2 * Microsoft Screen 3 (aka Microsoft ATC Screen) decoder
3 * Copyright (c) 2012 Konstantin Shishkov
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 * Microsoft Screen 3 (aka Microsoft ATC Screen) decoder
28 #include "bytestream.h"
32 #define HEADER_SIZE 27
34 #define MODEL2_SCALE 13
35 #define MODEL_SCALE 15
36 #define MODEL256_SEC_SCALE 9
38 typedef struct Model2 {
39 int upd_val, till_rescale;
40 unsigned zero_freq, zero_weight;
41 unsigned total_freq, total_weight;
44 typedef struct Model {
45 int weights[16], freqs[16];
48 int upd_val, max_upd_val, till_rescale;
51 typedef struct Model256 {
52 int weights[256], freqs[256];
56 int upd_val, max_upd_val, till_rescale;
59 #define RAC_BOTTOM 0x01000000
60 typedef struct RangeCoder {
61 const uint8_t *src, *src_end;
75 typedef struct BlockTypeContext {
80 typedef struct FillBlockCoder {
85 typedef struct ImageBlockCoder {
86 Model256 esc_model, vec_entry_model;
91 typedef struct DCTBlockCoder {
102 typedef struct HaarBlockCoder {
108 typedef struct MSS3Context {
109 AVCodecContext *avctx;
114 BlockTypeContext btype[3];
115 FillBlockCoder fill_coder[3];
116 ImageBlockCoder image_coder[3];
117 DCTBlockCoder dct_coder[3];
118 HaarBlockCoder haar_coder[3];
125 static void model2_reset(Model2 *m)
129 m->zero_freq = 0x1000;
130 m->total_freq = 0x2000;
135 static void model2_update(Model2 *m, int bit)
145 m->total_weight += m->upd_val;
146 if (m->total_weight > 0x2000) {
147 m->total_weight = (m->total_weight + 1) >> 1;
148 m->zero_weight = (m->zero_weight + 1) >> 1;
149 if (m->total_weight == m->zero_weight)
150 m->total_weight = m->zero_weight + 1;
152 m->upd_val = m->upd_val * 5 >> 2;
155 scale = 0x80000000u / m->total_weight;
156 m->zero_freq = m->zero_weight * scale >> 18;
157 m->total_freq = m->total_weight * scale >> 18;
158 m->till_rescale = m->upd_val;
161 static void model_update(Model *m, int val)
170 m->tot_weight += m->upd_val;
172 if (m->tot_weight > 0x8000) {
174 for (i = 0; i < m->num_syms; i++) {
175 m->weights[i] = (m->weights[i] + 1) >> 1;
176 m->tot_weight += m->weights[i];
179 scale = 0x80000000u / m->tot_weight;
180 for (i = 0; i < m->num_syms; i++) {
181 m->freqs[i] = sum * scale >> 16;
182 sum += m->weights[i];
185 m->upd_val = m->upd_val * 5 >> 2;
186 if (m->upd_val > m->max_upd_val)
187 m->upd_val = m->max_upd_val;
188 m->till_rescale = m->upd_val;
191 static void model_reset(Model *m)
196 for (i = 0; i < m->num_syms - 1; i++)
198 m->weights[m->num_syms - 1] = 0;
200 m->upd_val = m->num_syms;
202 model_update(m, m->num_syms - 1);
204 m->upd_val = (m->num_syms + 6) >> 1;
207 static av_cold void model_init(Model *m, int num_syms)
209 m->num_syms = num_syms;
210 m->max_upd_val = 8 * num_syms + 48;
215 static void model256_update(Model256 *m, int val)
225 m->tot_weight += m->upd_val;
227 if (m->tot_weight > 0x8000) {
229 for (i = 0; i < 256; i++) {
230 m->weights[i] = (m->weights[i] + 1) >> 1;
231 m->tot_weight += m->weights[i];
234 scale = 0x80000000u / m->tot_weight;
236 for (i = 0; i < 256; i++) {
