comparison sonic.c @ 2205:b90828c162f5 libavcodec

more decorrelation types

2204:4f8da6a9e6eb

 *  - selectable intlist writers/readers (bonk-style, golomb, cabac)
 */

#define MAX_CHANNELS 2

typedef struct SonicContext {
    int lossless, mid_side;
    
    int num_taps, downsampling;
    double quantization;
    
    int channels, samplerate, block_align, frame_size;

	av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n");
        return -1; /* only stereo or mono for now */
    }

    if (avctx->channels == 2)
	s->mid_side = 1;

    if (avctx->codec->id == CODEC_ID_SONIC_LS)
    {
	s->lossless = 1;
	s->num_taps = 32;

	put_bits(&pb, 4, code_samplerate(s->samplerate));
    }
    put_bits(&pb, 1, s->lossless);
    if (!s->lossless)
	put_bits(&pb, 3, SAMPLE_SHIFT); // XXX FIXME: sample precision
    put_bits(&pb, 1, s->mid_side);
    put_bits(&pb, 2, s->downsampling);
    put_bits(&pb, 5, (s->num_taps >> 5)-1); // 32..1024
    put_bits(&pb, 1, 0); // XXX FIXME: no custom tap quant table

    flush_put_bits(&pb);
    avctx->extradata_size = put_bits_count(&pb)/8;

    av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d ms: %d taps: %d block: %d frame: %d downsamp: %d\n",
	version, s->lossless, s->mid_side, s->num_taps, s->block_align, s->frame_size, s->downsampling);

    avctx->coded_frame = avcodec_alloc_frame();
    if (!avctx->coded_frame)
	return -ENOMEM;
    avctx->coded_frame->key_frame = 1;

    init_put_bits(&pb, buf, buf_size*8);

    // short -> internal
    for (i = 0; i < s->frame_size; i++)
    {
//	if (samples[i] < 0)
//	    s->int_samples[i] = samples[i]+32768;
//	else
//	    s->int_samples[i] = samples[i]-32768;
	s->int_samples[i] = samples[i];
//	av_log(NULL, AV_LOG_INFO, "%d\n", s->int_samples[i]);
    }

    if (!s->lossless)
	for (i = 0; i < s->frame_size; i++)
	    s->int_samples[i] = s->int_samples[i] << SAMPLE_SHIFT;

    if (s->mid_side)
	for (i = 0; i < s->frame_size; i += s->channels)
	{
	    s->int_samples[i] += s->int_samples[i+1];
	    s->int_samples[i+1] -= shift(s->int_samples[i], 1);
	}

    memset(s->window, 0, 4* s->window_size);
    
    for (i = 0; i < s->tail_size; i++)
	s->window[x++] = s->tail[i];

    }

    s->lossless = get_bits1(&gb);
    if (!s->lossless)
	skip_bits(&gb, 3); // XXX FIXME
    s->mid_side = get_bits1(&gb);

    s->downsampling = get_bits(&gb, 2);
    s->num_taps = (get_bits(&gb, 5)+1)<<5;
    if (get_bits1(&gb)) // XXX FIXME
	av_log(avctx, AV_LOG_INFO, "Custom quant table\n");
    
    s->block_align = (int)(2048.0*(s->samplerate/44100))/s->downsampling;
    s->frame_size = s->channels*s->block_align*s->downsampling;
//    avctx->frame_size = s->block_align;

    av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d ms: %d taps: %d block: %d frame: %d downsamp: %d\n",
	version, s->lossless, s->mid_side, s->num_taps, s->block_align, s->frame_size, s->downsampling);

    // generate taps
    s->tap_quant = av_mallocz(4* s->num_taps);
    for (i = 0; i < s->num_taps; i++)
	s->tap_quant[i] = (int)(sqrt(i+1));

	for (i = 0; i < s->num_taps; i++)
	    s->predictor_state[ch][i] = s->int_samples[s->frame_size - s->channels + ch - i*s->channels];
    }
    
    if (s->mid_side)
	for (i = 0; i < s->frame_size; i += s->channels)
	{
	    s->int_samples[i+1] += shift(s->int_samples[i], 1);
	    s->int_samples[i] -= s->int_samples[i+1];
	}

    if (!s->lossless)
	for (i = 0; i < s->frame_size; i++)
	    s->int_samples[i] = shift(s->int_samples[i], SAMPLE_SHIFT);

	else
	    samples[i] = s->int_samples[i];
    }

    align_get_bits(&gb);

//    if (buf_size != (get_bits_count(&gb)+7)/8)
//	av_log(NULL, AV_LOG_INFO, "buf_size (%d) and used bytes (%d) differs\n", buf_size, (get_bits_count(&gb)+7)/8);

