1/*
2 * Westwood Studios VQA Video Decoder
3 * Copyright (C) 2003 the ffmpeg project
4 *
5 * This file is part of Libav.
6 *
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.
11 *
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.
16 *
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
20 */
21
22/**
23 * @file
24 * VQA Video Decoder
25 * @author Mike Melanson (melanson@pcisys.net)
26 * @see 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
70#include "libavutil/intreadwrite.h"
71#include "libavutil/imgutils.h"
72#include "avcodec.h"
73#include "bytestream.h"
74
75#define PALETTE_COUNT 256
76#define VQA_HEADER_SIZE 0x2A
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
93typedef struct VqaContext {
94
95    AVCodecContext *avctx;
96    AVFrame frame;
97    GetByteContext gb;
98
99    uint32_t palette[PALETTE_COUNT];
100
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 */
106
107    unsigned char *codebook;         /* the current codebook */
108    int codebook_size;
109    unsigned char *next_codebook_buffer;  /* accumulator for next codebook */
110    int next_codebook_buffer_index;
111
112    unsigned char *decode_buffer;
113    int decode_buffer_size;
114
115    /* number of frames to go before replacing codebook */
116    int partial_countdown;
117    int partial_count;
118
119} VqaContext;
120
121static av_cold int vqa_decode_init(AVCodecContext *avctx)
122{
123    VqaContext *s = avctx->priv_data;
124    int i, j, codebook_index;
125
126    s->avctx = avctx;
127    avctx->pix_fmt = PIX_FMT_PAL8;
128
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 -1;
133    }
134
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(av_image_check_size(s->width, s->height, 0, avctx)){
140        s->width= s->height= 0;
141        return -1;
142    }
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];
146
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 -1;
152    }
153
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;
158    }
159
160    /* allocate codebooks */
161    s->codebook_size = MAX_CODEBOOK_SIZE;
162    s->codebook = av_malloc(s->codebook_size);
163    s->next_codebook_buffer = av_malloc(s->codebook_size);
164
165    /* initialize the solid-color vectors */
166    if (s->vector_height == 4) {
167        codebook_index = 0xFF00 * 16;
168        for (i = 0; i < 256; i++)
169            for (j = 0; j < 16; j++)
170                s->codebook[codebook_index++] = i;
171    } else {
172        codebook_index = 0xF00 * 8;
173        for (i = 0; i < 256; i++)
174            for (j = 0; j < 8; j++)
175                s->codebook[codebook_index++] = i;
176    }
177    s->next_codebook_buffer_index = 0;
178
179    /* allocate decode buffer */
180    s->decode_buffer_size = (s->width / s->vector_width) *
181        (s->height / s->vector_height) * 2;
182    s->decode_buffer = av_malloc(s->decode_buffer_size);
183
184    s->frame.data[0] = NULL;
185
186    return 0;
187}
188
189#define CHECK_COUNT() \
190    if (dest_index + count > dest_size) { \
191        av_log(NULL, AV_LOG_ERROR, "  VQA video: decode_format80 problem: next op would overflow dest_index\n"); \
192        av_log(NULL, AV_LOG_ERROR, "  VQA video: current dest_index = %d, count = %d, dest_size = %d\n", \
193            dest_index, count, dest_size); \
194        return AVERROR_INVALIDDATA; \
195    }
196
197#define CHECK_COPY(idx) \
198    if (idx < 0 || idx + count > dest_size) { \
199        av_log(NULL, AV_LOG_ERROR, "  VQA video: decode_format80 problem: next op would overflow dest_index\n"); \
200        av_log(NULL, AV_LOG_ERROR, "  VQA video: current src_pos = %d, count = %d, dest_size = %d\n", \
201            src_pos, count, dest_size); \
