interlaced_frame;
1473 if (cur->field_poc[0] != cur->field_poc[1]) {
1474 /* Derive top_field_first from field pocs. */
1475 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1477 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1478 /* Use picture timing SEI information. Even if it is a
1479 * information of a past frame, better than nothing. */
1480 if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
1481 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1482 cur->f.top_field_first = 1;
1484 cur->f.top_field_first = 0;
1486 /* Most likely progressive */
1487 cur->f.top_field_first = 0;
1491 cur->mmco_reset = h->mmco_reset;
1493 // FIXME do something with unavailable reference frames
1495 /* Sort B-frames into display order */
1497 if (h->sps.bitstream_restriction_flag &&
1498 s->avctx->has_b_frames < h->sps.num_reorder_frames) {
1499 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1503 if (s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
1504 !h->sps.bitstream_restriction_flag) {
1505 s->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
1509 for (i = 0; 1; i++) {
1510 if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
1512 h->last_pocs[i-1] = cur->poc;
1515 h->last_pocs[i-1]= h->last_pocs[i];
1518 out_of_order = MAX_DELAYED_PIC_COUNT - i;
1519 if( cur->f.pict_type == AV_PICTURE_TYPE_B
1520 || (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))
1521 out_of_order = FFMAX(out_of_order, 1);
1522 if (out_of_order == MAX_DELAYED_PIC_COUNT) {
1523 av_log(s->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
1524 for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
1525 h->last_pocs[i] = INT_MIN;
1526 h->last_pocs[0] = cur->poc;
1527 cur->mmco_reset = 1;
1528 } else if(s->avctx->has_b_frames < out_of_order && !h->sps.bitstream_restriction_flag){
1529 av_log(s->avctx, AV_LOG_VERBOSE, "Increasing reorder buffer to %d\n", out_of_order);
1530 s->avctx->has_b_frames = out_of_order;
1535 while (h->delayed_pic[pics])
1538 av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
1540 h->delayed_pic[pics++] = cur;
1541 if (cur->f.reference == 0)
1542 cur->f.reference = DELAYED_PIC_REF;
1544 out = h->delayed_pic[0];
1546 for (i = 1; h->delayed_pic[i] &&
1547 !h->delayed_pic[i]->f.key_frame &&
1548 !h->delayed_pic[i]->mmco_reset;
1550 if (h->delayed_pic[i]->poc < out->poc) {
1551 out = h->delayed_pic[i];
1554 if (s->avctx->has_b_frames == 0 &&
1555 (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
1556 h->next_outputed_poc = INT_MIN;
1557 out_of_order = out->poc < h->next_outputed_poc;
1559 if (out_of_order || pics > s->avctx->has_b_frames) {
1560 out->f.reference &= ~DELAYED_PIC_REF;
1561 // for frame threading, the owner must be the second field's thread or
1562 // else the first thread can release the picture and reuse it unsafely
1564 for (i = out_idx; h->delayed_pic[i]; i++)
1565 h->delayed_pic[i] = h->delayed_pic[i + 1];
1567 if (!out_of_order && pics > s->avctx->has_b_frames) {
1568 h->next_output_pic = out;
1569 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
1570 h->next_outputed_poc = INT_MIN;
1572 h->next_outputed_poc = out->poc;
1574 av_log(s->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
1577 if (h->next_output_pic && h->next_output_pic->sync) {
1582 ff_thread_finish_setup(s->avctx);
1585 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1586 uint8_t *src_cb, uint8_t *src_cr,
1587 int linesize, int uvlinesize,
1590 MpegEncContext *const s = &h->s;
1591 uint8_t *top_border;
1593 const int pixel_shift = h->pixel_shift;
1594 int chroma444 = CHROMA444;
1595 int chroma422 = CHROMA422;
1598 src_cb -= uvlinesize;
1599 src_cr -= uvlinesize;
1601 if (!simple && FRAME_MBAFF) {
1604 top_border = h->top_borders[0][s->mb_x];
1605 AV_COPY128(top_border, src_y + 15 * linesize);
1607 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
1608 if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1611 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
1612 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
1613 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
1614 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
1616 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
1617 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
1619 } else if (chroma422) {
1621 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
1622 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
1624 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
1625 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
1629 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
1630 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
1632 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
1633 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
1638 } else if (MB_MBAFF) {
1644 top_border = h->top_borders[top_idx][s->mb_x];
1645 /* There are two lines saved, the line above the top macroblock
1646 * of a pair, and the line above the bottom macroblock. */
1647 AV_COPY128(top_border, src_y + 16 * linesize);
1649 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
1651 if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1654 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
1655 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
1656 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
1657 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
1659 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
1660 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
1662 } else if (chroma422) {
1664 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
1665 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
1667 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
1668 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
1672 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
1673 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
1675 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
1676 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
1682 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1683 uint8_t *src_cb, uint8_t *src_cr,
1684 int linesize, int uvlinesize,
1685 int xchg, int chroma444,
1686 int simple, int pixel_shift)
1688 MpegEncContext *const s = &h->s;
1689 int deblock_topleft;
1692 uint8_t *top_border_m1;
1693 uint8_t *top_border;
1695 if (!simple && FRAME_MBAFF) {
1700 top_idx = MB_MBAFF ? 0 : 1;
1704 if (h->deblocking_filter == 2) {
1705 deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1706 deblock_top = h->top_type;
1708 deblock_topleft = (s->mb_x > 0);
1709 deblock_top = (s->mb_y > !!MB_FIELD);
1712 src_y -= linesize + 1 + pixel_shift;
1713 src_cb -= uvlinesize + 1 + pixel_shift;
1714 src_cr -= uvlinesize + 1 + pixel_shift;
1716 top_border_m1 = h->top_borders[top_idx][s->mb_x - 1];
1717 top_border = h->top_borders[top_idx][s->mb_x];
1719 #define XCHG(a, b, xchg) \
1720 if (pixel_shift) { \
1722 AV_SWAP64(b + 0, a + 0); \
1723 AV_SWAP64(b + 8, a + 8); \
1733 if (deblock_topleft) {
1734 XCHG(top_border_m1 + (8 << pixel_shift),
1735 src_y - (7 << pixel_shift), 1);
1737 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1738 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1739 if (s->mb_x + 1 < s->mb_width) {
1740 XCHG(h->top_borders[top_idx][s->mb_x + 1],
1741 src_y + (17 << pixel_shift), 1);
1744 if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1746 if (deblock_topleft) {
1747 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1748 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1750 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1751 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1752 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1753 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1754 if (s->mb_x + 1 < s->mb_width) {
1755 XCHG(h->top_borders[top_idx][s->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1756 XCHG(h->top_borders[top_idx][s->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1760 if (deblock_topleft) {
1761 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1762 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1764 XCHG(top_border + (16 << pixel_shift), src_cb + 1 + pixel_shift, 1);
1765 XCHG(top_border + (24 << pixel_shift), src_cr + 1 + pixel_shift, 1);
1771 static av_always_inline int dctcoef_get(int16_t *mb, int high_bit_depth,
1774 if (high_bit_depth) {
1775 return AV_RN32A(((int32_t *)mb) + index);
1777 return AV_RN16A(mb + index);
1780 static av_always_inline void dctcoef_set(int16_t *mb, int high_bit_depth,
1781 int index, int value)
1783 if (high_bit_depth) {
1784 AV_WN32A(((int32_t *)mb) + index, value);
1786 AV_WN16A(mb + index, value);
1789 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h,
1790 int mb_type, int is_h264,
1792 int transform_bypass,
1796 uint8_t *dest_y, int p)
1798 MpegEncContext *const s = &h->s;
1799 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
1800 void (*idct_dc_add)(uint8_t *dst, int16_t *block, int stride);
1802 int qscale = p == 0 ? s->qscale : h->chroma_qp[p - 1];
1803 block_offset += 16 * p;
1804 if (IS_INTRA4x4(mb_type)) {
1805 if (simple || !s->encoding) {
1806 if (IS_8x8DCT(mb_type)) {
1807 if (transform_bypass) {
1809 idct_add = s->dsp.add_pixels8;
1811 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1812 idct_add = h->h264dsp.h264_idct8_add;
1814 for (i = 0; i < 16; i += 4) {
1815 uint8_t *const ptr = dest_y + block_offset[i];
1816 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
1817 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1818 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1820 const int nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1821 h->hpc.pred8x8l[dir](ptr, (h->topleft_samples_available << i) & 0x8000,
1822 (h->topright_samples_available << i) & 0x4000, linesize);
1824 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1825 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1827 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1832 if (transform_bypass) {
1834 idct_add = s->dsp.add_pixels4;
1836 idct_dc_add = h->h264dsp.h264_idct_dc_add;
1837 idct_add = h->h264dsp.h264_idct_add;
1839 for (i = 0; i < 16; i++) {
1840 uint8_t *const ptr = dest_y + block_offset[i];
1841 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
1843 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1844 h->hpc.pred4x4_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1849 if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
1850 const int topright_avail = (h->topright_samples_available << i) & 0x8000;
1851 av_assert2(s->mb_y || linesize <= block_offset[i]);
1852 if (!topright_avail) {
1854 tr_high = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;
1855 topright = (uint8_t *)&tr_high;
1857 tr = ptr[3 - linesize] * 0x01010101u;
1858 topright = (uint8_t *)&tr;
1861 topright = ptr + (4 << pixel_shift) - linesize;
1865 h->hpc.pred4x4[dir](ptr, topright, linesize);
1866 nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1869 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1870 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1872 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1873 } else if (CONFIG_SVQ3_DECODER)
1874 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, qscale, 0);
1881 h->hpc.pred16x16[h->intra16x16_pred_mode](dest_y, linesize);
1883 if (h->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) {
1884 if (!transform_bypass)
1885 h->h264dsp.h264_luma_dc_dequant_idct(h->mb + (p * 256 << pixel_shift),
1887 h->dequant4_coeff[p][qscale][0]);
1889 static const uint8_t dc_mapping[16] = {
1890 0 * 16, 1 * 16, 4 * 16, 5 * 16,
1891 2 * 16, 3 * 16, 6 * 16, 7 * 16,
1892 8 * 16, 9 * 16, 12 * 16, 13 * 16,
1893 10 * 16, 11 * 16, 14 * 16, 15 * 16 };
1894 for (i = 0; i < 16; i++)
1895 dctcoef_set(h->mb + (p * 256 << pixel_shift),
1896 pixel_shift, dc_mapping[i],
1897 dctcoef_get(h->mb_luma_dc[p],
1901 } else if (CONFIG_SVQ3_DECODER)
1902 ff_svq3_luma_dc_dequant_idct_c(h->mb + p * 256,
1903 h->mb_luma_dc[p], qscale);
1907 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type,
1908 int is_h264, int simple,
1909 int transform_bypass,
1913 uint8_t *dest_y, int p)
1915 MpegEncContext *const s = &h->s;
