comparison vqavideo.c @ 1501:dc6f46cd4a7b libavcodec

added solid color vectors; basic PAL8, 4x2-vector video (as in Command & Conquer) looks great now; also added compressed codebook support, but files using these (Lands of Lore) are not correct yet

1500:726a81d83ddd

 * @file vqavideo.c
 * VQA Video Decoder by Mike Melanson ([e-mail])
 * For more information about the RPZA format, visit:
 *   http://www.pcisys.net/~melanson/codecs/
 *
 * The VQA video decoder outputs PAL8 colorspace data.
 *
 * This decoder needs the 42-byte VQHD header from the beginning
 * of the VQA file passed through the extradata field. The VQHD header
 * is laid out as:
 *
 *   bytes 0-3   chunk fourcc: 'VQHD'
 *   bytes 4-7   chunk size in big-endian format, should be 0x0000002A
 *   bytes 8-50  VQHD chunk data
 *
 * Bytes 8-50 are what this decoder expects to see.
 *
 * Briefly, VQA is a vector quantized animation format that operates in a
 * 6-bit VGA palettized colorspace. It operates on pixel vectors (blocks)
 * of either 4x2 or 4x4 in size. Compressed VQA chunks can contain vector
 * codebooks, palette information, and code maps for rendering vectors onto
 * frames. Any of these components can also be compressed with a run-length
 * encoding (RLE) algorithm commonly referred to as "format80".
 *

 * seeking difficult despite the fact that the frames are all intracoded.
 *
 * V1,2 VQA uses 12-bit codebook indices. If the 12-bit indices were
 * packed into bytes and then RLE compressed, bytewise, the results would
 * be poor. That is why the coding method divides each index into 2 parts,
 * the top 4 bits and the bottom 8 bits, the RL encodes the 4-bit pieces
 * together and the 8-bit pieces together. If most of the vectors are
 * clustered into one group of 256 vectors, most of the 4-bit index pieces
 * should be the same.
 */

#define PALETTE_COUNT 256
#define VQA_HEADER_SIZE 0x2A
#define CHUNK_PREAMBLE_SIZE 8

/* v1, v2 files: each vector codebook entry is 4x2=8 pixels = 8 bytes */
#define V1_2_VECTOR_SIZE 8
#define V1_2_MAX_VECTORS 0xF00
/* v3 files: each vector codebook entry is 4x4=16 pixels = 16 bytes */
#define V3_VECTOR_SIZE 16
#define V3_MAX_VECTORS 0xFF00

#define LE_16(x)  ((((uint8_t*)(x))[1] << 8) | ((uint8_t*)(x))[0])
#define BE_16(x)  ((((uint8_t*)(x))[0] << 8) | ((uint8_t*)(x))[1])
#define BE_32(x)  ((((uint8_t*)(x))[0] << 24) | \
                   (((uint8_t*)(x))[1] << 16) | \

    int vector_width;  /* width of individual vector */
    int vector_height;  /* height of individual vector */
    int vqa_version;  /* this should be either 1, 2 or 3 */

    unsigned char *codebook;         /* the current codebook */
    unsigned char *next_codebook_buffer;  /* accumulator for next codebook */
    int next_codebook_buffer_index;

    unsigned char *decode_buffer;
    int decode_buffer_size;

static int vqa_decode_init(AVCodecContext *avctx)
{
    VqaContext *s = (VqaContext *)avctx->priv_data;
    unsigned char *vqa_header;

    s->avctx = avctx;
    avctx->pix_fmt = PIX_FMT_PAL8;
    avctx->has_b_frames = 0;
    dsputil_init(&s->dsp, avctx);