202        return AVERROR_INVALIDDATA; \
203    }
204
205
206static int decode_format80(GetByteContext *gb, int src_size,
207    unsigned char *dest, int dest_size, int check_size) {
208
209    int dest_index = 0;
210    int count, opcode, start;
211    int src_pos;
212    unsigned char color;
213    int i;
214
215    start = bytestream2_tell(gb);
216    while (bytestream2_tell(gb) - start < src_size) {
217        opcode = bytestream2_get_byte(gb);
218        av_dlog(NULL, "      opcode %02X: ", opcode);
219
220        /* 0x80 means that frame is finished */
221        if (opcode == 0x80)
222            return 0;
223
224        if (dest_index >= dest_size) {
225            av_log(NULL, AV_LOG_ERROR, "  VQA video: decode_format80 problem: dest_index (%d) exceeded dest_size (%d)\n",
226                dest_index, dest_size);
227            return AVERROR_INVALIDDATA;
228        }
229
230        if (opcode == 0xFF) {
231
232            count   = bytestream2_get_le16(gb);
233            src_pos = bytestream2_get_le16(gb);
234            av_dlog(NULL, "(1) copy %X bytes from absolute pos %X\n", count, src_pos);
235            CHECK_COUNT();
236            CHECK_COPY(src_pos);
237            for (i = 0; i < count; i++)
238                dest[dest_index + i] = dest[src_pos + i];
239            dest_index += count;
240
241        } else if (opcode == 0xFE) {
242
243            count = bytestream2_get_le16(gb);
244            color = bytestream2_get_byte(gb);
245            av_dlog(NULL, "(2) set %X bytes to %02X\n", count, color);
246            CHECK_COUNT();
247            memset(&dest[dest_index], color, count);
248            dest_index += count;
249
250        } else if ((opcode & 0xC0) == 0xC0) {
251
252            count = (opcode & 0x3F) + 3;
253            src_pos = bytestream2_get_le16(gb);
254            av_dlog(NULL, "(3) copy %X bytes from absolute pos %X\n", count, src_pos);
255            CHECK_COUNT();
256            CHECK_COPY(src_pos);
257            for (i = 0; i < count; i++)
258                dest[dest_index + i] = dest[src_pos + i];
259            dest_index += count;
260
261        } else if (opcode > 0x80) {
262
263            count = opcode & 0x3F;
264            av_dlog(NULL, "(4) copy %X bytes from source to dest\n", count);
265            CHECK_COUNT();
266            bytestream2_get_buffer(gb, &dest[dest_index], count);
267            dest_index += count;
268
269        } else {
270
271            count = ((opcode & 0x70) >> 4) + 3;
272            src_pos = bytestream2_get_byte(gb) | ((opcode & 0x0F) << 8);
273            av_dlog(NULL, "(5) copy %X bytes from relpos %X\n", count, src_pos);
274            CHECK_COUNT();
275            CHECK_COPY(dest_index - src_pos);
276            for (i = 0; i < count; i++)
277                dest[dest_index + i] = dest[dest_index - src_pos + i];
278            dest_index += count;
279        }
280    }
281
282    /* validate that the entire destination buffer was filled; this is
283     * important for decoding frame maps since each vector needs to have a
284     * codebook entry; it is not important for compressed codebooks because
285     * not every entry needs to be filled */
286    if (check_size)
287        if (dest_index < dest_size)
288            av_log(NULL, AV_LOG_ERROR, "  VQA video: decode_format80 problem: decode finished with dest_index (%d) < dest_size (%d)\n",
289                dest_index, dest_size);
290
291    return 0; // let's display what we decoded anyway
292}
293
294static int 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    int res;
304
305    int cbf0_chunk = -1;
306    int cbfz_chunk = -1;
307    int cbp0_chunk = -1;
308    int cbpz_chunk = -1;