1916 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
1918 block_offset += 16 * p;
1919 if (!IS_INTRA4x4(mb_type)) {
1921 if (IS_INTRA16x16(mb_type)) {
1922 if (transform_bypass) {
1923 if (h->sps.profile_idc == 244 &&
1924 (h->intra16x16_pred_mode == VERT_PRED8x8 ||
1925 h->intra16x16_pred_mode == HOR_PRED8x8)) {
1926 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset,
1927 h->mb + (p * 256 << pixel_shift),
1930 for (i = 0; i < 16; i++)
1931 if (h->non_zero_count_cache[scan8[i + p * 16]] ||
1932 dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1933 s->dsp.add_pixels4(dest_y + block_offset[i],
1934 h->mb + (i * 16 + p * 256 << pixel_shift),
1938 h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
1939 h->mb + (p * 256 << pixel_shift),
1941 h->non_zero_count_cache + p * 5 * 8);
1943 } else if (h->cbp & 15) {
1944 if (transform_bypass) {
1945 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1946 idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8
1947 : s->dsp.add_pixels4;
1948 for (i = 0; i < 16; i += di)
1949 if (h->non_zero_count_cache[scan8[i + p * 16]])
1950 idct_add(dest_y + block_offset[i],
1951 h->mb + (i * 16 + p * 256 << pixel_shift),
1954 if (IS_8x8DCT(mb_type))
1955 h->h264dsp.h264_idct8_add4(dest_y, block_offset,
1956 h->mb + (p * 256 << pixel_shift),
1958 h->non_zero_count_cache + p * 5 * 8);
1960 h->h264dsp.h264_idct_add16(dest_y, block_offset,
1961 h->mb + (p * 256 << pixel_shift),
1963 h->non_zero_count_cache + p * 5 * 8);
1966 } else if (CONFIG_SVQ3_DECODER) {
1967 for (i = 0; i < 16; i++)
1968 if (h->non_zero_count_cache[scan8[i + p * 16]] || h->mb[i * 16 + p * 256]) {
1969 // FIXME benchmark weird rule, & below
1970 uint8_t *const ptr = dest_y + block_offset[i];
1971 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize,
1972 s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1980 #include "h264_mb_template.c"
1984 #include "h264_mb_template.c"
1988 #include "h264_mb_template.c"
1990 void ff_h264_hl_decode_mb(H264Context *h)
1992 MpegEncContext *const s = &h->s;
1993 const int mb_xy = h->mb_xy;
1994 const int mb_type = s->current_picture.f.mb_type[mb_xy];
1995 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1998 if (is_complex || h->pixel_shift)
1999 hl_decode_mb_444_complex(h);
2001 hl_decode_mb_444_simple_8(h);
2002 } else if (is_complex) {
2003 hl_decode_mb_complex(h);
2004 } else if (h->pixel_shift) {
2005 hl_decode_mb_simple_16(h);
2007 hl_decode_mb_simple_8(h);
2010 static int pred_weight_table(H264Context *h)
2012 MpegEncContext *const s = &h->s;
2014 int luma_def, chroma_def;
2017 h->use_weight_chroma = 0;
2018 h->luma_log2_weight_denom = get_ue_golomb(&s->gb);
2019 if (h->sps.chroma_format_idc)
2020 h->chroma_log2_weight_denom = get_ue_golomb(&s->gb);
2021 luma_def = 1 << h->luma_log2_weight_denom;
2022 chroma_def = 1 << h->chroma_log2_weight_denom;
2024 for (list = 0; list < 2; list++) {
2025 h->luma_weight_flag[list] = 0;
2026 h->chroma_weight_flag[list] = 0;
2027 for (i = 0; i < h->ref_count[list]; i++) {
2028 int luma_weight_flag, chroma_weight_flag;
2030 luma_weight_flag = get_bits1(&s->gb);
2031 if (luma_weight_flag) {
2032 h->luma_weight[i][list][0] = get_se_golomb(&s->gb);
2033 h->luma_weight[i][list][1] = get_se_golomb(&s->gb);
2034 if (h->luma_weight[i][list][0] != luma_def ||
2035 h->luma_weight[i][list][1] != 0) {
2037 h->luma_weight_flag[list] = 1;
2040 h->luma_weight[i][list][0] = luma_def;
2041 h->luma_weight[i][list][1] = 0;
2044 if (h->sps.chroma_format_idc) {
2045 chroma_weight_flag = get_bits1(&s->gb);
2046 if (chroma_weight_flag) {
2048 for (j = 0; j < 2; j++) {
2049 h->chroma_weight[i][list][j][0] = get_se_golomb(&s->gb);
2050 h->chroma_weight[i][list][j][1] = get_se_golomb(&s->gb);
2051 if (h->chroma_weight[i][list][j][0] != chroma_def ||
2052 h->chroma_weight[i][list][j][1] != 0) {
2053 h->use_weight_chroma = 1;
2054 h->chroma_weight_flag[list] = 1;
2059 for (j = 0; j < 2; j++) {
2060 h->chroma_weight[i][list][j][0] = chroma_def;
2061 h->chroma_weight[i][list][j][1] = 0;
2066 if (h->slice_type_nos != AV_PICTURE_TYPE_B)
2069 h->use_weight = h->use_weight || h->use_weight_chroma;
2074 * Initialize implicit_weight table.
2075 * @param field 0/1 initialize the weight for interlaced MBAFF
2076 * -1 initializes the rest
2078 static void implicit_weight_table(H264Context *h, int field)
2080 MpegEncContext *const s = &h->s;
2081 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2083 for (i = 0; i < 2; i++) {
2084 h->luma_weight_flag[i] = 0;
2085 h->chroma_weight_flag[i] = 0;
2089 if (s->picture_structure == PICT_FRAME) {
2090 cur_poc = s->current_picture_ptr->poc;
2092 cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2094 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF &&
2095 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
2097 h->use_weight_chroma = 0;
2101 ref_count0 = h->ref_count[0];
2102 ref_count1 = h->ref_count[1];
2104 cur_poc = s->current_picture_ptr->field_poc[field];
2106 ref_count0 = 16 + 2 * h->ref_count[0];
2107 ref_count1 = 16 + 2 * h->ref_count[1];
2111 h->use_weight_chroma = 2;
2112 h->luma_log2_weight_denom = 5;
2113 h->chroma_log2_weight_denom = 5;
2115 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
2116 int poc0 = h->ref_list[0][ref0].poc;
2117 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
2119 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2120 int poc1 = h->ref_list[1][ref1].poc;
2121 int td = av_clip(poc1 - poc0, -128, 127);
2123 int tb = av_clip(cur_poc - poc0, -128, 127);
2124 int tx = (16384 + (FFABS(td) >> 1)) / td;
2125 int dist_scale_factor = (tb * tx + 32) >> 8;
2126 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
2127 w = 64 - dist_scale_factor;
2131 h->implicit_weight[ref0][ref1][0] =
2132 h->implicit_weight[ref0][ref1][1] = w;
2134 h->implicit_weight[ref0][ref1][field] = w;
2141 * instantaneous decoder refresh.
2143 static void idr(H264Context *h)
2146 ff_h264_remove_all_refs(h);
2147 h->prev_frame_num = 0;
2148 h->prev_frame_num_offset = 0;
2149 h->prev_poc_msb = 1<<16;
2150 h->prev_poc_lsb = 0;
2151 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2152 h->last_pocs[i] = INT_MIN;
2155 /* forget old pics after a seek */
2156 static void flush_change(H264Context *h)
2160 h->outputed_poc = h->next_outputed_poc = INT_MIN;
2161 h->prev_interlaced_frame = 1;
2163 h->prev_frame_num = -1;
2164 if (h->s.current_picture_ptr) {
2165 h->s.current_picture_ptr->f.reference = 0;
2166 for (j=i=0; h->delayed_pic[i]; i++)
2167 if (h->delayed_pic[i] != h->s.current_picture_ptr)
2168 h->delayed_pic[j++] = h->delayed_pic[i];
2169 h->delayed_pic[j] = NULL;
2171 h->s.first_field = 0;
2172 memset(h->ref_list[0], 0, sizeof(h->ref_list[0]));
2173 memset(h->ref_list[1], 0, sizeof(h->ref_list[1]));
2174 memset(h->default_ref_list[0], 0, sizeof(h->default_ref_list[0]));
2175 memset(h->default_ref_list[1], 0, sizeof(h->default_ref_list[1]));
2176 ff_h264_reset_sei(h);
2177 h->recovery_frame= -1;
2180 h->current_slice = 0;
2183 /* forget old pics after a seek */
2184 static void flush_dpb(AVCodecContext *avctx)
2186 H264Context *h = avctx->priv_data;
2189 for (i = 0; i <= MAX_DELAYED_PIC_COUNT; i++) {
2190 if (h->delayed_pic[i])
2191 h->delayed_pic[i]->f.reference = 0;
2192 h->delayed_pic[i] = NULL;
2196 ff_mpeg_flush(avctx);
2199 static int init_poc(H264Context *h)
2201 MpegEncContext *const s = &h->s;
2202 const int max_frame_num = 1 << h->sps.log2_max_frame_num;
2204 Picture *cur = s->current_picture_ptr;
2206 h->frame_num_offset = h->prev_frame_num_offset;
2207 if (h->frame_num < h->prev_frame_num)
2208 h->frame_num_offset += max_frame_num;
2210 if (h->sps.poc_type == 0) {
2211 const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
2213 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
2214 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2215 else if (h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
2216 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2218 h->poc_msb = h->prev_poc_msb;
2220 field_poc[1] = h->poc_msb + h->poc_lsb;
2221 if (s->picture_structure == PICT_FRAME)
2222 field_poc[1] += h->delta_poc_bottom;
2223 } else if (h->sps.poc_type == 1) {
2224 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2227 if (h->sps.poc_cycle_length != 0)
2228 abs_frame_num = h->frame_num_offset + h->frame_num;
2232 if (h->nal_ref_idc == 0 && abs_frame_num > 0)
2235 expected_delta_per_poc_cycle = 0;
2236 for (i = 0; i < h->sps.poc_cycle_length; i++)
2237 // FIXME integrate during sps parse
2238 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
2240 if (abs_frame_num > 0) {
2241 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2242 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2244 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2245 for (i = 0; i <= frame_num_in_poc_cycle; i++)
2246 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
2250 if (h->nal_ref_idc == 0)
2251 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2253 field_poc[0] = expectedpoc + h->delta_poc[0];
2254 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2256 if (s->picture_structure == PICT_FRAME)
2257 field_poc[1] += h->delta_poc[1];
2259 int poc = 2 * (h->frame_num_offset + h->frame_num);
2261 if (!h->nal_ref_idc)
2268 if (s->picture_structure != PICT_BOTTOM_FIELD)
2269 s->current_picture_ptr->field_poc[0] = field_poc[0];
2270 if (s->picture_structure != PICT_TOP_FIELD)
2271 s->current_picture_ptr->field_poc[1] = field_poc[1];
2272 cur->poc = FFMIN(cur->field_poc[0], cur->field_poc[1]);
2278 * initialize scan tables
2280 static void init_scan_tables(H264Context *h)
2283 for (i = 0; i < 16; i++) {
2284 #define T(x) (x >> 2) | ((x << 2) & 0xF)
2285 h->zigzag_scan[i] = T(zigzag_scan[i]);
2286 h->field_scan[i] = T(field_scan[i]);
2289 for (i = 0; i < 64; i++) {
2290 #define T(x) (x >> 3) | ((x & 7) << 3)
2291 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2292 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2293 h->field_scan8x8[i] = T(field_scan8x8[i]);
2294 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
2297 if (h->sps.transform_bypass) { // FIXME same ugly
2298 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2299 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
2300 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2301 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
2302 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2303 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2305 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2306 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
2307 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2308 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
2309 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2310 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2314 static int field_end(H264Context *h, int in_setup)
2316 MpegEncContext *const s = &h->s;
2317 AVCodecContext *const avctx = s->avctx;
2321 if (!in_setup && !s->droppable)
2322 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
2323 s->picture_structure == PICT_BOTTOM_FIELD);
2325 if (CONFIG_H264_VDPAU_DECODER &&
2326 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2327 ff_vdpau_h264_set_reference_frames(s);
2329 if (in_setup || !(avctx->active_thread_type & FF_THREAD_FRAME)) {
2330 if (!s->droppable) {
2331 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2332 h->prev_poc_msb = h->poc_msb;
2333 h->prev_poc_lsb = h->poc_lsb;
2335 h->prev_frame_num_offset = h->frame_num_offset;
2336 h->prev_frame_num = h->frame_num;
2337 h->outputed_poc = h->next_outputed_poc;
2340 if (avctx->hwaccel) {
2341 if (avctx->hwaccel->end_frame(avctx) < 0)
2342 av_log(avctx, AV_LOG_ERROR,
2343 "hardware accelerator failed to decode picture\n");
2346 if (CONFIG_H264_VDPAU_DECODER &&
2347 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2348 ff_vdpau_h264_picture_complete(s);
2351 * FIXME: Error handling code does not seem to support interlaced
2352 * when slices span multiple rows
2353 * The ff_er_add_slice calls don't work right for bottom
2354 * fields; they cause massive erroneous error concealing
2355 * Error marking covers both fields (top and bottom).
