2 * VC-1 and WMV3 decoder
3 * Copyright (c) 2011 Mashiat Sarker Shakkhar
4 * Copyright (c) 2006-2007 Konstantin Shishkov
5 * Partly based on vc9.c (c) 2005 Anonymous, Alex Beregszaszi, Michael Niedermayer
7 * This file is part of FFmpeg.
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 * VC-1 and WMV3 decoder
34 #include "mpegvideo.h"
36 #include "msmpeg4data.h"
39 #include "vdpau_compat.h"
40 #include "libavutil/avassert.h"
43 #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER
45 typedef struct SpriteData {
47 * Transform coefficients for both sprites in 16.16 fixed point format,
48 * in the order they appear in the bitstream:
59 int effect_type, effect_flag;
60 int effect_pcount1, effect_pcount2; ///< amount of effect parameters stored in effect_params
61 int effect_params1[15], effect_params2[10]; ///< effect parameters in 16.16 fixed point format
64 static inline int get_fp_val(GetBitContext* gb)
66 return (get_bits_long(gb, 30) - (1 << 29)) << 1;
69 static void vc1_sprite_parse_transform(GetBitContext* gb, int c[7])
73 switch (get_bits(gb, 2)) {
76 c[2] = get_fp_val(gb);
80 c[0] = c[4] = get_fp_val(gb);
81 c[2] = get_fp_val(gb);
84 c[0] = get_fp_val(gb);
85 c[2] = get_fp_val(gb);
86 c[4] = get_fp_val(gb);
89 c[0] = get_fp_val(gb);
90 c[1] = get_fp_val(gb);
91 c[2] = get_fp_val(gb);
92 c[3] = get_fp_val(gb);
93 c[4] = get_fp_val(gb);
96 c[5] = get_fp_val(gb);
98 c[6] = get_fp_val(gb);
103 static int vc1_parse_sprites(VC1Context *v, GetBitContext* gb, SpriteData* sd)
105 AVCodecContext *avctx = v->s.avctx;
108 for (sprite = 0; sprite <= v->two_sprites; sprite++) {
109 vc1_sprite_parse_transform(gb, sd->coefs[sprite]);
110 if (sd->coefs[sprite][1] || sd->coefs[sprite][3])
111 avpriv_request_sample(avctx, "Non-zero rotation coefficients");
112 av_log(avctx, AV_LOG_DEBUG, sprite ? "S2:" : "S1:");
113 for (i = 0; i < 7; i++)
114 av_log(avctx, AV_LOG_DEBUG, " %d.%.3d",
115 sd->coefs[sprite][i] / (1<<16),
116 (abs(sd->coefs[sprite][i]) & 0xFFFF) * 1000 / (1 << 16));
117 av_log(avctx, AV_LOG_DEBUG, "\n");
121 if (sd->effect_type = get_bits_long(gb, 30)) {
122 switch (sd->effect_pcount1 = get_bits(gb, 4)) {
124 vc1_sprite_parse_transform(gb, sd->effect_params1);
127 vc1_sprite_parse_transform(gb, sd->effect_params1);
128 vc1_sprite_parse_transform(gb, sd->effect_params1 + 7);
131 for (i = 0; i < sd->effect_pcount1; i++)
132 sd->effect_params1[i] = get_fp_val(gb);
134 if (sd->effect_type != 13 || sd->effect_params1[0] != sd->coefs[0][6]) {
135 // effect 13 is simple alpha blending and matches the opacity above
136 av_log(avctx, AV_LOG_DEBUG, "Effect: %d; params: ", sd->effect_type);
137 for (i = 0; i < sd->effect_pcount1; i++)
138 av_log(avctx, AV_LOG_DEBUG, " %d.%.2d",
139 sd->effect_params1[i] / (1 << 16),
140 (abs(sd->effect_params1[i]) & 0xFFFF) * 1000 / (1 << 16));
141 av_log(avctx, AV_LOG_DEBUG, "\n");
144 sd->effect_pcount2 = get_bits(gb, 16);
145 if (sd->effect_pcount2 > 10) {
146 av_log(avctx, AV_LOG_ERROR, "Too many effect parameters\n");
147 return AVERROR_INVALIDDATA;
148 } else if (sd->effect_pcount2) {
150 av_log(avctx, AV_LOG_DEBUG, "Effect params 2: ");
151 while (++i < sd->effect_pcount2) {
152 sd->effect_params2[i] = get_fp_val(gb);
153 av_log(avctx, AV_LOG_DEBUG, " %d.%.2d",
154 sd->effect_params2[i] / (1 << 16),
155 (abs(sd->effect_params2[i]) & 0xFFFF) * 1000 / (1 << 16));
157 av_log(avctx, AV_LOG_DEBUG, "\n");
160 if (sd->effect_flag = get_bits1(gb))
161 av_log(avctx, AV_LOG_DEBUG, "Effect flag set\n");
163 if (get_bits_count(gb) >= gb->size_in_bits +
164 (avctx->codec_id == AV_CODEC_ID_WMV3IMAGE ? 64 : 0)) {
165 av_log(avctx, AV_LOG_ERROR, "Buffer overrun\n");
166 return AVERROR_INVALIDDATA;
168 if (get_bits_count(gb) < gb->size_in_bits - 8)
169 av_log(avctx, AV_LOG_WARNING, "Buffer not fully read\n");
174 static void vc1_draw_sprites(VC1Context *v, SpriteData* sd)
176 int i, plane, row, sprite;
