xref: /third_party/ffmpeg/libavcodec/mss3.c (revision cabdff1a)
1/*
2 * Microsoft Screen 3 (aka Microsoft ATC Screen) decoder
3 * Copyright (c) 2012 Konstantin Shishkov
4 *
5 * This file is part of FFmpeg.
6 *
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.
11 *
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.
16 *
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
20 */
21
22/**
23 * @file
24 * Microsoft Screen 3 (aka Microsoft ATC Screen) decoder
25 */
26
27#include "avcodec.h"
28#include "bytestream.h"
29#include "codec_internal.h"
30#include "internal.h"
31#include "mathops.h"
32#include "mss34dsp.h"
33
34#define HEADER_SIZE 27
35
36#define MODEL2_SCALE       13
37#define MODEL_SCALE        15
38#define MODEL256_SEC_SCALE  9
39
40typedef struct Model2 {
41    int      upd_val, till_rescale;
42    unsigned zero_freq,  zero_weight;
43    unsigned total_freq, total_weight;
44} Model2;
45
46typedef struct Model {
47    int weights[16], freqs[16];
48    int num_syms;
49    int tot_weight;
50    int upd_val, max_upd_val, till_rescale;
51} Model;
52
53typedef struct Model256 {
54    int weights[256], freqs[256];
55    int tot_weight;
56    int secondary[68];
57    int sec_size;
58    int upd_val, max_upd_val, till_rescale;
59} Model256;
60
61#define RAC_BOTTOM 0x01000000
62typedef struct RangeCoder {
63    const uint8_t *src, *src_end;
64
65    uint32_t range, low;
66    int got_error;
67} RangeCoder;
68
69enum BlockType {
70    FILL_BLOCK = 0,
71    IMAGE_BLOCK,
72    DCT_BLOCK,
73    HAAR_BLOCK,
74    SKIP_BLOCK
75};
76
77typedef struct BlockTypeContext {
78    int      last_type;
79    Model    bt_model[5];
80} BlockTypeContext;
81
82typedef struct FillBlockCoder {
83    int      fill_val;
84    Model    coef_model;
85} FillBlockCoder;
86
87typedef struct ImageBlockCoder {
88    Model256 esc_model, vec_entry_model;
89    Model    vec_size_model;
90    Model    vq_model[125];
91} ImageBlockCoder;
92
93typedef struct DCTBlockCoder {
94    int      *prev_dc;
95    ptrdiff_t prev_dc_stride;
96    int      prev_dc_height;
97    int      quality;
98    uint16_t qmat[64];
99    Model    dc_model;
100    Model2   sign_model;
101    Model256 ac_model;
102} DCTBlockCoder;
103
104typedef struct HaarBlockCoder {
105    int      quality, scale;
106    Model256 coef_model;
107    Model    coef_hi_model;
108} HaarBlockCoder;
109
110typedef struct MSS3Context {
111    AVCodecContext   *avctx;
112    AVFrame          *pic;
113
114    int              got_error;
115    RangeCoder       coder;
116    BlockTypeContext btype[3];
117    FillBlockCoder   fill_coder[3];
118    ImageBlockCoder  image_coder[3];
119    DCTBlockCoder    dct_coder[3];
120    HaarBlockCoder   haar_coder[3];
121
122    int              dctblock[64];
123    int              hblock[16 * 16];
124} MSS3Context;
125
126
127static void model2_reset(Model2 *m)
128{
129    m->zero_weight  = 1;
130    m->total_weight = 2;
131    m->zero_freq    = 0x1000;
132    m->total_freq   = 0x2000;
133    m->upd_val      = 4;
134    m->till_rescale = 4;
135}
136
137static void model2_update(Model2 *m, int bit)
138{
139    unsigned scale;
140
141    if (!bit)
142        m->zero_weight++;
143    m->till_rescale--;
144    if (m->till_rescale)
145        return;
146
147    m->total_weight += m->upd_val;
148    if (m->total_weight > 0x2000) {
149        m->total_weight = (m->total_weight + 1) >> 1;
150        m->zero_weight  = (m->zero_weight  + 1) >> 1;
151        if (m->total_weight == m->zero_weight)
152            m->total_weight = m->zero_weight + 1;
153    }
154    m->upd_val = m->upd_val * 5 >> 2;
155    if (m->upd_val > 64)
156        m->upd_val = 64;
