[FFmpeg-devel] [PATCH] Multi-Channel Correlation in ALS

[Thilo Borgmann]

Am 04.01.10 12:48, schrieb Michael Niedermayer:
> On Mon, Dec 21, 2009 at 01:24:07PM +0100, Thilo Borgmann wrote:
> [...]
>> +/** Recursively reverts the inter-channel correlation for a block.
>> + */
>> +static int revert_channel_correlation(ALSDecContext *ctx, ALSBlockData *bd,
>> +                                       ALSChannelData **cd, int *reverted,
>> +                                       unsigned int offset, int c)
>> +{
>> +    ALSChannelData *ch = cd[c];
>> +    unsigned int   dep = 0;
>> +    unsigned int channels = ctx->avctx->channels;
>> +
> 
>> +    if (reverted[c])
>> +        return 0;
>> +    else
> 
> {}
> 
> 
>> +        reverted[c] = 1;
>> +
>> +    while (dep < channels && !ch[dep].stop_flag) {
>> +        if (!reverted[ch[dep].master_channel])
>> +            revert_channel_correlation(ctx, bd, cd, reverted, offset,
>> +                                       ch[dep].master_channel);
>> +
>> +        dep++;
>> +    }
>> +
>> +    if (dep == channels) {
>> +        av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel correlation!\n");
>> +        return -1;
>> +    }
>> +
>> +    bd->use_ltp     = ctx->use_ltp + c;
>> +    bd->ltp_lag     = ctx->ltp_lag + c;
>> +    bd->ltp_gain    = ctx->ltp_gain[c];
>> +    bd->lpc_cof     = ctx->lpc_cof[c];
>> +    bd->quant_cof   = ctx->quant_cof[c];
>> +    bd->raw_samples = ctx->raw_samples[c] + offset;
>> +
>> +    dep = 0;
>> +    while (!ch[dep].stop_flag) {
>> +        unsigned int smp;
>> +        unsigned int begin = 1;
>> +        unsigned int end   = bd->block_length - 1;
>> +        int64_t y;
>> +        int32_t *master = ctx->raw_samples[ch[dep].master_channel] + offset;
>> +
>> +        if (ch[dep].time_diff_flag) {
>> +            int t = ch[dep].time_diff_index;
>> +
>> +            if (ch[dep].time_diff_sign)
>> +                begin -= t;
>> +            else
> 
> {}

I just removed the else.

> 
> 
> [...]
>> @@ -%ld,%ld +%ld,%ld @@
>>                      sizeof(*ctx->raw_samples[c]) * sconf->max_order);
>>          }
>>      } else { // multi-channel coding
>> +        ALSBlockData   bd;
>> +        int            b;
> 
>> +        int            reverted_channels[avctx->channels];
> 
> i think malloc() would be safer or check that channel is within a
> reasonabe range. Or make double sure the common code checks it
> 
> iam ok with the rest of the patch i think

Oh yes, and channels can become quite large, so moved into the context
and malloc'ed.

Revision 3 attached.

-Thilo
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: als_mcc.rev3.patch
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20100104/25e9c9d6/attachment.txt>