alac.c

mediastreamer2是开源的网络传输媒体流的库

    }

    if (predictor_coef_table == 8) {
        /* FIXME: optimized general case */
        return;
    }
#endif

    /* general case */
    if (predictor_coef_num > 0) {
        for (i = predictor_coef_num + 1; i < output_size; i++) {
            int j;
            int sum = 0;
            int outval;
            int error_val = error_buffer[i];

            for (j = 0; j < predictor_coef_num; j++) {
                sum += (buffer_out[predictor_coef_num-j] - buffer_out[0]) *
                       predictor_coef_table[j];
            }

            outval = (1 << (predictor_quantitization-1)) + sum;
            outval = outval >> predictor_quantitization;
            outval = outval + buffer_out[0] + error_val;
            outval = extend_sign32(outval, readsamplesize);

            buffer_out[predictor_coef_num+1] = outval;

            if (error_val > 0) {
                int predictor_num = predictor_coef_num - 1;

                while (predictor_num >= 0 && error_val > 0) {
                    int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
                    int sign = sign_only(val);

                    predictor_coef_table[predictor_num] -= sign;

                    val *= sign; /* absolute value */

                    error_val -= ((val >> predictor_quantitization) *
                                  (predictor_coef_num - predictor_num));

                    predictor_num--;
                }
            } else if (error_val < 0) {
                int predictor_num = predictor_coef_num - 1;

                while (predictor_num >= 0 && error_val < 0) {
                    int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
                    int sign = - sign_only(val);

                    predictor_coef_table[predictor_num] -= sign;

                    val *= sign; /* neg value */

                    error_val -= ((val >> predictor_quantitization) *
                                  (predictor_coef_num - predictor_num));

                    predictor_num--;
                }
            }

            buffer_out++;
        }
    }
}

static void reconstruct_stereo_16(int32_t *buffer[MAX_CHANNELS],
                                  int16_t *buffer_out,
                                  int numchannels, int numsamples,
                                  uint8_t interlacing_shift,
                                  uint8_t interlacing_leftweight)
{
    int i;
    if (numsamples <= 0)
        return;

    /* weighted interlacing */
    if (interlacing_leftweight) {
        for (i = 0; i < numsamples; i++) {
            int32_t a, b;

            a = buffer[0][i];
            b = buffer[1][i];

            a -= (b * interlacing_leftweight) >> interlacing_shift;
            b += a;

            buffer_out[i*numchannels] = b;
            buffer_out[i*numchannels + 1] = a;
        }

        return;
    }

    /* otherwise basic interlacing took place */
    for (i = 0; i < numsamples; i++) {
        int16_t left, right;

        left = buffer[0][i];
        right = buffer[1][i];

        buffer_out[i*numchannels] = left;
        buffer_out[i*numchannels + 1] = right;
    }
}

static int alac_decode_frame(AVCodecContext *avctx,
                             void *outbuffer, int *outputsize,
                             const uint8_t *inbuffer, int input_buffer_size)
{
    ALACContext *alac = avctx->priv_data;

    int channels;
    int32_t outputsamples;
    int hassize;
    int readsamplesize;
    int wasted_bytes;
    int isnotcompressed;
    uint8_t interlacing_shift;
    uint8_t interlacing_leftweight;

    /* short-circuit null buffers */
    if (!inbuffer || !input_buffer_size)
        return input_buffer_size;

    /* initialize from the extradata */
    if (!alac->context_initialized) {
        if (alac->avctx->extradata_size != ALAC_EXTRADATA_SIZE) {
            av_log(avctx, AV_LOG_ERROR, "alac: expected %d extradata bytes\n",
                ALAC_EXTRADATA_SIZE);
            return input_buffer_size;
        }
        if (alac_set_info(alac)) {
            av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
            return input_buffer_size;
        }
        alac->context_initialized = 1;
    }

    init_get_bits(&alac->gb, inbuffer, input_buffer_size * 8);

    channels = get_bits(&alac->gb, 3) + 1;
    if (channels > MAX_CHANNELS) {
        av_log(avctx, AV_LOG_ERROR, "channels > %d not supported\n",
               MAX_CHANNELS);
        return input_buffer_size;
    }

    /* 2^result = something to do with output waiting.
     * perhaps matters if we read > 1 frame in a pass?
     */
    skip_bits(&alac->gb, 4);

    skip_bits(&alac->gb, 12); /* unknown, skip 12 bits */

    /* the output sample size is stored soon */
    hassize = get_bits1(&alac->gb);

    wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */

    /* whether the frame is compressed */
    isnotcompressed = get_bits1(&alac->gb);

    if (hassize) {
        /* now read the number of samples as a 32bit integer */
        outputsamples = get_bits(&alac->gb, 32);
    } else
        outputsamples = alac->setinfo_max_samples_per_frame;

