2 * Microsoft Screen 2 (aka Windows Media Video V9 Screen) decoder
4 * This file is part of Libav.
6 * Libav is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * Libav is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * Microsoft Screen 2 (aka Windows Media Video V9 Screen) decoder
26 #include "libavutil/avassert.h"
27 #include "msmpeg4data.h"
32 typedef struct MSS2Context {
42 static void arith2_normalise(ArithCoder *c)
44 while ((c->high >> 15) - (c->low >> 15) < 2) {
45 if ((c->low ^ c->high) & 0x10000) {
50 c->high = c->high << 8 & 0xFFFFFF | 0xFF;
51 c->value = c->value << 8 & 0xFFFFFF | bytestream2_get_byte(c->gbc.gB);
52 c->low = c->low << 8 & 0xFFFFFF;
58 /* L. Stuiver and A. Moffat: "Piecewise Integer Mapping for Arithmetic Coding."
59 * In Proc. 8th Data Compression Conference (DCC '98), pp. 3-12, Mar. 1998 */
61 static int arith2_get_scaled_value(int value, int n, int range)
63 int split = (n << 1) - range;
66 return split + (value - split >> 1);
71 static void arith2_rescale_interval(ArithCoder *c, int range,
72 int low, int high, int n)
74 int split = (n << 1) - range;
77 c->high = split + (high - split << 1);
81 c->high += c->low - 1;
84 c->low += split + (low - split << 1);
89 static int arith2_get_number(ArithCoder *c, int n)
91 int range = c->high - c->low + 1;
92 int scale = av_log2(range) - av_log2(n);
95 if (n << scale > range)
100 val = arith2_get_scaled_value(c->value - c->low, n, range) >> scale;
102 arith2_rescale_interval(c, range, val << scale, (val + 1) << scale, n);
109 static int arith2_get_prob(ArithCoder *c, int16_t *probs)
111 int range = c->high - c->low + 1, n = *probs;
112 int scale = av_log2(range) - av_log2(n);
115 if (n << scale > range)
120 val = arith2_get_scaled_value(c->value - c->low, n, range) >> scale;
121 while (probs[++i] > val) ;
123 arith2_rescale_interval(c, range,
124 probs[i] << scale, probs[i - 1] << scale, n);
129 ARITH_GET_MODEL_SYM(2)
131 static int arith2_get_consumed_bytes(ArithCoder *c)
133 int diff = (c->high >> 16) - (c->low >> 16);
134 int bp = bytestream2_tell(c->gbc.gB) - 3 << 3;
137 while (!(diff & 0x80)) {
142 return (bits + bp + 7 >> 3) + ((c->low >> 16) + 1 == c->high >> 16);
145 static void arith2_init(ArithCoder *c, GetByteContext *gB)
149 c->value = bytestream2_get_be24(gB);
151 c->get_model_sym = arith2_get_model_sym;
152 c->get_number = arith2_get_number;
155 static int decode_pal_v2(MSS12Context *ctx, const uint8_t *buf, int buf_size)
158 uint32_t *pal = ctx->pal + 256 - ctx->free_colours;
160 if (!ctx->free_colours)
164 if (ncol > ctx->free_colours || buf_size < 2 + ncol * 3)
166 for (i = 0; i < ncol; i++)
167 *pal++ = AV_RB24(buf + 3 * i);
172 static int decode_555(GetByteContext *gB, uint16_t *dst, int stride,
173 int keyframe, int w, int h)
175 int last_symbol = 0, repeat = 0, prev_avail = 0;
178 int x, y, endx, endy, t;
180 #define READ_PAIR(a, b) \
