2 * Westwood Studios VQA Video Decoder
3 * Copyright (C) 2003 the ffmpeg project
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * VQA Video Decoder by Mike Melanson (melanson@pcisys.net)
24 * For more information about the RPZA format, visit:
25 * http://www.pcisys.net/~melanson/codecs/
27 * The VQA video decoder outputs PAL8 or RGB555 colorspace data, depending
28 * on the type of data in the file.
30 * This decoder needs the 42-byte VQHD header from the beginning
31 * of the VQA file passed through the extradata field. The VQHD header
34 * bytes 0-3 chunk fourcc: 'VQHD'
35 * bytes 4-7 chunk size in big-endian format, should be 0x0000002A
36 * bytes 8-49 VQHD chunk data
38 * Bytes 8-49 are what this decoder expects to see.
40 * Briefly, VQA is a vector quantized animation format that operates in a
41 * VGA palettized colorspace. It operates on pixel vectors (blocks)
42 * of either 4x2 or 4x4 in size. Compressed VQA chunks can contain vector
43 * codebooks, palette information, and code maps for rendering vectors onto
44 * frames. Any of these components can also be compressed with a run-length
45 * encoding (RLE) algorithm commonly referred to as "format80".
47 * VQA takes a novel approach to rate control. Each group of n frames
48 * (usually, n = 8) relies on a different vector codebook. Rather than
49 * transporting an entire codebook every 8th frame, the new codebook is
50 * broken up into 8 pieces and sent along with the compressed video chunks
51 * for each of the 8 frames preceding the 8 frames which require the
52 * codebook. A full codebook is also sent on the very first frame of a
53 * file. This is an interesting technique, although it makes random file
54 * seeking difficult despite the fact that the frames are all intracoded.
56 * V1,2 VQA uses 12-bit codebook indices. If the 12-bit indices were
57 * packed into bytes and then RLE compressed, bytewise, the results would
58 * be poor. That is why the coding method divides each index into 2 parts,
59 * the top 4 bits and the bottom 8 bits, then RL encodes the 4-bit pieces
60 * together and the 8-bit pieces together. If most of the vectors are
61 * clustered into one group of 256 vectors, most of the 4-bit index pieces
74 #define PALETTE_COUNT 256
75 #define VQA_HEADER_SIZE 0x2A
76 #define CHUNK_PREAMBLE_SIZE 8
78 /* allocate the maximum vector space, regardless of the file version:
79 * (0xFF00 codebook vectors + 0x100 solid pixel vectors) * (4x4 pixels/block) */
80 #define MAX_CODEBOOK_VECTORS 0xFF00
81 #define SOLID_PIXEL_VECTORS 0x100
82 #define MAX_VECTORS (MAX_CODEBOOK_VECTORS + SOLID_PIXEL_VECTORS)
83 #define MAX_CODEBOOK_SIZE (MAX_VECTORS * 4 * 4)
85 #define LE_16(x) ((((uint8_t*)(x))[1] << 8) | ((uint8_t*)(x))[0])
86 #define BE_16(x) ((((uint8_t*)(x))[0] << 8) | ((uint8_t*)(x))[1])
87 #define BE_32(x) ((((uint8_t*)(x))[0] << 24) | \
88 (((uint8_t*)(x))[1] << 16) | \
89 (((uint8_t*)(x))[2] << 8) | \
92 #define FOURCC_TAG( ch0, ch1, ch2, ch3 ) \
93 ( (long)(unsigned char)(ch3) | \
94 ( (long)(unsigned char)(ch2) << 8 ) | \
95 ( (long)(unsigned char)(ch1) << 16 ) | \
96 ( (long)(unsigned char)(ch0) << 24 ) )
98 #define CBF0_TAG FOURCC_TAG('C', 'B', 'F', '0')
99 #define CBFZ_TAG FOURCC_TAG('C', 'B', 'F', 'Z')
100 #define CBP0_TAG FOURCC_TAG('C', 'B', 'P', '0')
101 #define CBPZ_TAG FOURCC_TAG('C', 'B', 'P', 'Z')
102 #define CPL0_TAG FOURCC_TAG('C', 'P', 'L', '0')
103 #define CPLZ_TAG FOURCC_TAG('C', 'P', 'L', 'Z')
104 #define VPTZ_TAG FOURCC_TAG('V', 'P', 'T', 'Z')
109 #define vqa_debug printf
111 static inline void vqa_debug(const char *format, ...) { }
114 typedef struct VqaContext {
116 AVCodecContext *avctx;