    *outputsize = outputsamples * alac->bytespersample;
    readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1;

    if (!isnotcompressed) {
        /* so it is compressed */
        int16_t predictor_coef_table[channels][32];
        int predictor_coef_num[channels];
        int prediction_type[channels];
        int prediction_quantitization[channels];
        int ricemodifier[channels];
        int i, chan;

        interlacing_shift = get_bits(&alac->gb, 8);
        interlacing_leftweight = get_bits(&alac->gb, 8);

        for (chan = 0; chan < channels; chan++) {
            prediction_type[chan] = get_bits(&alac->gb, 4);
            prediction_quantitization[chan] = get_bits(&alac->gb, 4);

            ricemodifier[chan] = get_bits(&alac->gb, 3);
            predictor_coef_num[chan] = get_bits(&alac->gb, 5);

            /* read the predictor table */
            for (i = 0; i < predictor_coef_num[chan]; i++)
                predictor_coef_table[chan][i] = (int16_t)get_bits(&alac->gb, 16);
        }

        if (wasted_bytes)
            av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n");

        for (chan = 0; chan < channels; chan++) {
            bastardized_rice_decompress(alac,
                                        alac->predicterror_buffer[chan],
                                        outputsamples,
                                        readsamplesize,
                                        alac->setinfo_rice_initialhistory,
                                        alac->setinfo_rice_kmodifier,
                                        ricemodifier[chan] * alac->setinfo_rice_historymult / 4,
                                        (1 << alac->setinfo_rice_kmodifier) - 1);

            if (prediction_type[chan] == 0) {
                /* adaptive fir */
                predictor_decompress_fir_adapt(alac->predicterror_buffer[chan],
                                               alac->outputsamples_buffer[chan],
                                               outputsamples,
                                               readsamplesize,
                                               predictor_coef_table[chan],
                                               predictor_coef_num[chan],
                                               prediction_quantitization[chan]);
            } else {
                av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type[chan]);
                /* I think the only other prediction type (or perhaps this is
                 * just a boolean?) runs adaptive fir twice.. like:
                 * predictor_decompress_fir_adapt(predictor_error, tempout, ...)
                 * predictor_decompress_fir_adapt(predictor_error, outputsamples ...)
                 * little strange..
                 */
            }
        }
    } else {
        /* not compressed, easy case */
        if (alac->setinfo_sample_size <= 16) {
            int i, chan;
            for (chan = 0; chan < channels; chan++)
                for (i = 0; i < outputsamples; i++) {
                    int32_t audiobits;

                    audiobits = get_bits(&alac->gb, alac->setinfo_sample_size);
                    audiobits = extend_sign32(audiobits, readsamplesize);

                    alac->outputsamples_buffer[chan][i] = audiobits;
                }
        } else {
            int i, chan;
            for (chan = 0; chan < channels; chan++)
                for (i = 0; i < outputsamples; i++) {
                    int32_t audiobits;

                    audiobits = get_bits(&alac->gb, 16);
                    /* special case of sign extension..
                     * as we'll be ORing the low 16bits into this */
                    audiobits = audiobits << 16;
                    audiobits = audiobits >> (32 - alac->setinfo_sample_size);
                    audiobits |= get_bits(&alac->gb, alac->setinfo_sample_size - 16);

                    alac->outputsamples_buffer[chan][i] = audiobits;
                }
        }
        /* wasted_bytes = 0; */
        interlacing_shift = 0;
        interlacing_leftweight = 0;
    }

    switch(alac->setinfo_sample_size) {
    case 16:
        if (channels == 2) {
            reconstruct_stereo_16(alac->outputsamples_buffer,
                                  (int16_t*)outbuffer,
                                  alac->numchannels,
                                  outputsamples,
                                  interlacing_shift,
                                  interlacing_leftweight);
        } else {
            int i;
            for (i = 0; i < outputsamples; i++) {
                int16_t sample = alac->outputsamples_buffer[0][i];
                ((int16_t*)outbuffer)[i * alac->numchannels] = sample;
            }
        }
        break;
    case 20:
    case 24:
        // It is not clear if there exist any encoder that creates 24 bit ALAC
        // files. iTunes convert 24 bit raw files to 16 bit before encoding.
    case 32:
        av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size);
        break;
    default:
        break;
    }

    return input_buffer_size;
}

static int alac_decode_init(AVCodecContext * avctx)
{
    ALACContext *alac = avctx->priv_data;
    alac->avctx = avctx;
    alac->context_initialized = 0;

    alac->samplesize = alac->avctx->bits_per_sample;
    alac->numchannels = alac->avctx->channels;
    alac->bytespersample = (alac->samplesize / 8) * alac->numchannels;

    return 0;
}

static int alac_decode_close(AVCodecContext *avctx)
{
    ALACContext *alac = avctx->priv_data;

    int chan;
    for (chan = 0; chan < MAX_CHANNELS; chan++) {
        av_free(alac->predicterror_buffer[chan]);
        av_free(alac->outputsamples_buffer[chan]);
    }

    return 0;
}

AVCodec alac_decoder = {
    "alac",
    CODEC_TYPE_AUDIO,
    CODEC_ID_ALAC,
    sizeof(ALACContext),
    alac_decode_init,
    NULL,
    alac_decode_close,
    alac_decode_frame,
};