177 int sr_cache[2][2] = { { -1, -1 }, { -1, -1 } };
178 uint8_t* src_h[2][2];
179 int xoff[2], xadv[2], yoff[2], yadv[2], alpha;
181 MpegEncContext *s = &v->s;
183 for (i = 0; i <= v->two_sprites; i++) {
184 xoff[i] = av_clip(sd->coefs[i][2], 0, v->sprite_width-1 << 16);
185 xadv[i] = sd->coefs[i][0];
186 if (xadv[i] != 1<<16 || (v->sprite_width << 16) - (v->output_width << 16) - xoff[i])
187 xadv[i] = av_clip(xadv[i], 0, ((v->sprite_width<<16) - xoff[i] - 1) / v->output_width);
189 yoff[i] = av_clip(sd->coefs[i][5], 0, v->sprite_height-1 << 16);
190 yadv[i] = av_clip(sd->coefs[i][4], 0, ((v->sprite_height << 16) - yoff[i]) / v->output_height);
192 alpha = av_clip_uint16(sd->coefs[1][6]);
194 for (plane = 0; plane < (CONFIG_GRAY && s->avctx->flags & CODEC_FLAG_GRAY ? 1 : 3); plane++) {
195 int width = v->output_width>>!!plane;
197 for (row = 0; row < v->output_height>>!!plane; row++) {
198 uint8_t *dst = v->sprite_output_frame->data[plane] +
199 v->sprite_output_frame->linesize[plane] * row;
201 for (sprite = 0; sprite <= v->two_sprites; sprite++) {
202 uint8_t *iplane = s->current_picture.f->data[plane];
203 int iline = s->current_picture.f->linesize[plane];
204 int ycoord = yoff[sprite] + yadv[sprite] * row;
205 int yline = ycoord >> 16;
207 ysub[sprite] = ycoord & 0xFFFF;
209 iplane = s->last_picture.f->data[plane];
210 iline = s->last_picture.f->linesize[plane];
212 next_line = FFMIN(yline + 1, (v->sprite_height >> !!plane) - 1) * iline;
213 if (!(xoff[sprite] & 0xFFFF) && xadv[sprite] == 1 << 16) {
214 src_h[sprite][0] = iplane + (xoff[sprite] >> 16) + yline * iline;
216 src_h[sprite][1] = iplane + (xoff[sprite] >> 16) + next_line;
218 if (sr_cache[sprite][0] != yline) {
219 if (sr_cache[sprite][1] == yline) {
220 FFSWAP(uint8_t*, v->sr_rows[sprite][0], v->sr_rows[sprite][1]);
221 FFSWAP(int, sr_cache[sprite][0], sr_cache[sprite][1]);
223 v->vc1dsp.sprite_h(v->sr_rows[sprite][0], iplane + yline * iline, xoff[sprite], xadv[sprite], width);
224 sr_cache[sprite][0] = yline;
227 if (ysub[sprite] && sr_cache[sprite][1] != yline + 1) {
228 v->vc1dsp.sprite_h(v->sr_rows[sprite][1],
229 iplane + next_line, xoff[sprite],
230 xadv[sprite], width);
231 sr_cache[sprite][1] = yline + 1;
233 src_h[sprite][0] = v->sr_rows[sprite][0];
234 src_h[sprite][1] = v->sr_rows[sprite][1];
238 if (!v->two_sprites) {
240 v->vc1dsp.sprite_v_single(dst, src_h[0][0], src_h[0][1], ysub[0], width);
242 memcpy(dst, src_h[0][0], width);
245 if (ysub[0] && ysub[1]) {
246 v->vc1dsp.sprite_v_double_twoscale(dst, src_h[0][0], src_h[0][1], ysub[0],
247 src_h[1][0], src_h[1][1], ysub[1], alpha, width);
248 } else if (ysub[0]) {
249 v->vc1dsp.sprite_v_double_onescale(dst, src_h[0][0], src_h[0][1], ysub[0],
250 src_h[1][0], alpha, width);
251 } else if (ysub[1]) {
252 v->vc1dsp.sprite_v_double_onescale(dst, src_h[1][0], src_h[1][1], ysub[1],
253 src_h[0][0], (1<<16)-1-alpha, width);
255 v->vc1dsp.sprite_v_double_noscale(dst, src_h[0][0], src_h[1][0], alpha, width);
261 for (i = 0; i <= v->two_sprites; i++) {
271 static int vc1_decode_sprites(VC1Context *v, GetBitContext* gb)
274 MpegEncContext *s = &v->s;
275 AVCodecContext *avctx = s->avctx;
278 memset(&sd, 0, sizeof(sd));
280 ret = vc1_parse_sprites(v, gb, &sd);
284 if (!s->current_picture.f || !s->current_picture.f->data[0]) {
285 av_log(avctx, AV_LOG_ERROR, "Got no sprites\n");
289 if (v->two_sprites && (!s->last_picture_ptr || !s->last_picture.f->data[0])) {
290 av_log(avctx, AV_LOG_WARNING, "Need two sprites, only got one\n");
294 av_frame_unref(v->sprite_output_frame);
295 if ((ret = ff_get_buffer(avctx, v->sprite_output_frame, 0)) < 0)
298 vc1_draw_sprites(v, &sd);
303 static void vc1_sprite_flush(AVCodecContext *avctx)
305 VC1Context *v = avctx->priv_data;
306 MpegEncContext *s = &v->s;
307 AVFrame *f = s->current_picture.f;
310 /* Windows Media Image codecs have a convergence interval of two keyframes.