2356 * This causes a mismatched s->error_count
2357 * and a bad error table. Further, the error count goes to
2358 * INT_MAX when called for bottom field, because mb_y is
2359 * past end by one (callers fault) and resync_mb_y != 0
2360 * causes problems for the first MB line, too.
2362 if (!FIELD_PICTURE && h->current_slice)
2365 ff_MPV_frame_end(s);
2367 h->current_slice = 0;
2373 * Replicate H264 "master" context to thread contexts.
2375 static int clone_slice(H264Context *dst, H264Context *src)
2379 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2380 dst->s.current_picture_ptr = src->s.current_picture_ptr;
2381 dst->s.current_picture = src->s.current_picture;
2382 dst->s.linesize = src->s.linesize;
2383 dst->s.uvlinesize = src->s.uvlinesize;
2384 dst->s.first_field = src->s.first_field;
2386 if (!dst->s.edge_emu_buffer &&
2387 (ret = ff_mpv_frame_size_alloc(&dst->s, dst->s.linesize))) {
2388 av_log(dst->s.avctx, AV_LOG_ERROR,
2389 "Failed to allocate scratch buffers\n");
2393 dst->prev_poc_msb = src->prev_poc_msb;
2394 dst->prev_poc_lsb = src->prev_poc_lsb;
2395 dst->prev_frame_num_offset = src->prev_frame_num_offset;
2396 dst->prev_frame_num = src->prev_frame_num;
2397 dst->short_ref_count = src->short_ref_count;
2399 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2400 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2401 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2403 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2404 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
2410 * Compute profile from profile_idc and constraint_set?_flags.
2414 * @return profile as defined by FF_PROFILE_H264_*
2416 int ff_h264_get_profile(SPS *sps)
2418 int profile = sps->profile_idc;
2420 switch (sps->profile_idc) {
2421 case FF_PROFILE_H264_BASELINE:
2422 // constraint_set1_flag set to 1
2423 profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2425 case FF_PROFILE_H264_HIGH_10:
2426 case FF_PROFILE_H264_HIGH_422:
2427 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2428 // constraint_set3_flag set to 1
2429 profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
2436 static int h264_set_parameter_from_sps(H264Context *h)
2438 MpegEncContext *s = &h->s;
2440 if (s->flags & CODEC_FLAG_LOW_DELAY ||
2441 (h->sps.bitstream_restriction_flag &&
2442 !h->sps.num_reorder_frames)) {
2443 if (s->avctx->has_b_frames > 1 || h->delayed_pic[0])
2444 av_log(h->s.avctx, AV_LOG_WARNING, "Delayed frames seen. "
2445 "Reenabling low delay requires a codec flush.\n");
2450 if (s->avctx->has_b_frames < 2)
2451 s->avctx->has_b_frames = !s->low_delay;
2453 if (s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
2454 h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
2455 if (s->avctx->codec &&
2456 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU &&
2457 (h->sps.bit_depth_luma != 8 || h->sps.chroma_format_idc > 1)) {
2458 av_log(s->avctx, AV_LOG_ERROR,
2459 "VDPAU decoding does not support video colorspace.\n");
2460 return AVERROR_INVALIDDATA;
2462 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 14 &&
2463 h->sps.bit_depth_luma != 11 && h->sps.bit_depth_luma != 13 &&
2464 (h->sps.bit_depth_luma != 9 || !CHROMA422)) {
2465 s->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
2466 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
2467 h->pixel_shift = h->sps.bit_depth_luma > 8;
2469 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
2470 h->sps.chroma_format_idc);
2471 ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
2472 ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma,
2473 h->sps.chroma_format_idc);
2474 s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
2475 ff_dsputil_init(&s->dsp, s->avctx);
2476 ff_videodsp_init(&s->vdsp, h->sps.bit_depth_luma);
2478 av_log(s->avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n",
2479 h->sps.bit_depth_luma);
2480 return AVERROR_INVALIDDATA;
2486 static enum PixelFormat get_pixel_format(H264Context *h)
2488 MpegEncContext *const s = &h->s;
2489 switch (h->sps.bit_depth_luma) {
2492 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2493 return AV_PIX_FMT_GBRP9;
2495 return AV_PIX_FMT_YUV444P9;
2496 } else if (CHROMA422)
2497 return AV_PIX_FMT_YUV422P9;
2499 return AV_PIX_FMT_YUV420P9;
2503 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2504 return AV_PIX_FMT_GBRP10;
2506 return AV_PIX_FMT_YUV444P10;
2507 } else if (CHROMA422)
2508 return AV_PIX_FMT_YUV422P10;
2510 return AV_PIX_FMT_YUV420P10;
2514 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2515 return AV_PIX_FMT_GBRP12;
2517 return AV_PIX_FMT_YUV444P12;
2518 } else if (CHROMA422)
2519 return AV_PIX_FMT_YUV422P12;
2521 return AV_PIX_FMT_YUV420P12;
2525 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2526 return AV_PIX_FMT_GBRP14;
2528 return AV_PIX_FMT_YUV444P14;
2529 } else if (CHROMA422)
2530 return AV_PIX_FMT_YUV422P14;
2532 return AV_PIX_FMT_YUV420P14;
2536 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2537 av_log(h->s.avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
2538 return AV_PIX_FMT_GBR24P;
2539 } else if (s->avctx->colorspace == AVCOL_SPC_YCGCO) {
2540 av_log(h->s.avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
2542 return s->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
2543 : AV_PIX_FMT_YUV444P;
2544 } else if (CHROMA422) {
2545 return s->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
2546 : AV_PIX_FMT_YUV422P;
2548 return s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts ?
2549 s->avctx->codec->pix_fmts :
2550 s->avctx->color_range == AVCOL_RANGE_JPEG ?
2551 hwaccel_pixfmt_list_h264_jpeg_420 :
2552 ff_hwaccel_pixfmt_list_420);
2556 av_log(s->avctx, AV_LOG_ERROR,
2557 "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
2558 return AVERROR_INVALIDDATA;
2562 static int h264_slice_header_init(H264Context *h, int reinit)
2564 MpegEncContext *const s = &h->s;
2567 if( FFALIGN(s->avctx->width , 16 ) == s->width
2568 && FFALIGN(s->avctx->height, 16*(2 - h->sps.frame_mbs_only_flag)) == s->height
2569 && !h->sps.crop_right && !h->sps.crop_bottom
2570 && (s->avctx->width != s->width || s->avctx->height && s->height)
2572 av_log(h->s.avctx, AV_LOG_DEBUG, "Using externally provided dimensions\n");
2573 s->avctx->coded_width = s->width;
2574 s->avctx->coded_height = s->height;
2576 avcodec_set_dimensions(s->avctx, s->width, s->height);
2577 s->avctx->width -= (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2578 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);
2581 s->avctx->sample_aspect_ratio = h->sps.sar;
2582 av_assert0(s->avctx->sample_aspect_ratio.den);
2584 if (h->sps.timing_info_present_flag) {
2585 int64_t den = h->sps.time_scale;
2586 if (h->x264_build < 44U)
2588 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2589 h->sps.num_units_in_tick, den, 1 << 30);
2592 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2596 if ((ret = ff_MPV_common_frame_size_change(s)) < 0) {
2597 av_log(h->s.avctx, AV_LOG_ERROR, "ff_MPV_common_frame_size_change() failed.\n");
2601 if ((ret = ff_MPV_common_init(s)) < 0) {
2602 av_log(h->s.avctx, AV_LOG_ERROR, "ff_MPV_common_init() failed.\n");
2607 h->prev_interlaced_frame = 1;
2609 init_scan_tables(h);
2610 if (ff_h264_alloc_tables(h) < 0) {
2611 av_log(h->s.avctx, AV_LOG_ERROR,
2612 "Could not allocate memory for h264\n");
2613 return AVERROR(ENOMEM);
2616 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_SLICE)) {
2617 if (context_init(h) < 0) {
2618 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2622 for (i = 1; i < s->slice_context_count; i++) {
2624 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2625 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2626 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2627 c->h264dsp = h->h264dsp;
2628 c->h264qpel = h->h264qpel;
2631 c->pixel_shift = h->pixel_shift;
2632 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
2633 init_scan_tables(c);
2634 clone_tables(c, h, i);
2637 for (i = 0; i < s->slice_context_count; i++)
2638 if (context_init(h->thread_context[i]) < 0) {
2639 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2648 * Decode a slice header.
2649 * This will also call ff_MPV_common_init() and frame_start() as needed.