157    scale = 0x80000000u / m->total_weight;
158    m->zero_freq    = m->zero_weight  * scale >> 18;
159    m->total_freq   = m->total_weight * scale >> 18;
160    m->till_rescale = m->upd_val;
161}
162
163static void model_update(Model *m, int val)
164{
165    int i, sum = 0;
166    unsigned scale;
167
168    m->weights[val]++;
169    m->till_rescale--;
170    if (m->till_rescale)
171        return;
172    m->tot_weight += m->upd_val;
173
174    if (m->tot_weight > 0x8000) {
175        m->tot_weight = 0;
176        for (i = 0; i < m->num_syms; i++) {
177            m->weights[i]  = (m->weights[i] + 1) >> 1;
178            m->tot_weight +=  m->weights[i];
179        }
180    }
181    scale = 0x80000000u / m->tot_weight;
182    for (i = 0; i < m->num_syms; i++) {
183        m->freqs[i] = sum * scale >> 16;
184        sum += m->weights[i];
185    }
186
187    m->upd_val = m->upd_val * 5 >> 2;
188    if (m->upd_val > m->max_upd_val)
189        m->upd_val = m->max_upd_val;
190    m->till_rescale = m->upd_val;
191}
192
193static void model_reset(Model *m)
194{
195    int i;
196
197    m->tot_weight   = 0;
198    for (i = 0; i < m->num_syms - 1; i++)
199        m->weights[i] = 1;
200    m->weights[m->num_syms - 1] = 0;
201
202    m->upd_val      = m->num_syms;
203    m->till_rescale = 1;
204    model_update(m, m->num_syms - 1);
205    m->till_rescale =
206    m->upd_val      = (m->num_syms + 6) >> 1;
207}
208
209static av_cold void model_init(Model *m, int num_syms)
210{
211    m->num_syms    = num_syms;
212    m->max_upd_val = 8 * num_syms + 48;
213
214    model_reset(m);
215}
216
217static void model256_update(Model256 *m, int val)
218{
219    int i, sum = 0;
220    unsigned scale;
221    int send, sidx = 1;
222
223    m->weights[val]++;
224    m->till_rescale--;
225    if (m->till_rescale)
226        return;
227    m->tot_weight += m->upd_val;
228
229    if (m->tot_weight > 0x8000) {
230        m->tot_weight = 0;
231        for (i = 0; i < 256; i++) {
232            m->weights[i]  = (m->weights[i] + 1) >> 1;
233            m->tot_weight +=  m->weights[i];
234        }
235    }
236    scale = 0x80000000u / m->tot_weight;
237    m->secondary[0] = 0;
238    for (i = 0; i < 256; i++) {
239        m->freqs[i] = sum * scale >> 16;
240        sum += m->weights[i];
241        send = m->freqs[i] >> MODEL256_SEC_SCALE;
242        while (sidx <= send)
243            m->secondary[sidx++] = i - 1;
244    }
245    while (sidx < m->sec_size)
246        m->secondary[sidx++] = 255;
247
248    m->upd_val = m->upd_val * 5 >> 2;
249    if (m->upd_val > m->max_upd_val)
250        m->upd_val = m->max_upd_val;
251    m->till_rescale = m->upd_val;
252}
253
254static void model256_reset(Model256 *m)
255{
256    int i;
257
258    for (i = 0; i < 255; i++)
259        m->weights[i] = 1;
260    m->weights[255] = 0;
261
262    m->tot_weight   = 0;
263    m->upd_val      = 256;
264    m->till_rescale = 1;
265    model256_update(m, 255);
266    m->till_rescale =
267    m->upd_val      = (256 + 6) >> 1;
268}
269
270static av_cold void model256_init(Model256 *m)
271{
272    m->max_upd_val = 8 * 256 + 48;
273    m->sec_size    = (1 << 6) + 2;
274
275    model256_reset(m);
276}
277
278static void rac_init(RangeCoder *c, const uint8_t *src, int size)
279{
280    int i;
281
282    c->src       = src;
283    c->src_end   = src + size;
284    c->low       = 0;
285    for (i = 0; i < FFMIN(size, 4); i++)
286        c->low = (c->low << 8) | *c->src++;
287    c->range     = 0xFFFFFFFF;
288    c->got_error = 0;
289}
290
291static void rac_normalise(RangeCoder *c)
292{
293    for (;;) {
294        c->range <<= 8;
295        c->low   <<= 8;
296        if (c->src < c->src_end) {