309    int cpl0_chunk = -1;
310    int cplz_chunk = -1;
311    int vptz_chunk = -1;
312
313    int x, y;
314    int lines = 0;
315    int pixel_ptr;
316    int vector_index = 0;
317    int lobyte = 0;
318    int hibyte = 0;
319    int lobytes = 0;
320    int hibytes = s->decode_buffer_size / 2;
321
322    /* first, traverse through the frame and find the subchunks */
323    while (bytestream2_get_bytes_left(&s->gb) >= 8) {
324
325        chunk_type = bytestream2_get_be32u(&s->gb);
326        index      = bytestream2_tell(&s->gb);
327        chunk_size = bytestream2_get_be32u(&s->gb);
328
329        switch (chunk_type) {
330
331        case CBF0_TAG:
332            cbf0_chunk = index;
333            break;
334
335        case CBFZ_TAG:
336            cbfz_chunk = index;
337            break;
338
339        case CBP0_TAG:
340            cbp0_chunk = index;
341            break;
342
343        case CBPZ_TAG:
344            cbpz_chunk = index;
345            break;
346
347        case CPL0_TAG:
348            cpl0_chunk = index;
349            break;
350
351        case CPLZ_TAG:
352            cplz_chunk = index;
353            break;
354
355        case VPTZ_TAG:
356            vptz_chunk = index;
357            break;
358
359        default:
360            av_log(s->avctx, AV_LOG_ERROR, "  VQA video: Found unknown chunk type: %c%c%c%c (%08X)\n",
361            (chunk_type >> 24) & 0xFF,
362            (chunk_type >> 16) & 0xFF,
363            (chunk_type >>  8) & 0xFF,
364            (chunk_type >>  0) & 0xFF,
365            chunk_type);
366            break;
367        }
368
369        byte_skip = chunk_size & 0x01;
370        bytestream2_skip(&s->gb, chunk_size + byte_skip);
371    }
372
373    /* next, deal with the palette */
374    if ((cpl0_chunk != -1) && (cplz_chunk != -1)) {
375
376        /* a chunk should not have both chunk types */
377        av_log(s->avctx, AV_LOG_ERROR, "  VQA video: problem: found both CPL0 and CPLZ chunks\n");
378        return AVERROR_INVALIDDATA;
379    }
380
381    /* decompress the palette chunk */
382    if (cplz_chunk != -1) {
383
384/* yet to be handled */
385
386    }
387
388    /* convert the RGB palette into the machine's endian format */
389    if (cpl0_chunk != -1) {
390
391        bytestream2_seek(&s->gb, cpl0_chunk, SEEK_SET);
392        chunk_size = bytestream2_get_be32(&s->gb);
393        /* sanity check the palette size */
394        if (chunk_size / 3 > 256 || chunk_size > bytestream2_get_bytes_left(&s->gb)) {
395            av_log(s->avctx, AV_LOG_ERROR, "  VQA video: problem: found a palette chunk with %d colors\n",
396                chunk_size / 3);
397            return AVERROR_INVALIDDATA;
398        }
399        for (i = 0; i < chunk_size / 3; i++) {
400            /* scale by 4 to transform 6-bit palette -> 8-bit */
401            r = bytestream2_get_byteu(&s->gb) * 4;
402            g = bytestream2_get_byteu(&s->gb) * 4;
403            b = bytestream2_get_byteu(&s->gb) * 4;
404            s->palette[i] = (r << 16) | (g << 8) | (b);
405        }
406    }
407
408    /* next, look for a full codebook */
409    if ((cbf0_chunk != -1) && (cbfz_chunk != -1)) {
410
411        /* a chunk should not have both chunk types */
412        av_log(s->avctx, AV_LOG_ERROR, "  VQA video: problem: found both CBF0 and CBFZ chunks\n");
413        return AVERROR_INVALIDDATA;
414    }
415
416    /* decompress the full codebook chunk */
417    if (cbfz_chunk != -1) {
418
419        bytestream2_seek(&s->gb, cbfz_chunk, SEEK_SET);
420        chunk_size = bytestream2_get_be32(&s->gb);
421        if ((res = decode_format80(&s->gb, chunk_size, s->codebook,
422                                   s->codebook_size, 0)) < 0)