2651 * @param h h264context
2652 * @param h0 h264 master context (differs from 'h' when doing sliced based
2653 * parallel decoding)
2655 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2657 static int decode_slice_header(H264Context *h, H264Context *h0)
2659 MpegEncContext *const s = &h->s;
2660 MpegEncContext *const s0 = &h0->s;
2661 unsigned int first_mb_in_slice;
2662 unsigned int pps_id;
2663 int num_ref_idx_active_override_flag, ret;
2664 unsigned int slice_type, tmp, i, j;
2665 int default_ref_list_done = 0;
2666 int last_pic_structure, last_pic_droppable;
2668 int needs_reinit = 0;
2670 s->me.qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
2671 s->me.qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
2673 first_mb_in_slice = get_ue_golomb_long(&s->gb);
2675 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
2676 if (h0->current_slice && FIELD_PICTURE) {
2680 h0->current_slice = 0;
2681 if (!s0->first_field) {
2682 if (s->current_picture_ptr && !s->droppable &&
2683 s->current_picture_ptr->owner2 == s) {
2684 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
2685 s->picture_structure == PICT_BOTTOM_FIELD);
2687 s->current_picture_ptr = NULL;
2691 slice_type = get_ue_golomb_31(&s->gb);
2692 if (slice_type > 9) {
2693 av_log(h->s.avctx, AV_LOG_ERROR,
2694 "slice type too large (%d) at %d %d\n",
2695 slice_type, s->mb_x, s->mb_y);
2698 if (slice_type > 4) {
2700 h->slice_type_fixed = 1;
2702 h->slice_type_fixed = 0;
2704 slice_type = golomb_to_pict_type[slice_type];
2705 if (slice_type == AV_PICTURE_TYPE_I ||
2706 (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
2707 default_ref_list_done = 1;
2709 h->slice_type = slice_type;
2710 h->slice_type_nos = slice_type & 3;
2712 // to make a few old functions happy, it's wrong though
2713 s->pict_type = h->slice_type;
2715 pps_id = get_ue_golomb(&s->gb);
2716 if (pps_id >= MAX_PPS_COUNT) {
2717 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id %d out of range\n", pps_id);
2720 if (!h0->pps_buffers[pps_id]) {
2721 av_log(h->s.avctx, AV_LOG_ERROR,
2722 "non-existing PPS %u referenced\n",
2726 h->pps = *h0->pps_buffers[pps_id];
2728 if (!h0->sps_buffers[h->pps.sps_id]) {
2729 av_log(h->s.avctx, AV_LOG_ERROR,
2730 "non-existing SPS %u referenced\n",
2735 if (h->pps.sps_id != h->current_sps_id ||
2736 h->context_reinitialized ||
2737 h0->sps_buffers[h->pps.sps_id]->new) {
2738 SPS *new_sps = h0->sps_buffers[h->pps.sps_id];
2740 h0->sps_buffers[h->pps.sps_id]->new = 0;
2742 if (h->sps.chroma_format_idc != new_sps->chroma_format_idc ||
2743 h->sps.bit_depth_luma != new_sps->bit_depth_luma)
2746 h->current_sps_id = h->pps.sps_id;
2747 h->sps = *h0->sps_buffers[h->pps.sps_id];
2749 if (s->mb_width != h->sps.mb_width ||
2750 s->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
2751 s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
2752 h->cur_chroma_format_idc != h->sps.chroma_format_idc
2756 if ((ret = h264_set_parameter_from_sps(h)) < 0)
2760 s->avctx->profile = ff_h264_get_profile(&h->sps);
2761 s->avctx->level = h->sps.level_idc;
2762 s->avctx->refs = h->sps.ref_frame_count;
2764 must_reinit = (s->context_initialized &&
2765 ( 16*h->sps.mb_width != s->avctx->coded_width
2766 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != s->avctx->coded_height
2767 || s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
2768 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
2769 || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio)));
2772 s->mb_width = h->sps.mb_width;
2773 s->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2775 h->b_stride = s->mb_width * 4;
2777 s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2779 s->width = 16 * s->mb_width;
2780 s->height = 16 * s->mb_height;
2782 if (h->sps.video_signal_type_present_flag) {
2783 s->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
2785 if (h->sps.colour_description_present_flag) {
2786 if (s->avctx->colorspace != h->sps.colorspace)
2788 s->avctx->color_primaries = h->sps.color_primaries;
2789 s->avctx->color_trc = h->sps.color_trc;
2790 s->avctx->colorspace = h->sps.colorspace;
2794 if (s->context_initialized &&
2800 av_log(s->avctx, AV_LOG_ERROR, "changing width/height on "
2801 "slice %d\n", h0->current_slice + 1);
2802 return AVERROR_INVALIDDATA;
2807 if ((ret = get_pixel_format(h)) < 0)
2809 s->avctx->pix_fmt = ret;
2811 av_log(h->s.avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
2812 "pix_fmt: %d\n", s->width, s->height, s->avctx->pix_fmt);
2814 if ((ret = h264_slice_header_init(h, 1)) < 0) {
2815 av_log(h->s.avctx, AV_LOG_ERROR,
2816 "h264_slice_header_init() failed\n");
2819 h->context_reinitialized = 1;
2821 if (!s->context_initialized) {
2823 av_log(h->s.avctx, AV_LOG_ERROR,
2824 "Cannot (re-)initialize context during parallel decoding.\n");
2828 if ((ret = get_pixel_format(h)) < 0)
2830 s->avctx->pix_fmt = ret;
2832 if ((ret = h264_slice_header_init(h, 0)) < 0) {
2833 av_log(h->s.avctx, AV_LOG_ERROR,
2834 "h264_slice_header_init() failed\n");
2839 if (h == h0 && h->dequant_coeff_pps != pps_id) {
2840 h->dequant_coeff_pps = pps_id;
2841 init_dequant_tables(h);
2844 h->frame_num = get_bits(&s->gb, h->sps.log2_max_frame_num);
2847 h->mb_aff_frame = 0;
2848 last_pic_structure = s0->picture_structure;
2849 last_pic_droppable = s0->droppable;
2850 s->droppable = h->nal_ref_idc == 0;
2851 if (h->sps.frame_mbs_only_flag) {
2852 s->picture_structure = PICT_FRAME;
2854 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
2855 av_log(h->s.avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
2858 if (get_bits1(&s->gb)) { // field_pic_flag
2859 s->picture_structure = PICT_TOP_FIELD + get_bits1(&s->gb); // bottom_field_flag
2861 s->picture_structure = PICT_FRAME;
2862 h->mb_aff_frame = h->sps.mb_aff;
2865 h->mb_field_decoding_flag = s->picture_structure != PICT_FRAME;
2867 if (h0->current_slice != 0) {
2868 if (last_pic_structure != s->picture_structure ||
2869 last_pic_droppable != s->droppable) {
2870 av_log(h->s.avctx, AV_LOG_ERROR,
2871 "Changing field mode (%d -> %d) between slices is not allowed\n",
2872 last_pic_structure, s->picture_structure);
2873 s->picture_structure = last_pic_structure;
2874 s->droppable = last_pic_droppable;
2875 return AVERROR_INVALIDDATA;
2876 } else if (!s0->current_picture_ptr) {
2877 av_log(s->avctx, AV_LOG_ERROR,
2878 "unset current_picture_ptr on %d. slice\n",
2879 h0->current_slice + 1);
2880 return AVERROR_INVALIDDATA;
2883 /* Shorten frame num gaps so we don't have to allocate reference
2884 * frames just to throw them away */
2885 if (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0) {
2886 int unwrap_prev_frame_num = h->prev_frame_num;
2887 int max_frame_num = 1 << h->sps.log2_max_frame_num;
2889 if (unwrap_prev_frame_num > h->frame_num)
2890 unwrap_prev_frame_num -= max_frame_num;
2892 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2893 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2894 if (unwrap_prev_frame_num < 0)
2895 unwrap_prev_frame_num += max_frame_num;
2897 h->prev_frame_num = unwrap_prev_frame_num;
2901 /* See if we have a decoded first field looking for a pair...
2902 * Here, we're using that to see if we should mark previously
2903 * decode frames as "finished".
2904 * We have to do that before the "dummy" in-between frame allocation,
2905 * since that can modify s->current_picture_ptr. */
2906 if (s0->first_field) {
2907 assert(s0->current_picture_ptr);
2908 assert(s0->current_picture_ptr->f.data[0]);
2909 assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2911 /* Mark old field/frame as completed */
2912 if (!last_pic_droppable && s0->current_picture_ptr->owner2 == s0) {
2913 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2914 last_pic_structure == PICT_BOTTOM_FIELD);
2917 /* figure out if we have a complementary field pair */
2918 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2919 /* Previous field is unmatched. Don't display it, but let it
2920 * remain for reference if marked as such. */
2921 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
2922 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2923 last_pic_structure == PICT_TOP_FIELD);
2926 if (s0->current_picture_ptr->frame_num != h->frame_num) {
2927 /* This and previous field were reference, but had
2928 * different frame_nums. Consider this field first in
2929 * pair. Throw away previous field except for reference
2931 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
2932 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2933 last_pic_structure == PICT_TOP_FIELD);
2936 /* Second field in complementary pair */
2937 if (!((last_pic_structure == PICT_TOP_FIELD &&
2938 s->picture_structure == PICT_BOTTOM_FIELD) ||
2939 (last_pic_structure == PICT_BOTTOM_FIELD &&
2940 s->picture_structure == PICT_TOP_FIELD))) {
2941 av_log(s->avctx, AV_LOG_ERROR,
2942 "Invalid field mode combination %d/%d\n",
2943 last_pic_structure, s->picture_structure);
2944 s->picture_structure = last_pic_structure;
2945 s->droppable = last_pic_droppable;
2946 return AVERROR_INVALIDDATA;
2947 } else if (last_pic_droppable != s->droppable) {
2948 av_log(s->avctx, AV_LOG_ERROR,
2949 "Cannot combine reference and non-reference fields in the same frame\n");
2950 av_log_ask_for_sample(s->avctx, NULL);
2951 s->picture_structure = last_pic_structure;
2952 s->droppable = last_pic_droppable;
2953 return AVERROR_PATCHWELCOME;
2956 /* Take ownership of this buffer. Note that if another thread owned
2957 * the first field of this buffer, we're not operating on that pointer,
2958 * so the original thread is still responsible for reporting progress
2959 * on that first field (or if that was us, we just did that above).
2960 * By taking ownership, we assign responsibility to ourselves to
2961 * report progress on the second field. */
2962 s0->current_picture_ptr->owner2 = s0;
2967 while (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0 && !s0->first_field &&
2968 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
2969 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2970 av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
2971 h->frame_num, h->prev_frame_num);
2972 if (ff_h264_frame_start(h) < 0)
2974 h->prev_frame_num++;
2975 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
2976 s->current_picture_ptr->frame_num = h->prev_frame_num;
2977 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
2978 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 1);
2979 if ((ret = ff_generate_sliding_window_mmcos(h, 1)) < 0 &&
2980 s->avctx->err_recognition & AV_EF_EXPLODE)
2982 if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2983 (s->avctx->err_recognition & AV_EF_EXPLODE))
2984 return AVERROR_INVALIDDATA;
2985 /* Error concealment: if a ref is missing, copy the previous ref in its place.
2986 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2987 * about there being no actual duplicates.
