2 * Westwood Studios VQA Video Decoder
3 * Copyright (C) 2003 the ffmpeg project
5 * This file is part of Libav.
7 * Libav 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 * Libav 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 Libav; 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, " VQA video: 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 s->width = AV_RL16(&s->avctx->extradata[6]);
138 s->height = AV_RL16(&s->avctx->extradata[8]);
139 if ((ret = av_image_check_size(s->width, s->height, 0, avctx)) < 0) {
140 s->width= s->height= 0;
143 s->vector_width = s->avctx->extradata[10];
144 s->vector_height = s->avctx->extradata[11];
145 s->partial_count = s->partial_countdown = s->avctx->extradata[13];
147 /* the vector dimensions have to meet very stringent requirements */
148 if ((s->vector_width != 4) ||
149 ((s->vector_height != 2) && (s->vector_height != 4))) {
150 /* return without further initialization */
151 return AVERROR_INVALIDDATA;
154 if (s->width & (s->vector_width - 1) ||
155 s->height & (s->vector_height - 1)) {
156 av_log(avctx, AV_LOG_ERROR, "Image size not multiple of block size\n");
157 return AVERROR_INVALIDDATA;
160 /* allocate codebooks */
161 s->codebook_size = MAX_CODEBOOK_SIZE;
162 s->codebook = av_malloc(s->codebook_size);
165 s->next_codebook_buffer = av_malloc(s->codebook_size);
166 if (!s->next_codebook_buffer)
169 /* allocate decode buffer */
170 s->decode_buffer_size = (s->width / s->vector_width) *
171 (s->height / s->vector_height) * 2;
172 s->decode_buffer = av_malloc(s->decode_buffer_size);
173 if (!s->decode_buffer)
176 /* initialize the solid-color vectors */
177 if (s->vector_height == 4) {
178 codebook_index = 0xFF00 * 16;
179 for (i = 0; i < 256; i++)
180 for (j = 0; j < 16; j++)
181 s->codebook[codebook_index++] = i;
183 codebook_index = 0xF00 * 8;
184 for (i = 0; i < 256; i++)
185 for (j = 0; j < 8; j++)
186 s->codebook[codebook_index++] = i;
188 s->next_codebook_buffer_index = 0;
192 av_freep(&s->codebook);
193 av_freep(&s->next_codebook_buffer);
194 av_freep(&s->decode_buffer);
195 return AVERROR(ENOMEM);
198 #define CHECK_COUNT() \
199 if (dest_index + count > dest_size) { \
200 av_log(NULL, AV_LOG_ERROR, " VQA video: decode_format80 problem: next op would overflow dest_index\n"); \
201 av_log(NULL, AV_LOG_ERROR, " VQA video: current dest_index = %d, count = %d, dest_size = %d\n", \
202 dest_index, count, dest_size); \
203 return AVERROR_INVALIDDATA; \
206 #define CHECK_COPY(idx) \
207 if (idx < 0 || idx + count > dest_size) { \
208 av_log(NULL, AV_LOG_ERROR, " VQA video: decode_format80 problem: next op would overflow dest_index\n"); \
209 av_log(NULL, AV_LOG_ERROR, " VQA video: current src_pos = %d, count = %d, dest_size = %d\n", \
210 src_pos, count, dest_size); \
211 return AVERROR_INVALIDDATA; \
215 static int decode_format80(GetByteContext *gb, int src_size,
216 unsigned char *dest, int dest_size, int check_size) {
219 int count, opcode, start;
224 start = bytestream2_tell(gb);
225 while (bytestream2_tell(gb) - start < src_size) {
226 opcode = bytestream2_get_byte(gb);
227 av_dlog(NULL, " opcode %02X: ", opcode);
229 /* 0x80 means that frame is finished */
233 if (dest_index >= dest_size) {
234 av_log(NULL, AV_LOG_ERROR, " VQA video: decode_format80 problem: dest_index (%d) exceeded dest_size (%d)\n",
235 dest_index, dest_size);
236 return AVERROR_INVALIDDATA;
239 if (opcode == 0xFF) {
241 count = bytestream2_get_le16(gb);
242 src_pos = bytestream2_get_le16(gb);
243 av_dlog(NULL, "(1) copy %X bytes from absolute pos %X\n", count, src_pos);
246 for (i = 0; i < count; i++)
247 dest[dest_index + i] = dest[src_pos + i];
250 } else if (opcode == 0xFE) {
252 count = bytestream2_get_le16(gb);
253 color = bytestream2_get_byte(gb);
254 av_dlog(NULL, "(2) set %X bytes to %02X\n", count, color);
256 memset(&dest[dest_index], color, count);
259 } else if ((opcode & 0xC0) == 0xC0) {
261 count = (opcode & 0x3F) + 3;
262 src_pos = bytestream2_get_le16(gb);
263 av_dlog(NULL, "(3) copy %X bytes from absolute pos %X\n", count, src_pos);