237 m->freqs[i] = sum * scale >> 16;
238 sum += m->weights[i];
239 send = m->freqs[i] >> MODEL256_SEC_SCALE;
241 m->secondary[sidx++] = i - 1;
243 while (sidx < m->sec_size)
244 m->secondary[sidx++] = 255;
246 m->upd_val = m->upd_val * 5 >> 2;
247 if (m->upd_val > m->max_upd_val)
248 m->upd_val = m->max_upd_val;
249 m->till_rescale = m->upd_val;
252 static void model256_reset(Model256 *m)
256 for (i = 0; i < 255; i++)
263 model256_update(m, 255);
265 m->upd_val = (256 + 6) >> 1;
268 static av_cold void model256_init(Model256 *m)
270 m->max_upd_val = 8 * 256 + 48;
271 m->sec_size = (1 << 6) + 2;
276 static void rac_init(RangeCoder *c, const uint8_t *src, int size)
281 c->src_end = src + size;
283 for (i = 0; i < FFMIN(size, 4); i++)
284 c->low = (c->low << 8) | *c->src++;
285 c->range = 0xFFFFFFFF;
289 static void rac_normalise(RangeCoder *c)
294 if (c->src < c->src_end) {
296 } else if (!c->low) {
300 if (c->range >= RAC_BOTTOM)
305 static int rac_get_bit(RangeCoder *c)
311 bit = (c->range <= c->low);
315 if (c->range < RAC_BOTTOM)
321 static int rac_get_bits(RangeCoder *c, int nbits)
326 val = c->low / c->range;
327 c->low -= c->range * val;
329 if (c->range < RAC_BOTTOM)
335 static int rac_get_model2_sym(RangeCoder *c, Model2 *m)
339 helper = m->zero_freq * (c->range >> MODEL2_SCALE);
340 bit = (c->low >= helper);
348 if (c->range < RAC_BOTTOM)
351 model2_update(m, bit);
356 static int rac_get_model_sym(RangeCoder *c, Model *m)
358 int prob, prob2, helper, val;
363 c->range >>= MODEL_SCALE;
365 end = m->num_syms >> 1;
368 helper = m->freqs[end] * c->range;
369 if (helper <= c->low) {
376 end = (end2 + val) >> 1;
377 } while (end != val);
379 c->range = prob2 - prob;
380 if (c->range < RAC_BOTTOM)
383 model_update(m, val);
388 static int rac_get_model256_sym(RangeCoder *c, Model256 *m)
390 int prob, prob2, helper, val;
395 c->range >>= MODEL_SCALE;
397 helper = c->low / c->range;
398 ssym = helper >> MODEL256_SEC_SCALE;
399 val = m->secondary[ssym];
401 end = start = m->secondary[ssym + 1] + 1;
402 while (end > val + 1) {
403 ssym = (end + val) >> 1;
404 if (m->freqs[ssym] <= helper) {
408 end = (end + val) >> 1;
412 prob = m->freqs[val] * c->range;
414 prob2 = m->freqs[val + 1] * c->range;
417 c->range = prob2 - prob;
418 if (c->range < RAC_BOTTOM)
421 model256_update(m, val);
426 static int decode_block_type(RangeCoder *c, BlockTypeContext *bt)
428 bt->last_type = rac_get_model_sym(c, &bt->bt_model[bt->last_type]);
430 return bt->last_type;
433 static int decode_coeff(RangeCoder *c, Model *m)
437 val = rac_get_model_sym(c, m);
439 sign = rac_get_bit(c);
442 val = (1 << val) + rac_get_bits(c, val);
451 static void decode_fill_block(RangeCoder *c, FillBlockCoder *fc,
452 uint8_t *dst, int stride, int block_size)
456 fc->fill_val += decode_coeff(c, &fc->coef_model);
458 for (i = 0; i < block_size; i++, dst += stride)
459 memset(dst, fc->fill_val, block_size);
462 static void decode_image_block(RangeCoder *c, ImageBlockCoder *ic,
463 uint8_t *dst, int stride, int block_size)