311 Since we can't enforce it, clear to black the missing sprite. This is
312 wrong but it looks better than doing nothing. */
315 for (plane = 0; plane < (CONFIG_GRAY && s->avctx->flags & CODEC_FLAG_GRAY ? 1 : 3); plane++)
316 for (i = 0; i < v->sprite_height>>!!plane; i++)
317 memset(f->data[plane] + i * f->linesize[plane],
318 plane ? 128 : 0, f->linesize[plane]);
323 av_cold int ff_vc1_decode_init_alloc_tables(VC1Context *v)
325 MpegEncContext *s = &v->s;
327 int mb_height = FFALIGN(s->mb_height, 2);
329 /* Allocate mb bitplanes */
330 v->mv_type_mb_plane = av_malloc (s->mb_stride * mb_height);
331 v->direct_mb_plane = av_malloc (s->mb_stride * mb_height);
332 v->forward_mb_plane = av_malloc (s->mb_stride * mb_height);
333 v->fieldtx_plane = av_mallocz(s->mb_stride * mb_height);
334 v->acpred_plane = av_malloc (s->mb_stride * mb_height);
335 v->over_flags_plane = av_malloc (s->mb_stride * mb_height);
337 v->n_allocated_blks = s->mb_width + 2;
338 v->block = av_malloc(sizeof(*v->block) * v->n_allocated_blks);
339 v->cbp_base = av_malloc(sizeof(v->cbp_base[0]) * 2 * s->mb_stride);
340 v->cbp = v->cbp_base + s->mb_stride;
341 v->ttblk_base = av_malloc(sizeof(v->ttblk_base[0]) * 2 * s->mb_stride);
342 v->ttblk = v->ttblk_base + s->mb_stride;
343 v->is_intra_base = av_mallocz(sizeof(v->is_intra_base[0]) * 2 * s->mb_stride);
344 v->is_intra = v->is_intra_base + s->mb_stride;
345 v->luma_mv_base = av_mallocz(sizeof(v->luma_mv_base[0]) * 2 * s->mb_stride);
346 v->luma_mv = v->luma_mv_base + s->mb_stride;
348 /* allocate block type info in that way so it could be used with s->block_index[] */
349 v->mb_type_base = av_malloc(s->b8_stride * (mb_height * 2 + 1) + s->mb_stride * (mb_height + 1) * 2);
350 v->mb_type[0] = v->mb_type_base + s->b8_stride + 1;
351 v->mb_type[1] = v->mb_type_base + s->b8_stride * (mb_height * 2 + 1) + s->mb_stride + 1;
352 v->mb_type[2] = v->mb_type[1] + s->mb_stride * (mb_height + 1);
354 /* allocate memory to store block level MV info */
355 v->blk_mv_type_base = av_mallocz( s->b8_stride * (mb_height * 2 + 1) + s->mb_stride * (mb_height + 1) * 2);
356 v->blk_mv_type = v->blk_mv_type_base + s->b8_stride + 1;
357 v->mv_f_base = av_mallocz(2 * (s->b8_stride * (mb_height * 2 + 1) + s->mb_stride * (mb_height + 1) * 2));
358 v->mv_f[0] = v->mv_f_base + s->b8_stride + 1;
359 v->mv_f[1] = v->mv_f[0] + (s->b8_stride * (mb_height * 2 + 1) + s->mb_stride * (mb_height + 1) * 2);
360 v->mv_f_next_base = av_mallocz(2 * (s->b8_stride * (mb_height * 2 + 1) + s->mb_stride * (mb_height + 1) * 2));
361 v->mv_f_next[0] = v->mv_f_next_base + s->b8_stride + 1;
362 v->mv_f_next[1] = v->mv_f_next[0] + (s->b8_stride * (mb_height * 2 + 1) + s->mb_stride * (mb_height + 1) * 2);
364 /* Init coded blocks info */
365 if (v->profile == PROFILE_ADVANCED) {
366 // if (alloc_bitplane(&v->over_flags_plane, s->mb_width, s->mb_height) < 0)
368 // if (alloc_bitplane(&v->ac_pred_plane, s->mb_width, s->mb_height) < 0)
372 ff_intrax8_common_init(&v->x8,s);
374 if (s->avctx->codec_id == AV_CODEC_ID_WMV3IMAGE || s->avctx->codec_id == AV_CODEC_ID_VC1IMAGE) {
375 for (i = 0; i < 4; i++)
376 if (!(v->sr_rows[i >> 1][i & 1] = av_malloc(v->output_width)))
377 return AVERROR(ENOMEM);
380 if (!v->mv_type_mb_plane || !v->direct_mb_plane || !v->acpred_plane || !v->over_flags_plane ||
381 !v->block || !v->cbp_base || !v->ttblk_base || !v->is_intra_base || !v->luma_mv_base ||
383 av_freep(&v->mv_type_mb_plane);
384 av_freep(&v->direct_mb_plane);
385 av_freep(&v->acpred_plane);
386 av_freep(&v->over_flags_plane);
388 av_freep(&v->cbp_base);
389 av_freep(&v->ttblk_base);
390 av_freep(&v->is_intra_base);
391 av_freep(&v->luma_mv_base);
392 av_freep(&v->mb_type_base);
393 return AVERROR(ENOMEM);
399 av_cold void ff_vc1_init_transposed_scantables(VC1Context *v)
402 for (i = 0; i < 64; i++) {
403 #define transpose(x) (((x) >> 3) | (((x) & 7) << 3))
404 v->zz_8x8[0][i] = transpose(ff_wmv1_scantable[0][i]);
405 v->zz_8x8[1][i] = transpose(ff_wmv1_scantable[1][i]);
406 v->zz_8x8[2][i] = transpose(ff_wmv1_scantable[2][i]);
407 v->zz_8x8[3][i] = transpose(ff_wmv1_scantable[3][i]);
408 v->zzi_8x8[i] = transpose(ff_vc1_adv_interlaced_8x8_zz[i]);
414 /** Initialize a VC1/WMV3 decoder
415 * @todo TODO: Handle VC-1 IDUs (Transport level?)
