1/*
2 * Westwood Studios VQA Video Decoder
3 * Copyright (C) 2003 the ffmpeg project
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 libavcodec/vqavideo.c
24 * VQA Video Decoder by Mike Melanson (melanson@pcisys.net)
25 * For more information about the VQA format, visit:
26 *   http://wiki.multimedia.cx/index.php?title=VQA
27 *
28 * The VQA video decoder outputs PAL8 or RGB555 colorspace data, depending
29 * on the type of data in the file.
30 *
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
33 * is laid out as:
34 *
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
38 *
39 * Bytes 8-49 are what this decoder expects to see.
40 *
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".
47 *
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.
56 *
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
63 * should be the same.
64 */
65
66#include <stdio.h>
67#include <stdlib.h>
68#include <string.h>
69#include <unistd.h>
70
71#include "libavutil/intreadwrite.h"
72#include "avcodec.h"
73
74#define PALETTE_COUNT 256
75#define VQA_HEADER_SIZE 0x2A
76#define CHUNK_PREAMBLE_SIZE 8
77
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)
84
85#define CBF0_TAG MKBETAG('C', 'B', 'F', '0')
86#define CBFZ_TAG MKBETAG('C', 'B', 'F', 'Z')
87#define CBP0_TAG MKBETAG('C', 'B', 'P', '0')
88#define CBPZ_TAG MKBETAG('C', 'B', 'P', 'Z')
89#define CPL0_TAG MKBETAG('C', 'P', 'L', '0')
90#define CPLZ_TAG MKBETAG('C', 'P', 'L', 'Z')
91#define VPTZ_TAG MKBETAG('V', 'P', 'T', 'Z')
92
93#define VQA_DEBUG 0
94
95#if VQA_DEBUG
96#define vqa_debug printf
97#else
98static inline void vqa_debug(const char *format, ...) { }
99#endif
100
101typedef struct VqaContext {
102
103    AVCodecContext *avctx;
104    AVFrame frame;
105
106    const unsigned char *buf;
107    int size;
108
109    uint32_t palette[PALETTE_COUNT];
110
111    int width;   /* width of a frame */
112    int height;   /* height of a frame */
113    int vector_width;  /* width of individual vector */
114    int vector_height;  /* height of individual vector */
115    int vqa_version;  /* this should be either 1, 2 or 3 */
116
117    unsigned char *codebook;         /* the current codebook */
118    int codebook_size;
119    unsigned char *next_codebook_buffer;  /* accumulator for next codebook */
120    int next_codebook_buffer_index;
121
122    unsigned char *decode_buffer;
123    int decode_buffer_size;
124
125    /* number of frames to go before replacing codebook */
126    int partial_countdown;
127    int partial_count;
128
129} VqaContext;
130
131static av_cold int vqa_decode_init(AVCodecContext *avctx)
132{
133    VqaContext *s = avctx->priv_data;
134    unsigned char *vqa_header;
135    int i, j, codebook_index;
136
137    s->avctx = avctx;
138    avctx->pix_fmt = PIX_FMT_PAL8;
139
140    /* make sure the extradata made it */
141    if (s->avctx->extradata_size != VQA_HEADER_SIZE) {
142        av_log(s->avctx, AV_LOG_ERROR, "  VQA video: expected extradata size of %d\n", VQA_HEADER_SIZE);
143        return -1;
144    }
145
146    /* load up the VQA parameters from the header */
147    vqa_header = (unsigned char *)s->avctx->extradata;
148    s->vqa_version = vqa_header[0];
149    s->width = AV_RL16(&vqa_header[6]);
150    s->height = AV_RL16(&vqa_header[8]);
151    if(avcodec_check_dimensions(avctx, s->width, s->height)){
152        s->width= s->height= 0;
153        return -1;
154    }
155    s->vector_width = vqa_header[10];
156    s->vector_height = vqa_header[11];
157    s->partial_count = s->partial_countdown = vqa_header[13];
158
159    /* the vector dimensions have to meet very stringent requirements */
160    if ((s->vector_width != 4) ||
161        ((s->vector_height != 2) && (s->vector_height != 4))) {
162        /* return without further initialization */
163        return -1;
164    }
165
166    /* allocate codebooks */