2988 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
2989 * concealing a lost frame, this probably isn't noticeable by comparison, but it should
2991 if (h->short_ref_count) {
2993 av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2994 (const uint8_t **)prev->f.data, prev->f.linesize,
2995 s->avctx->pix_fmt, s->mb_width * 16, s->mb_height * 16);
2996 h->short_ref[0]->poc = prev->poc + 2;
2998 h->short_ref[0]->frame_num = h->prev_frame_num;
3002 /* See if we have a decoded first field looking for a pair...
3003 * We're using that to see whether to continue decoding in that
3004 * frame, or to allocate a new one. */
3005 if (s0->first_field) {
3006 assert(s0->current_picture_ptr);
3007 assert(s0->current_picture_ptr->f.data[0]);
3008 assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
3010 /* figure out if we have a complementary field pair */
3011 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
3012 /* Previous field is unmatched. Don't display it, but let it
3013 * remain for reference if marked as such. */
3014 s0->current_picture_ptr = NULL;
3015 s0->first_field = FIELD_PICTURE;
3017 if (s0->current_picture_ptr->frame_num != h->frame_num) {
3018 ff_thread_report_progress((AVFrame*)s0->current_picture_ptr, INT_MAX,
3019 s0->picture_structure==PICT_BOTTOM_FIELD);
3020 /* This and the previous field had different frame_nums.
3021 * Consider this field first in pair. Throw away previous
3022 * one except for reference purposes. */
3023 s0->first_field = 1;
3024 s0->current_picture_ptr = NULL;
3026 /* Second field in complementary pair */
3027 s0->first_field = 0;
3031 /* Frame or first field in a potentially complementary pair */
3032 s0->first_field = FIELD_PICTURE;
3035 if (!FIELD_PICTURE || s0->first_field) {
3036 if (ff_h264_frame_start(h) < 0) {
3037 s0->first_field = 0;
3041 ff_release_unused_pictures(s, 0);
3044 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
3047 s->current_picture_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
3049 av_assert1(s->mb_num == s->mb_width * s->mb_height);
3050 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
3051 first_mb_in_slice >= s->mb_num) {
3052 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3055 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3056 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
3057 if (s->picture_structure == PICT_BOTTOM_FIELD)
3058 s->resync_mb_y = s->mb_y = s->mb_y + 1;
3059 av_assert1(s->mb_y < s->mb_height);
3061 if (s->picture_structure == PICT_FRAME) {
3062 h->curr_pic_num = h->frame_num;
3063 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
3065 h->curr_pic_num = 2 * h->frame_num + 1;
3066 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
3069 if (h->nal_unit_type == NAL_IDR_SLICE)
3070 get_ue_golomb(&s->gb); /* idr_pic_id */
3072 if (h->sps.poc_type == 0) {
3073 h->poc_lsb = get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3075 if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
3076 h->delta_poc_bottom = get_se_golomb(&s->gb);
3079 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
3080 h->delta_poc[0] = get_se_golomb(&s->gb);
3082 if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
3083 h->delta_poc[1] = get_se_golomb(&s->gb);
3088 if (h->pps.redundant_pic_cnt_present)
3089 h->redundant_pic_count = get_ue_golomb(&s->gb);
3091 // set defaults, might be overridden a few lines later
3092 h->ref_count[0] = h->pps.ref_count[0];
3093 h->ref_count[1] = h->pps.ref_count[1];
3095 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
3097 max[0] = max[1] = s->picture_structure == PICT_FRAME ? 15 : 31;
3099 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
3100 h->direct_spatial_mv_pred = get_bits1(&s->gb);
3101 num_ref_idx_active_override_flag = get_bits1(&s->gb);
3103 if (num_ref_idx_active_override_flag) {
3104 h->ref_count[0] = get_ue_golomb(&s->gb) + 1;
3105 if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
3106 h->ref_count[1] = get_ue_golomb(&s->gb) + 1;
3108 // full range is spec-ok in this case, even for frames
3109 h->ref_count[1] = 1;
3112 if (h->ref_count[0]-1 > max[0] || h->ref_count[1]-1 > max[1]){
3113 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]);
3114 h->ref_count[0] = h->ref_count[1] = 1;
3115 return AVERROR_INVALIDDATA;
3118 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
3123 h->ref_count[1]= h->ref_count[0]= h->list_count= 0;
3125 if (!default_ref_list_done)
3126 ff_h264_fill_default_ref_list(h);
3128 if (h->slice_type_nos != AV_PICTURE_TYPE_I &&
3129 ff_h264_decode_ref_pic_list_reordering(h) < 0) {
3130 h->ref_count[1] = h->ref_count[0] = 0;
3134 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
3135 s->last_picture_ptr = &h->ref_list[0][0];
3136 s->last_picture_ptr->owner2 = s;
3137 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3139 if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
3140 s->next_picture_ptr = &h->ref_list[1][0];
3141 s->next_picture_ptr->owner2 = s;
3142 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3145 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
3146 (h->pps.weighted_bipred_idc == 1 &&
3147 h->slice_type_nos == AV_PICTURE_TYPE_B))
3148 pred_weight_table(h);
3149 else if (h->pps.weighted_bipred_idc == 2 &&
3150 h->slice_type_nos == AV_PICTURE_TYPE_B) {
3151 implicit_weight_table(h, -1);
3154 for (i = 0; i < 2; i++) {
3155 h->luma_weight_flag[i] = 0;
3156 h->chroma_weight_flag[i] = 0;
3160 // If frame-mt is enabled, only update mmco tables for the first slice
3161 // in a field. Subsequent slices can temporarily clobber h->mmco_index
3162 // or h->mmco, which will cause ref list mix-ups and decoding errors
3163 // further down the line. This may break decoding if the first slice is
3164 // corrupt, thus we only do this if frame-mt is enabled.
3165 if (h->nal_ref_idc &&
3166 ff_h264_decode_ref_pic_marking(h0, &s->gb,
3167 !(s->avctx->active_thread_type & FF_THREAD_FRAME) ||
3168 h0->current_slice == 0) < 0 &&
3169 (s->avctx->err_recognition & AV_EF_EXPLODE))
3170 return AVERROR_INVALIDDATA;
3173 ff_h264_fill_mbaff_ref_list(h);
3175 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
3176 implicit_weight_table(h, 0);
3177 implicit_weight_table(h, 1);
3181 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3182 ff_h264_direct_dist_scale_factor(h);
3183 ff_h264_direct_ref_list_init(h);
3185 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
3186 tmp = get_ue_golomb_31(&s->gb);
3188 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3191 h->cabac_init_idc = tmp;
3194 h->last_qscale_diff = 0;
3195 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3196 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
3197 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3201 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3202 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3203 // FIXME qscale / qp ... stuff
3204 if (h->slice_type == AV_PICTURE_TYPE_SP)
3205 get_bits1(&s->gb); /* sp_for_switch_flag */
3206 if (h->slice_type == AV_PICTURE_TYPE_SP ||
3207 h->slice_type == AV_PICTURE_TYPE_SI)
3208 get_se_golomb(&s->gb); /* slice_qs_delta */
3210 h->deblocking_filter = 1;
3211 h->slice_alpha_c0_offset = 52;
3212 h->slice_beta_offset = 52;
3213 if (h->pps.deblocking_filter_parameters_present) {
3214 tmp = get_ue_golomb_31(&s->gb);
3216 av_log(s->avctx, AV_LOG_ERROR,
3217 "deblocking_filter_idc %u out of range\n", tmp);
3220 h->deblocking_filter = tmp;
3221 if (h->deblocking_filter < 2)
3222 h->deblocking_filter ^= 1; // 1<->0
3224 if (h->deblocking_filter) {
3225 h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
3226 h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
3227 if (h->slice_alpha_c0_offset > 104U ||
3228 h->slice_beta_offset > 104U) {
3229 av_log(s->avctx, AV_LOG_ERROR,
3230 "deblocking filter parameters %d %d out of range\n",
3231 h->slice_alpha_c0_offset, h->slice_beta_offset);
3237 if (s->avctx->skip_loop_filter >= AVDISCARD_ALL ||
3238 (s->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
3239 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
3240 (s->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
3241 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
3242 (s->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
3243 h->nal_ref_idc == 0))
3244 h->deblocking_filter = 0;
3246 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
3247 if (s->avctx->flags2 & CODEC_FLAG2_FAST) {
3248 /* Cheat slightly for speed:
3249 * Do not bother to deblock across slices. */
3250 h->deblocking_filter = 2;
3252 h0->max_contexts = 1;
3253 if (!h0->single_decode_warning) {
3254 av_log(s->avctx, AV_LOG_INFO,
3255 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3256 h0->single_decode_warning = 1;
3259 av_log(h->s.avctx, AV_LOG_ERROR,
3260 "Deblocking switched inside frame.\n");
3265 h->qp_thresh = 15 + 52 -
3266 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
3268 h->pps.chroma_qp_index_offset[0],
3269 h->pps.chroma_qp_index_offset[1]) +
3270 6 * (h->sps.bit_depth_luma - 8);
3272 h0->last_slice_type = slice_type;
3273 h->slice_num = ++h0->current_slice;
3276 h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= s->resync_mb_y;
3277 if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= s->resync_mb_y
3278 && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= s->resync_mb_y
3279 && h->slice_num >= MAX_SLICES) {
3280 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
3281 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);
3284 for (j = 0; j < 2; j++) {
3286 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
3287 for (i = 0; i < 16; i++) {
3289 if (h->ref_list[j][i].f.data[0]) {
3291 uint8_t *base = h->ref_list[j][i].f.base[0];
3292 for (k = 0; k < h->short_ref_count; k++)
3293 if (h->short_ref[k]->f.base[0] == base) {
3297 for (k = 0; k < h->long_ref_count; k++)
3298 if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
3299 id_list[i] = h->short_ref_count + k;
3307 for (i = 0; i < 16; i++)
3308 ref2frm[i + 2] = 4 * id_list[i] +
3309 (h->ref_list[j][i].f.reference & 3);
3311 ref2frm[18 + 1] = -1;
3312 for (i = 16; i < 48; i++)
3313 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
3314 (h->ref_list[j][i].f.reference & 3);
3317 // FIXME: fix draw_edges + PAFF + frame threads
3318 h->emu_edge_width = (s->flags & CODEC_FLAG_EMU_EDGE ||
3319 (!h->sps.frame_mbs_only_flag &&
3320 s->avctx->active_thread_type))
3322 h->emu_edge_height = (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3324 if (s->avctx->debug & FF_DEBUG_PICT_INFO) {
3325 av_log(h->s.avctx, AV_LOG_DEBUG,
3326 "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",
3328 (s->picture_structure == PICT_FRAME ? "F" : s->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
3330 av_get_picture_type_char(h->slice_type),
3331 h->slice_type_fixed ? " fix" : "",
3332 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3333 pps_id, h->frame_num,
