2 * Microsoft Screen 4 (aka Microsoft Expression Encoder 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 4 (aka Microsoft Titanium Screen 2,
25 * aka Microsoft Expression Encoder Screen) decoder
29 #include "bytestream.h"
32 #include "jpegtables.h"
56 static const uint8_t mss4_dc_vlc_lens[2][16] = {
57 { 0, 1, 5, 1, 1, 1, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0 },
58 { 0, 3, 1, 1, 1, 1, 1, 1, 1, 2, 0, 0, 0, 0, 0, 0 }
61 static const uint8_t vec_len_syms[2][4] = {
66 static const uint8_t mss4_vec_entry_vlc_lens[2][16] = {
67 { 0, 2, 2, 3, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
68 { 0, 1, 5, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }
71 static const uint8_t mss4_vec_entry_vlc_syms[2][9] = {
72 { 0, 7, 6, 5, 8, 4, 3, 1, 2 },
73 { 0, 2, 3, 4, 5, 6, 7, 1, 8 }
76 #define MAX_ENTRIES 162
78 typedef struct MSS4Context {
81 VLC dc_vlc[2], ac_vlc[2];
84 uint8_t imgbuf[3][16 * 16];
87 uint16_t quant_mat[2][64];
90 ptrdiff_t dc_stride[3];
96 static av_cold int mss4_init_vlc(VLC *vlc, const uint8_t *lens,
99 uint8_t bits[MAX_ENTRIES];
103 for (i = 0; i < 16; i++) {
104 for (j = 0; j < lens[i]; j++) {
110 return ff_init_vlc_from_lengths(vlc, FFMIN(bits[idx - 1], 9), idx,
111 bits, 1, syms, 1, 1, 0, 0, NULL);
114 static av_cold int mss4_init_vlcs(MSS4Context *ctx)
118 for (i = 0; i < 2; i++) {
119 ret = mss4_init_vlc(&ctx->dc_vlc[i], mss4_dc_vlc_lens[i], NULL);
122 ret = mss4_init_vlc(&ctx->ac_vlc[i],
123 i ? avpriv_mjpeg_bits_ac_chrominance + 1
124 : avpriv_mjpeg_bits_ac_luminance + 1,
125 i ? avpriv_mjpeg_val_ac_chrominance
126 : avpriv_mjpeg_val_ac_luminance);
129 ret = mss4_init_vlc(&ctx->vec_entry_vlc[i], mss4_vec_entry_vlc_lens[i],
130 mss4_vec_entry_vlc_syms[i]);
137 static av_cold void mss4_free_vlcs(MSS4Context *ctx)
141 for (i = 0; i < 2; i++) {
142 ff_free_vlc(&ctx->dc_vlc[i]);
143 ff_free_vlc(&ctx->ac_vlc[i]);
144 ff_free_vlc(&ctx->vec_entry_vlc[i]);
148 /* This function returns values in the range
149 * (-range + 1; -range/2] U [range/2; range - 1)
153 * nbits = 2 -> -3, -2, 2, 3
155 static av_always_inline int get_coeff_bits(GetBitContext *gb, int nbits)
162 val = get_bits(gb, nbits);
163 if (val < (1 << (nbits - 1)))
164 val -= (1 << nbits) - 1;
169 static inline int get_coeff(GetBitContext *gb, VLC *vlc)
171 int val = get_vlc2(gb, vlc->table, vlc->bits, 2);
173 return get_coeff_bits(gb, val);
176 static int mss4_decode_dct(GetBitContext *gb, VLC *dc_vlc, VLC *ac_vlc,
177 int *block, int *dc_cache,
178 int bx, int by, uint16_t *quant_mat)
180 int skip, val, pos = 1, zz_pos, dc;
182 memset(block, 0, sizeof(*block) * 64);
184 dc = get_coeff(gb, dc_vlc);
185 // DC prediction is the same as in MSS3
191 tl = dc_cache[TOP_LEFT];
194 if (FFABS(t - tl) <= FFABS(l - tl))
202 dc += dc_cache[LEFT];
205 block[0] = dc * quant_mat[0];
208 val = get_vlc2(gb, ac_vlc->table, 9, 2);
218 val = get_coeff_bits(gb, val & 0xF);
223 zz_pos = ff_zigzag_direct[pos];
224 block[zz_pos] = val * quant_mat[zz_pos];
228 return pos == 64 ? 0 : -1;
231 static int mss4_decode_dct_block(MSS4Context *c, GetBitContext *gb,