167    s->codebook_size = MAX_CODEBOOK_SIZE;
168    s->codebook = av_malloc(s->codebook_size);
169    s->next_codebook_buffer = av_malloc(s->codebook_size);
170
171    /* initialize the solid-color vectors */
172    if (s->vector_height == 4) {
173        codebook_index = 0xFF00 * 16;
174        for (i = 0; i < 256; i++)
175            for (j = 0; j < 16; j++)
176                s->codebook[codebook_index++] = i;
177    } else {
178        codebook_index = 0xF00 * 8;
179        for (i = 0; i < 256; i++)
180            for (j = 0; j < 8; j++)
181                s->codebook[codebook_index++] = i;
182    }
183    s->next_codebook_buffer_index = 0;
184
185    /* allocate decode buffer */
186    s->decode_buffer_size = (s->width / s->vector_width) *
187        (s->height / s->vector_height) * 2;
188    s->decode_buffer = av_malloc(s->decode_buffer_size);
189
190    s->frame.data[0] = NULL;
191
192    return 0;
193}
194
195#define CHECK_COUNT() \
196    if (dest_index + count > dest_size) { \
197        av_log(NULL, AV_LOG_ERROR, "  VQA video: decode_format80 problem: next op would overflow dest_index\n"); \
198        av_log(NULL, AV_LOG_ERROR, "  VQA video: current dest_index = %d, count = %d, dest_size = %d\n", \
199            dest_index, count, dest_size); \
200        return; \
201    }
202
203static void decode_format80(const unsigned char *src, int src_size,
204    unsigned char *dest, int dest_size, int check_size) {
205
206    int src_index = 0;
207    int dest_index = 0;
208    int count;
209    int src_pos;
210    unsigned char color;
211    int i;
212
213    while (src_index < src_size) {
214
215        vqa_debug("      opcode %02X: ", src[src_index]);
216
217        /* 0x80 means that frame is finished */
218        if (src[src_index] == 0x80)
219            return;
220
221        if (dest_index >= dest_size) {
222            av_log(NULL, AV_LOG_ERROR, "  VQA video: decode_format80 problem: dest_index (%d) exceeded dest_size (%d)\n",
223                dest_index, dest_size);
224            return;
225        }
226
227        if (src[src_index] == 0xFF) {
228
229            src_index++;
230            count = AV_RL16(&src[src_index]);
231            src_index += 2;
232            src_pos = AV_RL16(&src[src_index]);
233            src_index += 2;
234            vqa_debug("(1) copy %X bytes from absolute pos %X\n", count, src_pos);
235            CHECK_COUNT();
236            for (i = 0; i < count; i++)
237                dest[dest_index + i] = dest[src_pos + i];
238            dest_index += count;
239
240        } else if (src[src_index] == 0xFE) {
241
242            src_index++;
243            count = AV_RL16(&src[src_index]);
244            src_index += 2;
245            color = src[src_index++];
246            vqa_debug("(2) set %X bytes to %02X\n", count, color);
247            CHECK_COUNT();
248            memset(&dest[dest_index], color, count);
249            dest_index += count;
250
251        } else if ((src[src_index] & 0xC0) == 0xC0) {
252
253            count = (src[src_index++] & 0x3F) + 3;
254            src_pos = AV_RL16(&src[src_index]);
255            src_index += 2;
256            vqa_debug("(3) copy %X bytes from absolute pos %X\n", count, src_pos);
257            CHECK_COUNT();
258            for (i = 0; i < count; i++)
259                dest[dest_index + i] = dest[src_pos + i];
260            dest_index += count;
261
262        } else if (src[src_index] > 0x80) {
263
264            count = src[src_index++] & 0x3F;
265            vqa_debug("(4) copy %X bytes from source to dest\n", count);
266            CHECK_COUNT();
267            memcpy(&dest[dest_index], &src[src_index], count);
268            src_index += count;
269            dest_index += count;
270
271        } else {
272
273            count = ((src[src_index] & 0x70) >> 4) + 3;
274            src_pos = AV_RB16(&src[src_index]) & 0x0FFF;