123 unsigned int palette[PALETTE_COUNT];
125 int width; /* width of a frame */
126 int height; /* height of a frame */
127 int vector_width; /* width of individual vector */
128 int vector_height; /* height of individual vector */
129 int vqa_version; /* this should be either 1, 2 or 3 */
131 unsigned char *codebook; /* the current codebook */
133 unsigned char *next_codebook_buffer; /* accumulator for next codebook */
134 int next_codebook_buffer_index;
136 unsigned char *decode_buffer;
137 int decode_buffer_size;
139 /* number of frames to go before replacing codebook */
140 int partial_countdown;
145 static int vqa_decode_init(AVCodecContext *avctx)
147 VqaContext *s = (VqaContext *)avctx->priv_data;
148 unsigned char *vqa_header;
149 int i, j, codebook_index;;
152 avctx->pix_fmt = PIX_FMT_PAL8;
153 avctx->has_b_frames = 0;
154 dsputil_init(&s->dsp, avctx);
156 /* make sure the extradata made it */
157 if (s->avctx->extradata_size != VQA_HEADER_SIZE) {
158 printf(" VQA video: expected extradata size of %d\n", VQA_HEADER_SIZE);
162 /* load up the VQA parameters from the header */
163 vqa_header = (unsigned char *)s->avctx->extradata;
164 s->vqa_version = vqa_header[0];
165 s->width = LE_16(&vqa_header[6]);
166 s->height = LE_16(&vqa_header[8]);
167 s->vector_width = vqa_header[10];
168 s->vector_height = vqa_header[11];
169 s->partial_count = s->partial_countdown = vqa_header[13];
171 /* the vector dimensions have to meet very stringent requirements */
172 if ((s->vector_width != 4) ||
173 ((s->vector_height != 2) && (s->vector_height != 4))) {
174 /* return without further initialization */
178 /* allocate codebooks */
179 s->codebook_size = MAX_CODEBOOK_SIZE;
180 s->codebook = av_malloc(s->codebook_size);
181 s->next_codebook_buffer = av_malloc(s->codebook_size);
183 /* initialize the solid-color vectors */
184 if (s->vector_height == 4) {
185 codebook_index = 0xFF00 * 16;
186 for (i = 0; i < 256; i++)
187 for (j = 0; j < 16; j++)
188 s->codebook[codebook_index++] = i;
190 codebook_index = 0xF00 * 8;
191 for (i = 0; i < 256; i++)
192 for (j = 0; j < 8; j++)
193 s->codebook[codebook_index++] = i;
195 s->next_codebook_buffer_index = 0;
197 /* allocate decode buffer */
198 s->decode_buffer_size = (s->width / s->vector_width) *
199 (s->height / s->vector_height) * 2;
200 s->decode_buffer = av_malloc(s->decode_buffer_size);
202 s->frame.data[0] = NULL;
207 #define CHECK_COUNT() \
208 if (dest_index + count > dest_size) { \
209 printf (" VQA video: decode_format80 problem: next op would overflow dest_index\n"); \
210 printf (" VQA video: current dest_index = %d, count = %d, dest_size = %d\n", \
211 dest_index, count, dest_size); \
215 static void decode_format80(unsigned char *src, int src_size,
216 unsigned char *dest, int dest_size, int check_size) {
225 while (src_index < src_size) {
227 vqa_debug(" opcode %02X: ", src[src_index]);
229 /* 0x80 means that frame is finished */
230 if (src[src_index] == 0x80)
233 if (dest_index >= dest_size) {
234 printf (" VQA video: decode_format80 problem: dest_index (%d) exceeded dest_size (%d)\n",
235 dest_index, dest_size);
239 if (src[src_index] == 0xFF) {
242 count = LE_16(&src[src_index]);
244 src_pos = LE_16(&src[src_index]);
246 vqa_debug("(1) copy %X bytes from absolute pos %X\n", count, src_pos);
248 for (i = 0; i < count; i++)
249 dest[dest_index + i] = dest[src_pos + i];
252 } else if (src[src_index] == 0xFE) {
255 count = LE_16(&src[src_index]);
257 color = src[src_index++];
258 vqa_debug("(2) set %X bytes to %02X\n", count, color);
260 memset(&dest[dest_index], color, count);
263 } else if ((src[src_index] & 0xC0) == 0xC0) {