3334 s->current_picture_ptr->field_poc[0],
3335 s->current_picture_ptr->field_poc[1],
3336 h->ref_count[0], h->ref_count[1],
3338 h->deblocking_filter,
3339 h->slice_alpha_c0_offset / 2 - 26, h->slice_beta_offset / 2 - 26,
3341 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
3342 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
3348 int ff_h264_get_slice_type(const H264Context *h)
3350 switch (h->slice_type) {
3351 case AV_PICTURE_TYPE_P:
3353 case AV_PICTURE_TYPE_B:
3355 case AV_PICTURE_TYPE_I:
3357 case AV_PICTURE_TYPE_SP:
3359 case AV_PICTURE_TYPE_SI:
3366 static av_always_inline void fill_filter_caches_inter(H264Context *h,
3367 MpegEncContext *const s,
3368 int mb_type, int top_xy,
3369 int left_xy[LEFT_MBS],
3371 int left_type[LEFT_MBS],
3372 int mb_xy, int list)
3374 int b_stride = h->b_stride;
3375 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3376 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3377 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
3378 if (USES_LIST(top_type, list)) {
3379 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
3380 const int b8_xy = 4 * top_xy + 2;
3381 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3382 AV_COPY128(mv_dst - 1 * 8, s->current_picture.f.motion_val[list][b_xy + 0]);
3383 ref_cache[0 - 1 * 8] =
3384 ref_cache[1 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3385 ref_cache[2 - 1 * 8] =
3386 ref_cache[3 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3388 AV_ZERO128(mv_dst - 1 * 8);
3389 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3392 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
3393 if (USES_LIST(left_type[LTOP], list)) {
3394 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
3395 const int b8_xy = 4 * left_xy[LTOP] + 1;
3396 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3397 AV_COPY32(mv_dst - 1 + 0, s->current_picture.f.motion_val[list][b_xy + b_stride * 0]);
3398 AV_COPY32(mv_dst - 1 + 8, s->current_picture.f.motion_val[list][b_xy + b_stride * 1]);
3399 AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride * 2]);
3400 AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride * 3]);
3402 ref_cache[-1 + 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 0]];
3403 ref_cache[-1 + 16] =
3404 ref_cache[-1 + 24] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 1]];
3406 AV_ZERO32(mv_dst - 1 + 0);
3407 AV_ZERO32(mv_dst - 1 + 8);
3408 AV_ZERO32(mv_dst - 1 + 16);
3409 AV_ZERO32(mv_dst - 1 + 24);
3412 ref_cache[-1 + 16] =
3413 ref_cache[-1 + 24] = LIST_NOT_USED;
3418 if (!USES_LIST(mb_type, list)) {
3419 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
3420 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3421 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3422 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3423 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3428 int8_t *ref = &s->current_picture.f.ref_index[list][4 * mb_xy];
3429 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3430 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
3431 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
3432 AV_WN32A(&ref_cache[0 * 8], ref01);
3433 AV_WN32A(&ref_cache[1 * 8], ref01);
3434 AV_WN32A(&ref_cache[2 * 8], ref23);
3435 AV_WN32A(&ref_cache[3 * 8], ref23);
3439 int16_t(*mv_src)[2] = &s->current_picture.f.motion_val[list][4 * s->mb_x + 4 * s->mb_y * b_stride];
3440 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
3441 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
3442 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
3443 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
3449 * @return non zero if the loop filter can be skipped
3451 static int fill_filter_caches(H264Context *h, int mb_type)
3453 MpegEncContext *const s = &h->s;
3454 const int mb_xy = h->mb_xy;
3455 int top_xy, left_xy[LEFT_MBS];
3456 int top_type, left_type[LEFT_MBS];
3460 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
3462 /* Wow, what a mess, why didn't they simplify the interlacing & intra
3463 * stuff, I can't imagine that these complex rules are worth it. */
3465 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
3467 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
3468 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
3470 if (left_mb_field_flag != curr_mb_field_flag)
3471 left_xy[LTOP] -= s->mb_stride;
3473 if (curr_mb_field_flag)
3474 top_xy += s->mb_stride &
3475 (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
3476 if (left_mb_field_flag != curr_mb_field_flag)
3477 left_xy[LBOT] += s->mb_stride;
3481 h->top_mb_xy = top_xy;
3482 h->left_mb_xy[LTOP] = left_xy[LTOP];
3483 h->left_mb_xy[LBOT] = left_xy[LBOT];
3485 /* For sufficiently low qp, filtering wouldn't do anything.
3486 * This is a conservative estimate: could also check beta_offset
3487 * and more accurate chroma_qp. */
3488 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
3489 int qp = s->current_picture.f.qscale_table[mb_xy];
3490 if (qp <= qp_thresh &&
3491 (left_xy[LTOP] < 0 ||
3492 ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
3494 ((qp + s->current_picture.f.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
3497 if ((left_xy[LTOP] < 0 ||
3498 ((qp + s->current_picture.f.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
3499 (top_xy < s->mb_stride ||
3500 ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
3505 top_type = s->current_picture.f.mb_type[top_xy];
3506 left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
3507 left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
3508 if (h->deblocking_filter == 2) {
3509 if (h->slice_table[top_xy] != h->slice_num)
3511 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
3512 left_type[LTOP] = left_type[LBOT] = 0;
3514 if (h->slice_table[top_xy] == 0xFFFF)
3516 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
3517 left_type[LTOP] = left_type[LBOT] = 0;
3519 h->top_type = top_type;
3520 h->left_type[LTOP] = left_type[LTOP];
3521 h->left_type[LBOT] = left_type[LBOT];
3523 if (IS_INTRA(mb_type))
3526 fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy,
3527 top_type, left_type, mb_xy, 0);
3528 if (h->list_count == 2)
3529 fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy,
3530 top_type, left_type, mb_xy, 1);
3532 nnz = h->non_zero_count[mb_xy];
3533 nnz_cache = h->non_zero_count_cache;
3534 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
3535 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
3536 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
3537 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
3538 h->cbp = h->cbp_table[mb_xy];
3541 nnz = h->non_zero_count[top_xy];
3542 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
3545 if (left_type[LTOP]) {
3546 nnz = h->non_zero_count[left_xy[LTOP]];
3547 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
3548 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
3549 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
3550 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
3553 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
3554 * from what the loop filter needs */
3555 if (!CABAC && h->pps.transform_8x8_mode) {
3556 if (IS_8x8DCT(top_type)) {
3557 nnz_cache[4 + 8 * 0] =
3558 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
3559 nnz_cache[6 + 8 * 0] =
3560 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
3562 if (IS_8x8DCT(left_type[LTOP])) {
3563 nnz_cache[3 + 8 * 1] =
3564 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
3566 if (IS_8x8DCT(left_type[LBOT])) {
3567 nnz_cache[3 + 8 * 3] =
3568 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
3571 if (IS_8x8DCT(mb_type)) {
3572 nnz_cache[scan8[0]] =
3573 nnz_cache[scan8[1]] =
3574 nnz_cache[scan8[2]] =
3575 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
3577 nnz_cache[scan8[0 + 4]] =
3578 nnz_cache[scan8[1 + 4]] =
3579 nnz_cache[scan8[2 + 4]] =
3580 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
3582 nnz_cache[scan8[0 + 8]] =
3583 nnz_cache[scan8[1 + 8]] =
3584 nnz_cache[scan8[2 + 8]] =
3585 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
3587 nnz_cache[scan8[0 + 12]] =
3588 nnz_cache[scan8[1 + 12]] =
3589 nnz_cache[scan8[2 + 12]] =
3590 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
3597 static void loop_filter(H264Context *h, int start_x, int end_x)
3599 MpegEncContext *const s = &h->s;
3600 uint8_t *dest_y, *dest_cb, *dest_cr;
3601 int linesize, uvlinesize, mb_x, mb_y;
3602 const int end_mb_y = s->mb_y + FRAME_MBAFF;
3603 const int old_slice_type = h->slice_type;
3604 const int pixel_shift = h->pixel_shift;
3605 const int block_h = 16 >> s->chroma_y_shift;
3607 if (h->deblocking_filter) {
3608 for (mb_x = start_x; mb_x < end_x; mb_x++)
3609 for (mb_y = end_mb_y - FRAME_MBAFF; mb_y <= end_mb_y; mb_y++) {
3611 mb_xy = h->mb_xy = mb_x + mb_y * s->mb_stride;
3612 h->slice_num = h->slice_table[mb_xy];
3613 mb_type = s->current_picture.f.mb_type[mb_xy];
3614 h->list_count = h->list_counts[mb_xy];
3618 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3622 dest_y = s->current_picture.f.data[0] +
3623 ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
3624 dest_cb = s->current_picture.f.data[1] +
3625 (mb_x << pixel_shift) * (8 << CHROMA444) +
3626 mb_y * s->uvlinesize * block_h;
3627 dest_cr = s->current_picture.f.data[2] +
3628 (mb_x << pixel_shift) * (8 << CHROMA444) +
3629 mb_y * s->uvlinesize * block_h;
3630 // FIXME simplify above
3633 linesize = h->mb_linesize = s->linesize * 2;
3634 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3635 if (mb_y & 1) { // FIXME move out of this function?
3636 dest_y -= s->linesize * 15;
3637 dest_cb -= s->uvlinesize * (block_h - 1);
3638 dest_cr -= s->uvlinesize * (block_h - 1);
3641 linesize = h->mb_linesize = s->linesize;
3642 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3644 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
3646 if (fill_filter_caches(h, mb_type))
3648 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
3649 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
3652 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
3653 linesize, uvlinesize);
3655 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
3656 dest_cr, linesize, uvlinesize);
3660 h->slice_type = old_slice_type;
3662 s->mb_y = end_mb_y - FRAME_MBAFF;
3663 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3664 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3667 static void predict_field_decoding_flag(H264Context *h)
3669 MpegEncContext *const s = &h->s;
3670 const int mb_xy = s->mb_x + s->mb_y * s->mb_stride;
3671 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
3672 s->current_picture.f.mb_type[mb_xy - 1] :
3673 (h->slice_table[mb_xy - s->mb_stride] == h->slice_num) ?
3674 s->current_picture.f.mb_type[mb_xy - s->mb_stride] : 0;
3675 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3679 * Draw edges and report progress for the last MB row.