232 uint8_t *dst[3], int mb_x, int mb_y)
235 uint8_t *out = dst[0];
237 for (j = 0; j < 2; j++) {
238 for (i = 0; i < 2; i++) {
239 int xpos = mb_x * 2 + i;
240 c->dc_cache[j][TOP_LEFT] = c->dc_cache[j][TOP];
241 c->dc_cache[j][TOP] = c->prev_dc[0][mb_x * 2 + i];
242 ret = mss4_decode_dct(gb, c->dc_vlc, c->ac_vlc, c->block,
244 xpos, mb_y * 2 + j, c->quant_mat[0]);
247 c->prev_dc[0][mb_x * 2 + i] = c->dc_cache[j][LEFT];
249 ff_mss34_dct_put(out + xpos * 8, c->pic->linesize[0],
252 out += 8 * c->pic->linesize[0];
255 for (i = 1; i < 3; i++) {
256 c->dc_cache[i + 1][TOP_LEFT] = c->dc_cache[i + 1][TOP];
257 c->dc_cache[i + 1][TOP] = c->prev_dc[i][mb_x];
258 ret = mss4_decode_dct(gb, c->dc_vlc + 1, c->ac_vlc + 1,
259 c->block, c->dc_cache[i + 1], mb_x, mb_y,
263 c->prev_dc[i][mb_x] = c->dc_cache[i + 1][LEFT];
265 ff_mss34_dct_put(c->imgbuf[i], 8, c->block);
266 out = dst[i] + mb_x * 16;
267 // Since the DCT block is coded as YUV420 and the whole frame as YUV444,
268 // we need to scale chroma.
269 for (j = 0; j < 16; j++) {
270 for (k = 0; k < 8; k++)
271 AV_WN16A(out + k * 2, c->imgbuf[i][k + (j & ~1) * 4] * 0x101);
272 out += c->pic->linesize[i];
279 static void read_vec_pos(GetBitContext *gb, int *vec_pos, int *sel_flag,
280 int *sel_len, int *prev)
284 for (i = 2; i >= 0; i--) {
289 if ((!i && !y_flag) || get_bits1(gb)) {
290 if (sel_len[i] > 0) {
292 vec_pos[i] = get_bits(gb, sel_len[i]);
293 if (vec_pos[i] >= pval)
296 vec_pos[i] = !prev[i];
300 vec_pos[i] = prev[i];
305 static int get_value_cached(GetBitContext *gb, int vec_pos, uint8_t *vec,
306 int vec_size, int component, int shift, int *prev)
308 if (vec_pos < vec_size)
311 return prev[component];
312 prev[component] = get_bits(gb, 8 - shift) << shift;
313 return prev[component];
316 #define MKVAL(vals) ((vals)[0] | ((vals)[1] << 3) | ((vals)[2] << 6))
318 /* Image mode - the hardest to comprehend MSS4 coding mode.
320 * In this mode all three 16x16 blocks are coded together with a method
321 * remotely similar to the methods employed in MSS1-MSS3.
322 * The idea is that every component has a vector of 1-4 most common symbols
323 * and an escape mode for reading new value from the bitstream. Decoding
324 * consists of retrieving pixel values from the vector or reading new ones
325 * from the bitstream; depending on flags read from the bitstream, these vector
326 * positions can be updated or reused from the state of the previous line
329 static int mss4_decode_image_block(MSS4Context *ctx, GetBitContext *gb,
330 uint8_t *picdst[3], int mb_x, int mb_y)
334 int sel_len[3], sel_flag[3];
335 int i, j, k, mode, split;
336 int prev_vec1 = 0, prev_split = 0;
338 int prev_pix[3] = { 0 };
339 int prev_mode[16] = { 0 };
342 const int val_shift = ctx->quality == 100 ? 0 : 2;
344 for (i = 0; i < 3; i++)
345 dst[i] = ctx->imgbuf[i];
347 for (i = 0; i < 3; i++) {
348 vec_len[i] = vec_len_syms[!!i][get_unary(gb, 0, 3)];
349 for (j = 0; j < vec_len[i]; j++) {
350 vec[i][j] = get_coeff(gb, &ctx->vec_entry_vlc[!!i]);
351 vec[i][j] += ctx->prev_vec[i][j];
352 ctx->prev_vec[i][j] = vec[i][j];
