[FFmpeg-devel] Fixpoint FFT optimization, with MDCT and IMDCT wrappers for audio optimization

[Diego Biurrun]

On Fri, Jul 27, 2007 at 06:56:14AM -0400, mmh wrote:
> 
> I could really use some help on the integration aspects of this patch.
> The intent of this is to be able to implement optimized audio codecs
> for architectures with out floating point operations.  This is part of
> the cook fixpoint patch but its a little more general.
> 
> --- Makefile	(revision 9807)
> +++ Makefile	(working copy)
> @@ -460,6 +460,6 @@
>  
> -fft-test: fft-test.o $(LIB)
> +fft-test: fft-test.o $(LIB) fft_fixpoint.o

nit: Keep the $(LIB) at the end.

> --- fft-test.c	(revision 9807)
> +++ fft-test.c	(working copy)
> @@ -153,13 +153,47 @@
>  
> +    if (do_fixpoint) {
> +        for(i=0;i<n;i++) {

I think leaving a space after the for is slightly preferred.

>      av_log(NULL, AV_LOG_INFO,"usage: fft-test [-h] [-s] [-i] [-n b]\n"
>             "-h     print this help\n"
>             "-s     speed test\n"
>             "-m     (I)MDCT test\n"
> +           "-f     check fixpoint\n"

fixEDpoint :)

> @@ -171,11 +205,13 @@
>      int do_inverse = 0;
> +    int do_fixpoint = 0;

ditto

> +            ff_fixpoint_mdct_init (m, fft_nbits, do_inverse);

same here

> +            ff_fixpoint_fft_init (s, fft_nbits, do_inverse);

and here

> --- fft_fixpoint.c	(revision 0)
> +++ fft_fixpoint.c	(revision 0)
> @@ -0,0 +1,294 @@
> +
> +/**
> + * @file fft_fixpoint.c

Please rename this file to fft_fixedpoint.c.

> + * This module defines several fixpoint spectrum analysis tools.

And replace all the occurrences of fixpoint with fixedpoint in this
file.

Diego