        /* return without further initialization */
        return -1;
    }

    /* allocate codebooks */
    if (s->vqa_version == 3) {
        s->codebook = av_malloc(V3_VECTOR_SIZE * V3_MAX_VECTORS);
        s->next_codebook_buffer = av_malloc(V3_VECTOR_SIZE * V3_MAX_VECTORS);
    } else {
        s->codebook = av_malloc(V1_2_VECTOR_SIZE * V1_2_MAX_VECTORS);
        s->next_codebook_buffer = av_malloc(V1_2_VECTOR_SIZE * V1_2_MAX_VECTORS);
    }
    s->next_codebook_buffer_index = 0;

    /* allocate decode buffer */
    s->decode_buffer_size = (s->width / s->vector_width) *

    return 0;
}

#define CHECK_COUNT() \
    if (dest_index + count > dest_size) { \
        printf ("vqavideo: decode_format80 problem: next op would overflow dest_index\n"); \
        printf ("vqavideo: current dest_index = %d, count = %d, dest_size = %d\n", \
            dest_index, count, dest_size); \
        return; \
    }

static void decode_format80(unsigned char *src, int src_size,
    unsigned char *dest, int dest_size) {

    int src_index = 0;
    int dest_index = 0;
    int count;
    int src_pos;

        /* 0x80 means that frame is finished */
        if (src[src_index] == 0x80)
            return;

        if (dest_index >= dest_size) {
            printf ("vqavideo: decode_format80 problem: dest_index (%d) exceeded dest_size (%d)\n",
                dest_index, dest_size);
            return;
        }

        if (src[src_index] == 0xFF) {

                dest[dest_index + i] = dest[dest_index - src_pos + i];
            dest_index += count;
        }
    }

    if (dest_index < dest_size)
        printf ("vqavideo: decode_format80 problem: decode finished with dest_index (%d) < dest_size (%d)\n",
            dest_index, dest_size);
}

static void vqa_decode_chunk(VqaContext *s)
{

//static int frame = 0;
    unsigned int chunk_type;
    unsigned int chunk_size;
    int byte_skip;
    unsigned int index = 0;
    int i;

    int lobyte = 0;
    int hibyte = 0;
    int lobytes = 0;
    int hibytes = s->decode_buffer_size / 2;

//printf (" **** decoding frame #%d, stride = %d\n", frame++, s->frame.linesize[0]);
    /* first, traverse through the frame and find the subchunks */
    while (index < s->size) {

        chunk_type = BE_32(&s->buf[index]);
        chunk_size = BE_32(&s->buf[index + 4]);

    if (cpl0_chunk != -1) {

        chunk_size = BE_32(&s->buf[cpl0_chunk + 4]);
        /* sanity check the palette size */
        if (chunk_size / 3 > 256) {
            printf ("vqavideo: problem: found a palette chunk with %d colors\n",
                chunk_size / 3);
            return;
        }
        cpl0_chunk += CHUNK_PREAMBLE_SIZE;
        palette32 = (unsigned int *)s->palette;

        /* a chunk should not have both chunk types */
        printf ("  VQA video: problem: found both CBF0 and CBFZ chunks\n");
        return;
    }

    /* decompress the full codebook chunk into the codebook accumulation
     * buffer; this is safe since only the first frame is supposed to have 
     * a full codebook */
    if (cbfz_chunk != -1) {

/* yet to be handled */

    }

    /* copy a full codebook */
    if (cbf0_chunk != -1) {

        index = cbf0_chunk;

        chunk_size = BE_32(&s->buf[cbf0_chunk + 4]);
        /* sanity check the full codebook size */
        if (s->vqa_version == 3) {
            if (chunk_size / (V3_VECTOR_SIZE) > V3_MAX_VECTORS) {
                printf ("  VQA video: problem: CBF0 chunk too large (0x%X bytes)\n",
                    chunk_size);
                return;
            }
        } else {
            if (chunk_size / (V1_2_VECTOR_SIZE) > V1_2_MAX_VECTORS) {
                printf ("  VQA video: problem: CBF0 chunk too large (0x%X bytes)\n",
                    chunk_size);
                return;
            }
        }
        cbf0_chunk += CHUNK_PREAMBLE_SIZE;

        memcpy(s->codebook, &s->buf[cbf0_chunk], chunk_size);
    }

        /* something is wrong if there is no VPTZ chunk */
        printf ("  VQA video: problem: no VPTZ chunk found\n");
        return;
    }