297            c->low |= *c->src++;
298        } else if (!c->low) {
299            c->got_error = 1;
300            c->low = 1;
301        }
302        if (c->low > c->range) {
303            c->got_error = 1;
304            c->low = 1;
305        }
306        if (c->range >= RAC_BOTTOM)
307            return;
308    }
309}
310
311static int rac_get_bit(RangeCoder *c)
312{
313    int bit;
314
315    c->range >>= 1;
316
317    bit = (c->range <= c->low);
318    if (bit)
319        c->low -= c->range;
320
321    if (c->range < RAC_BOTTOM)
322        rac_normalise(c);
323
324    return bit;
325}
326
327static int rac_get_bits(RangeCoder *c, int nbits)
328{
329    int val;
330
331    c->range >>= nbits;
332    val = c->low / c->range;
333    c->low -= c->range * val;
334
335    if (c->range < RAC_BOTTOM)
336        rac_normalise(c);
337
338    return val;
339}
340
341static int rac_get_model2_sym(RangeCoder *c, Model2 *m)
342{
343    int bit, helper;
344
345    helper = m->zero_freq * (c->range >> MODEL2_SCALE);
346    bit    = (c->low >= helper);
347    if (bit) {
348        c->low   -= helper;
349        c->range -= helper;
350    } else {
351        c->range  = helper;
352    }
353
354    if (c->range < RAC_BOTTOM)
355        rac_normalise(c);
356
357    model2_update(m, bit);
358
359    return bit;
360}
361
362static int rac_get_model_sym(RangeCoder *c, Model *m)
363{
364    int val;
365    int end, end2;
366    unsigned prob, prob2, helper;
367
368    prob       = 0;
369    prob2      = c->range;
370    c->range >>= MODEL_SCALE;
371    val        = 0;
372    end        = m->num_syms >> 1;
373    end2       = m->num_syms;
374    do {
375        helper = m->freqs[end] * c->range;
376        if (helper <= c->low) {
377            val   = end;
378            prob  = helper;
379        } else {
380            end2  = end;
381            prob2 = helper;
382        }
383        end = (end2 + val) >> 1;
384    } while (end != val);
385    c->low  -= prob;
386    c->range = prob2 - prob;
387    if (c->range < RAC_BOTTOM)
388        rac_normalise(c);
389
390    model_update(m, val);
391
392    return val;
393}
394
395static int rac_get_model256_sym(RangeCoder *c, Model256 *m)
396{
397    int val;
398    int start, end;
399    int ssym;
400    unsigned prob, prob2, helper;
401
402    prob2      = c->range;
403    c->range >>= MODEL_SCALE;
404
405    helper     = c->low / c->range;
406    ssym       = helper >> MODEL256_SEC_SCALE;
407    val        = m->secondary[ssym];
408
409    end = start = m->secondary[ssym + 1] + 1;
410    while (end > val + 1) {
411        ssym = (end + val) >> 1;
412        if (m->freqs[ssym] <= helper) {
413            end = start;
414            val = ssym;
415        } else {
416            end   = (end + val) >> 1;
417            start = ssym;
418        }
419    }
420    prob = m->freqs[val] * c->range;
421    if (val != 255)
422        prob2 = m->freqs[val + 1] * c->range;
423
424    c->low  -= prob;
425    c->range = prob2 - prob;
426    if (c->range < RAC_BOTTOM)
427        rac_normalise(c);
428
429    model256_update(m, val);
430
431    return val;
432}
433
434static int decode_block_type(RangeCoder *c, BlockTypeContext *bt)
435{
436    bt->last_type = rac_get_model_sym(c, &bt->bt_model[bt->last_type]);
437
438    return bt->last_type;
439}
440
441static int decode_coeff(RangeCoder *c, Model *m)
442{
443    int val, sign;
444
445    val = rac_get_model_sym(c, m);
446    if (val) {
447        sign = rac_get_bit(c);
448        if (val > 1) {
449            val--;
450            val = (1 << val) + rac_get_bits(c, val);
451        }
452        if (!sign)
453            val = -val;