471 vec_size = rac_get_model_sym(c, &ic->vec_size_model) + 2;
472 for (i = 0; i < vec_size; i++)
473 vec[i] = rac_get_model256_sym(c, &ic->vec_entry_model);
476 memset(prev_line, 0, sizeof(prev_line));
478 for (j = 0; j < block_size; j++) {
481 for (i = 0; i < block_size; i++) {
484 A = rac_get_model_sym(c, &ic->vq_model[A + B * 5 + C * 25]);
490 dst[i] = rac_get_model256_sym(c, &ic->esc_model);
496 static int decode_dct(RangeCoder *c, DCTBlockCoder *bc, int *block,
499 int skip, val, sign, pos = 1, zz_pos, dc;
500 int blk_pos = bx + by * bc->prev_dc_stride;
502 memset(block, 0, sizeof(*block) * 64);
504 dc = decode_coeff(c, &bc->dc_model);
509 l = bc->prev_dc[blk_pos - 1];
510 tl = bc->prev_dc[blk_pos - 1 - bc->prev_dc_stride];
511 t = bc->prev_dc[blk_pos - bc->prev_dc_stride];
513 if (FFABS(t - tl) <= FFABS(l - tl))
518 dc += bc->prev_dc[blk_pos - bc->prev_dc_stride];
521 dc += bc->prev_dc[bx - 1];
523 bc->prev_dc[blk_pos] = dc;
524 block[0] = dc * bc->qmat[0];
527 val = rac_get_model256_sym(c, &bc->ac_model);
542 sign = rac_get_model2_sym(c, &bc->sign_model);
545 val = (1 << val) + rac_get_bits(c, val);
550 zz_pos = ff_zigzag_direct[pos];
551 block[zz_pos] = val * bc->qmat[zz_pos];
555 return pos == 64 ? 0 : -1;
558 static void decode_dct_block(RangeCoder *c, DCTBlockCoder *bc,
559 uint8_t *dst, int stride, int block_size,
560 int *block, int mb_x, int mb_y)
564 int nblocks = block_size >> 3;
569 for (j = 0; j < nblocks; j++) {
570 for (i = 0; i < nblocks; i++) {
571 if (decode_dct(c, bc, block, bx + i, by + j)) {
575 ff_mss34_dct_put(dst + i * 8, stride, block);
581 static void decode_haar_block(RangeCoder *c, HaarBlockCoder *hc,
582 uint8_t *dst, int stride, int block_size,
585 const int hsize = block_size >> 1;
586 int A, B, C, D, t1, t2, t3, t4;
589 for (j = 0; j < block_size; j++) {
590 for (i = 0; i < block_size; i++) {
591 if (i < hsize && j < hsize)
592 block[i] = rac_get_model256_sym(c, &hc->coef_model);
594 block[i] = decode_coeff(c, &hc->coef_hi_model);
595 block[i] *= hc->scale;
599 block -= block_size * block_size;
601 for (j = 0; j < hsize; j++) {
602 for (i = 0; i < hsize; i++) {
604 B = block[i + hsize];
605 C = block[i + hsize * block_size];
606 D = block[i + hsize * block_size + hsize];
612 dst[i * 2] = av_clip_uint8(t1 - t2);
613 dst[i * 2 + stride] = av_clip_uint8(t1 + t2);
614 dst[i * 2 + 1] = av_clip_uint8(t3 - t4);
615 dst[i * 2 + 1 + stride] = av_clip_uint8(t3 + t4);
622 static void reset_coders(MSS3Context *ctx, int quality)
626 for (i = 0; i < 3; i++) {
627 ctx->btype[i].last_type = SKIP_BLOCK;
628 for (j = 0; j < 5; j++)
629 model_reset(&ctx->btype[i].bt_model[j]);
630 ctx->fill_coder[i].fill_val = 0;
631 model_reset(&ctx->fill_coder[i].coef_model);
632 model256_reset(&ctx->image_coder[i].esc_model);
633 model256_reset(&ctx->image_coder[i].vec_entry_model);
634 model_reset(&ctx->image_coder[i].vec_size_model);
635 for (j = 0; j < 125; j++)
636 model_reset(&ctx->image_coder[i].vq_model[j]);
637 if (ctx->dct_coder[i].quality != quality) {
638 ctx->dct_coder[i].quality = quality;