    /* decode frame */
    chunk_size = BE_32(&s->buf[vptz_chunk + 4]);
    vptz_chunk += CHUNK_PREAMBLE_SIZE;
    decode_format80(&s->buf[vptz_chunk], chunk_size,
        s->decode_buffer, s->decode_buffer_size);


if (0)
{
int count = 0;
printf (" finished decoding frame...");
for (index = 8000; index < 16000; index++)
  if (s->decode_buffer[index] > 0x0F)
    count++;
printf ("  %d/8000 hibytes exceeded 0x0F\n", count);
}


    /* render the final PAL8 frame */
    for (y = 0; y < s->frame.linesize[0] * s->height; 
        y += s->frame.linesize[0] * s->vector_height) {

                break;

            case 2:
                lobyte = s->decode_buffer[lobytes];
                hibyte = s->decode_buffer[hibytes];
                if (hibyte > 0x0F) {
                    printf ("  VQA video: problem: vector #%d/%d high byte out of range (0x%X >= 0x0F)\n",
                        lobytes, s->decode_buffer_size / 2, hibyte);
                    hibyte = lobyte = 0;
                } else if (hibyte == 0x0F)
                    hibyte = 0;
                vector_index = (hibyte << 8) | lobyte;
                vector_index *= V1_2_VECTOR_SIZE;
                lines = 2;
                break;

            case 3:
/* not implemented yet */
                lines = 0;

        }
    }

    if (cbpz_chunk != -1) {

/* more partial codebook handling ... */

    }
}

static int vqa_decode_frame(AVCodecContext *avctx,
                            void *data, int *data_size,

1501:dc6f46cd4a7b

 * @file vqavideo.c
 * VQA Video Decoder by Mike Melanson ([e-mail])
 * For more information about the RPZA format, visit:
 *   http://www.pcisys.net/~melanson/codecs/
 *
 * The VQA video decoder outputs PAL8 or RGB555 colorspace data, depending
 * on the type of data in the file.
 *
 * This decoder needs the 42-byte VQHD header from the beginning
 * of the VQA file passed through the extradata field. The VQHD header
 * is laid out as:
 *
 *   bytes 0-3   chunk fourcc: 'VQHD'
 *   bytes 4-7   chunk size in big-endian format, should be 0x0000002A
 *   bytes 8-49  VQHD chunk data
 *
 * Bytes 8-49 are what this decoder expects to see.
 *
 * Briefly, VQA is a vector quantized animation format that operates in a
 * VGA palettized colorspace. It operates on pixel vectors (blocks)
 * of either 4x2 or 4x4 in size. Compressed VQA chunks can contain vector
 * codebooks, palette information, and code maps for rendering vectors onto
 * frames. Any of these components can also be compressed with a run-length
 * encoding (RLE) algorithm commonly referred to as "format80".
 *

 * seeking difficult despite the fact that the frames are all intracoded.
 *
 * V1,2 VQA uses 12-bit codebook indices. If the 12-bit indices were
 * packed into bytes and then RLE compressed, bytewise, the results would
 * be poor. That is why the coding method divides each index into 2 parts,
 * the top 4 bits and the bottom 8 bits, then RL encodes the 4-bit pieces
 * together and the 8-bit pieces together. If most of the vectors are
 * clustered into one group of 256 vectors, most of the 4-bit index pieces
 * should be the same.
 */

#define PALETTE_COUNT 256
#define VQA_HEADER_SIZE 0x2A
#define CHUNK_PREAMBLE_SIZE 8

