2 * Westwood Studios VQA Video Decoder
3 * Copyright (C) 2003 the ffmpeg project
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
25 * @author Mike Melanson (melanson@pcisys.net)
26 * @see http://wiki.multimedia.cx/index.php?title=VQA
28 * The VQA video decoder outputs PAL8 or RGB555 colorspace data, depending
29 * on the type of data in the file.
31 * This decoder needs the 42-byte VQHD header from the beginning
32 * of the VQA file passed through the extradata field. The VQHD header
35 * bytes 0-3 chunk fourcc: 'VQHD'
36 * bytes 4-7 chunk size in big-endian format, should be 0x0000002A
37 * bytes 8-49 VQHD chunk data
39 * Bytes 8-49 are what this decoder expects to see.
41 * Briefly, VQA is a vector quantized animation format that operates in a
42 * VGA palettized colorspace. It operates on pixel vectors (blocks)
43 * of either 4x2 or 4x4 in size. Compressed VQA chunks can contain vector
44 * codebooks, palette information, and code maps for rendering vectors onto
45 * frames. Any of these components can also be compressed with a run-length
46 * encoding (RLE) algorithm commonly referred to as "format80".
48 * VQA takes a novel approach to rate control. Each group of n frames
49 * (usually, n = 8) relies on a different vector codebook. Rather than
50 * transporting an entire codebook every 8th frame, the new codebook is
51 * broken up into 8 pieces and sent along with the compressed video chunks
52 * for each of the 8 frames preceding the 8 frames which require the
53 * codebook. A full codebook is also sent on the very first frame of a
54 * file. This is an interesting technique, although it makes random file
55 * seeking difficult despite the fact that the frames are all intracoded.
57 * V1,2 VQA uses 12-bit codebook indexes. If the 12-bit indexes were
58 * packed into bytes and then RLE compressed, bytewise, the results would
59 * be poor. That is why the coding method divides each index into 2 parts,
60 * the top 4 bits and the bottom 8 bits, then RL encodes the 4-bit pieces
61 * together and the 8-bit pieces together. If most of the vectors are
62 * clustered into one group of 256 vectors, most of the 4-bit index pieces
70 #include "libavutil/intreadwrite.h"
71 #include "libavutil/imgutils.h"
73 #include "bytestream.h"
76 #define PALETTE_COUNT 256
77 #define VQA_HEADER_SIZE 0x2A
79 /* allocate the maximum vector space, regardless of the file version:
80 * (0xFF00 codebook vectors + 0x100 solid pixel vectors) * (4x4 pixels/block) */
81 #define MAX_CODEBOOK_VECTORS 0xFF00
82 #define SOLID_PIXEL_VECTORS 0x100
83 #define MAX_VECTORS (MAX_CODEBOOK_VECTORS + SOLID_PIXEL_VECTORS)
84 #define MAX_CODEBOOK_SIZE (MAX_VECTORS * 4 * 4)
86 #define CBF0_TAG MKBETAG('C', 'B', 'F', '0')
87 #define CBFZ_TAG MKBETAG('C', 'B', 'F', 'Z')
88 #define CBP0_TAG MKBETAG('C', 'B', 'P', '0')
89 #define CBPZ_TAG MKBETAG('C', 'B', 'P', 'Z')
90 #define CPL0_TAG MKBETAG('C', 'P', 'L', '0')
91 #define CPLZ_TAG MKBETAG('C', 'P', 'L', 'Z')
92 #define VPTZ_TAG MKBETAG('V', 'P', 'T', 'Z')
94 typedef struct VqaContext {
96 AVCodecContext *avctx;
99 uint32_t palette[PALETTE_COUNT];
101 int width; /* width of a frame */
102 int height; /* height of a frame */
103 int vector_width; /* width of individual vector */
104 int vector_height; /* height of individual vector */
105 int vqa_version; /* this should be either 1, 2 or 3 */