3681 static void decode_finish_row(H264Context *h)
3683 MpegEncContext *const s = &h->s;
3684 int top = 16 * (s->mb_y >> FIELD_PICTURE);
3685 int pic_height = 16 * s->mb_height >> FIELD_PICTURE;
3686 int height = 16 << FRAME_MBAFF;
3687 int deblock_border = (16 + 4) << FRAME_MBAFF;
3689 if (h->deblocking_filter) {
3690 if ((top + height) >= pic_height)
3691 height += deblock_border;
3692 top -= deblock_border;
3695 if (top >= pic_height || (top + height) < h->emu_edge_height)
3698 height = FFMIN(height, pic_height - top);
3699 if (top < h->emu_edge_height) {
3700 height = top + height;
3704 ff_draw_horiz_band(s, top, height);
3709 ff_thread_report_progress(&s->current_picture_ptr->f, top + height - 1,
3710 s->picture_structure == PICT_BOTTOM_FIELD);
3713 static int decode_slice(struct AVCodecContext *avctx, void *arg)
3715 H264Context *h = *(void **)arg;
3716 MpegEncContext *const s = &h->s;
3717 const int part_mask = s->partitioned_frame ? (ER_AC_END | ER_AC_ERROR)
3719 int lf_x_start = s->mb_x;
3721 s->mb_skip_run = -1;
3723 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * s->linesize * ((scan8[15] - scan8[0]) >> 3));
3725 h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME ||
3726 s->codec_id != AV_CODEC_ID_H264 ||
3727 (CONFIG_GRAY && (s->flags & CODEC_FLAG_GRAY));
3731 align_get_bits(&s->gb);
3734 ff_init_cabac_decoder(&h->cabac,
3735 s->gb.buffer + get_bits_count(&s->gb) / 8,
3736 (get_bits_left(&s->gb) + 7) / 8);
3738 ff_h264_init_cabac_states(h);
3742 int ret = ff_h264_decode_mb_cabac(h);
3744 // STOP_TIMER("decode_mb_cabac")
3747 ff_h264_hl_decode_mb(h);
3749 // FIXME optimal? or let mb_decode decode 16x32 ?
3750 if (ret >= 0 && FRAME_MBAFF) {
3753 ret = ff_h264_decode_mb_cabac(h);
3756 ff_h264_hl_decode_mb(h);
3759 eos = get_cabac_terminate(&h->cabac);
3761 if ((s->workaround_bugs & FF_BUG_TRUNCATED) &&
3762 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3763 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x - 1,
3764 s->mb_y, ER_MB_END & part_mask);
3765 if (s->mb_x >= lf_x_start)
3766 loop_filter(h, lf_x_start, s->mb_x + 1);
3769 if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
3770 av_log(h->s.avctx, AV_LOG_DEBUG, "bytestream overread %td\n", h->cabac.bytestream_end - h->cabac.bytestream);
3771 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
3772 av_log(h->s.avctx, AV_LOG_ERROR,
3773 "error while decoding MB %d %d, bytestream (%td)\n",
3775 h->cabac.bytestream_end - h->cabac.bytestream);
3776 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3777 s->mb_y, ER_MB_ERROR & part_mask);
3781 if (++s->mb_x >= s->mb_width) {
3782 loop_filter(h, lf_x_start, s->mb_x);
3783 s->mb_x = lf_x_start = 0;
3784 decode_finish_row(h);
3786 if (FIELD_OR_MBAFF_PICTURE) {
3788 if (FRAME_MBAFF && s->mb_y < s->mb_height)
3789 predict_field_decoding_flag(h);
3793 if (eos || s->mb_y >= s->mb_height) {
3794 tprintf(s->avctx, "slice end %d %d\n",
3795 get_bits_count(&s->gb), s->gb.size_in_bits);
3796 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x - 1,
3797 s->mb_y, ER_MB_END & part_mask);
3798 if (s->mb_x > lf_x_start)
3799 loop_filter(h, lf_x_start, s->mb_x);
3805 int ret = ff_h264_decode_mb_cavlc(h);
3808 ff_h264_hl_decode_mb(h);
3810 // FIXME optimal? or let mb_decode decode 16x32 ?
3811 if (ret >= 0 && FRAME_MBAFF) {
3813 ret = ff_h264_decode_mb_cavlc(h);
3816 ff_h264_hl_decode_mb(h);
3821 av_log(h->s.avctx, AV_LOG_ERROR,
3822 "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3823 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3824 s->mb_y, ER_MB_ERROR & part_mask);
3828 if (++s->mb_x >= s->mb_width) {
3829 loop_filter(h, lf_x_start, s->mb_x);
3830 s->mb_x = lf_x_start = 0;
3831 decode_finish_row(h);
3833 if (FIELD_OR_MBAFF_PICTURE) {
3835 if (FRAME_MBAFF && s->mb_y < s->mb_height)
3836 predict_field_decoding_flag(h);
3838 if (s->mb_y >= s->mb_height) {
3839 tprintf(s->avctx, "slice end %d %d\n",
3840 get_bits_count(&s->gb), s->gb.size_in_bits);
3842 if ( get_bits_left(&s->gb) == 0
3843 || get_bits_left(&s->gb) > 0 && !(s->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
3844 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3845 s->mb_x - 1, s->mb_y,
3846 ER_MB_END & part_mask);
3850 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3852 ER_MB_END & part_mask);
3859 if (get_bits_left(&s->gb) <= 0 && s->mb_skip_run <= 0) {
3860 tprintf(s->avctx, "slice end %d %d\n",
3861 get_bits_count(&s->gb), s->gb.size_in_bits);
3862 if (get_bits_left(&s->gb) == 0) {
3863 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3864 s->mb_x - 1, s->mb_y,
3865 ER_MB_END & part_mask);
3866 if (s->mb_x > lf_x_start)
3867 loop_filter(h, lf_x_start, s->mb_x);
3871 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3872 s->mb_y, ER_MB_ERROR & part_mask);
3882 * Call decode_slice() for each context.
3884 * @param h h264 master context
3885 * @param context_count number of contexts to execute
3887 static int execute_decode_slices(H264Context *h, int context_count)
3889 MpegEncContext *const s = &h->s;
3890 AVCodecContext *const avctx = s->avctx;
3894 if (s->avctx->hwaccel ||
3895 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
3897 if (context_count == 1) {
3898 return decode_slice(avctx, &h);
3900 for (i = 1; i < context_count; i++) {
3901 hx = h->thread_context[i];
3902 hx->s.err_recognition = avctx->err_recognition;
3903 hx->s.error_count = 0;
3904 hx->x264_build = h->x264_build;
3907 avctx->execute(avctx, decode_slice, h->thread_context,
3908 NULL, context_count, sizeof(void *));
3910 /* pull back stuff from slices to master context */
3911 hx = h->thread_context[context_count - 1];
3912 s->mb_x = hx->s.mb_x;
3913 s->mb_y = hx->s.mb_y;
3914 s->droppable = hx->s.droppable;
3915 s->picture_structure = hx->s.picture_structure;
3916 for (i = 1; i < context_count; i++)
3917 h->s.error_count += h->thread_context[i]->s.error_count;
3923 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
3924 int parse_extradata)
3926 MpegEncContext *const s = &h->s;
3927 AVCodecContext *const avctx = s->avctx;
3928 H264Context *hx; ///< thread context
3932 int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3933 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
3936 int first_slice = 0;
3938 h->nal_unit_type= 0;
3940 if(!s->slice_context_count)
3941 s->slice_context_count= 1;
3942 h->max_contexts = s->slice_context_count;
3943 if (!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3944 h->current_slice = 0;
3945 if (!s->first_field)
3946 s->current_picture_ptr = NULL;
3947 ff_h264_reset_sei(h);
3950 if (h->nal_length_size == 4) {
3951 if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
3953 }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
3957 for (; pass <= 1; pass++) {
3960 next_avc = h->is_avc ? 0 : buf_size;
3970 if (buf_index >= next_avc) {
3971 if (buf_index >= buf_size - h->nal_length_size)
3974 for (i = 0; i < h->nal_length_size; i++)
3975 nalsize = (nalsize << 8) | buf[buf_index++];
3976 if (nalsize <= 0 || nalsize > buf_size - buf_index) {
3977 av_log(h->s.avctx, AV_LOG_ERROR,
3978 "AVC: nal size %d\n", nalsize);
3981 next_avc = buf_index + nalsize;
3983 // start code prefix search
3984 for (; buf_index + 3 < next_avc; buf_index++)
3985 // This should always succeed in the first iteration.
3986 if (buf[buf_index] == 0 &&
3987 buf[buf_index + 1] == 0 &&
3988 buf[buf_index + 2] == 1)
3991 if (buf_index + 3 >= buf_size) {
3992 buf_index = buf_size;
3997 if (buf_index >= next_avc)
4001 hx = h->thread_context[context_count];
4003 ptr = ff_h264_decode_nal(hx, buf + buf_index, &dst_length,
4004 &consumed, next_avc - buf_index);
4005 if (ptr == NULL || dst_length < 0) {
4009 i = buf_index + consumed;
4010 if ((s->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
4011 buf[i] == 0x00 && buf[i + 1] == 0x00 &&
4012 buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
4013 s->workaround_bugs |= FF_BUG_TRUNCATED;
4015 if (!(s->workaround_bugs & FF_BUG_TRUNCATED))
4016 while(dst_length > 0 && ptr[dst_length - 1] == 0)
4018 bit_length = !dst_length ? 0
4020 decode_rbsp_trailing(h, ptr + dst_length - 1));
4022 if (s->avctx->debug & FF_DEBUG_STARTCODE)
4023 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);
4025 if (h->is_avc && (nalsize != consumed) && nalsize)
4026 av_log(h->s.avctx, AV_LOG_DEBUG,
4027 "AVC: Consumed only %d bytes instead of %d\n",
4030 buf_index += consumed;
4034 /* packets can sometimes contain multiple PPS/SPS,
4035 * e.g. two PAFF field pictures in one packet, or a demuxer
4036 * which splits NALs strangely if so, when frame threading we
4037 * can't start the next thread until we've read all of them */
4038 switch (hx->nal_unit_type) {
4041 nals_needed = nal_index;
4046 init_get_bits(&hx->s.gb, ptr, bit_length);
4047 if (!get_ue_golomb(&hx->s.gb) || !first_slice)
4048 nals_needed = nal_index;
4050 first_slice = hx->nal_unit_type;
4056 switch (hx->nal_unit_type) {
4060 first_slice = hx->nal_unit_type;
4063 // FIXME do not discard SEI id
4064 if (avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
4068 /* Ignore per frame NAL unit type during extradata
4069 * parsing. Decoding slices is not possible in codec init
4071 if (parse_extradata) {
4072 switch (hx->nal_unit_type) {
4078 case NAL_AUXILIARY_SLICE:
4079 av_log(h->s.avctx, AV_LOG_WARNING, "Ignoring NAL %d in global header/extradata\n", hx->nal_unit_type);
4080 hx->nal_unit_type = NAL_FF_IGNORE;
4086 switch (hx->nal_unit_type) {
4088 if (first_slice != NAL_IDR_SLICE) {
4089 av_log(h->s.avctx, AV_LOG_ERROR,
4090 "Invalid mix of idr and non-idr slices\n");
4095 idr(h); // FIXME ensure we don't lose some frames if there is reordering
4098 init_get_bits(&hx->s.gb, ptr, bit_length);
4100 hx->inter_gb_ptr = &hx->s.gb;
4101 hx->s.data_partitioning = 0;
4103 if ((err = decode_slice_header(hx, h)))