/* allocate the maximum vector space, regardless of the file version:
 * (0xFF00 codebook vectors + 0x100 solid pixel vectors) * (4x4 pixels/block) */
#define MAX_CODEBOOK_VECTORS 0xFF00
#define SOLID_PIXEL_VECTORS 0x100
#define MAX_VECTORS (MAX_CODEBOOK_VECTORS + SOLID_PIXEL_VECTORS)
#define MAX_CODEBOOK_SIZE (MAX_VECTORS * 4 * 4)

#define LE_16(x)  ((((uint8_t*)(x))[1] << 8) | ((uint8_t*)(x))[0])
#define BE_16(x)  ((((uint8_t*)(x))[0] << 8) | ((uint8_t*)(x))[1])
#define BE_32(x)  ((((uint8_t*)(x))[0] << 24) | \
                   (((uint8_t*)(x))[1] << 16) | \

    int vector_width;  /* width of individual vector */
    int vector_height;  /* height of individual vector */
    int vqa_version;  /* this should be either 1, 2 or 3 */

    unsigned char *codebook;         /* the current codebook */
    int codebook_size;
    unsigned char *next_codebook_buffer;  /* accumulator for next codebook */
    int next_codebook_buffer_index;

    unsigned char *decode_buffer;
    int decode_buffer_size;

static int vqa_decode_init(AVCodecContext *avctx)
{
    VqaContext *s = (VqaContext *)avctx->priv_data;
    unsigned char *vqa_header;
    int i, j, codebook_index;;

    s->avctx = avctx;
    avctx->pix_fmt = PIX_FMT_PAL8;
    avctx->has_b_frames = 0;
    dsputil_init(&s->dsp, avctx);

        /* return without further initialization */
        return -1;
    }

    /* allocate codebooks */
    s->codebook_size = MAX_CODEBOOK_SIZE;
    s->codebook = av_malloc(s->codebook_size);
    s->next_codebook_buffer = av_malloc(s->codebook_size);

    /* initialize the solid-color vectors */
    if (s->vector_height == 4) {
        codebook_index = 0xFF00 * 16;
        for (i = 0; i < 256; i++)
            for (j = 0; j < 16; j++)
                s->codebook[codebook_index++] = i;
    } else {
        codebook_index = 0xF00 * 8;
        for (i = 0; i < 256; i++)
            for (j = 0; j < 8; j++)
                s->codebook[codebook_index++] = i;
    }
    s->next_codebook_buffer_index = 0;

    /* allocate decode buffer */
    s->decode_buffer_size = (s->width / s->vector_width) *

    return 0;
}

#define CHECK_COUNT() \
    if (dest_index + count > dest_size) { \
        printf ("  VQA video: decode_format80 problem: next op would overflow dest_index\n"); \
        printf ("  VQA video: current dest_index = %d, count = %d, dest_size = %d\n", \
            dest_index, count, dest_size); \
        return; \
    }

static void decode_format80(unsigned char *src, int src_size,
    unsigned char *dest, int dest_size, int check_size) {

    int src_index = 0;
    int dest_index = 0;
    int count;
    int src_pos;

        /* 0x80 means that frame is finished */
        if (src[src_index] == 0x80)
            return;

        if (dest_index >= dest_size) {
            printf ("  VQA video: decode_format80 problem: dest_index (%d) exceeded dest_size (%d)\n",
                dest_index, dest_size);
            return;
        }

        if (src[src_index] == 0xFF) {

                dest[dest_index + i] = dest[dest_index - src_pos + i];
            dest_index += count;
        }
    }

    /* validate that the entire destination buffer was filled; this is
     * important for decoding frame maps since each vector needs to have a
     * codebook entry; it is not important for compressed codebooks because
     * not every entry needs to be filled */
    if (check_size)
        if (dest_index < dest_size)
            printf ("  VQA video: decode_format80 problem: decode finished with dest_index (%d) < dest_size (%d)\n",
                dest_index, dest_size);
}

static void vqa_decode_chunk(VqaContext *s)
{
    unsigned int chunk_type;
    unsigned int chunk_size;
    int byte_skip;
    unsigned int index = 0;
    int i;

    int lobyte = 0;
    int hibyte = 0;
    int lobytes = 0;
    int hibytes = s->decode_buffer_size / 2;

