[FFmpeg-devel] [PATCH 1/2] mpegaudiodec: move imdct36() to MPADSPContext

Vitor Sessak vitor1001 at gmail.com
Sun Aug 21 21:55:03 CEST 2011


On Sun, Aug 21, 2011 at 9:19 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
> Hi Vitor
>
> On Sun, Aug 21, 2011 at 04:51:48PM +0200, Vitor Sessak wrote:
>> $subj, should allow SIMD optimizations.
> [...]
>>          /* NOTE: we do frequency inversion adter the MDCT by changing
>>             the sign of the right window coefs */
>>          for(j=0;j<4;j++) {
>> -            for(i=0;i<36;i+=2) {
>
>> +            mdct_win[j    ][18] = mdct_win[j    ][19] = 0;
>> +            mdct_win[j + 4][18] = mdct_win[j + 4][19] = 0;
>
> these seem unneeded, or maybe i miss something ?

Indeed, I just forgot static buffer are already initialized with zeros.

> [...]
>> @@ -1403,15 +1294,15 @@ static void compute_imdct(MPADecodeContext *s,
>>          else
>>              win1 = mdct_win[g->block_type];
>>          /* select frequency inversion */
>> -        win = win1 + ((4 * 36) & -(j & 1));
>> -        imdct36(out_ptr, buf, ptr, win);
>> +        win = win1 + ((4 * 40) & -(j & 1));
>> +     s->mpadsp.RENAME(imdct36)(out_ptr, buf, ptr, win);
>
> tabs
>
>
> [...]
>> +/* using Lee like decomposition followed by hand coded 9 points DCT */
>> +void RENAME(ff_imdct36)(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in,
>
> trailing whitespace

Fixed both locally (and together with it my .emacs).

-Vitor


More information about the ffmpeg-devel mailing list