275            src_index += 2;
276            vqa_debug("(5) copy %X bytes from relpos %X\n", count, src_pos);
277            CHECK_COUNT();
278            for (i = 0; i < count; i++)
279                dest[dest_index + i] = dest[dest_index - src_pos + i];
280            dest_index += count;
281        }
282    }
283
284    /* validate that the entire destination buffer was filled; this is
285     * important for decoding frame maps since each vector needs to have a
286     * codebook entry; it is not important for compressed codebooks because
287     * not every entry needs to be filled */
288    if (check_size)
289        if (dest_index < dest_size)
290            av_log(NULL, AV_LOG_ERROR, "  VQA video: decode_format80 problem: decode finished with dest_index (%d) < dest_size (%d)\n",
291                dest_index, dest_size);
292}
293
294static void vqa_decode_chunk(VqaContext *s)
295{
296    unsigned int chunk_type;
297    unsigned int chunk_size;
298    int byte_skip;
299    unsigned int index = 0;
300    int i;
301    unsigned char r, g, b;
302    int index_shift;
303
304    int cbf0_chunk = -1;
305    int cbfz_chunk = -1;
306    int cbp0_chunk = -1;
307    int cbpz_chunk = -1;
308    int cpl0_chunk = -1;
309    int cplz_chunk = -1;
310    int vptz_chunk = -1;
311
312    int x, y;
313    int lines = 0;
314    int pixel_ptr;
315    int vector_index = 0;
316    int lobyte = 0;
317    int hibyte = 0;
318    int lobytes = 0;
319    int hibytes = s->decode_buffer_size / 2;
320
321    /* first, traverse through the frame and find the subchunks */
322    while (index < s->size) {
323
324        chunk_type = AV_RB32(&s->buf[index]);
325        chunk_size = AV_RB32(&s->buf[index + 4]);
326
327        switch (chunk_type) {
328
329        case CBF0_TAG:
330            cbf0_chunk = index;
331            break;
332
333        case CBFZ_TAG:
334            cbfz_chunk = index;
335            break;
336
337        case CBP0_TAG:
338            cbp0_chunk = index;
339            break;
340
341        case CBPZ_TAG:
342            cbpz_chunk = index;
343            break;
344
345        case CPL0_TAG:
346            cpl0_chunk = index;
347            break;
348
349        case CPLZ_TAG:
350            cplz_chunk = index;
351            break;
352
353        case VPTZ_TAG:
354            vptz_chunk = index;
355            break;
356
357        default:
358            av_log(s->avctx, AV_LOG_ERROR, "  VQA video: Found unknown chunk type: %c%c%c%c (%08X)\n",
359            (chunk_type >> 24) & 0xFF,
360            (chunk_type >> 16) & 0xFF,
361            (chunk_type >>  8) & 0xFF,
362            (chunk_type >>  0) & 0xFF,
363            chunk_type);
364            break;
365        }
366
367        byte_skip = chunk_size & 0x01;
368        index += (CHUNK_PREAMBLE_SIZE + chunk_size + byte_skip);
369    }
370
371    /* next, deal with the palette */
372    if ((cpl0_chunk != -1) && (cplz_chunk != -1)) {
373
374        /* a chunk should not have both chunk types */
375        av_log(s->avctx, AV_LOG_ERROR, "  VQA video: problem: found both CPL0 and CPLZ chunks\n");
376        return;
377    }
378
379    /* decompress the palette chunk */
380    if (cplz_chunk != -1) {
381
382/* yet to be handled */
383
384    }
385
386    /* convert the RGB palette into the machine's endian format */
387    if (cpl0_chunk != -1) {
388
389        chunk_size = AV_RB32(&s->buf[cpl0_chunk + 4]);
390        /* sanity check the palette size */
391        if (chunk_size / 3 > 256) {
392            av_log(s->avctx, AV_LOG_ERROR, "  VQA video: problem: found a palette chunk with %d colors\n",
393                chunk_size / 3);
394            return;
395        }
396        cpl0_chunk += CHUNK_PREAMBLE_SIZE;
397        for (i = 0; i < chunk_size / 3; i++) {
398            /* scale by 4 to transform 6-bit palette -> 8-bit */
399            r = s->buf[cpl0_chunk++] * 4;