107 unsigned char *codebook; /* the current codebook */
109 unsigned char *next_codebook_buffer; /* accumulator for next codebook */
110 int next_codebook_buffer_index;
112 unsigned char *decode_buffer;
113 int decode_buffer_size;
115 /* number of frames to go before replacing codebook */
116 int partial_countdown;
121 static av_cold int vqa_decode_init(AVCodecContext *avctx)
123 VqaContext *s = avctx->priv_data;
124 int i, j, codebook_index, ret;
127 avctx->pix_fmt = AV_PIX_FMT_PAL8;
129 /* make sure the extradata made it */
130 if (s->avctx->extradata_size != VQA_HEADER_SIZE) {
131 av_log(s->avctx, AV_LOG_ERROR, "expected extradata size of %d\n", VQA_HEADER_SIZE);
132 return AVERROR(EINVAL);
135 /* load up the VQA parameters from the header */
136 s->vqa_version = s->avctx->extradata[0];
137 if (s->vqa_version < 1 || s->vqa_version > 3) {
138 av_log(s->avctx, AV_LOG_ERROR, "unsupported version %d\n", s->vqa_version);
139 return AVERROR_PATCHWELCOME;
141 s->width = AV_RL16(&s->avctx->extradata[6]);
142 s->height = AV_RL16(&s->avctx->extradata[8]);
143 if ((ret = av_image_check_size(s->width, s->height, 0, avctx)) < 0) {
144 s->width= s->height= 0;
147 s->vector_width = s->avctx->extradata[10];
148 s->vector_height = s->avctx->extradata[11];
149 s->partial_count = s->partial_countdown = s->avctx->extradata[13];
151 /* the vector dimensions have to meet very stringent requirements */
152 if ((s->vector_width != 4) ||
153 ((s->vector_height != 2) && (s->vector_height != 4))) {
154 /* return without further initialization */
155 return AVERROR_INVALIDDATA;
158 if (s->width % s->vector_width || s->height % s->vector_height) {
159 av_log(avctx, AV_LOG_ERROR, "Image size not multiple of block size\n");
160 return AVERROR_INVALIDDATA;
163 /* allocate codebooks */
164 s->codebook_size = MAX_CODEBOOK_SIZE;
165 s->codebook = av_malloc(s->codebook_size);
168 s->next_codebook_buffer = av_malloc(s->codebook_size);
169 if (!s->next_codebook_buffer)
172 /* allocate decode buffer */
173 s->decode_buffer_size = (s->width / s->vector_width) *
174 (s->height / s->vector_height) * 2;
175 s->decode_buffer = av_malloc(s->decode_buffer_size);
176 if (!s->decode_buffer)
179 /* initialize the solid-color vectors */
180 if (s->vector_height == 4) {
181 codebook_index = 0xFF00 * 16;
182 for (i = 0; i < 256; i++)
183 for (j = 0; j < 16; j++)
184 s->codebook[codebook_index++] = i;
186 codebook_index = 0xF00 * 8;
187 for (i = 0; i < 256; i++)
188 for (j = 0; j < 8; j++)
189 s->codebook[codebook_index++] = i;
191 s->next_codebook_buffer_index = 0;
195 av_freep(&s->codebook);
196 av_freep(&s->next_codebook_buffer);
197 av_freep(&s->decode_buffer);
198 return AVERROR(ENOMEM);
201 #define CHECK_COUNT() \
202 if (dest_index + count > dest_size) { \
203 av_log(s->avctx, AV_LOG_ERROR, "decode_format80 problem: next op would overflow dest_index\n"); \
204 av_log(s->avctx, AV_LOG_ERROR, "current dest_index = %d, count = %d, dest_size = %d\n", \
205 dest_index, count, dest_size); \
206 return AVERROR_INVALIDDATA; \
209 #define CHECK_COPY(idx) \
210 if (idx < 0 || idx + count > dest_size) { \
211 av_log(s->avctx, AV_LOG_ERROR, "decode_format80 problem: next op would overflow dest_index\n"); \
212 av_log(s->avctx, AV_LOG_ERROR, "current src_pos = %d, count = %d, dest_size = %d\n", \
213 src_pos, count, dest_size); \