265 count = (src[src_index++] & 0x3F) + 3;
266 src_pos = LE_16(&src[src_index]);
268 vqa_debug("(3) copy %X bytes from absolute pos %X\n", count, src_pos);
270 for (i = 0; i < count; i++)
271 dest[dest_index + i] = dest[src_pos + i];
274 } else if (src[src_index] > 0x80) {
276 count = src[src_index++] & 0x3F;
277 vqa_debug("(4) copy %X bytes from source to dest\n", count);
279 memcpy(&dest[dest_index], &src[src_index], count);
285 count = ((src[src_index] & 0x70) >> 4) + 3;
286 src_pos = BE_16(&src[src_index]) & 0x0FFF;
288 vqa_debug("(5) copy %X bytes from relpos %X\n", count, src_pos);
290 for (i = 0; i < count; i++)
291 dest[dest_index + i] = dest[dest_index - src_pos + i];
296 /* validate that the entire destination buffer was filled; this is
297 * important for decoding frame maps since each vector needs to have a
298 * codebook entry; it is not important for compressed codebooks because
299 * not every entry needs to be filled */
301 if (dest_index < dest_size)
302 printf (" VQA video: decode_format80 problem: decode finished with dest_index (%d) < dest_size (%d)\n",
303 dest_index, dest_size);
306 static void vqa_decode_chunk(VqaContext *s)
308 unsigned int chunk_type;
309 unsigned int chunk_size;
311 unsigned int index = 0;
313 unsigned char r, g, b;
327 int vector_index = 0;
331 int hibytes = s->decode_buffer_size / 2;
333 /* first, traverse through the frame and find the subchunks */
334 while (index < s->size) {
336 chunk_type = BE_32(&s->buf[index]);
337 chunk_size = BE_32(&s->buf[index + 4]);
339 switch (chunk_type) {
370 printf (" VQA video: Found unknown chunk type: %c%c%c%c (%08X)\n",
371 (chunk_type >> 24) & 0xFF,
372 (chunk_type >> 16) & 0xFF,
373 (chunk_type >> 8) & 0xFF,
374 (chunk_type >> 0) & 0xFF,
379 byte_skip = chunk_size & 0x01;
380 index += (CHUNK_PREAMBLE_SIZE + chunk_size + byte_skip);
383 /* next, deal with the palette */
384 if ((cpl0_chunk != -1) && (cplz_chunk != -1)) {
386 /* a chunk should not have both chunk types */
387 printf (" VQA video: problem: found both CPL0 and CPLZ chunks\n");
391 /* decompress the palette chunk */
392 if (cplz_chunk != -1) {
394 /* yet to be handled */
398 /* convert the RGB palette into the machine's endian format */
399 if (cpl0_chunk != -1) {
401 chunk_size = BE_32(&s->buf[cpl0_chunk + 4]);
402 /* sanity check the palette size */
403 if (chunk_size / 3 > 256) {
404 printf (" VQA video: problem: found a palette chunk with %d colors\n",
408 cpl0_chunk += CHUNK_PREAMBLE_SIZE;
409 for (i = 0; i < chunk_size / 3; i++) {
410 /* scale by 4 to transform 6-bit palette -> 8-bit */
411 r = s->buf[cpl0_chunk++] * 4;
412 g = s->buf[cpl0_chunk++] * 4;
413 b = s->buf[cpl0_chunk++] * 4;
414 s->palette[i] = (r << 16) | (g << 8) | (b);
418 /* next, look for a full codebook */
419 if ((cbf0_chunk != -1) && (cbfz_chunk != -1)) {
421 /* a chunk should not have both chunk types */
422 printf (" VQA video: problem: found both CBF0 and CBFZ chunks\n");
426 /* decompress the full codebook chunk */
427 if (cbfz_chunk != -1) {
429 chunk_size = BE_32(&s->buf[cbfz_chunk + 4]);
430 cbfz_chunk += CHUNK_PREAMBLE_SIZE;
431 decode_format80(&s->buf[cbfz_chunk], chunk_size,
432 s->codebook, s->codebook_size, 0);
435 /* copy a full codebook */
436 if (cbf0_chunk != -1) {
438 chunk_size = BE_32(&s->buf[cbf0_chunk + 4]);
439 /* sanity check the full codebook size */
440 if (chunk_size > MAX_CODEBOOK_SIZE) {
441 printf (" VQA video: problem: CBF0 chunk too large (0x%X bytes)\n",
445 cbf0_chunk += CHUNK_PREAMBLE_SIZE;
447 memcpy(s->codebook, &s->buf[cbf0_chunk], chunk_size);
450 /* decode the frame */
451 if (vptz_chunk == -1) {
453 /* something is wrong if there is no VPTZ chunk */