416 * @todo TODO: Decypher remaining bits in extra_data
418 static av_cold int vc1_decode_init(AVCodecContext *avctx)
420 VC1Context *v = avctx->priv_data;
421 MpegEncContext *s = &v->s;
425 /* save the container output size for WMImage */
426 v->output_width = avctx->width;
427 v->output_height = avctx->height;
429 if (!avctx->extradata_size || !avctx->extradata)
431 if (!CONFIG_GRAY || !(avctx->flags & CODEC_FLAG_GRAY))
432 avctx->pix_fmt = ff_get_format(avctx, avctx->codec->pix_fmts);
434 avctx->pix_fmt = AV_PIX_FMT_GRAY8;
435 if (avctx->color_range == AVCOL_RANGE_UNSPECIFIED)
436 avctx->color_range = AVCOL_RANGE_MPEG;
440 if ((ret = ff_vc1_init_common(v)) < 0)
442 // ensure static VLC tables are initialized
443 if ((ret = ff_msmpeg4_decode_init(avctx)) < 0)
445 if ((ret = ff_vc1_decode_init_alloc_tables(v)) < 0)
447 // Hack to ensure the above functions will be called
448 // again once we know all necessary settings.
449 // That this is necessary might indicate a bug.
450 ff_vc1_decode_end(avctx);
452 ff_blockdsp_init(&s->bdsp, avctx);
453 ff_h264chroma_init(&v->h264chroma, 8);
454 ff_qpeldsp_init(&s->qdsp);
456 if (avctx->codec_id == AV_CODEC_ID_WMV3 || avctx->codec_id == AV_CODEC_ID_WMV3IMAGE) {
459 // looks like WMV3 has a sequence header stored in the extradata
460 // advanced sequence header may be before the first frame
461 // the last byte of the extradata is a version number, 1 for the
462 // samples we can decode
464 init_get_bits(&gb, avctx->extradata, avctx->extradata_size*8);
466 if ((ret = ff_vc1_decode_sequence_header(avctx, v, &gb)) < 0)
469 count = avctx->extradata_size*8 - get_bits_count(&gb);
471 av_log(avctx, AV_LOG_INFO, "Extra data: %i bits left, value: %X\n",
472 count, get_bits(&gb, count));
473 } else if (count < 0) {
474 av_log(avctx, AV_LOG_INFO, "Read %i bits in overflow\n", -count);
476 } else { // VC1/WVC1/WVP2
477 const uint8_t *start = avctx->extradata;
478 uint8_t *end = avctx->extradata + avctx->extradata_size;
481 uint8_t *buf2 = NULL;
482 int seq_initialized = 0, ep_initialized = 0;
484 if (avctx->extradata_size < 16) {
485 av_log(avctx, AV_LOG_ERROR, "Extradata size too small: %i\n", avctx->extradata_size);
489 buf2 = av_mallocz(avctx->extradata_size + FF_INPUT_BUFFER_PADDING_SIZE);
491 return AVERROR(ENOMEM);
493 start = find_next_marker(start, end); // in WVC1 extradata first byte is its size, but can be 0 in mkv
495 for (; next < end; start = next) {
496 next = find_next_marker(start + 4, end);
497 size = next - start - 4;
500 buf2_size = vc1_unescape_buffer(start + 4, size, buf2);
501 init_get_bits(&gb, buf2, buf2_size * 8);
502 switch (AV_RB32(start)) {
503 case VC1_CODE_SEQHDR:
504 if ((ret = ff_vc1_decode_sequence_header(avctx, v, &gb)) < 0) {
510 case VC1_CODE_ENTRYPOINT:
511 if ((ret = ff_vc1_decode_entry_point(avctx, v, &gb)) < 0) {
520 if (!seq_initialized || !ep_initialized) {
521 av_log(avctx, AV_LOG_ERROR, "Incomplete extradata\n");
524 v->res_sprite = (avctx->codec_id == AV_CODEC_ID_VC1IMAGE);
527 v->sprite_output_frame = av_frame_alloc();
528 if (!v->sprite_output_frame)
529 return AVERROR(ENOMEM);
531 avctx->profile = v->profile;
532 if (v->profile == PROFILE_ADVANCED)
533 avctx->level = v->level;
535 avctx->has_b_frames = !!avctx->max_b_frames;
537 if (v->color_prim == 1 || v->color_prim == 5 || v->color_prim == 6)
538 avctx->color_primaries = v->color_prim;
539 if (v->transfer_char == 1 || v->transfer_char == 7)
540 avctx->color_trc = v->transfer_char;
541 if (v->matrix_coef == 1 || v->matrix_coef == 6 || v->matrix_coef == 7)
542 avctx->colorspace = v->matrix_coef;
544 s->mb_width = (avctx->coded_width + 15) >> 4;
545 s->mb_height = (avctx->coded_height + 15) >> 4;
547 if (v->profile == PROFILE_ADVANCED || v->res_fasttx) {
548 ff_vc1_init_transposed_scantables(v);
550 memcpy(v->zz_8x8, ff_wmv1_scantable, 4*64);
555 if (avctx->codec_id == AV_CODEC_ID_WMV3IMAGE || avctx->codec_id == AV_CODEC_ID_VC1IMAGE) {
556 v->sprite_width = avctx->coded_width;
557 v->sprite_height = avctx->coded_height;
559 avctx->coded_width = avctx->width = v->output_width;
560 avctx->coded_height = avctx->height = v->output_height;
562 // prevent 16.16 overflows
563 if (v->sprite_width > 1 << 14 ||
564 v->sprite_height > 1 << 14 ||
565 v->output_width > 1 << 14 ||
566 v->output_height > 1 << 14) return -1;
568 if ((v->sprite_width&1) || (v->sprite_height&1)) {
569 avpriv_request_sample(avctx, "odd sprites support");
570 return AVERROR_PATCHWELCOME;
576 /** Close a VC1/WMV3 decoder
577 * @warning Initial try at using MpegEncContext stuff
579 av_cold int ff_vc1_decode_end(AVCodecContext *avctx)
581 VC1Context *v = avctx->priv_data;
584 av_frame_free(&v->sprite_output_frame);
586 for (i = 0; i < 4; i++)
587 av_freep(&v->sr_rows[i >> 1][i & 1]);
588 av_freep(&v->hrd_rate);
589 av_freep(&v->hrd_buffer);
590 ff_mpv_common_end(&v->s);
591 av_freep(&v->mv_type_mb_plane);
592 av_freep(&v->direct_mb_plane);
593 av_freep(&v->forward_mb_plane);
594 av_freep(&v->fieldtx_plane);
595 av_freep(&v->acpred_plane);
596 av_freep(&v->over_flags_plane);
597 av_freep(&v->mb_type_base);
598 av_freep(&v->blk_mv_type_base);
599 av_freep(&v->mv_f_base);
600 av_freep(&v->mv_f_next_base);
602 av_freep(&v->cbp_base);
603 av_freep(&v->ttblk_base);
604 av_freep(&v->is_intra_base); // FIXME use v->mb_type[]
605 av_freep(&v->luma_mv_base);
606 ff_intrax8_common_end(&v->x8);
611 /** Decode a VC1/WMV3 frame
612 * @todo TODO: Handle VC-1 IDUs (Transport level?)