181 a = bytestream2_get_byte(gB) << 4; \
182 t = bytestream2_get_byte(gB); \
184 b = (t & 0xF) << 8; \
185 b |= bytestream2_get_byte(gB); \
190 if (endx >= w || endy >= h || x > endx || y > endy)
192 dst += x + stride * y;
203 int b = bytestream2_get_byte(gB);
205 last_symbol = b << 8 | bytestream2_get_byte(gB);
209 repeat = (repeat << 8) + bytestream2_get_byte(gB) + 1;
210 if (last_symbol == -2) {
211 int skip = FFMIN((unsigned)repeat, dst + w - p);
216 last_symbol = 127 - b;
218 if (last_symbol >= 0)
220 else if (last_symbol == -1 && prev_avail)
222 } while (++p < dst + w);
230 static int decode_rle(GetBitContext *gb, uint8_t *pal_dst, int pal_stride,
231 uint8_t *rgb_dst, int rgb_stride, uint32_t *pal,
232 int keyframe, int kf_slipt, int slice, int w, int h)
234 uint8_t bits[270] = { 0 };
238 int current_length = 0, read_codes = 0, next_code = 0, current_codes = 0;
239 int remaining_codes, surplus_codes, i;
241 const int alphabet_size = 270 - keyframe;
243 int last_symbol = 0, repeat = 0, prev_avail = 0;
246 int x, y, clipw, cliph;
248 x = get_bits(gb, 12);
249 y = get_bits(gb, 12);
250 clipw = get_bits(gb, 12) + 1;
251 cliph = get_bits(gb, 12) + 1;
253 if (x + clipw > w || y + cliph > h)
254 return AVERROR_INVALIDDATA;
255 pal_dst += pal_stride * y + x;
256 rgb_dst += rgb_stride * y + x * 3;
263 pal_dst += pal_stride * kf_slipt;
264 rgb_dst += rgb_stride * kf_slipt;
271 /* read explicit codes */
273 while (current_codes--) {
274 int symbol = get_bits(gb, 8);
275 if (symbol >= 204 - keyframe)
276 symbol += 14 - keyframe;
277 else if (symbol > 189)
278 symbol = get_bits1(gb) + (symbol << 1) - 190;
280 return AVERROR_INVALIDDATA;
281 bits[symbol] = current_length;
282 codes[symbol] = next_code++;
287 remaining_codes = (1 << current_length) - next_code;
288 current_codes = get_bits(gb, av_ceil_log2(remaining_codes + 1));
289 if (current_length > 22 || current_codes > remaining_codes)
290 return AVERROR_INVALIDDATA;
291 } while (current_codes != remaining_codes);
293 remaining_codes = alphabet_size - read_codes;
295 /* determine the minimum length to fit the rest of the alphabet */
296 while ((surplus_codes = (2 << current_length) -
297 (next_code << 1) - remaining_codes) < 0) {
302 /* add the rest of the symbols lexicographically */
303 for (i = 0; i < alphabet_size; i++)
305 if (surplus_codes-- == 0) {
309 bits[i] = current_length;
310 codes[i] = next_code++;
313 if (next_code != 1 << current_length)
314 return AVERROR_INVALIDDATA;
316 if (i = init_vlc(&vlc, 9, alphabet_size, bits, 1, 1, codes, 4, 4, 0))
321 uint8_t *pp = pal_dst;
322 uint8_t *rp = rgb_dst;
325 int b = get_vlc2(gb, vlc.table, 9, 3);
331 b = get_bits(gb, 4) + 10;
336 repeat = get_bits(gb, b);
341 if (last_symbol == -2) {
342 int skip = FFMIN(repeat, pal_dst + w - pp);
348 last_symbol = 267 - b;
350 if (last_symbol >= 0) {
352 AV_WB24(rp, pal[last_symbol]);
353 } else if (last_symbol == -1 && prev_avail) {