639 ff_mss34_gen_quant_mat(ctx->dct_coder[i].qmat, quality, !i);
641 memset(ctx->dct_coder[i].prev_dc, 0,
642 sizeof(*ctx->dct_coder[i].prev_dc) *
643 ctx->dct_coder[i].prev_dc_stride *
644 ctx->dct_coder[i].prev_dc_height);
645 model_reset(&ctx->dct_coder[i].dc_model);
646 model2_reset(&ctx->dct_coder[i].sign_model);
647 model256_reset(&ctx->dct_coder[i].ac_model);
648 if (ctx->haar_coder[i].quality != quality) {
649 ctx->haar_coder[i].quality = quality;
650 ctx->haar_coder[i].scale = 17 - 7 * quality / 50;
652 model_reset(&ctx->haar_coder[i].coef_hi_model);
653 model256_reset(&ctx->haar_coder[i].coef_model);
657 static av_cold void init_coders(MSS3Context *ctx)
661 for (i = 0; i < 3; i++) {
662 for (j = 0; j < 5; j++)
663 model_init(&ctx->btype[i].bt_model[j], 5);
664 model_init(&ctx->fill_coder[i].coef_model, 12);
665 model256_init(&ctx->image_coder[i].esc_model);
666 model256_init(&ctx->image_coder[i].vec_entry_model);
667 model_init(&ctx->image_coder[i].vec_size_model, 3);
668 for (j = 0; j < 125; j++)
669 model_init(&ctx->image_coder[i].vq_model[j], 5);
670 model_init(&ctx->dct_coder[i].dc_model, 12);
671 model256_init(&ctx->dct_coder[i].ac_model);
672 model_init(&ctx->haar_coder[i].coef_hi_model, 12);
673 model256_init(&ctx->haar_coder[i].coef_model);
677 static int mss3_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
680 const uint8_t *buf = avpkt->data;
681 int buf_size = avpkt->size;
682 MSS3Context *c = avctx->priv_data;
683 RangeCoder *acoder = &c->coder;
686 int dec_width, dec_height, dec_x, dec_y, quality, keyframe;
687 int x, y, i, mb_width, mb_height, blk_size, btype;
690 if (buf_size < HEADER_SIZE) {
691 av_log(avctx, AV_LOG_ERROR,
692 "Frame should have at least %d bytes, got %d instead\n",
693 HEADER_SIZE, buf_size);
694 return AVERROR_INVALIDDATA;
697 bytestream2_init(&gb, buf, buf_size);
698 keyframe = bytestream2_get_be32(&gb);
699 if (keyframe & ~0x301) {
700 av_log(avctx, AV_LOG_ERROR, "Invalid frame type %X\n", keyframe);
701 return AVERROR_INVALIDDATA;
703 keyframe = !(keyframe & 1);
704 bytestream2_skip(&gb, 6);
705 dec_x = bytestream2_get_be16(&gb);
706 dec_y = bytestream2_get_be16(&gb);
707 dec_width = bytestream2_get_be16(&gb);
708 dec_height = bytestream2_get_be16(&gb);
710 if (dec_x + dec_width > avctx->width ||
711 dec_y + dec_height > avctx->height ||
712 (dec_width | dec_height) & 0xF) {
713 av_log(avctx, AV_LOG_ERROR, "Invalid frame dimensions %dx%d +%d,%d\n",
714 dec_width, dec_height, dec_x, dec_y);
715 return AVERROR_INVALIDDATA;
717 bytestream2_skip(&gb, 4);
718 quality = bytestream2_get_byte(&gb);
719 if (quality < 1 || quality > 100) {
720 av_log(avctx, AV_LOG_ERROR, "Invalid quality setting %d\n", quality);
721 return AVERROR_INVALIDDATA;
723 bytestream2_skip(&gb, 4);
725 if (keyframe && !bytestream2_get_bytes_left(&gb)) {
726 av_log(avctx, AV_LOG_ERROR, "Keyframe without data found\n");
727 return AVERROR_INVALIDDATA;
729 if (!keyframe && c->got_error)
733 c->pic.reference = 3;
734 c->pic.buffer_hints = FF_BUFFER_HINTS_VALID | FF_BUFFER_HINTS_PRESERVE |
