[FFmpeg-devel] [PATCH] aacdec: buf_mdct data array reduction

[Alex Converse]

On Sat, Feb 19, 2011 at 8:18 PM, Young Han Lee <cpumaker at gmail.com> wrote:
> On Thu, Feb 17, 2011 at 4:44 PM, Alex Converse <alex.converse at gmail.com>wrote:
>
> Hi,
>
>
>> This is getting pretty close to dsp.vector_fmul_window(). Is there a
>> way to further combine things and the the dsp function?
>>
>>
> I think it is different from vector_fmul_window().
> it is slightly different from vector_fmul() or vector_fmul_reverse()
> because it is reversely stored.
> Can I make new one in dsp function?
> If it is allowed, I can make it
> but it seems it is only used here.
>
>
>>
>> Otherwise looks good.
>>
>> If you are looking for other was to reduce LTP memory usage, LTP
>> variables can live in a union with AOT_AAC_MAIN prediction variables.
>>
>> As you told me, ltp_state data array is changed to the pointer using the
> prediction variables.
> Thank you for comment :)
>

I would rather see a union used than completely lose all type
information. This is especially important if one set of prediction
information gets bigger or smaller.

e.g.

union {
PredictorState predictor_state[MAX_PREDICTORS];
DECLARE_ALIGNED(16, int16_t, ltp_state)[3072];
} prediction;

I've gone ahead and pushed the first patch though.

Thanks,
Alex