454    }
455
456    return val;
457}
458
459static void decode_fill_block(RangeCoder *c, FillBlockCoder *fc,
460                              uint8_t *dst, ptrdiff_t stride, int block_size)
461{
462    int i;
463
464    fc->fill_val += decode_coeff(c, &fc->coef_model);
465
466    for (i = 0; i < block_size; i++, dst += stride)
467        memset(dst, fc->fill_val, block_size);
468}
469
470static void decode_image_block(RangeCoder *c, ImageBlockCoder *ic,
471                               uint8_t *dst, ptrdiff_t stride, int block_size)
472{
473    int i, j;
474    int vec_size;
475    int vec[4];
476    int prev_line[16];
477    int A, B, C;
478
479    vec_size = rac_get_model_sym(c, &ic->vec_size_model) + 2;
480    for (i = 0; i < vec_size; i++)
481        vec[i] = rac_get_model256_sym(c, &ic->vec_entry_model);
482    for (; i < 4; i++)
483        vec[i] = 0;
484    memset(prev_line, 0, sizeof(prev_line));
485
486    for (j = 0; j < block_size; j++) {
487        A = 0;
488        B = 0;
489        for (i = 0; i < block_size; i++) {
490            C = B;
491            B = prev_line[i];
492            A = rac_get_model_sym(c, &ic->vq_model[A + B * 5 + C * 25]);
493
494            prev_line[i] = A;
495            if (A < 4)
496               dst[i] = vec[A];
497            else
498               dst[i] = rac_get_model256_sym(c, &ic->esc_model);
499        }
500        dst += stride;
501    }
502}
503
504static int decode_dct(RangeCoder *c, DCTBlockCoder *bc, int *block,
505                      int bx, int by)
506{
507    int skip, val, sign, pos = 1, zz_pos, dc;
508    int blk_pos = bx + by * bc->prev_dc_stride;
509
510    memset(block, 0, sizeof(*block) * 64);
511
512    dc = decode_coeff(c, &bc->dc_model);
513    if (by) {
514        if (bx) {
515            int l, tl, t;
516
517            l  = bc->prev_dc[blk_pos - 1];
518            tl = bc->prev_dc[blk_pos - 1 - bc->prev_dc_stride];
519            t  = bc->prev_dc[blk_pos     - bc->prev_dc_stride];
520
521            if (FFABS(t - tl) <= FFABS(l - tl))
522                dc += l;
523            else
524                dc += t;
525        } else {
526            dc += bc->prev_dc[blk_pos - bc->prev_dc_stride];
527        }
528    } else if (bx) {
529        dc += bc->prev_dc[bx - 1];
530    }
531    bc->prev_dc[blk_pos] = dc;
532    block[0]             = dc * bc->qmat[0];
533
534    while (pos < 64) {
535        val = rac_get_model256_sym(c, &bc->ac_model);
536        if (!val)
537            return 0;
538        if (val == 0xF0) {
539            pos += 16;
540            continue;
541        }
542        skip = val >> 4;
543        val  = val & 0xF;
544        if (!val)
545            return -1;
546        pos += skip;
547        if (pos >= 64)
548            return -1;
549
550        sign = rac_get_model2_sym(c, &bc->sign_model);
551        if (val > 1) {
552            val--;
553            val = (1 << val) + rac_get_bits(c, val);
554        }
555        if (!sign)
556            val = -val;
557
558        zz_pos = ff_zigzag_direct[pos];
559        block[zz_pos] = val * bc->qmat[zz_pos];
560        pos++;
561    }
562
563    return pos == 64 ? 0 : -1;
564}
565
566static void decode_dct_block(RangeCoder *c, DCTBlockCoder *bc,
567                             uint8_t *dst, ptrdiff_t stride, int block_size,
568                             int *block, int mb_x, int mb_y)
569{
570    int i, j;
571    int bx, by;
572    int nblocks = block_size >> 3;
573
574    bx = mb_x * nblocks;
575    by = mb_y * nblocks;
576
577    for (j = 0; j < nblocks; j++) {
578        for (i = 0; i < nblocks; i++) {
579            if (decode_dct(c, bc, block, bx + i, by + j)) {