400            g = s->buf[cpl0_chunk++] * 4;
401            b = s->buf[cpl0_chunk++] * 4;
402            s->palette[i] = (r << 16) | (g << 8) | (b);
403        }
404    }
405
406    /* next, look for a full codebook */
407    if ((cbf0_chunk != -1) && (cbfz_chunk != -1)) {
408
409        /* a chunk should not have both chunk types */
410        av_log(s->avctx, AV_LOG_ERROR, "  VQA video: problem: found both CBF0 and CBFZ chunks\n");
411        return;
412    }
413
414    /* decompress the full codebook chunk */
415    if (cbfz_chunk != -1) {
416
417        chunk_size = AV_RB32(&s->buf[cbfz_chunk + 4]);
418        cbfz_chunk += CHUNK_PREAMBLE_SIZE;
419        decode_format80(&s->buf[cbfz_chunk], chunk_size,
420            s->codebook, s->codebook_size, 0);
421    }
422
423    /* copy a full codebook */
424    if (cbf0_chunk != -1) {
425
426        chunk_size = AV_RB32(&s->buf[cbf0_chunk + 4]);
427        /* sanity check the full codebook size */
428        if (chunk_size > MAX_CODEBOOK_SIZE) {
429            av_log(s->avctx, AV_LOG_ERROR, "  VQA video: problem: CBF0 chunk too large (0x%X bytes)\n",
430                chunk_size);
431            return;
432        }
433        cbf0_chunk += CHUNK_PREAMBLE_SIZE;
434
435        memcpy(s->codebook, &s->buf[cbf0_chunk], chunk_size);
436    }
437
438    /* decode the frame */
439    if (vptz_chunk == -1) {
440
441        /* something is wrong if there is no VPTZ chunk */
442        av_log(s->avctx, AV_LOG_ERROR, "  VQA video: problem: no VPTZ chunk found\n");
443        return;
444    }
445
446    chunk_size = AV_RB32(&s->buf[vptz_chunk + 4]);
447    vptz_chunk += CHUNK_PREAMBLE_SIZE;
448    decode_format80(&s->buf[vptz_chunk], chunk_size,
449        s->decode_buffer, s->decode_buffer_size, 1);
450
451    /* render the final PAL8 frame */
452    if (s->vector_height == 4)
453        index_shift = 4;
454    else
455        index_shift = 3;
456    for (y = 0; y < s->frame.linesize[0] * s->height;
457        y += s->frame.linesize[0] * s->vector_height) {
458
459        for (x = y; x < y + s->width; x += 4, lobytes++, hibytes++) {
460            pixel_ptr = x;
461
462            /* get the vector index, the method for which varies according to
463             * VQA file version */
464            switch (s->vqa_version) {
465
466            case 1:
467/* still need sample media for this case (only one game, "Legend of
468 * Kyrandia III : Malcolm's Revenge", is known to use this version) */
469                lobyte = s->decode_buffer[lobytes * 2];
470                hibyte = s->decode_buffer[(lobytes * 2) + 1];
471                vector_index = ((hibyte << 8) | lobyte) >> 3;
472                vector_index <<= index_shift;
473                lines = s->vector_height;
474                /* uniform color fill - a quick hack */
475                if (hibyte == 0xFF) {
476                    while (lines--) {
477                        s->frame.data[0][pixel_ptr + 0] = 255 - lobyte;
478                        s->frame.data[0][pixel_ptr + 1] = 255 - lobyte;
479                        s->frame.data[0][pixel_ptr + 2] = 255 - lobyte;
480                        s->frame.data[0][pixel_ptr + 3] = 255 - lobyte;
481                        pixel_ptr += s->frame.linesize[0];
482                    }
483                    lines=0;
484                }
485                break;
486
487            case 2:
488                lobyte = s->decode_buffer[lobytes];
489                hibyte = s->decode_buffer[hibytes];
490                vector_index = (hibyte << 8) | lobyte;
491                vector_index <<= index_shift;
492                lines = s->vector_height;
493                break;
494
495            case 3:
496/* not implemented yet */
497                lines = 0;
498                break;
499            }
500
501            while (lines--) {