354 sel_flag[i] = vec_len[i] > 1;
355 sel_len[i] = vec_len[i] > 2 ? vec_len[i] - 2 : 0;
358 for (j = 0; j < 16; j++) {
363 vals[0] = vals[1] = vals[2] = 0;
366 mode = get_bits1(gb);
368 split = get_bits(gb, 4);
370 for (i = 0; i < 16; i++) {
372 vals[0] = prev_mode[i] & 7;
373 vals[1] = (prev_mode[i] >> 3) & 7;
374 vals[2] = prev_mode[i] >> 6;
375 if (mode == 1 && i == split) {
376 read_vec_pos(gb, vals, sel_flag, sel_len, vals);
378 } else if (mode == 2) {
380 read_vec_pos(gb, vals, sel_flag, sel_len, vals);
382 for (k = 0; k < 3; k++)
383 *dst[k]++ = get_value_cached(gb, vals[k], vec[k],
385 val_shift, prev_pix);
386 prev_mode[i] = MKVAL(vals);
390 split = get_bits(gb, 4);
391 if (split >= prev_split)
398 vals[0] = prev_mode[0] & 7;
399 vals[1] = (prev_mode[0] >> 3) & 7;
400 vals[2] = prev_mode[0] >> 6;
401 for (i = 0; i < 3; i++) {
402 for (k = 0; k < split; k++) {
403 *dst[i]++ = get_value_cached(gb, vals[i], vec[i],
404 vec_len[i], i, val_shift,
406 prev_mode[k] = MKVAL(vals);
412 vals[0] = prev_vec1 & 7;
413 vals[1] = (prev_vec1 >> 3) & 7;
414 vals[2] = prev_vec1 >> 6;
416 read_vec_pos(gb, vals, sel_flag, sel_len, vals);
417 prev_vec1 = MKVAL(vals);
419 for (i = 0; i < 3; i++) {
420 for (k = 0; k < 16 - split; k++) {
421 *dst[i]++ = get_value_cached(gb, vals[i], vec[i],
422 vec_len[i], i, val_shift,
424 prev_mode[split + k] = MKVAL(vals);
431 for (i = 0; i < 3; i++)
432 for (j = 0; j < 16; j++)
433 memcpy(picdst[i] + mb_x * 16 + j * ctx->pic->linesize[i],
434 ctx->imgbuf[i] + j * 16, 16);
439 static inline void mss4_update_dc_cache(MSS4Context *c, int mb_x)
443 c->dc_cache[0][TOP] = c->prev_dc[0][mb_x * 2 + 1];
444 c->dc_cache[0][LEFT] = 0;
445 c->dc_cache[1][TOP] = 0;
446 c->dc_cache[1][LEFT] = 0;
448 for (i = 0; i < 2; i++)
449 c->prev_dc[0][mb_x * 2 + i] = 0;
451 for (i = 1; i < 3; i++) {
452 c->dc_cache[i + 1][TOP] = c->prev_dc[i][mb_x];
453 c->dc_cache[i + 1][LEFT] = 0;
454 c->prev_dc[i][mb_x] = 0;
458 static int mss4_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
461 const uint8_t *buf = avpkt->data;
462 int buf_size = avpkt->size;
463 MSS4Context *c = avctx->priv_data;
467 int width, height, quality, frame_type;
468 int x, y, i, mb_width, mb_height, blk_type;
471 if (buf_size < HEADER_SIZE) {
472 av_log(avctx, AV_LOG_ERROR,
473 "Frame should have at least %d bytes, got %d instead\n",
474 HEADER_SIZE, buf_size);
475 return AVERROR_INVALIDDATA;
478 bytestream2_init(&bc, buf, buf_size);
479 width = bytestream2_get_be16(&bc);
480 height = bytestream2_get_be16(&bc);
481 bytestream2_skip(&bc, 2);
482 quality = bytestream2_get_byte(&bc);
483 frame_type = bytestream2_get_byte(&bc);
485 if (width > avctx->width ||
486 height != avctx->height) {
487 av_log(avctx, AV_LOG_ERROR, "Invalid frame dimensions %dx%d\n",
489 return AVERROR_INVALIDDATA;
491 if (quality < 1 || quality > 100) {
492 av_log(avctx, AV_LOG_ERROR, "Invalid quality setting %d\n", quality);
493 return AVERROR_INVALIDDATA;
495 if ((frame_type & ~3) || frame_type == 3) {
496 av_log(avctx, AV_LOG_ERROR, "Invalid frame type %d\n", frame_type);
497 return AVERROR_INVALIDDATA;