354 *pp = *(pp - pal_stride);
355 memcpy(rp, rp - rgb_stride, 3);
358 } while (++pp < pal_dst + w);
359 pal_dst += pal_stride;
360 rgb_dst += rgb_stride;
368 static int decode_wmv9(AVCodecContext *avctx, const uint8_t *buf, int buf_size,
369 int x, int y, int w, int h, int wmv9_mask)
371 MSS2Context *ctx = avctx->priv_data;
372 MSS12Context *c = &ctx->c;
373 VC1Context *v = avctx->priv_data;
374 MpegEncContext *s = &v->s;
377 ff_mpeg_flush(avctx);
379 if (s->current_picture_ptr == NULL || s->current_picture_ptr->f.data[0]) {
380 int i = ff_find_unused_picture(s, 0);
383 s->current_picture_ptr = &s->picture[i];
386 init_get_bits(&s->gb, buf, buf_size * 8);
388 s->loop_filter = avctx->skip_loop_filter < AVDISCARD_ALL;
390 if (ff_vc1_parse_frame_header(v, &s->gb) == -1) {
391 av_log(v->s.avctx, AV_LOG_ERROR, "header error\n");
392 return AVERROR_INVALIDDATA;
395 if (s->pict_type != AV_PICTURE_TYPE_I) {
396 av_log(v->s.avctx, AV_LOG_ERROR, "expected I-frame\n");
397 return AVERROR_INVALIDDATA;
400 avctx->pix_fmt = PIX_FMT_YUV420P;
402 if (ff_MPV_frame_start(s, avctx) < 0) {
403 av_log(v->s.avctx, AV_LOG_ERROR, "ff_MPV_frame_start error\n");
404 avctx->pix_fmt = PIX_FMT_RGB24;
408 ff_er_frame_start(s);
410 v->bits = buf_size * 8;
412 v->end_mb_x = (w + 15) >> 4;
413 s->end_mb_y = (h + 15) >> 4;
415 v->end_mb_x = v->end_mb_x + 1 >> 1;
417 s->end_mb_y = s->end_mb_y + 1 >> 1;
419 ff_vc1_decode_blocks(v);
425 f = &s->current_picture.f;
427 if (v->respic == 3) {
428 ctx->dsp.upsample_plane(f->data[0], f->linesize[0], w, h);
429 ctx->dsp.upsample_plane(f->data[1], f->linesize[1], w >> 1, h >> 1);
430 ctx->dsp.upsample_plane(f->data[2], f->linesize[2], w >> 1, h >> 1);
431 } else if (v->respic)
432 av_log_ask_for_sample(v->s.avctx,
433 "Asymmetric WMV9 rectangle subsampling\n");
435 av_assert0(f->linesize[1] == f->linesize[2]);
438 ctx->dsp.mss2_blit_wmv9_masked(c->rgb_pic + y * c->rgb_stride + x * 3,
439 c->rgb_stride, wmv9_mask,
440 c->pal_pic + y * c->pal_stride + x,
442 f->data[0], f->linesize[0],
443 f->data[1], f->data[2], f->linesize[1],
446 ctx->dsp.mss2_blit_wmv9(c->rgb_pic + y * c->rgb_stride + x * 3,
448 f->data[0], f->linesize[0],
449 f->data[1], f->data[2], f->linesize[1],
452 avctx->pix_fmt = PIX_FMT_RGB24;
457 typedef struct Rectangle {
458 int coded, x, y, w, h;
461 #define MAX_WMV9_RECTANGLES 20
462 #define ARITH2_PADDING 2
464 static int mss2_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
467 const uint8_t *buf = avpkt->data;
468 int buf_size = avpkt->size;
469 MSS2Context *ctx = avctx->priv_data;
470 MSS12Context *c = &ctx->c;
475 int keyframe, has_wmv9, has_mv, is_rle, is_555, ret;
477 Rectangle wmv9rects[MAX_WMV9_RECTANGLES], *r;
478 int used_rects = 0, i, implicit_rect, av_uninit(wmv9_mask);
480 av_assert0(FF_INPUT_BUFFER_PADDING_SIZE >=
481 ARITH2_PADDING + (MIN_CACHE_BITS + 7) / 8);
483 init_get_bits(&gb, buf, buf_size * 8);