580                c->got_error = 1;
581                return;
582            }
583            ff_mss34_dct_put(dst + i * 8, stride, block);
584        }
585        dst += 8 * stride;
586    }
587}
588
589static void decode_haar_block(RangeCoder *c, HaarBlockCoder *hc,
590                              uint8_t *dst, ptrdiff_t stride,
591                              int block_size, int *block)
592{
593    const int hsize = block_size >> 1;
594    int A, B, C, D, t1, t2, t3, t4;
595    int i, j;
596
597    for (j = 0; j < block_size; j++) {
598        for (i = 0; i < block_size; i++) {
599            if (i < hsize && j < hsize)
600                block[i] = rac_get_model256_sym(c, &hc->coef_model);
601            else
602                block[i] = decode_coeff(c, &hc->coef_hi_model);
603            block[i] *= hc->scale;
604        }
605        block += block_size;
606    }
607    block -= block_size * block_size;
608
609    for (j = 0; j < hsize; j++) {
610        for (i = 0; i < hsize; i++) {
611            A = block[i];
612            B = block[i + hsize];
613            C = block[i + hsize * block_size];
614            D = block[i + hsize * block_size + hsize];
615
616            t1 = A - B;
617            t2 = C - D;
618            t3 = A + B;
619            t4 = C + D;
620            dst[i * 2]              = av_clip_uint8(t1 - t2);
621            dst[i * 2 + stride]     = av_clip_uint8(t1 + t2);
622            dst[i * 2 + 1]          = av_clip_uint8(t3 - t4);
623            dst[i * 2 + 1 + stride] = av_clip_uint8(t3 + t4);
624        }
625        block += block_size;
626        dst   += stride * 2;
627    }
628}
629
630static void reset_coders(MSS3Context *ctx, int quality)
631{
632    int i, j;
633
634    for (i = 0; i < 3; i++) {
635        ctx->btype[i].last_type = SKIP_BLOCK;
636        for (j = 0; j < 5; j++)
637            model_reset(&ctx->btype[i].bt_model[j]);
638        ctx->fill_coder[i].fill_val = 0;
639        model_reset(&ctx->fill_coder[i].coef_model);
640        model256_reset(&ctx->image_coder[i].esc_model);
641        model256_reset(&ctx->image_coder[i].vec_entry_model);
642        model_reset(&ctx->image_coder[i].vec_size_model);
643        for (j = 0; j < 125; j++)
644            model_reset(&ctx->image_coder[i].vq_model[j]);
645        if (ctx->dct_coder[i].quality != quality) {
646            ctx->dct_coder[i].quality = quality;
647            ff_mss34_gen_quant_mat(ctx->dct_coder[i].qmat, quality, !i);
648        }
649        memset(ctx->dct_coder[i].prev_dc, 0,
650               sizeof(*ctx->dct_coder[i].prev_dc) *
651               ctx->dct_coder[i].prev_dc_stride *
652               ctx->dct_coder[i].prev_dc_height);
653        model_reset(&ctx->dct_coder[i].dc_model);
654        model2_reset(&ctx->dct_coder[i].sign_model);
655        model256_reset(&ctx->dct_coder[i].ac_model);
656        if (ctx->haar_coder[i].quality != quality) {
657            ctx->haar_coder[i].quality = quality;
658            ctx->haar_coder[i].scale   = 17 - 7 * quality / 50;
659        }
660        model_reset(&ctx->haar_coder[i].coef_hi_model);
661        model256_reset(&ctx->haar_coder[i].coef_model);
662    }
663}
664
665static av_cold void init_coders(MSS3Context *ctx)
666{
667    int i, j;
668
669    for (i = 0; i < 3; i++) {
670        for (j = 0; j < 5; j++)
671            model_init(&ctx->btype[i].bt_model[j], 5);
672        model_init(&ctx->fill_coder[i].coef_model, 12);
673        model256_init(&ctx->image_coder[i].esc_model);
674        model256_init(&ctx->image_coder[i].vec_entry_model);
675        model_init(&ctx->image_coder[i].vec_size_model, 3);
676        for (j = 0; j < 125; j++)