    /* first, traverse through the frame and find the subchunks */
    while (index < s->size) {

        chunk_type = BE_32(&s->buf[index]);
        chunk_size = BE_32(&s->buf[index + 4]);

    if (cpl0_chunk != -1) {

        chunk_size = BE_32(&s->buf[cpl0_chunk + 4]);
        /* sanity check the palette size */
        if (chunk_size / 3 > 256) {
            printf ("  VQA video: problem: found a palette chunk with %d colors\n",
                chunk_size / 3);
            return;
        }
        cpl0_chunk += CHUNK_PREAMBLE_SIZE;
        palette32 = (unsigned int *)s->palette;

        /* a chunk should not have both chunk types */
        printf ("  VQA video: problem: found both CBF0 and CBFZ chunks\n");
        return;
    }

    /* decompress the full codebook chunk */
    if (cbfz_chunk != -1) {

        chunk_size = BE_32(&s->buf[cbfz_chunk + 4]);
        cbfz_chunk += CHUNK_PREAMBLE_SIZE;
        decode_format80(&s->buf[cbfz_chunk], chunk_size,
            s->codebook, s->codebook_size, 0);
    }

    /* copy a full codebook */
    if (cbf0_chunk != -1) {

        chunk_size = BE_32(&s->buf[cbf0_chunk + 4]);
        /* sanity check the full codebook size */
        if (chunk_size > MAX_CODEBOOK_SIZE) {
            printf ("  VQA video: problem: CBF0 chunk too large (0x%X bytes)\n",
                chunk_size);
            return;
        }
        cbf0_chunk += CHUNK_PREAMBLE_SIZE;

        memcpy(s->codebook, &s->buf[cbf0_chunk], chunk_size);
    }

        /* something is wrong if there is no VPTZ chunk */
        printf ("  VQA video: problem: no VPTZ chunk found\n");
        return;
    }

    chunk_size = BE_32(&s->buf[vptz_chunk + 4]);
    vptz_chunk += CHUNK_PREAMBLE_SIZE;
    decode_format80(&s->buf[vptz_chunk], chunk_size,
        s->decode_buffer, s->decode_buffer_size, 1);

    /* render the final PAL8 frame */
    for (y = 0; y < s->frame.linesize[0] * s->height; 
        y += s->frame.linesize[0] * s->vector_height) {

                break;

            case 2:
                lobyte = s->decode_buffer[lobytes];
                hibyte = s->decode_buffer[hibytes];
                vector_index = (hibyte << 8) | lobyte;
                vector_index *= 8;
                lines = s->vector_height;
                break;

            case 3:
/* not implemented yet */
                lines = 0;

        }
    }

    if (cbpz_chunk != -1) {

        chunk_size = BE_32(&s->buf[cbpz_chunk + 4]);
        cbpz_chunk += CHUNK_PREAMBLE_SIZE;

        /* accumulate partial codebook */
        memcpy(&s->next_codebook_buffer[s->next_codebook_buffer_index],
            &s->buf[cbpz_chunk], chunk_size);
        s->next_codebook_buffer_index += chunk_size;

        s->partial_countdown--;
        if (s->partial_countdown == 0) {

            /* decompress codebook */
            decode_format80(s->next_codebook_buffer, 
                s->next_codebook_buffer_index, 
                s->codebook, s->codebook_size, 0);

            /* reset accounting */
            s->next_codebook_buffer_index = 0;
            s->partial_countdown = s->partial_count;
        }
    }
}

static int vqa_decode_frame(AVCodecContext *avctx,
                            void *data, int *data_size,