485 if (keyframe = get_bits1(&gb))
487 has_wmv9 = get_bits1(&gb);
488 has_mv = keyframe ? 0 : get_bits1(&gb);
489 is_rle = get_bits1(&gb);
490 is_555 = is_rle && get_bits1(&gb);
491 if (c->slice_split > 0)
492 ctx->split_position = c->slice_split;
493 else if (c->slice_split < 0) {
494 if (get_bits1(&gb)) {
495 if (get_bits1(&gb)) {
497 ctx->split_position = get_bits(&gb, 16);
499 ctx->split_position = get_bits(&gb, 12);
501 ctx->split_position = get_bits(&gb, 8) << 4;
504 ctx->split_position = avctx->height / 2;
507 ctx->split_position = avctx->height;
509 if (c->slice_split && (ctx->split_position < 1 - is_555 ||
510 ctx->split_position > avctx->height - 1))
511 return AVERROR_INVALIDDATA;
514 buf += get_bits_count(&gb) >> 3;
515 buf_size -= get_bits_count(&gb) >> 3;
518 return AVERROR_INVALIDDATA;
520 if (is_555 && (has_wmv9 || has_mv || c->slice_split && ctx->split_position))
521 return AVERROR_INVALIDDATA;
523 avctx->pix_fmt = is_555 ? PIX_FMT_RGB555 : PIX_FMT_RGB24;
524 if (ctx->pic.data[0] && ctx->pic.format != avctx->pix_fmt)
525 avctx->release_buffer(avctx, &ctx->pic);
528 bytestream2_init(&gB, buf, buf_size + ARITH2_PADDING);
529 arith2_init(&acoder, &gB);
531 implicit_rect = !arith2_get_bit(&acoder);
533 while (arith2_get_bit(&acoder)) {
534 if (used_rects == MAX_WMV9_RECTANGLES)
535 return AVERROR_INVALIDDATA;
536 r = &wmv9rects[used_rects];
538 r->x = arith2_get_number(&acoder, avctx->width);
540 r->x = arith2_get_number(&acoder, avctx->width -
541 wmv9rects[used_rects - 1].x) +
542 wmv9rects[used_rects - 1].x;
543 r->y = arith2_get_number(&acoder, avctx->height);
544 r->w = arith2_get_number(&acoder, avctx->width - r->x) + 1;
545 r->h = arith2_get_number(&acoder, avctx->height - r->y) + 1;
549 if (implicit_rect && used_rects) {
550 av_log(avctx, AV_LOG_ERROR, "implicit_rect && used_rects > 0\n");
551 return AVERROR_INVALIDDATA;
557 wmv9rects[0].w = avctx->width;
558 wmv9rects[0].h = avctx->height;
562 for (i = 0; i < used_rects; i++) {
563 if (!implicit_rect && arith2_get_bit(&acoder)) {
564 av_log(avctx, AV_LOG_ERROR, "Unexpected grandchildren\n");
565 return AVERROR_INVALIDDATA;
568 wmv9_mask = arith2_get_bit(&acoder) - 1;
570 wmv9_mask = arith2_get_number(&acoder, 256);
572 wmv9rects[i].coded = arith2_get_number(&acoder, 2);
575 buf += arith2_get_consumed_bytes(&acoder);
576 buf_size -= arith2_get_consumed_bytes(&acoder);
578 return AVERROR_INVALIDDATA;
582 if (keyframe && !is_555) {
583 if ((i = decode_pal_v2(c, buf, buf_size)) < 0)
584 return AVERROR_INVALIDDATA;
591 return AVERROR_INVALIDDATA;
592 c->mvX = AV_RB16(buf - 4) - avctx->width;
593 c->mvY = AV_RB16(buf - 2) - avctx->height;
596 if (c->mvX < 0 || c->mvY < 0) {
597 FFSWAP(AVFrame, ctx->pic, ctx->last_pic);
598 FFSWAP(uint8_t *, c->pal_pic, c->last_pal_pic);
600 if (ctx->pic.data[0])
601 avctx->release_buffer(avctx, &ctx->pic);
603 ctx->pic.reference = 3;