735 FF_BUFFER_HINTS_REUSABLE;
736 if ((ret = avctx->reget_buffer(avctx, &c->pic)) < 0) {
737 av_log(avctx, AV_LOG_ERROR, "reget_buffer() failed\n");
740 c->pic.key_frame = keyframe;
741 c->pic.pict_type = keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
742 if (!bytestream2_get_bytes_left(&gb)) {
743 *data_size = sizeof(AVFrame);
744 *(AVFrame*)data = c->pic;
749 reset_coders(c, quality);
751 rac_init(acoder, buf + HEADER_SIZE, buf_size - HEADER_SIZE);
753 mb_width = dec_width >> 4;
754 mb_height = dec_height >> 4;
755 dst[0] = c->pic.data[0] + dec_x + dec_y * c->pic.linesize[0];
756 dst[1] = c->pic.data[1] + dec_x / 2 + (dec_y / 2) * c->pic.linesize[1];
757 dst[2] = c->pic.data[2] + dec_x / 2 + (dec_y / 2) * c->pic.linesize[2];
758 for (y = 0; y < mb_height; y++) {
759 for (x = 0; x < mb_width; x++) {
760 for (i = 0; i < 3; i++) {
763 btype = decode_block_type(acoder, c->btype + i);
766 decode_fill_block(acoder, c->fill_coder + i,
767 dst[i] + x * blk_size,
768 c->pic.linesize[i], blk_size);
771 decode_image_block(acoder, c->image_coder + i,
772 dst[i] + x * blk_size,
773 c->pic.linesize[i], blk_size);
776 decode_dct_block(acoder, c->dct_coder + i,
777 dst[i] + x * blk_size,
778 c->pic.linesize[i], blk_size,
782 decode_haar_block(acoder, c->haar_coder + i,
783 dst[i] + x * blk_size,
784 c->pic.linesize[i], blk_size,
788 if (c->got_error || acoder->got_error) {
789 av_log(avctx, AV_LOG_ERROR, "Error decoding block %d,%d\n",
792 return AVERROR_INVALIDDATA;
796 dst[0] += c->pic.linesize[0] * 16;
797 dst[1] += c->pic.linesize[1] * 8;
798 dst[2] += c->pic.linesize[2] * 8;
801 *data_size = sizeof(AVFrame);
802 *(AVFrame*)data = c->pic;
807 static av_cold int mss3_decode_init(AVCodecContext *avctx)
809 MSS3Context * const c = avctx->priv_data;
814 if ((avctx->width & 0xF) || (avctx->height & 0xF)) {
815 av_log(avctx, AV_LOG_ERROR,
816 "Image dimensions should be a multiple of 16.\n");
817 return AVERROR_INVALIDDATA;
821 for (i = 0; i < 3; i++) {
822 int b_width = avctx->width >> (2 + !!i);
823 int b_height = avctx->height >> (2 + !!i);
824 c->dct_coder[i].prev_dc_stride = b_width;
825 c->dct_coder[i].prev_dc_height = b_height;
826 c->dct_coder[i].prev_dc = av_malloc(sizeof(*c->dct_coder[i].prev_dc) *
828 if (!c->dct_coder[i].prev_dc) {
829 av_log(avctx, AV_LOG_ERROR, "Cannot allocate buffer\n");
831 av_freep(&c->dct_coder[i].prev_dc);
834 return AVERROR(ENOMEM);
838 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
839 avctx->coded_frame = &c->pic;
846 static av_cold int mss3_decode_end(AVCodecContext *avctx)
848 MSS3Context * const c = avctx->priv_data;
852 avctx->release_buffer(avctx, &c->pic);
853 for (i = 0; i < 3; i++)
854 av_freep(&c->dct_coder[i].prev_dc);
859 AVCodec ff_msa1_decoder = {
861 .type = AVMEDIA_TYPE_VIDEO,
862 .id = AV_CODEC_ID_MSA1,
863 .priv_data_size = sizeof(MSS3Context),
864 .init = mss3_decode_init,
865 .close = mss3_decode_end,
866 .decode = mss3_decode_frame,
867 .capabilities = CODEC_CAP_DR1,
868 .long_name = NULL_IF_CONFIG_SMALL("MS ATC Screen"),