614 static int vc1_decode_frame(AVCodecContext *avctx, void *data,
615 int *got_frame, AVPacket *avpkt)
617 const uint8_t *buf = avpkt->data;
618 int buf_size = avpkt->size, n_slices = 0, i, ret;
619 VC1Context *v = avctx->priv_data;
620 MpegEncContext *s = &v->s;
621 AVFrame *pict = data;
622 uint8_t *buf2 = NULL;
623 const uint8_t *buf_start = buf, *buf_start_second_field = NULL;
624 int mb_height, n_slices1=-1;
629 } *slices = NULL, *tmp;
633 if(s->avctx->flags & CODEC_FLAG_LOW_DELAY)
636 /* no supplementary picture */
637 if (buf_size == 0 || (buf_size == 4 && AV_RB32(buf) == VC1_CODE_ENDOFSEQ)) {
638 /* special case for last picture */
639 if (s->low_delay == 0 && s->next_picture_ptr) {
640 if ((ret = av_frame_ref(pict, s->next_picture_ptr->f)) < 0)
642 s->next_picture_ptr = NULL;
650 if (s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU) {
651 if (v->profile < PROFILE_ADVANCED)
652 avctx->pix_fmt = AV_PIX_FMT_VDPAU_WMV3;
654 avctx->pix_fmt = AV_PIX_FMT_VDPAU_VC1;
657 //for advanced profile we may need to parse and unescape data
658 if (avctx->codec_id == AV_CODEC_ID_VC1 || avctx->codec_id == AV_CODEC_ID_VC1IMAGE) {
660 buf2 = av_mallocz(buf_size + FF_INPUT_BUFFER_PADDING_SIZE);
662 return AVERROR(ENOMEM);
664 if (IS_MARKER(AV_RB32(buf))) { /* frame starts with marker and needs to be parsed */
665 const uint8_t *start, *end, *next;
669 for (start = buf, end = buf + buf_size; next < end; start = next) {
670 next = find_next_marker(start + 4, end);
671 size = next - start - 4;
672 if (size <= 0) continue;
673 switch (AV_RB32(start)) {
675 if (avctx->hwaccel ||
676 s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
678 buf_size2 = vc1_unescape_buffer(start + 4, size, buf2);
680 case VC1_CODE_FIELD: {
682 if (avctx->hwaccel ||
683 s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
684 buf_start_second_field = start;
685 tmp = av_realloc_array(slices, sizeof(*slices), (n_slices+1));
689 slices[n_slices].buf = av_mallocz(buf_size + FF_INPUT_BUFFER_PADDING_SIZE);
690 if (!slices[n_slices].buf)
692 buf_size3 = vc1_unescape_buffer(start + 4, size,
693 slices[n_slices].buf);
694 init_get_bits(&slices[n_slices].gb, slices[n_slices].buf,
696 /* assuming that the field marker is at the exact middle,
698 slices[n_slices].mby_start = s->mb_height + 1 >> 1;
699 n_slices1 = n_slices - 1; // index of the last slice of the first field
703 case VC1_CODE_ENTRYPOINT: /* it should be before frame data */
704 buf_size2 = vc1_unescape_buffer(start + 4, size, buf2);
705 init_get_bits(&s->gb, buf2, buf_size2 * 8);
706 ff_vc1_decode_entry_point(avctx, v, &s->gb);
708 case VC1_CODE_SLICE: {
710 tmp = av_realloc_array(slices, sizeof(*slices), (n_slices+1));
714 slices[n_slices].buf = av_mallocz(buf_size + FF_INPUT_BUFFER_PADDING_SIZE);
715 if (!slices[n_slices].buf)
717 buf_size3 = vc1_unescape_buffer(start + 4, size,
718 slices[n_slices].buf);
719 init_get_bits(&slices[n_slices].gb, slices[n_slices].buf,
721 slices[n_slices].mby_start = get_bits(&slices[n_slices].gb, 9);
727 } else if (v->interlace && ((buf[0] & 0xC0) == 0xC0)) { /* WVC1 interlaced stores both fields divided by marker */
728 const uint8_t *divider;
731 divider = find_next_marker(buf, buf + buf_size);
732 if ((divider == (buf + buf_size)) || AV_RB32(divider) != VC1_CODE_FIELD) {
733 av_log(avctx, AV_LOG_ERROR, "Error in WVC1 interlaced frame\n");
735 } else { // found field marker, unescape second field
736 if (avctx->hwaccel ||
737 s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
738 buf_start_second_field = divider;
739 tmp = av_realloc_array(slices, sizeof(*slices), (n_slices+1));