500 if (frame_type != SKIP_FRAME && !bytestream2_get_bytes_left(&bc)) {
501 av_log(avctx, AV_LOG_ERROR,
502 "Empty frame found but it is not a skip frame.\n");
503 return AVERROR_INVALIDDATA;
505 mb_width = FFALIGN(width, 16) >> 4;
506 mb_height = FFALIGN(height, 16) >> 4;
508 if (frame_type != SKIP_FRAME && 8*buf_size < 8*HEADER_SIZE + mb_width*mb_height)
509 return AVERROR_INVALIDDATA;
511 if ((ret = ff_reget_buffer(avctx, c->pic, 0)) < 0)
513 c->pic->key_frame = (frame_type == INTRA_FRAME);
514 c->pic->pict_type = (frame_type == INTRA_FRAME) ? AV_PICTURE_TYPE_I
516 if (frame_type == SKIP_FRAME) {
518 if ((ret = av_frame_ref(data, c->pic)) < 0)
524 if (c->quality != quality) {
525 c->quality = quality;
526 for (i = 0; i < 2; i++)
527 ff_mss34_gen_quant_mat(c->quant_mat[i], quality, !i);
530 if ((ret = init_get_bits8(&gb, buf + HEADER_SIZE, buf_size - HEADER_SIZE)) < 0)
532 dst[0] = c->pic->data[0];
533 dst[1] = c->pic->data[1];
534 dst[2] = c->pic->data[2];
536 memset(c->prev_vec, 0, sizeof(c->prev_vec));
537 for (y = 0; y < mb_height; y++) {
538 memset(c->dc_cache, 0, sizeof(c->dc_cache));
539 for (x = 0; x < mb_width; x++) {
540 blk_type = decode012(&gb);
543 if (mss4_decode_dct_block(c, &gb, dst, x, y) < 0) {
544 av_log(avctx, AV_LOG_ERROR,
545 "Error decoding DCT block %d,%d\n",
547 return AVERROR_INVALIDDATA;
551 if (mss4_decode_image_block(c, &gb, dst, x, y) < 0) {
552 av_log(avctx, AV_LOG_ERROR,
553 "Error decoding VQ block %d,%d\n",
555 return AVERROR_INVALIDDATA;
559 if (frame_type == INTRA_FRAME) {
560 av_log(avctx, AV_LOG_ERROR, "Skip block in intra frame\n");
561 return AVERROR_INVALIDDATA;
565 if (blk_type != DCT_BLOCK)
566 mss4_update_dc_cache(c, x);
568 dst[0] += c->pic->linesize[0] * 16;
569 dst[1] += c->pic->linesize[1] * 16;
570 dst[2] += c->pic->linesize[2] * 16;
573 if ((ret = av_frame_ref(data, c->pic)) < 0)
581 static av_cold int mss4_decode_end(AVCodecContext *avctx)
583 MSS4Context * const c = avctx->priv_data;
586 av_frame_free(&c->pic);
587 for (i = 0; i < 3; i++)
588 av_freep(&c->prev_dc[i]);
594 static av_cold int mss4_decode_init(AVCodecContext *avctx)
596 MSS4Context * const c = avctx->priv_data;
599 if (mss4_init_vlcs(c)) {
600 av_log(avctx, AV_LOG_ERROR, "Cannot initialise VLCs\n");
601 return AVERROR(ENOMEM);
603 for (i = 0; i < 3; i++) {
604 c->dc_stride[i] = FFALIGN(avctx->width, 16) >> (2 + !!i);
605 c->prev_dc[i] = av_malloc_array(c->dc_stride[i], sizeof(**c->prev_dc));
606 if (!c->prev_dc[i]) {
607 av_log(avctx, AV_LOG_ERROR, "Cannot allocate buffer\n");
608 return AVERROR(ENOMEM);
612 c->pic = av_frame_alloc();
614 return AVERROR(ENOMEM);
616 avctx->pix_fmt = AV_PIX_FMT_YUV444P;
621 AVCodec ff_mts2_decoder = {
623 .long_name = NULL_IF_CONFIG_SMALL("MS Expression Encoder Screen"),
624 .type = AVMEDIA_TYPE_VIDEO,
625 .id = AV_CODEC_ID_MTS2,
626 .priv_data_size = sizeof(MSS4Context),
627 .init = mss4_decode_init,
628 .close = mss4_decode_end,
629 .decode = mss4_decode_frame,
630 .capabilities = AV_CODEC_CAP_DR1,
631 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP | FF_CODEC_CAP_INIT_THREADSAFE,