604 ctx->pic.buffer_hints = FF_BUFFER_HINTS_VALID |
605 FF_BUFFER_HINTS_READABLE |
606 FF_BUFFER_HINTS_PRESERVE |
607 FF_BUFFER_HINTS_REUSABLE;
609 if ((ret = avctx->get_buffer(avctx, &ctx->pic)) < 0) {
610 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
614 if (ctx->last_pic.data[0]) {
615 av_assert0(ctx->pic.linesize[0] == ctx->last_pic.linesize[0]);
616 c->last_rgb_pic = ctx->last_pic.data[0] +
617 ctx->last_pic.linesize[0] * (avctx->height - 1);
619 av_log(avctx, AV_LOG_ERROR, "Missing keyframe\n");
623 if (ctx->last_pic.data[0])
624 avctx->release_buffer(avctx, &ctx->last_pic);
626 ctx->pic.reference = 3;
627 ctx->pic.buffer_hints = FF_BUFFER_HINTS_VALID |
628 FF_BUFFER_HINTS_READABLE |
629 FF_BUFFER_HINTS_PRESERVE |
630 FF_BUFFER_HINTS_REUSABLE;
632 if ((ret = avctx->reget_buffer(avctx, &ctx->pic)) < 0) {
633 av_log(avctx, AV_LOG_ERROR, "reget_buffer() failed\n");
637 c->last_rgb_pic = NULL;
639 c->rgb_pic = ctx->pic.data[0] +
640 ctx->pic.linesize[0] * (avctx->height - 1);
641 c->rgb_stride = -ctx->pic.linesize[0];
643 ctx->pic.key_frame = keyframe;
644 ctx->pic.pict_type = keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
647 bytestream2_init(&gB, buf, buf_size);
649 if (decode_555(&gB, (uint16_t *)c->rgb_pic, c->rgb_stride >> 1,
650 keyframe, avctx->width, avctx->height))
651 return AVERROR_INVALIDDATA;
653 buf_size -= bytestream2_tell(&gB);
655 init_get_bits(&gb, buf, buf_size * 8);
656 if (ret = decode_rle(&gb, c->pal_pic, c->pal_stride,
657 c->rgb_pic, c->rgb_stride, c->pal, keyframe,
658 ctx->split_position, 0,
659 avctx->width, avctx->height))
664 if (ret = decode_rle(&gb, c->pal_pic, c->pal_stride,
665 c->rgb_pic, c->rgb_stride, c->pal, keyframe,
666 ctx->split_position, 1,
667 avctx->width, avctx->height))
671 buf += get_bits_count(&gb) >> 3;
672 buf_size -= get_bits_count(&gb) >> 3;
676 ff_mss12_slicecontext_reset(&ctx->sc[0]);
678 ff_mss12_slicecontext_reset(&ctx->sc[1]);
680 else if (c->corrupted)
681 return AVERROR_INVALIDDATA;
682 bytestream2_init(&gB, buf, buf_size + ARITH2_PADDING);
683 arith2_init(&acoder, &gB);
684 c->keyframe = keyframe;
685 if (c->corrupted = ff_mss12_decode_rect(&ctx->sc[0], &acoder, 0, 0,
687 ctx->split_position))
688 return AVERROR_INVALIDDATA;
690 buf += arith2_get_consumed_bytes(&acoder);
691 buf_size -= arith2_get_consumed_bytes(&acoder);
692 if (c->slice_split) {
694 return AVERROR_INVALIDDATA;
695 bytestream2_init(&gB, buf, buf_size + ARITH2_PADDING);
696 arith2_init(&acoder, &gB);
697 if (c->corrupted = ff_mss12_decode_rect(&ctx->sc[1], &acoder, 0,
700 avctx->height - ctx->split_position))
701 return AVERROR_INVALIDDATA;
703 buf += arith2_get_consumed_bytes(&acoder);
704 buf_size -= arith2_get_consumed_bytes(&acoder);
709 for (i = 0; i < used_rects; i++) {
710 int x = wmv9rects[i].x;
711 int y = wmv9rects[i].y;
712 int w = wmv9rects[i].w;
713 int h = wmv9rects[i].h;
714 if (wmv9rects[i].coded) {