743 slices[n_slices].buf = av_mallocz(buf_size + FF_INPUT_BUFFER_PADDING_SIZE);
744 if (!slices[n_slices].buf)
746 buf_size3 = vc1_unescape_buffer(divider + 4, buf + buf_size - divider - 4, slices[n_slices].buf);
747 init_get_bits(&slices[n_slices].gb, slices[n_slices].buf,
749 slices[n_slices].mby_start = s->mb_height + 1 >> 1;
750 n_slices1 = n_slices - 1;
753 buf_size2 = vc1_unescape_buffer(buf, divider - buf, buf2);
755 buf_size2 = vc1_unescape_buffer(buf, buf_size, buf2);
757 init_get_bits(&s->gb, buf2, buf_size2*8);
759 init_get_bits(&s->gb, buf, buf_size*8);
762 v->new_sprite = !get_bits1(&s->gb);
763 v->two_sprites = get_bits1(&s->gb);
764 /* res_sprite means a Windows Media Image stream, AV_CODEC_ID_*IMAGE means
765 we're using the sprite compositor. These are intentionally kept separate
766 so you can get the raw sprites by using the wmv3 decoder for WMVP or
767 the vc1 one for WVP2 */
768 if (avctx->codec_id == AV_CODEC_ID_WMV3IMAGE || avctx->codec_id == AV_CODEC_ID_VC1IMAGE) {
770 // switch AVCodecContext parameters to those of the sprites
771 avctx->width = avctx->coded_width = v->sprite_width;
772 avctx->height = avctx->coded_height = v->sprite_height;
779 if (s->context_initialized &&
780 (s->width != avctx->coded_width ||
781 s->height != avctx->coded_height)) {
782 ff_vc1_decode_end(avctx);
785 if (!s->context_initialized) {
786 if (ff_msmpeg4_decode_init(avctx) < 0)
788 if (ff_vc1_decode_init_alloc_tables(v) < 0) {
789 ff_mpv_common_end(s);
793 s->low_delay = !avctx->has_b_frames || v->res_sprite;
795 if (v->profile == PROFILE_ADVANCED) {
796 if(avctx->coded_width<=1 || avctx->coded_height<=1)
798 s->h_edge_pos = avctx->coded_width;
799 s->v_edge_pos = avctx->coded_height;
803 // do parse frame header
804 v->pic_header_flag = 0;
805 v->first_pic_header_flag = 1;
806 if (v->profile < PROFILE_ADVANCED) {
807 if (ff_vc1_parse_frame_header(v, &s->gb) < 0) {
811 if (ff_vc1_parse_frame_header_adv(v, &s->gb) < 0) {
815 v->first_pic_header_flag = 0;
817 if (avctx->debug & FF_DEBUG_PICT_INFO)
818 av_log(v->s.avctx, AV_LOG_DEBUG, "pict_type: %c\n", av_get_picture_type_char(s->pict_type));
820 if ((avctx->codec_id == AV_CODEC_ID_WMV3IMAGE || avctx->codec_id == AV_CODEC_ID_VC1IMAGE)
821 && s->pict_type != AV_PICTURE_TYPE_I) {
822 av_log(v->s.avctx, AV_LOG_ERROR, "Sprite decoder: expected I-frame\n");
826 if ((s->mb_height >> v->field_mode) == 0) {
827 av_log(v->s.avctx, AV_LOG_ERROR, "image too short\n");
831 // for skipping the frame
832 s->current_picture.f->pict_type = s->pict_type;
833 s->current_picture.f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
835 /* skip B-frames if we don't have reference frames */
836 if (!s->last_picture_ptr && (s->pict_type == AV_PICTURE_TYPE_B || s->droppable)) {
837 av_log(v->s.avctx, AV_LOG_DEBUG, "Skipping B frame without reference frames\n");
840 if ((avctx->skip_frame >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B) ||
841 (avctx->skip_frame >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I) ||
842 avctx->skip_frame >= AVDISCARD_ALL) {
846 if (s->next_p_frame_damaged) {
847 if (s->pict_type == AV_PICTURE_TYPE_B)
850 s->next_p_frame_damaged = 0;
853 if (ff_mpv_frame_start(s, avctx) < 0) {
857 v->s.current_picture_ptr->field_picture = v->field_mode;
858 v->s.current_picture_ptr->f->interlaced_frame = (v->fcm != PROGRESSIVE);
859 v->s.current_picture_ptr->f->top_field_first = v->tff;
861 // process pulldown flags
862 s->current_picture_ptr->f->repeat_pict = 0;
863 // Pulldown flags are only valid when 'broadcast' has been set.