423            return res;
424    }
425
426    /* copy a full codebook */
427    if (cbf0_chunk != -1) {
428
429        bytestream2_seek(&s->gb, cbf0_chunk, SEEK_SET);
430        chunk_size = bytestream2_get_be32(&s->gb);
431        /* sanity check the full codebook size */
432        if (chunk_size > MAX_CODEBOOK_SIZE) {
433            av_log(s->avctx, AV_LOG_ERROR, "  VQA video: problem: CBF0 chunk too large (0x%X bytes)\n",
434                chunk_size);
435            return AVERROR_INVALIDDATA;
436        }
437
438        bytestream2_get_buffer(&s->gb, s->codebook, chunk_size);
439    }
440
441    /* decode the frame */
442    if (vptz_chunk == -1) {
443
444        /* something is wrong if there is no VPTZ chunk */
445        av_log(s->avctx, AV_LOG_ERROR, "  VQA video: problem: no VPTZ chunk found\n");
446        return AVERROR_INVALIDDATA;
447    }
448
449    bytestream2_seek(&s->gb, vptz_chunk, SEEK_SET);
450    chunk_size = bytestream2_get_be32(&s->gb);
451    if ((res = decode_format80(&s->gb, chunk_size,
452                               s->decode_buffer, s->decode_buffer_size, 1)) < 0)
453        return res;
454
455    /* render the final PAL8 frame */
456    if (s->vector_height == 4)
457        index_shift = 4;
458    else
459        index_shift = 3;
460    for (y = 0; y < s->frame.linesize[0] * s->height;
461        y += s->frame.linesize[0] * s->vector_height) {
462
463        for (x = y; x < y + s->width; x += 4, lobytes++, hibytes++) {
464            pixel_ptr = x;
465
466            /* get the vector index, the method for which varies according to
467             * VQA file version */
468            switch (s->vqa_version) {
469
470            case 1:
471                lobyte = s->decode_buffer[lobytes * 2];
472                hibyte = s->decode_buffer[(lobytes * 2) + 1];
473                vector_index = ((hibyte << 8) | lobyte) >> 3;
474                vector_index <<= index_shift;
475                lines = s->vector_height;
476                /* uniform color fill - a quick hack */
477                if (hibyte == 0xFF) {
478                    while (lines--) {
479                        s->frame.data[0][pixel_ptr + 0] = 255 - lobyte;
480                        s->frame.data[0][pixel_ptr + 1] = 255 - lobyte;
481                        s->frame.data[0][pixel_ptr + 2] = 255 - lobyte;
482                        s->frame.data[0][pixel_ptr + 3] = 255 - lobyte;
483                        pixel_ptr += s->frame.linesize[0];
484                    }
485                    lines=0;
486                }
487                break;
488
489            case 2:
490                lobyte = s->decode_buffer[lobytes];
491                hibyte = s->decode_buffer[hibytes];
492                vector_index = (hibyte << 8) | lobyte;
493                vector_index <<= index_shift;
494                lines = s->vector_height;
495                break;
496
497            case 3:
498/* not implemented yet */
499                lines = 0;
500                break;
501            }
502
503            while (lines--) {
504                s->frame.data[0][pixel_ptr + 0] = s->codebook[vector_index++];
505                s->frame.data[0][pixel_ptr + 1] = s->codebook[vector_index++];
506                s->frame.data[0][pixel_ptr + 2] = s->codebook[vector_index++];
507                s->frame.data[0][pixel_ptr + 3] = s->codebook[vector_index++];
508                pixel_ptr += s->frame.linesize[0];
509            }
510        }
511    }
512
513    /* handle partial codebook */
514    if ((cbp0_chunk != -1) && (cbpz_chunk != -1)) {
515        /* a chunk should not have both chunk types */
516        av_log(s->avctx, AV_LOG_ERROR, "  VQA video: problem: found both CBP0 and CBPZ chunks\n");