715 int WMV9codedFrameSize;
716 if (buf_size < 4 || !(WMV9codedFrameSize = AV_RL24(buf)))
717 return AVERROR_INVALIDDATA;
718 if (ret = decode_wmv9(avctx, buf + 3, buf_size - 3,
719 x, y, w, h, wmv9_mask))
721 buf += WMV9codedFrameSize + 3;
722 buf_size -= WMV9codedFrameSize + 3;
724 uint8_t *dst = c->rgb_pic + y * c->rgb_stride + x * 3;
725 if (wmv9_mask != -1) {
726 ctx->dsp.mss2_gray_fill_masked(dst, c->rgb_stride,
728 c->pal_pic + y * c->pal_stride + x,
733 memset(dst, 0x80, w * 3);
734 dst += c->rgb_stride;
742 av_log(avctx, AV_LOG_WARNING, "buffer not fully consumed\n");
744 *data_size = sizeof(AVFrame);
745 *(AVFrame *)data = ctx->pic;
750 static av_cold int wmv9_init(AVCodecContext *avctx)
752 VC1Context *v = avctx->priv_data;
755 avctx->flags |= CODEC_FLAG_EMU_EDGE;
756 v->s.flags |= CODEC_FLAG_EMU_EDGE;
758 if (avctx->idct_algo == FF_IDCT_AUTO)
759 avctx->idct_algo = FF_IDCT_WMV2;
761 if (ff_vc1_init_common(v) < 0)
763 ff_vc1dsp_init(&v->vc1dsp);
765 v->profile = PROFILE_MAIN;
767 v->zz_8x4 = ff_wmv2_scantableA;
768 v->zz_4x8 = ff_wmv2_scantableB;
772 v->frmrtq_postproc = 7;
773 v->bitrtq_postproc = 31;
790 v->s.resync_marker = 0;
793 v->s.max_b_frames = avctx->max_b_frames = 0;
794 v->quantizer_mode = 0;
800 ff_vc1_init_transposed_scantables(v);
802 if (ff_msmpeg4_decode_init(avctx) < 0 ||
803 ff_vc1_decode_init_alloc_tables(v) < 0)
806 /* error concealment */
807 v->s.me.qpel_put = v->s.dsp.put_qpel_pixels_tab;
808 v->s.me.qpel_avg = v->s.dsp.avg_qpel_pixels_tab;
813 static av_cold int mss2_decode_end(AVCodecContext *avctx)
815 MSS2Context *const ctx = avctx->priv_data;
817 if (ctx->pic.data[0])
818 avctx->release_buffer(avctx, &ctx->pic);
819 if (ctx->last_pic.data[0])
820 avctx->release_buffer(avctx, &ctx->last_pic);
822 ff_mss12_decode_end(&ctx->c);
823 av_freep(&ctx->c.pal_pic);
824 av_freep(&ctx->c.last_pal_pic);
825 ff_vc1_decode_end(avctx);
830 static av_cold int mss2_decode_init(AVCodecContext *avctx)
832 MSS2Context * const ctx = avctx->priv_data;
833 MSS12Context *c = &ctx->c;
836 avctx->coded_frame = &ctx->pic;
837 if (ret = ff_mss12_decode_init(c, 1, &ctx->sc[0], &ctx->sc[1]))
839 c->pal_stride = c->mask_stride;
840 c->pal_pic = av_malloc(c->pal_stride * avctx->height);
841 c->last_pal_pic = av_malloc(c->pal_stride * avctx->height);
842 if (!c->pal_pic || !c->last_pal_pic) {
843 mss2_decode_end(avctx);
844 return AVERROR(ENOMEM);
846 if (ret = wmv9_init(avctx)) {
847 mss2_decode_end(avctx);
850 ff_mss2dsp_init(&ctx->dsp);
852 avctx->pix_fmt = c->free_colours == 127 ? PIX_FMT_RGB555
858 AVCodec ff_mss2_decoder = {
860 .type = AVMEDIA_TYPE_VIDEO,
861 .id = AV_CODEC_ID_MSS2,
862 .priv_data_size = sizeof(MSS2Context),
863 .init = mss2_decode_init,
864 .close = mss2_decode_end,
865 .decode = mss2_decode_frame,
866 .capabilities = CODEC_CAP_DR1,
867 .long_name = NULL_IF_CONFIG_SMALL("MS Windows Media Video V9 Screen"),