454 printf (" VQA video: problem: no VPTZ chunk found\n");
458 chunk_size = BE_32(&s->buf[vptz_chunk + 4]);
459 vptz_chunk += CHUNK_PREAMBLE_SIZE;
460 decode_format80(&s->buf[vptz_chunk], chunk_size,
461 s->decode_buffer, s->decode_buffer_size, 1);
463 /* render the final PAL8 frame */
464 if (s->vector_height == 4)
468 for (y = 0; y < s->frame.linesize[0] * s->height;
469 y += s->frame.linesize[0] * s->vector_height) {
471 for (x = y; x < y + s->width; x += 4, lobytes++, hibytes++) {
474 /* get the vector index, the method for which varies according to
475 * VQA file version */
476 switch (s->vqa_version) {
479 /* still need sample media for this case (only one game, "Legend of
480 * Kyrandia III : Malcolm's Revenge", is known to use this version) */
485 lobyte = s->decode_buffer[lobytes];
486 hibyte = s->decode_buffer[hibytes];
487 vector_index = (hibyte << 8) | lobyte;
488 vector_index <<= index_shift;
489 lines = s->vector_height;
493 /* not implemented yet */
499 s->frame.data[0][pixel_ptr + 0] = s->codebook[vector_index++];
500 s->frame.data[0][pixel_ptr + 1] = s->codebook[vector_index++];
501 s->frame.data[0][pixel_ptr + 2] = s->codebook[vector_index++];
502 s->frame.data[0][pixel_ptr + 3] = s->codebook[vector_index++];
503 pixel_ptr += s->frame.linesize[0];
508 /* handle partial codebook */
509 if ((cbp0_chunk != -1) && (cbpz_chunk != -1)) {
510 /* a chunk should not have both chunk types */
511 printf (" VQA video: problem: found both CBP0 and CBPZ chunks\n");
515 if (cbp0_chunk != -1) {
517 chunk_size = BE_32(&s->buf[cbp0_chunk + 4]);
518 cbp0_chunk += CHUNK_PREAMBLE_SIZE;
520 /* accumulate partial codebook */
521 memcpy(&s->next_codebook_buffer[s->next_codebook_buffer_index],
522 &s->buf[cbp0_chunk], chunk_size);
523 s->next_codebook_buffer_index += chunk_size;
525 s->partial_countdown--;
526 if (s->partial_countdown == 0) {
528 /* time to replace codebook */
529 memcpy(s->codebook, s->next_codebook_buffer,
530 s->next_codebook_buffer_index);
532 /* reset accounting */
533 s->next_codebook_buffer_index = 0;
534 s->partial_countdown = s->partial_count;
538 if (cbpz_chunk != -1) {
540 chunk_size = BE_32(&s->buf[cbpz_chunk + 4]);
541 cbpz_chunk += CHUNK_PREAMBLE_SIZE;
543 /* accumulate partial codebook */
544 memcpy(&s->next_codebook_buffer[s->next_codebook_buffer_index],
545 &s->buf[cbpz_chunk], chunk_size);
546 s->next_codebook_buffer_index += chunk_size;
548 s->partial_countdown--;
549 if (s->partial_countdown == 0) {
551 /* decompress codebook */
552 decode_format80(s->next_codebook_buffer,
553 s->next_codebook_buffer_index,
554 s->codebook, s->codebook_size, 0);
556 /* reset accounting */
557 s->next_codebook_buffer_index = 0;
558 s->partial_countdown = s->partial_count;
563 static int vqa_decode_frame(AVCodecContext *avctx,
564 void *data, int *data_size,
565 uint8_t *buf, int buf_size)
567 VqaContext *s = (VqaContext *)avctx->priv_data;
572 if (s->frame.data[0])
573 avctx->release_buffer(avctx, &s->frame);
575 if (avctx->get_buffer(avctx, &s->frame)) {
576 printf (" VQA Video: get_buffer() failed\n");
582 /* make the palette available on the way out */
583 memcpy(s->frame.data[1], s->palette, PALETTE_COUNT * 4);
584 s->frame.palette_has_changed = 1;
586 *data_size = sizeof(AVFrame);
587 *(AVFrame*)data = s->frame;
589 /* report that the buffer was completely consumed */
593 static int vqa_decode_end(AVCodecContext *avctx)
595 VqaContext *s = (VqaContext *)avctx->priv_data;
597 av_free(s->codebook);
598 av_free(s->next_codebook_buffer);
599 av_free(s->decode_buffer);
601 if (s->frame.data[0])
602 avctx->release_buffer(avctx, &s->frame);
607 AVCodec vqa_decoder = {