677            model_init(&ctx->image_coder[i].vq_model[j], 5);
678        model_init(&ctx->dct_coder[i].dc_model, 12);
679        model256_init(&ctx->dct_coder[i].ac_model);
680        model_init(&ctx->haar_coder[i].coef_hi_model, 12);
681        model256_init(&ctx->haar_coder[i].coef_model);
682    }
683}
684
685static int mss3_decode_frame(AVCodecContext *avctx, AVFrame *rframe,
686                             int *got_frame, AVPacket *avpkt)
687{
688    const uint8_t *buf = avpkt->data;
689    int buf_size = avpkt->size;
690    MSS3Context *c = avctx->priv_data;
691    RangeCoder *acoder = &c->coder;
692    GetByteContext gb;
693    uint8_t *dst[3];
694    int dec_width, dec_height, dec_x, dec_y, quality, keyframe;
695    int x, y, i, mb_width, mb_height, blk_size, btype;
696    int ret;
697
698    if (buf_size < HEADER_SIZE) {
699        av_log(avctx, AV_LOG_ERROR,
700               "Frame should have at least %d bytes, got %d instead\n",
701               HEADER_SIZE, buf_size);
702        return AVERROR_INVALIDDATA;
703    }
704
705    bytestream2_init(&gb, buf, buf_size);
706    keyframe   = bytestream2_get_be32(&gb);
707    if (keyframe & ~0x301) {
708        av_log(avctx, AV_LOG_ERROR, "Invalid frame type %X\n", keyframe);
709        return AVERROR_INVALIDDATA;
710    }
711    keyframe   = !(keyframe & 1);
712    bytestream2_skip(&gb, 6);
713    dec_x      = bytestream2_get_be16(&gb);
714    dec_y      = bytestream2_get_be16(&gb);
715    dec_width  = bytestream2_get_be16(&gb);
716    dec_height = bytestream2_get_be16(&gb);
717
718    if (dec_x + dec_width > avctx->width ||
719        dec_y + dec_height > avctx->height ||
720        (dec_width | dec_height) & 0xF) {
721        av_log(avctx, AV_LOG_ERROR, "Invalid frame dimensions %dx%d +%d,%d\n",
722               dec_width, dec_height, dec_x, dec_y);
723        return AVERROR_INVALIDDATA;
724    }
725    bytestream2_skip(&gb, 4);
726    quality    = bytestream2_get_byte(&gb);
727    if (quality < 1 || quality > 100) {
728        av_log(avctx, AV_LOG_ERROR, "Invalid quality setting %d\n", quality);
729        return AVERROR_INVALIDDATA;
730    }
731    bytestream2_skip(&gb, 4);
732
733    if (keyframe && !bytestream2_get_bytes_left(&gb)) {
734        av_log(avctx, AV_LOG_ERROR, "Keyframe without data found\n");
735        return AVERROR_INVALIDDATA;
736    }
737    if (!keyframe && c->got_error)
738        return buf_size;
739    c->got_error = 0;
740
741    if ((ret = ff_reget_buffer(avctx, c->pic, 0)) < 0)
742        return ret;
743    c->pic->key_frame = keyframe;
744    c->pic->pict_type = keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
745    if (!bytestream2_get_bytes_left(&gb)) {
746        if ((ret = av_frame_ref(rframe, c->pic)) < 0)
747            return ret;
748        *got_frame      = 1;
749
750        return buf_size;
751    }
752
753    reset_coders(c, quality);
754
755    rac_init(acoder, buf + HEADER_SIZE, buf_size - HEADER_SIZE);
756
757    mb_width  = dec_width  >> 4;
758    mb_height = dec_height >> 4;
759    dst[0] = c->pic->data[0] + dec_x     +  dec_y      * c->pic->linesize[0];
760    dst[1] = c->pic->data[1] + dec_x / 2 + (dec_y / 2) * c->pic->linesize[1];
761    dst[2] = c->pic->data[2] + dec_x / 2 + (dec_y / 2) * c->pic->linesize[2];
762    for (y = 0; y < mb_height; y++) {
763        for (x = 0; x < mb_width; x++) {
764            for (i = 0; i < 3; i++) {
765                blk_size = 8 << !i;
766
767                btype = decode_block_type(acoder, c->btype + i);
768                switch (btype) {
769                case FILL_BLOCK:
770                    decode_fill_block(acoder, c->fill_coder + i,