4106 if (h->sei_recovery_frame_cnt >= 0 && (h->frame_num != h->sei_recovery_frame_cnt || hx->slice_type_nos != AV_PICTURE_TYPE_I))
4107 h->valid_recovery_point = 1;
4109 if ( h->sei_recovery_frame_cnt >= 0
4110 && ( h->recovery_frame<0
4111 || ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt)) {
4112 h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) %
4113 (1 << h->sps.log2_max_frame_num);
4115 if (!h->valid_recovery_point)
4116 h->recovery_frame = h->frame_num;
4119 s->current_picture_ptr->f.key_frame |=
4120 (hx->nal_unit_type == NAL_IDR_SLICE);
4122 if (h->recovery_frame == h->frame_num) {
4123 s->current_picture_ptr->sync |= 1;
4124 h->recovery_frame = -1;
4127 h->sync |= !!s->current_picture_ptr->f.key_frame;
4128 h->sync |= 3*!!(s->flags2 & CODEC_FLAG2_SHOW_ALL);
4129 s->current_picture_ptr->sync |= h->sync;
4131 if (h->current_slice == 1) {
4132 if (!(s->flags2 & CODEC_FLAG2_CHUNKS))
4133 decode_postinit(h, nal_index >= nals_needed);
4135 if (s->avctx->hwaccel &&
4136 s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
4138 if (CONFIG_H264_VDPAU_DECODER &&
4139 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
4140 ff_vdpau_h264_picture_start(s);
4143 if (hx->redundant_pic_count == 0 &&
4144 (avctx->skip_frame < AVDISCARD_NONREF ||
4146 (avctx->skip_frame < AVDISCARD_BIDIR ||
4147 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4148 (avctx->skip_frame < AVDISCARD_NONKEY ||
4149 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4150 avctx->skip_frame < AVDISCARD_ALL) {
4151 if (avctx->hwaccel) {
4152 if (avctx->hwaccel->decode_slice(avctx,
4153 &buf[buf_index - consumed],
4156 } else if (CONFIG_H264_VDPAU_DECODER &&
4157 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) {
4158 static const uint8_t start_code[] = {
4160 ff_vdpau_add_data_chunk(s, start_code,
4161 sizeof(start_code));
4162 ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed],
4169 init_get_bits(&hx->s.gb, ptr, bit_length);
4171 hx->inter_gb_ptr = NULL;
4173 if ((err = decode_slice_header(hx, h)) < 0)
4176 hx->s.data_partitioning = 1;
4179 init_get_bits(&hx->intra_gb, ptr, bit_length);
4180 hx->intra_gb_ptr = &hx->intra_gb;
4183 init_get_bits(&hx->inter_gb, ptr, bit_length);
4184 hx->inter_gb_ptr = &hx->inter_gb;
4186 av_log(h->s.avctx, AV_LOG_ERROR, "Partitioned H.264 support is incomplete\n");
4189 if (hx->redundant_pic_count == 0 &&
4191 hx->s.data_partitioning &&
4192 s->current_picture_ptr &&
4193 s->context_initialized &&
4194 (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc) &&
4195 (avctx->skip_frame < AVDISCARD_BIDIR ||
4196 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4197 (avctx->skip_frame < AVDISCARD_NONKEY ||
4198 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4199 avctx->skip_frame < AVDISCARD_ALL)
4203 init_get_bits(&s->gb, ptr, bit_length);
4204 ff_h264_decode_sei(h);
4207 init_get_bits(&s->gb, ptr, bit_length);
4208 if (ff_h264_decode_seq_parameter_set(h) < 0 && (h->is_avc ? (nalsize != consumed) && nalsize : 1)) {
4209 av_log(h->s.avctx, AV_LOG_DEBUG,
4210 "SPS decoding failure, trying again with the complete NAL\n");
4212 av_assert0(next_avc - buf_index + consumed == nalsize);
4213 if ((next_avc - buf_index + consumed - 1) >= INT_MAX/8)
4215 init_get_bits(&s->gb, &buf[buf_index + 1 - consumed],
4216 8*(next_avc - buf_index + consumed - 1));
4217 ff_h264_decode_seq_parameter_set(h);
4222 init_get_bits(&s->gb, ptr, bit_length);
4223 ff_h264_decode_picture_parameter_set(h, bit_length);
4226 case NAL_END_SEQUENCE:
4227 case NAL_END_STREAM:
4228 case NAL_FILLER_DATA:
4230 case NAL_AUXILIARY_SLICE:
4235 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
4236 hx->nal_unit_type, bit_length);
4239 if (context_count == h->max_contexts) {
4240 execute_decode_slices(h, context_count);
4245 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
4246 else if (err == 1) {
4247 /* Slice could not be decoded in parallel mode, copy down
4248 * NAL unit stuff to context 0 and restart. Note that
4249 * rbsp_buffer is not transferred, but since we no longer
4250 * run in parallel mode this should not be an issue. */
4251 h->nal_unit_type = hx->nal_unit_type;
4252 h->nal_ref_idc = hx->nal_ref_idc;
4259 execute_decode_slices(h, context_count);
4263 if (s->current_picture_ptr && s->current_picture_ptr->owner2 == s &&
4265 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
4266 s->picture_structure == PICT_BOTTOM_FIELD);
4273 * Return the number of bytes consumed for building the current frame.
4275 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size)
4278 pos = 1; // avoid infinite loops (i doubt that is needed but ...)
4279 if (pos + 10 > buf_size)
4280 pos = buf_size; // oops ;)
4285 static int decode_frame(AVCodecContext *avctx, void *data,
4286 int *got_frame, AVPacket *avpkt)
4288 const uint8_t *buf = avpkt->data;
4289 int buf_size = avpkt->size;
4290 H264Context *h = avctx->priv_data;
4291 MpegEncContext *s = &h->s;
4292 AVFrame *pict = data;
4297 s->flags = avctx->flags;
4298 s->flags2 = avctx->flags2;
4300 /* end of stream, output what is still in the buffers */
4301 if (buf_size == 0) {
4304 s->current_picture_ptr = NULL;
4307 // FIXME factorize this with the output code below
4308 out = h->delayed_pic[0];
4311 h->delayed_pic[i] &&
4312 !h->delayed_pic[i]->f.key_frame &&
4313 !h->delayed_pic[i]->mmco_reset;
4315 if (h->delayed_pic[i]->poc < out->poc) {
4316 out = h->delayed_pic[i];
4320 for (i = out_idx; h->delayed_pic[i]; i++)
4321 h->delayed_pic[i] = h->delayed_pic[i + 1];
4324 out->f.reference &= ~DELAYED_PIC_REF;
4331 if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
4332 int cnt= buf[5]&0x1f;
4333 const uint8_t *p= buf+6;
4335 int nalsize= AV_RB16(p) + 2;
4336 if(nalsize > buf_size - (p-buf) || p[2]!=0x67)
4344 int nalsize= AV_RB16(p) + 2;
4345 if(nalsize > buf_size - (p-buf) || p[2]!=0x68)
4350 return ff_h264_decode_extradata(h, buf, buf_size);
4354 buf_index = decode_nal_units(h, buf, buf_size, 0);
4358 if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
4359 av_assert0(buf_index <= buf_size);
4363 if (!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr) {
4364 if (avctx->skip_frame >= AVDISCARD_NONREF ||
4365 buf_size >= 4 && !memcmp("Q264", buf, 4))
4367 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
4371 if (!(s->flags2 & CODEC_FLAG2_CHUNKS) ||
4372 (s->mb_y >= s->mb_height && s->mb_height)) {
4373 if (s->flags2 & CODEC_FLAG2_CHUNKS)
4374 decode_postinit(h, 1);
4377 h->context_reinitialized = 0;
4379 /* Wait for second field. */
4381 if (h->next_output_pic && (h->next_output_pic->sync || h->sync>1)) {
4383 *pict = h->next_output_pic->f;
4387 assert(pict->data[0] || !*got_frame);
4388 ff_print_debug_info(s, pict);
4390 return get_consumed_bytes(s, buf_index, buf_size);
4393 av_cold void ff_h264_free_context(H264Context *h)
4397 free_tables(h, 1); // FIXME cleanup init stuff perhaps
4399 for (i = 0; i < MAX_SPS_COUNT; i++)
4400 av_freep(h->sps_buffers + i);
4402 for (i = 0; i < MAX_PPS_COUNT; i++)
4403 av_freep(h->pps_buffers + i);
4406 static av_cold int h264_decode_end(AVCodecContext *avctx)
4408 H264Context *h = avctx->priv_data;
4409 MpegEncContext *s = &h->s;
4411 ff_h264_remove_all_refs(h);
4412 ff_h264_free_context(h);
4414 ff_MPV_common_end(s);
4416 // memset(h, 0, sizeof(H264Context));
4421 static const AVProfile profiles[] = {
4422 { FF_PROFILE_H264_BASELINE, "Baseline" },
4423 { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
4424 { FF_PROFILE_H264_MAIN, "Main" },
4425 { FF_PROFILE_H264_EXTENDED, "Extended" },
4426 { FF_PROFILE_H264_HIGH, "High" },
4427 { FF_PROFILE_H264_HIGH_10, "High 10" },
4428 { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
4429 { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
4430 { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
4431 { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
4432 { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
4433 { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
4434 { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
4435 { FF_PROFILE_UNKNOWN },
4438 static const AVOption h264_options[] = {
4439 {"is_avc", "is avc", offsetof(H264Context, is_avc), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 1, 0},
4440 {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0},
4444 static const AVClass h264_class = {
4445 .class_name = "H264 Decoder",
4446 .item_name = av_default_item_name,
4447 .option = h264_options,
4448 .version = LIBAVUTIL_VERSION_INT,
4451 static const AVClass h264_vdpau_class = {
4452 .class_name = "H264 VDPAU Decoder",
4453 .item_name = av_default_item_name,
4454 .option = h264_options,
4455 .version = LIBAVUTIL_VERSION_INT,
4458 AVCodec ff_h264_decoder = {
4460 .type = AVMEDIA_TYPE_VIDEO,
4461 .id = AV_CODEC_ID_H264,
4462 .priv_data_size = sizeof(H264Context),
4463 .init = ff_h264_decode_init,
4464 .close = h264_decode_end,
4465 .decode = decode_frame,
4466 .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 |
4467 CODEC_CAP_DELAY | CODEC_CAP_SLICE_THREADS |
4468 CODEC_CAP_FRAME_THREADS,
4470 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
4471 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
4472 .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
4473 .profiles = NULL_IF_CONFIG_SMALL(profiles),
4474 .priv_class = &h264_class,
4477 #if CONFIG_H264_VDPAU_DECODER
4478 AVCodec ff_h264_vdpau_decoder = {
4479 .name = "h264_vdpau",
4480 .type = AVMEDIA_TYPE_VIDEO,
4481 .id = AV_CODEC_ID_H264,
4482 .priv_data_size = sizeof(H264Context),
4483 .init = ff_h264_decode_init,
4484 .close = h264_decode_end,
4485 .decode = decode_frame,
4486 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
4488 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
4489 .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_VDPAU_H264,
4491 .profiles = NULL_IF_CONFIG_SMALL(profiles),
4492 .priv_class = &h264_vdpau_class,