214 return AVERROR_INVALIDDATA; \
218 static int decode_format80(VqaContext *s, int src_size,
219 unsigned char *dest, int dest_size, int check_size) {
222 int count, opcode, start;
227 start = bytestream2_tell(&s->gb);
228 while (bytestream2_tell(&s->gb) - start < src_size) {
229 opcode = bytestream2_get_byte(&s->gb);
230 av_dlog(s->avctx, "opcode %02X: ", opcode);
232 /* 0x80 means that frame is finished */
236 if (dest_index >= dest_size) {
237 av_log(s->avctx, AV_LOG_ERROR, "decode_format80 problem: dest_index (%d) exceeded dest_size (%d)\n",
238 dest_index, dest_size);
239 return AVERROR_INVALIDDATA;
242 if (opcode == 0xFF) {
244 count = bytestream2_get_le16(&s->gb);
245 src_pos = bytestream2_get_le16(&s->gb);
246 av_dlog(s->avctx, "(1) copy %X bytes from absolute pos %X\n", count, src_pos);
249 for (i = 0; i < count; i++)
250 dest[dest_index + i] = dest[src_pos + i];
253 } else if (opcode == 0xFE) {
255 count = bytestream2_get_le16(&s->gb);
256 color = bytestream2_get_byte(&s->gb);
257 av_dlog(s->avctx, "(2) set %X bytes to %02X\n", count, color);
259 memset(&dest[dest_index], color, count);
262 } else if ((opcode & 0xC0) == 0xC0) {
264 count = (opcode & 0x3F) + 3;
265 src_pos = bytestream2_get_le16(&s->gb);
266 av_dlog(s->avctx, "(3) copy %X bytes from absolute pos %X\n", count, src_pos);
269 for (i = 0; i < count; i++)
270 dest[dest_index + i] = dest[src_pos + i];
273 } else if (opcode > 0x80) {
275 count = opcode & 0x3F;
276 av_dlog(s->avctx, "(4) copy %X bytes from source to dest\n", count);
278 bytestream2_get_buffer(&s->gb, &dest[dest_index], count);
283 count = ((opcode & 0x70) >> 4) + 3;
284 src_pos = bytestream2_get_byte(&s->gb) | ((opcode & 0x0F) << 8);
285 av_dlog(s->avctx, "(5) copy %X bytes from relpos %X\n", count, src_pos);
287 CHECK_COPY(dest_index - src_pos);
288 for (i = 0; i < count; i++)
289 dest[dest_index + i] = dest[dest_index - src_pos + i];
294 /* validate that the entire destination buffer was filled; this is
295 * important for decoding frame maps since each vector needs to have a
296 * codebook entry; it is not important for compressed codebooks because
297 * not every entry needs to be filled */
299 if (dest_index < dest_size)
300 av_log(s->avctx, AV_LOG_ERROR, "decode_format80 problem: decode finished with dest_index (%d) < dest_size (%d)\n",
301 dest_index, dest_size);
303 return 0; // let's display what we decoded anyway
306 static int vqa_decode_chunk(VqaContext *s, AVFrame *frame)
308 unsigned int chunk_type;
309 unsigned int chunk_size;
311 unsigned int index = 0;
313 unsigned char r, g, b;
328 int vector_index = 0;
332 int hibytes = s->decode_buffer_size / 2;
334 /* first, traverse through the frame and find the subchunks */
335 while (bytestream2_get_bytes_left(&s->gb) >= 8) {
337 chunk_type = bytestream2_get_be32u(&s->gb);
338 index = bytestream2_tell(&s->gb);
339 chunk_size = bytestream2_get_be32u(&s->gb);
341 switch (chunk_type) {
372 av_log(s->avctx, AV_LOG_ERROR, "Found unknown chunk type: %c%c%c%c (%08X)\n",
373 (chunk_type >> 24) & 0xFF,
374 (chunk_type >> 16) & 0xFF,
375 (chunk_type >> 8) & 0xFF,
376 (chunk_type >> 0) & 0xFF,
381 byte_skip = chunk_size & 0x01;
382 bytestream2_skip(&s->gb, chunk_size + byte_skip);
385 /* next, deal with the palette */
386 if ((cpl0_chunk != -1) && (cplz_chunk != -1)) {
388 /* a chunk should not have both chunk types */