502                s->frame.data[0][pixel_ptr + 0] = s->codebook[vector_index++];
503                s->frame.data[0][pixel_ptr + 1] = s->codebook[vector_index++];
504                s->frame.data[0][pixel_ptr + 2] = s->codebook[vector_index++];
505                s->frame.data[0][pixel_ptr + 3] = s->codebook[vector_index++];
506                pixel_ptr += s->frame.linesize[0];
507            }
508        }
509    }
510
511    /* handle partial codebook */
512    if ((cbp0_chunk != -1) && (cbpz_chunk != -1)) {
513        /* a chunk should not have both chunk types */
514        av_log(s->avctx, AV_LOG_ERROR, "  VQA video: problem: found both CBP0 and CBPZ chunks\n");
515        return;
516    }
517
518    if (cbp0_chunk != -1) {
519
520        chunk_size = AV_RB32(&s->buf[cbp0_chunk + 4]);
521        cbp0_chunk += CHUNK_PREAMBLE_SIZE;
522
523        /* accumulate partial codebook */
524        memcpy(&s->next_codebook_buffer[s->next_codebook_buffer_index],
525            &s->buf[cbp0_chunk], chunk_size);
526        s->next_codebook_buffer_index += chunk_size;
527
528        s->partial_countdown--;
529        if (s->partial_countdown == 0) {
530
531            /* time to replace codebook */
532            memcpy(s->codebook, s->next_codebook_buffer,
533                s->next_codebook_buffer_index);
534
535            /* reset accounting */
536            s->next_codebook_buffer_index = 0;
537            s->partial_countdown = s->partial_count;
538        }
539    }
540
541    if (cbpz_chunk != -1) {
542
543        chunk_size = AV_RB32(&s->buf[cbpz_chunk + 4]);
544        cbpz_chunk += CHUNK_PREAMBLE_SIZE;
545
546        /* accumulate partial codebook */
547        memcpy(&s->next_codebook_buffer[s->next_codebook_buffer_index],
548            &s->buf[cbpz_chunk], chunk_size);
549        s->next_codebook_buffer_index += chunk_size;
550
551        s->partial_countdown--;
552        if (s->partial_countdown == 0) {
553
554            /* decompress codebook */
555            decode_format80(s->next_codebook_buffer,
556                s->next_codebook_buffer_index,
557                s->codebook, s->codebook_size, 0);
558
559            /* reset accounting */
560            s->next_codebook_buffer_index = 0;
561            s->partial_countdown = s->partial_count;
562        }
563    }
564}
565
566static int vqa_decode_frame(AVCodecContext *avctx,
567                            void *data, int *data_size,
568                            const uint8_t *buf, int buf_size)
569{
570    VqaContext *s = avctx->priv_data;
571
572    s->buf = buf;
573    s->size = buf_size;
574
575    if (s->frame.data[0])
576        avctx->release_buffer(avctx, &s->frame);
577
578    if (avctx->get_buffer(avctx, &s->frame)) {
579        av_log(s->avctx, AV_LOG_ERROR, "  VQA Video: get_buffer() failed\n");
580        return -1;
581    }
582
583    vqa_decode_chunk(s);
584
585    /* make the palette available on the way out */
586    memcpy(s->frame.data[1], s->palette, PALETTE_COUNT * 4);
587    s->frame.palette_has_changed = 1;
588
589    *data_size = sizeof(AVFrame);
590    *(AVFrame*)data = s->frame;
591
592    /* report that the buffer was completely consumed */
593    return buf_size;
594}
595
596static av_cold int vqa_decode_end(AVCodecContext *avctx)
597{
598    VqaContext *s = avctx->priv_data;
599
600    av_free(s->codebook);
601    av_free(s->next_codebook_buffer);
602    av_free(s->decode_buffer);
603
604    if (s->frame.data[0])
605        avctx->release_buffer(avctx, &s->frame);
606
607    return 0;
608}
609
610AVCodec vqa_decoder = {
611    "vqavideo",
612    CODEC_TYPE_VIDEO,
613    CODEC_ID_WS_VQA,
614    sizeof(VqaContext),
615    vqa_decode_init,
616    NULL,
617    vqa_decode_end,
618    vqa_decode_frame,
619    CODEC_CAP_DR1,
620    .long_name = NULL_IF_CONFIG_SMALL("Westwood Studios VQA (Vector Quantized Animation) video"),
621};
622