771                                      dst[i] + x * blk_size,
772                                      c->pic->linesize[i], blk_size);
773                    break;
774                case IMAGE_BLOCK:
775                    decode_image_block(acoder, c->image_coder + i,
776                                       dst[i] + x * blk_size,
777                                       c->pic->linesize[i], blk_size);
778                    break;
779                case DCT_BLOCK:
780                    decode_dct_block(acoder, c->dct_coder + i,
781                                     dst[i] + x * blk_size,
782                                     c->pic->linesize[i], blk_size,
783                                     c->dctblock, x, y);
784                    break;
785                case HAAR_BLOCK:
786                    decode_haar_block(acoder, c->haar_coder + i,
787                                      dst[i] + x * blk_size,
788                                      c->pic->linesize[i], blk_size,
789                                      c->hblock);
790                    break;
791                }
792                if (c->got_error || acoder->got_error) {
793                    av_log(avctx, AV_LOG_ERROR, "Error decoding block %d,%d\n",
794                           x, y);
795                    c->got_error = 1;
796                    return AVERROR_INVALIDDATA;
797                }
798            }
799        }
800        dst[0] += c->pic->linesize[0] * 16;
801        dst[1] += c->pic->linesize[1] * 8;
802        dst[2] += c->pic->linesize[2] * 8;
803    }
804
805    if ((ret = av_frame_ref(rframe, c->pic)) < 0)
806        return ret;
807
808    *got_frame      = 1;
809
810    return buf_size;
811}
812
813static av_cold int mss3_decode_end(AVCodecContext *avctx)
814{
815    MSS3Context * const c = avctx->priv_data;
816    int i;
817
818    av_frame_free(&c->pic);
819    for (i = 0; i < 3; i++)
820        av_freep(&c->dct_coder[i].prev_dc);
821
822    return 0;
823}
824
825static av_cold int mss3_decode_init(AVCodecContext *avctx)
826{
827    MSS3Context * const c = avctx->priv_data;
828    int i;
829
830    c->avctx = avctx;
831
832    if ((avctx->width & 0xF) || (avctx->height & 0xF)) {
833        av_log(avctx, AV_LOG_ERROR,
834               "Image dimensions should be a multiple of 16.\n");
835        return AVERROR_INVALIDDATA;
836    }
837
838    c->got_error = 0;
839    for (i = 0; i < 3; i++) {
840        int b_width  = avctx->width  >> (2 + !!i);
841        int b_height = avctx->height >> (2 + !!i);
842        c->dct_coder[i].prev_dc_stride = b_width;
843        c->dct_coder[i].prev_dc_height = b_height;
844        c->dct_coder[i].prev_dc = av_malloc(sizeof(*c->dct_coder[i].prev_dc) *
845                                            b_width * b_height);
846        if (!c->dct_coder[i].prev_dc) {
847            av_log(avctx, AV_LOG_ERROR, "Cannot allocate buffer\n");
848            return AVERROR(ENOMEM);
849        }
850    }
851
852    c->pic = av_frame_alloc();
853    if (!c->pic)
854        return AVERROR(ENOMEM);
855
856    avctx->pix_fmt     = AV_PIX_FMT_YUV420P;
857
858    init_coders(c);
859
860    return 0;
861}
862
863const FFCodec ff_msa1_decoder = {
864    .p.name         = "msa1",
865    .p.long_name    = NULL_IF_CONFIG_SMALL("MS ATC Screen"),
866    .p.type         = AVMEDIA_TYPE_VIDEO,
867    .p.id           = AV_CODEC_ID_MSA1,
868    .priv_data_size = sizeof(MSS3Context),
869    .init           = mss3_decode_init,
870    .close          = mss3_decode_end,
871    FF_CODEC_DECODE_CB(mss3_decode_frame),
872    .p.capabilities = AV_CODEC_CAP_DR1,
873    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
874};
875