517        return AVERROR_INVALIDDATA;
518    }
519
520    if (cbp0_chunk != -1) {
521
522        bytestream2_seek(&s->gb, cbp0_chunk, SEEK_SET);
523        chunk_size = bytestream2_get_be32(&s->gb);
524
525        /* accumulate partial codebook */
526        bytestream2_get_buffer(&s->gb, &s->next_codebook_buffer[s->next_codebook_buffer_index],
527                               chunk_size);
528        s->next_codebook_buffer_index += chunk_size;
529
530        s->partial_countdown--;
531        if (s->partial_countdown == 0) {
532
533            /* time to replace codebook */
534            memcpy(s->codebook, s->next_codebook_buffer,
535                s->next_codebook_buffer_index);
536
537            /* reset accounting */
538            s->next_codebook_buffer_index = 0;
539            s->partial_countdown = s->partial_count;
540        }
541    }
542
543    if (cbpz_chunk != -1) {
544
545        bytestream2_seek(&s->gb, cbpz_chunk, SEEK_SET);
546        chunk_size = bytestream2_get_be32(&s->gb);
547
548        /* accumulate partial codebook */
549        bytestream2_get_buffer(&s->gb, &s->next_codebook_buffer[s->next_codebook_buffer_index],
550                               chunk_size);
551        s->next_codebook_buffer_index += chunk_size;
552
553        s->partial_countdown--;
554        if (s->partial_countdown == 0) {
555            GetByteContext gb;
556
557            bytestream2_init(&gb, s->next_codebook_buffer, s->next_codebook_buffer_index);
558            /* decompress codebook */
559            if ((res = decode_format80(&gb, s->next_codebook_buffer_index,
560                                       s->codebook, s->codebook_size, 0)) < 0)
561                return res;
562
563            /* reset accounting */
564            s->next_codebook_buffer_index = 0;
565            s->partial_countdown = s->partial_count;
566        }
567    }
568
569    return 0;
570}
571
572static int vqa_decode_frame(AVCodecContext *avctx,
573                            void *data, int *data_size,
574                            AVPacket *avpkt)
575{
576    VqaContext *s = avctx->priv_data;
577    int res;
578
579    if (s->frame.data[0])
580        avctx->release_buffer(avctx, &s->frame);
581
582    if (avctx->get_buffer(avctx, &s->frame)) {
583        av_log(s->avctx, AV_LOG_ERROR, "  VQA Video: get_buffer() failed\n");
584        return -1;
585    }
586
587    bytestream2_init(&s->gb, avpkt->data, avpkt->size);
588    if ((res = vqa_decode_chunk(s)) < 0)
589        return res;
590
591    /* make the palette available on the way out */
592    memcpy(s->frame.data[1], s->palette, PALETTE_COUNT * 4);
593    s->frame.palette_has_changed = 1;
594
595    *data_size = sizeof(AVFrame);
596    *(AVFrame*)data = s->frame;
597
598    /* report that the buffer was completely consumed */
599    return avpkt->size;
600}
601
602static av_cold int vqa_decode_end(AVCodecContext *avctx)
603{
604    VqaContext *s = avctx->priv_data;
605
606    av_free(s->codebook);
607    av_free(s->next_codebook_buffer);
608    av_free(s->decode_buffer);
609
610    if (s->frame.data[0])
611        avctx->release_buffer(avctx, &s->frame);
612
613    return 0;
614}
615
616AVCodec ff_vqa_decoder = {
617    .name           = "vqavideo",
618    .type           = AVMEDIA_TYPE_VIDEO,
619    .id             = CODEC_ID_WS_VQA,
620    .priv_data_size = sizeof(VqaContext),
621    .init           = vqa_decode_init,
622    .close          = vqa_decode_end,
623    .decode         = vqa_decode_frame,
624    .capabilities   = CODEC_CAP_DR1,
625    .long_name = NULL_IF_CONFIG_SMALL("Westwood Studios VQA (Vector Quantized Animation) video"),
626};
627