864 // So ticks_per_frame will be 2
867 s->current_picture_ptr->f->repeat_pict = 1;
868 } else if (v->rptfrm) {
870 s->current_picture_ptr->f->repeat_pict = v->rptfrm * 2;
873 s->me.qpel_put = s->qdsp.put_qpel_pixels_tab;
874 s->me.qpel_avg = s->qdsp.avg_qpel_pixels_tab;
876 if ((CONFIG_VC1_VDPAU_DECODER)
877 &&s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU) {
878 if (v->field_mode && buf_start_second_field) {
879 ff_vdpau_vc1_decode_picture(s, buf_start, buf_start_second_field - buf_start);
880 ff_vdpau_vc1_decode_picture(s, buf_start_second_field, (buf + buf_size) - buf_start_second_field);
882 ff_vdpau_vc1_decode_picture(s, buf_start, (buf + buf_size) - buf_start);
884 } else if (avctx->hwaccel) {
885 if (v->field_mode && buf_start_second_field) {
886 // decode first field
887 s->picture_structure = PICT_BOTTOM_FIELD - v->tff;
888 if (avctx->hwaccel->start_frame(avctx, buf_start, buf_start_second_field - buf_start) < 0)
890 if (avctx->hwaccel->decode_slice(avctx, buf_start, buf_start_second_field - buf_start) < 0)
892 if (avctx->hwaccel->end_frame(avctx) < 0)
895 // decode second field
896 s->gb = slices[n_slices1 + 1].gb;
897 s->picture_structure = PICT_TOP_FIELD + v->tff;
899 v->pic_header_flag = 0;
900 if (ff_vc1_parse_frame_header_adv(v, &s->gb) < 0) {
901 av_log(avctx, AV_LOG_ERROR, "parsing header for second field failed");
904 v->s.current_picture_ptr->f->pict_type = v->s.pict_type;
906 if (avctx->hwaccel->start_frame(avctx, buf_start_second_field, (buf + buf_size) - buf_start_second_field) < 0)
908 if (avctx->hwaccel->decode_slice(avctx, buf_start_second_field, (buf + buf_size) - buf_start_second_field) < 0)
910 if (avctx->hwaccel->end_frame(avctx) < 0)
913 s->picture_structure = PICT_FRAME;
914 if (avctx->hwaccel->start_frame(avctx, buf_start, (buf + buf_size) - buf_start) < 0)
916 if (avctx->hwaccel->decode_slice(avctx, buf_start, (buf + buf_size) - buf_start) < 0)
918 if (avctx->hwaccel->end_frame(avctx) < 0)
924 ff_mpeg_er_frame_start(s);
926 v->bits = buf_size * 8;
927 v->end_mb_x = s->mb_width;
929 s->current_picture.f->linesize[0] <<= 1;
930 s->current_picture.f->linesize[1] <<= 1;
931 s->current_picture.f->linesize[2] <<= 1;
935 mb_height = s->mb_height >> v->field_mode;
937 av_assert0 (mb_height > 0);
939 for (i = 0; i <= n_slices; i++) {
940 if (i > 0 && slices[i - 1].mby_start >= mb_height) {
941 if (v->field_mode <= 0) {
942 av_log(v->s.avctx, AV_LOG_ERROR, "Slice %d starts beyond "
943 "picture boundary (%d >= %d)\n", i,
944 slices[i - 1].mby_start, mb_height);
948 av_assert0((s->mb_height & 1) == 0);
949 v->blocks_off = s->b8_stride * (s->mb_height&~1);
950 v->mb_off = s->mb_stride * s->mb_height >> 1;
957 v->pic_header_flag = 0;
958 if (v->field_mode && i == n_slices1 + 2) {
959 if ((header_ret = ff_vc1_parse_frame_header_adv(v, &s->gb)) < 0) {
960 av_log(v->s.avctx, AV_LOG_ERROR, "Field header damaged\n");
961 if (avctx->err_recognition & AV_EF_EXPLODE)
965 } else if (get_bits1(&s->gb)) {
966 v->pic_header_flag = 1;
967 if ((header_ret = ff_vc1_parse_frame_header_adv(v, &s->gb)) < 0) {
968 av_log(v->s.avctx, AV_LOG_ERROR, "Slice header damaged\n");
969 if (avctx->err_recognition & AV_EF_EXPLODE)
977 s->start_mb_y = (i == 0) ? 0 : FFMAX(0, slices[i-1].mby_start % mb_height);
978 if (!v->field_mode || v->second_field)
979 s->end_mb_y = (i == n_slices ) ? mb_height : FFMIN(mb_height, slices[i].mby_start % mb_height);
982 av_log(v->s.avctx, AV_LOG_ERROR, "first field slice count too large\n");
985 s->end_mb_y = (i <= n_slices1 + 1) ? mb_height : FFMIN(mb_height, slices[i].mby_start % mb_height);
987 if (s->end_mb_y <= s->start_mb_y) {
988 av_log(v->s.avctx, AV_LOG_ERROR, "end mb y %d %d invalid\n", s->end_mb_y, s->start_mb_y);
991 if (!v->p_frame_skipped && s->pict_type != AV_PICTURE_TYPE_I && !v->cbpcy_vlc) {
992 av_log(v->s.avctx, AV_LOG_ERROR, "missing cbpcy_vlc\n");
995 ff_vc1_decode_blocks(v);
997 s->gb = slices[i].gb;
1000 v->second_field = 0;
1001 s->current_picture.f->linesize[0] >>= 1;
1002 s->current_picture.f->linesize[1] >>= 1;
1003 s->current_picture.f->linesize[2] >>= 1;
1005 s->uvlinesize >>= 1;
1006 if (v->s.pict_type != AV_PICTURE_TYPE_BI && v->s.pict_type != AV_PICTURE_TYPE_B) {
1007 FFSWAP(uint8_t *, v->mv_f_next[0], v->mv_f[0]);
1008 FFSWAP(uint8_t *, v->mv_f_next[1], v->mv_f[1]);
1011 ff_dlog(s->avctx, "Consumed %i/%i bits\n",
1012 get_bits_count(&s->gb), s->gb.size_in_bits);
1013 // if (get_bits_count(&s->gb) > buf_size * 8)
1015 if(s->er.error_occurred && s->pict_type == AV_PICTURE_TYPE_B)