266 for (i = 0; i < count; i++)
267 dest[dest_index + i] = dest[src_pos + i];
270 } else if (opcode > 0x80) {
272 count = opcode & 0x3F;
273 av_dlog(NULL, "(4) copy %X bytes from source to dest\n", count);
275 bytestream2_get_buffer(gb, &dest[dest_index], count);
280 count = ((opcode & 0x70) >> 4) + 3;
281 src_pos = bytestream2_get_byte(gb) | ((opcode & 0x0F) << 8);
282 av_dlog(NULL, "(5) copy %X bytes from relpos %X\n", count, src_pos);
284 CHECK_COPY(dest_index - src_pos);
285 for (i = 0; i < count; i++)
286 dest[dest_index + i] = dest[dest_index - src_pos + i];
291 /* validate that the entire destination buffer was filled; this is
292 * important for decoding frame maps since each vector needs to have a
293 * codebook entry; it is not important for compressed codebooks because
294 * not every entry needs to be filled */
296 if (dest_index < dest_size)
297 av_log(NULL, AV_LOG_ERROR, " VQA video: decode_format80 problem: decode finished with dest_index (%d) < dest_size (%d)\n",
298 dest_index, dest_size);
300 return 0; // let's display what we decoded anyway
303 static int vqa_decode_chunk(VqaContext *s, AVFrame *frame)
305 unsigned int chunk_type;
306 unsigned int chunk_size;
308 unsigned int index = 0;
310 unsigned char r, g, b;
325 int vector_index = 0;
329 int hibytes = s->decode_buffer_size / 2;
331 /* first, traverse through the frame and find the subchunks */
332 while (bytestream2_get_bytes_left(&s->gb) >= 8) {
334 chunk_type = bytestream2_get_be32u(&s->gb);
335 index = bytestream2_tell(&s->gb);
336 chunk_size = bytestream2_get_be32u(&s->gb);
338 switch (chunk_type) {
369 av_log(s->avctx, AV_LOG_ERROR, " VQA video: Found unknown chunk type: %c%c%c%c (%08X)\n",
370 (chunk_type >> 24) & 0xFF,
371 (chunk_type >> 16) & 0xFF,
372 (chunk_type >> 8) & 0xFF,
373 (chunk_type >> 0) & 0xFF,
378 byte_skip = chunk_size & 0x01;
379 bytestream2_skip(&s->gb, chunk_size + byte_skip);
382 /* next, deal with the palette */
383 if ((cpl0_chunk != -1) && (cplz_chunk != -1)) {
385 /* a chunk should not have both chunk types */
386 av_log(s->avctx, AV_LOG_ERROR, " VQA video: problem: found both CPL0 and CPLZ chunks\n");
387 return AVERROR_INVALIDDATA;
390 /* decompress the palette chunk */
391 if (cplz_chunk != -1) {
393 /* yet to be handled */
397 /* convert the RGB palette into the machine's endian format */
398 if (cpl0_chunk != -1) {
400 bytestream2_seek(&s->gb, cpl0_chunk, SEEK_SET);
401 chunk_size = bytestream2_get_be32(&s->gb);
402 /* sanity check the palette size */
403 if (chunk_size / 3 > 256 || chunk_size > bytestream2_get_bytes_left(&s->gb)) {
404 av_log(s->avctx, AV_LOG_ERROR, " VQA video: problem: found a palette chunk with %d colors\n",
406 return AVERROR_INVALIDDATA;
408 for (i = 0; i < chunk_size / 3; i++) {
409 /* scale by 4 to transform 6-bit palette -> 8-bit */
410 r = bytestream2_get_byteu(&s->gb) * 4;
411 g = bytestream2_get_byteu(&s->gb) * 4;
412 b = bytestream2_get_byteu(&s->gb) * 4;
413 s->palette[i] = (r << 16) | (g << 8) | (b);
417 /* next, look for a full codebook */
418 if ((cbf0_chunk != -1) && (cbfz_chunk != -1)) {
420 /* a chunk should not have both chunk types */
421 av_log(s->avctx, AV_LOG_ERROR, " VQA video: problem: found both CBF0 and CBFZ chunks\n");
422 return AVERROR_INVALIDDATA;
425 /* decompress the full codebook chunk */
426 if (cbfz_chunk != -1) {
428 bytestream2_seek(&s->gb, cbfz_chunk, SEEK_SET);
429 chunk_size = bytestream2_get_be32(&s->gb);
430 if ((res = decode_format80(&s->gb, chunk_size, s->codebook,
431 s->codebook_size, 0)) < 0)
435 /* copy a full codebook */
436 if (cbf0_chunk != -1) {
438 bytestream2_seek(&s->gb, cbf0_chunk, SEEK_SET);
439 chunk_size = bytestream2_get_be32(&s->gb);
440 /* sanity check the full codebook size */
441 if (chunk_size > MAX_CODEBOOK_SIZE) {
442 av_log(s->avctx, AV_LOG_ERROR, " VQA video: problem: CBF0 chunk too large (0x%X bytes)\n",
444 return AVERROR_INVALIDDATA;
447 bytestream2_get_buffer(&s->gb, s->codebook, chunk_size);
450 /* decode the frame */