389 av_log(s->avctx, AV_LOG_ERROR, "problem: found both CPL0 and CPLZ chunks\n");
390 return AVERROR_INVALIDDATA;
393 /* decompress the palette chunk */
394 if (cplz_chunk != -1) {
396 /* yet to be handled */
400 /* convert the RGB palette into the machine's endian format */
401 if (cpl0_chunk != -1) {
403 bytestream2_seek(&s->gb, cpl0_chunk, SEEK_SET);
404 chunk_size = bytestream2_get_be32(&s->gb);
405 /* sanity check the palette size */
406 if (chunk_size / 3 > 256 || chunk_size > bytestream2_get_bytes_left(&s->gb)) {
407 av_log(s->avctx, AV_LOG_ERROR, "problem: found a palette chunk with %d colors\n",
409 return AVERROR_INVALIDDATA;
411 for (i = 0; i < chunk_size / 3; i++) {
412 /* scale by 4 to transform 6-bit palette -> 8-bit */
413 r = bytestream2_get_byteu(&s->gb) * 4;
414 g = bytestream2_get_byteu(&s->gb) * 4;
415 b = bytestream2_get_byteu(&s->gb) * 4;
416 s->palette[i] = 0xFFU << 24 | r << 16 | g << 8 | b;
417 s->palette[i] |= s->palette[i] >> 6 & 0x30303;
421 /* next, look for a full codebook */
422 if ((cbf0_chunk != -1) && (cbfz_chunk != -1)) {
424 /* a chunk should not have both chunk types */
425 av_log(s->avctx, AV_LOG_ERROR, "problem: found both CBF0 and CBFZ chunks\n");
426 return AVERROR_INVALIDDATA;
429 /* decompress the full codebook chunk */
430 if (cbfz_chunk != -1) {
432 bytestream2_seek(&s->gb, cbfz_chunk, SEEK_SET);
433 chunk_size = bytestream2_get_be32(&s->gb);
434 if ((res = decode_format80(s, chunk_size, s->codebook,
435 s->codebook_size, 0)) < 0)
439 /* copy a full codebook */
440 if (cbf0_chunk != -1) {
442 bytestream2_seek(&s->gb, cbf0_chunk, SEEK_SET);
443 chunk_size = bytestream2_get_be32(&s->gb);
444 /* sanity check the full codebook size */
445 if (chunk_size > MAX_CODEBOOK_SIZE) {
446 av_log(s->avctx, AV_LOG_ERROR, "problem: CBF0 chunk too large (0x%X bytes)\n",
448 return AVERROR_INVALIDDATA;
451 bytestream2_get_buffer(&s->gb, s->codebook, chunk_size);
454 /* decode the frame */
455 if (vptz_chunk == -1) {
457 /* something is wrong if there is no VPTZ chunk */
458 av_log(s->avctx, AV_LOG_ERROR, "problem: no VPTZ chunk found\n");
459 return AVERROR_INVALIDDATA;
462 bytestream2_seek(&s->gb, vptz_chunk, SEEK_SET);
463 chunk_size = bytestream2_get_be32(&s->gb);
464 if ((res = decode_format80(s, chunk_size,
465 s->decode_buffer, s->decode_buffer_size, 1)) < 0)
468 /* render the final PAL8 frame */
469 if (s->vector_height == 4)
473 for (y = 0; y < s->height; y += s->vector_height) {
474 for (x = 0; x < s->width; x += 4, lobytes++, hibytes++) {
475 pixel_ptr = y * frame->linesize[0] + x;
477 /* get the vector index, the method for which varies according to
478 * VQA file version */
479 switch (s->vqa_version) {
482 lobyte = s->decode_buffer[lobytes * 2];
483 hibyte = s->decode_buffer[(lobytes * 2) + 1];
484 vector_index = ((hibyte << 8) | lobyte) >> 3;
485 vector_index <<= index_shift;
486 lines = s->vector_height;
487 /* uniform color fill - a quick hack */
488 if (hibyte == 0xFF) {
490 frame->data[0][pixel_ptr + 0] = 255 - lobyte;
491 frame->data[0][pixel_ptr + 1] = 255 - lobyte;
492 frame->data[0][pixel_ptr + 2] = 255 - lobyte;
493 frame->data[0][pixel_ptr + 3] = 255 - lobyte;
494 pixel_ptr += frame->linesize[0];
501 lobyte = s->decode_buffer[lobytes];
502 hibyte = s->decode_buffer[hibytes];
503 vector_index = (hibyte << 8) | lobyte;
504 vector_index <<= index_shift;