1018 ff_er_frame_end(&s->er);
1021 ff_mpv_frame_end(s);
1023 if (avctx->codec_id == AV_CODEC_ID_WMV3IMAGE || avctx->codec_id == AV_CODEC_ID_VC1IMAGE) {
1025 avctx->width = avctx->coded_width = v->output_width;
1026 avctx->height = avctx->coded_height = v->output_height;
1027 if (avctx->skip_frame >= AVDISCARD_NONREF)
1029 #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER
1030 if (vc1_decode_sprites(v, &s->gb))
1033 if ((ret = av_frame_ref(pict, v->sprite_output_frame)) < 0)
1037 if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
1038 if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
1040 ff_print_debug_info(s, s->current_picture_ptr, pict);
1042 } else if (s->last_picture_ptr) {
1043 if ((ret = av_frame_ref(pict, s->last_picture_ptr->f)) < 0)
1045 ff_print_debug_info(s, s->last_picture_ptr, pict);
1052 for (i = 0; i < n_slices; i++)
1053 av_free(slices[i].buf);
1059 for (i = 0; i < n_slices; i++)
1060 av_free(slices[i].buf);
1066 static const AVProfile profiles[] = {
1067 { FF_PROFILE_VC1_SIMPLE, "Simple" },
1068 { FF_PROFILE_VC1_MAIN, "Main" },
1069 { FF_PROFILE_VC1_COMPLEX, "Complex" },
1070 { FF_PROFILE_VC1_ADVANCED, "Advanced" },
1071 { FF_PROFILE_UNKNOWN },
1074 static const enum AVPixelFormat vc1_hwaccel_pixfmt_list_420[] = {
1075 #if CONFIG_VC1_DXVA2_HWACCEL
1076 AV_PIX_FMT_DXVA2_VLD,
1078 #if CONFIG_VC1_D3D11VA_HWACCEL
1079 AV_PIX_FMT_D3D11VA_VLD,
1081 #if CONFIG_VC1_VAAPI_HWACCEL
1082 AV_PIX_FMT_VAAPI_VLD,
1084 #if CONFIG_VC1_VDPAU_HWACCEL
1091 AVCodec ff_vc1_decoder = {
1093 .long_name = NULL_IF_CONFIG_SMALL("SMPTE VC-1"),
1094 .type = AVMEDIA_TYPE_VIDEO,
1095 .id = AV_CODEC_ID_VC1,
1096 .priv_data_size = sizeof(VC1Context),
1097 .init = vc1_decode_init,
1098 .close = ff_vc1_decode_end,
1099 .decode = vc1_decode_frame,
1100 .flush = ff_mpeg_flush,
1101 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1102 .pix_fmts = vc1_hwaccel_pixfmt_list_420,
1103 .profiles = NULL_IF_CONFIG_SMALL(profiles)
1106 #if CONFIG_WMV3_DECODER
1107 AVCodec ff_wmv3_decoder = {
1109 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 9"),
1110 .type = AVMEDIA_TYPE_VIDEO,
1111 .id = AV_CODEC_ID_WMV3,
1112 .priv_data_size = sizeof(VC1Context),
1113 .init = vc1_decode_init,
1114 .close = ff_vc1_decode_end,
1115 .decode = vc1_decode_frame,
1116 .flush = ff_mpeg_flush,
1117 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1118 .pix_fmts = vc1_hwaccel_pixfmt_list_420,
1119 .profiles = NULL_IF_CONFIG_SMALL(profiles)
1123 #if CONFIG_WMV3_VDPAU_DECODER
1124 AVCodec ff_wmv3_vdpau_decoder = {
1125 .name = "wmv3_vdpau",
1126 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 9 VDPAU"),
1127 .type = AVMEDIA_TYPE_VIDEO,
1128 .id = AV_CODEC_ID_WMV3,
1129 .priv_data_size = sizeof(VC1Context),
1130 .init = vc1_decode_init,
1131 .close = ff_vc1_decode_end,
1132 .decode = vc1_decode_frame,
1133 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
1134 .pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_VDPAU_WMV3, AV_PIX_FMT_NONE },
1135 .profiles = NULL_IF_CONFIG_SMALL(profiles)
1139 #if CONFIG_VC1_VDPAU_DECODER
1140 AVCodec ff_vc1_vdpau_decoder = {
1141 .name = "vc1_vdpau",
1142 .long_name = NULL_IF_CONFIG_SMALL("SMPTE VC-1 VDPAU"),
1143 .type = AVMEDIA_TYPE_VIDEO,
1144 .id = AV_CODEC_ID_VC1,
1145 .priv_data_size = sizeof(VC1Context),
1146 .init = vc1_decode_init,
1147 .close = ff_vc1_decode_end,
1148 .decode = vc1_decode_frame,
1149 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
1150 .pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_VDPAU_VC1, AV_PIX_FMT_NONE },
1151 .profiles = NULL_IF_CONFIG_SMALL(profiles)
1155 #if CONFIG_WMV3IMAGE_DECODER
1156 AVCodec ff_wmv3image_decoder = {
1157 .name = "wmv3image",
1158 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 9 Image"),
1159 .type = AVMEDIA_TYPE_VIDEO,
1160 .id = AV_CODEC_ID_WMV3IMAGE,
1161 .priv_data_size = sizeof(VC1Context),
1162 .init = vc1_decode_init,
1163 .close = ff_vc1_decode_end,
1164 .decode = vc1_decode_frame,
1165 .capabilities = CODEC_CAP_DR1,
1166 .flush = vc1_sprite_flush,
1167 .pix_fmts = (const enum AVPixelFormat[]) {
1174 #if CONFIG_VC1IMAGE_DECODER
1175 AVCodec ff_vc1image_decoder = {
1177 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 9 Image v2"),
1178 .type = AVMEDIA_TYPE_VIDEO,
1179 .id = AV_CODEC_ID_VC1IMAGE,
1180 .priv_data_size = sizeof(VC1Context),
1181 .init = vc1_decode_init,
1182 .close = ff_vc1_decode_end,
1183 .decode = vc1_decode_frame,
1184 .capabilities = CODEC_CAP_DR1,
1185 .flush = vc1_sprite_flush,
1186 .pix_fmts = (const enum AVPixelFormat[]) {