451 if (vptz_chunk == -1) {
453 /* something is wrong if there is no VPTZ chunk */
454 av_log(s->avctx, AV_LOG_ERROR, " VQA video: problem: no VPTZ chunk found\n");
455 return AVERROR_INVALIDDATA;
458 bytestream2_seek(&s->gb, vptz_chunk, SEEK_SET);
459 chunk_size = bytestream2_get_be32(&s->gb);
460 if ((res = decode_format80(&s->gb, chunk_size,
461 s->decode_buffer, s->decode_buffer_size, 1)) < 0)
464 /* render the final PAL8 frame */
465 if (s->vector_height == 4)
469 for (y = 0; y < s->height; y += s->vector_height) {
470 for (x = 0; x < s->width; x += 4, lobytes++, hibytes++) {
471 pixel_ptr = y * frame->linesize[0] + x;
473 /* get the vector index, the method for which varies according to
474 * VQA file version */
475 switch (s->vqa_version) {
478 lobyte = s->decode_buffer[lobytes * 2];
479 hibyte = s->decode_buffer[(lobytes * 2) + 1];
480 vector_index = ((hibyte << 8) | lobyte) >> 3;
481 vector_index <<= index_shift;
482 lines = s->vector_height;
483 /* uniform color fill - a quick hack */
484 if (hibyte == 0xFF) {
486 frame->data[0][pixel_ptr + 0] = 255 - lobyte;
487 frame->data[0][pixel_ptr + 1] = 255 - lobyte;
488 frame->data[0][pixel_ptr + 2] = 255 - lobyte;
489 frame->data[0][pixel_ptr + 3] = 255 - lobyte;
490 pixel_ptr += frame->linesize[0];
497 lobyte = s->decode_buffer[lobytes];
498 hibyte = s->decode_buffer[hibytes];
499 vector_index = (hibyte << 8) | lobyte;
500 vector_index <<= index_shift;
501 lines = s->vector_height;
505 /* not implemented yet */
511 frame->data[0][pixel_ptr + 0] = s->codebook[vector_index++];
512 frame->data[0][pixel_ptr + 1] = s->codebook[vector_index++];
513 frame->data[0][pixel_ptr + 2] = s->codebook[vector_index++];
514 frame->data[0][pixel_ptr + 3] = s->codebook[vector_index++];
515 pixel_ptr += frame->linesize[0];
520 /* handle partial codebook */
521 if ((cbp0_chunk != -1) && (cbpz_chunk != -1)) {
522 /* a chunk should not have both chunk types */
523 av_log(s->avctx, AV_LOG_ERROR, " VQA video: problem: found both CBP0 and CBPZ chunks\n");
524 return AVERROR_INVALIDDATA;
527 if (cbp0_chunk != -1) {
529 bytestream2_seek(&s->gb, cbp0_chunk, SEEK_SET);
530 chunk_size = bytestream2_get_be32(&s->gb);
532 if (chunk_size > MAX_CODEBOOK_SIZE - s->next_codebook_buffer_index) {
533 av_log(s->avctx, AV_LOG_ERROR, "cbp0 chunk too large (%u bytes)\n",
535 return AVERROR_INVALIDDATA;
538 /* accumulate partial codebook */
539 bytestream2_get_buffer(&s->gb, &s->next_codebook_buffer[s->next_codebook_buffer_index],
541 s->next_codebook_buffer_index += chunk_size;
543 s->partial_countdown--;
544 if (s->partial_countdown == 0) {
546 /* time to replace codebook */
547 memcpy(s->codebook, s->next_codebook_buffer,
548 s->next_codebook_buffer_index);
550 /* reset accounting */
551 s->next_codebook_buffer_index = 0;
552 s->partial_countdown = s->partial_count;
556 if (cbpz_chunk != -1) {
558 bytestream2_seek(&s->gb, cbpz_chunk, SEEK_SET);
559 chunk_size = bytestream2_get_be32(&s->gb);
561 if (chunk_size > MAX_CODEBOOK_SIZE - s->next_codebook_buffer_index) {
562 av_log(s->avctx, AV_LOG_ERROR, "cbpz chunk too large (%u bytes)\n",
564 return AVERROR_INVALIDDATA;
567 /* accumulate partial codebook */
568 bytestream2_get_buffer(&s->gb, &s->next_codebook_buffer[s->next_codebook_buffer_index],
570 s->next_codebook_buffer_index += chunk_size;
572 s->partial_countdown--;
573 if (s->partial_countdown == 0) {
576 bytestream2_init(&gb, s->next_codebook_buffer, s->next_codebook_buffer_index);
577 /* decompress codebook */
578 if ((res = decode_format80(&gb, s->next_codebook_buffer_index,
579 s->codebook, s->codebook_size, 0)) < 0)
582 /* reset accounting */
583 s->next_codebook_buffer_index = 0;
584 s->partial_countdown = s->partial_count;
591 static int vqa_decode_frame(AVCodecContext *avctx,
592 void *data, int *got_frame,
595 VqaContext *s = avctx->priv_data;
596 AVFrame *frame = data;
599 if ((res = ff_get_buffer(avctx, frame, 0)) < 0) {
600 av_log(s->avctx, AV_LOG_ERROR, " VQA Video: get_buffer() failed\n");
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"),