505 lines = s->vector_height;
509 /* not implemented yet */
515 frame->data[0][pixel_ptr + 0] = s->codebook[vector_index++];
516 frame->data[0][pixel_ptr + 1] = s->codebook[vector_index++];
517 frame->data[0][pixel_ptr + 2] = s->codebook[vector_index++];
518 frame->data[0][pixel_ptr + 3] = s->codebook[vector_index++];
519 pixel_ptr += frame->linesize[0];
524 /* handle partial codebook */
525 if ((cbp0_chunk != -1) && (cbpz_chunk != -1)) {
526 /* a chunk should not have both chunk types */
527 av_log(s->avctx, AV_LOG_ERROR, "problem: found both CBP0 and CBPZ chunks\n");
528 return AVERROR_INVALIDDATA;
531 if (cbp0_chunk != -1) {
533 bytestream2_seek(&s->gb, cbp0_chunk, SEEK_SET);
534 chunk_size = bytestream2_get_be32(&s->gb);
536 if (chunk_size > MAX_CODEBOOK_SIZE - s->next_codebook_buffer_index) {
537 av_log(s->avctx, AV_LOG_ERROR, "cbp0 chunk too large (%u bytes)\n",
539 return AVERROR_INVALIDDATA;
542 /* accumulate partial codebook */
543 bytestream2_get_buffer(&s->gb, &s->next_codebook_buffer[s->next_codebook_buffer_index],
545 s->next_codebook_buffer_index += chunk_size;
547 s->partial_countdown--;
548 if (s->partial_countdown <= 0) {
550 /* time to replace codebook */
551 memcpy(s->codebook, s->next_codebook_buffer,
552 s->next_codebook_buffer_index);
554 /* reset accounting */
555 s->next_codebook_buffer_index = 0;
556 s->partial_countdown = s->partial_count;
560 if (cbpz_chunk != -1) {
562 bytestream2_seek(&s->gb, cbpz_chunk, SEEK_SET);
563 chunk_size = bytestream2_get_be32(&s->gb);
565 if (chunk_size > MAX_CODEBOOK_SIZE - s->next_codebook_buffer_index) {
566 av_log(s->avctx, AV_LOG_ERROR, "cbpz chunk too large (%u bytes)\n",
568 return AVERROR_INVALIDDATA;
571 /* accumulate partial codebook */
572 bytestream2_get_buffer(&s->gb, &s->next_codebook_buffer[s->next_codebook_buffer_index],
574 s->next_codebook_buffer_index += chunk_size;
576 s->partial_countdown--;
577 if (s->partial_countdown <= 0) {
578 bytestream2_init(&s->gb, s->next_codebook_buffer, s->next_codebook_buffer_index);
579 /* decompress codebook */
580 if ((res = decode_format80(s, s->next_codebook_buffer_index,
581 s->codebook, s->codebook_size, 0)) < 0)
584 /* reset accounting */
585 s->next_codebook_buffer_index = 0;
586 s->partial_countdown = s->partial_count;
593 static int vqa_decode_frame(AVCodecContext *avctx,
594 void *data, int *got_frame,
597 VqaContext *s = avctx->priv_data;
598 AVFrame *frame = data;
601 if ((res = ff_get_buffer(avctx, frame, 0)) < 0)
604 bytestream2_init(&s->gb, avpkt->data, avpkt->size);
605 if ((res = vqa_decode_chunk(s, frame)) < 0)
608 /* make the palette available on the way out */
609 memcpy(frame->data[1], s->palette, PALETTE_COUNT * 4);
610 frame->palette_has_changed = 1;
614 /* report that the buffer was completely consumed */
618 static av_cold int vqa_decode_end(AVCodecContext *avctx)
620 VqaContext *s = avctx->priv_data;
622 av_freep(&s->codebook);
623 av_freep(&s->next_codebook_buffer);
624 av_freep(&s->decode_buffer);
629 AVCodec ff_vqa_decoder = {
631 .type = AVMEDIA_TYPE_VIDEO,
632 .id = AV_CODEC_ID_WS_VQA,
633 .priv_data_size = sizeof(VqaContext),
634 .init = vqa_decode_init,
635 .close = vqa_decode_end,
636 .decode = vqa_decode_frame,
637 .capabilities = CODEC_CAP_DR1,
638 .long_name = NULL_IF_CONFIG_SMALL("Westwood Studios VQA (Vector Quantized Animation) video"),