    *data_size = s->frame_size * 2;

    return (get_bits_count(&gb)+7)/8;
}

2205:b90828c162f5

 *  - selectable intlist writers/readers (bonk-style, golomb, cabac)
 */

#define MAX_CHANNELS 2

#define MID_SIDE 0
#define LEFT_SIDE 1
#define RIGHT_SIDE 2

typedef struct SonicContext {
    int lossless, decorrelation;
    
    int num_taps, downsampling;
    double quantization;
    
    int channels, samplerate, block_align, frame_size;

	av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n");
        return -1; /* only stereo or mono for now */
    }

    if (avctx->channels == 2)
	s->decorrelation = MID_SIDE;

    if (avctx->codec->id == CODEC_ID_SONIC_LS)
    {
	s->lossless = 1;
	s->num_taps = 32;

	put_bits(&pb, 4, code_samplerate(s->samplerate));
    }
    put_bits(&pb, 1, s->lossless);
    if (!s->lossless)
	put_bits(&pb, 3, SAMPLE_SHIFT); // XXX FIXME: sample precision
    put_bits(&pb, 2, s->decorrelation);
    put_bits(&pb, 2, s->downsampling);
    put_bits(&pb, 5, (s->num_taps >> 5)-1); // 32..1024
    put_bits(&pb, 1, 0); // XXX FIXME: no custom tap quant table

    flush_put_bits(&pb);
    avctx->extradata_size = put_bits_count(&pb)/8;

    av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n",
	version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling);

    avctx->coded_frame = avcodec_alloc_frame();
    if (!avctx->coded_frame)
	return -ENOMEM;
    avctx->coded_frame->key_frame = 1;

    init_put_bits(&pb, buf, buf_size*8);

    // short -> internal
    for (i = 0; i < s->frame_size; i++)
	s->int_samples[i] = samples[i];

    if (!s->lossless)
	for (i = 0; i < s->frame_size; i++)
	    s->int_samples[i] = s->int_samples[i] << SAMPLE_SHIFT;

    switch(s->decorrelation)
    {
	case MID_SIDE:
	    for (i = 0; i < s->frame_size; i += s->channels)
	    {
		s->int_samples[i] += s->int_samples[i+1];
		s->int_samples[i+1] -= shift(s->int_samples[i], 1);
	    }
	    break;
	case LEFT_SIDE:
	    for (i = 0; i < s->frame_size; i += s->channels)
		s->int_samples[i+1] -= s->int_samples[i];
	    break;
	case RIGHT_SIDE:
	    for (i = 0; i < s->frame_size; i += s->channels)
		s->int_samples[i] -= s->int_samples[i+1];
	    break;
    }

    memset(s->window, 0, 4* s->window_size);
    
    for (i = 0; i < s->tail_size; i++)
	s->window[x++] = s->tail[i];

    }

    s->lossless = get_bits1(&gb);
    if (!s->lossless)
	skip_bits(&gb, 3); // XXX FIXME
    s->decorrelation = get_bits(&gb, 2);

    s->downsampling = get_bits(&gb, 2);
    s->num_taps = (get_bits(&gb, 5)+1)<<5;
    if (get_bits1(&gb)) // XXX FIXME
	av_log(avctx, AV_LOG_INFO, "Custom quant table\n");
    
    s->block_align = (int)(2048.0*(s->samplerate/44100))/s->downsampling;
    s->frame_size = s->channels*s->block_align*s->downsampling;
//    avctx->frame_size = s->block_align;

    av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n",
	version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling);

    // generate taps
    s->tap_quant = av_mallocz(4* s->num_taps);
    for (i = 0; i < s->num_taps; i++)
	s->tap_quant[i] = (int)(sqrt(i+1));

	for (i = 0; i < s->num_taps; i++)
	    s->predictor_state[ch][i] = s->int_samples[s->frame_size - s->channels + ch - i*s->channels];
    }
    
    switch(s->decorrelation)
    {
	case MID_SIDE:
	    for (i = 0; i < s->frame_size; i += s->channels)
	    {
		s->int_samples[i+1] += shift(s->int_samples[i], 1);
		s->int_samples[i] -= s->int_samples[i+1];
	    }
	    break;
	case LEFT_SIDE:
	    for (i = 0; i < s->frame_size; i += s->channels)
		s->int_samples[i+1] += s->int_samples[i];
	    break;
	case RIGHT_SIDE:
	    for (i = 0; i < s->frame_size; i += s->channels)
		s->int_samples[i] += s->int_samples[i+1];
	    break;
    }

    if (!s->lossless)
	for (i = 0; i < s->frame_size; i++)
	    s->int_samples[i] = shift(s->int_samples[i], SAMPLE_SHIFT);

	else
	    samples[i] = s->int_samples[i];
    }

    align_get_bits(&gb);

    *data_size = s->frame_size * 2;

    return (get_bits_count(&gb)+7)/8;
}