[FFmpeg-devel] [PATCH] Optimization of AC3 floating point decoder for MIPS

[Babic, Nedeljko]

You are correct. This is not needed here and I'll remove it for next patch.

However there is another problem now.

Last couple of days fate tests are failing on fft tests (and some other tests) although
I didn't made any functional changes in code.
I tried to checkout to an older version, but they are failing also.
Is there a way to revert complete fate test system (including samples) so I can check
what is going on?

Thanks,
Nedeljko
________________________________________
From: Vitor Sessak [vitor1001 at gmail.com]
Sent: Monday, August 27, 2012 18:28
To: FFmpeg development discussions and patches
Cc: Babic, Nedeljko; Lukac, Zeljko
Subject: Re: [FFmpeg-devel] [PATCH] Optimization of AC3 floating point decoder for MIPS

On 08/23/2012 04:35 PM, Nedeljko Babic wrote:
> FFT in MIPS implementation is working iteratively instead
>   of "recursively" calling functions for smaller FFT sizes.
> Some of DSP and format convert utils functions are also optimized.

One last comment:

>   doc/mips.txt                      |    1 +
>   libavcodec/dsputil.c              |    1 +
>   libavcodec/dsputil.h              |    1 +
>   libavcodec/fft.c                  |    2 +
>   libavcodec/fft.h                  |    1 +
>   libavcodec/fmtconvert.c           |    1 +
>   libavcodec/fmtconvert.h           |    1 +
>   libavcodec/mips/Makefile          |    4 +
>   libavcodec/mips/dsputil_mips.c    |  164 ++++++++++++
>   libavcodec/mips/fft_init_table.c  |   67 +++++
>   libavcodec/mips/fft_mips.c        |  529 +++++++++++++++++++++++++++++++++++++
>   libavcodec/mips/fft_table.h       |   63 +++++
>   libavcodec/mips/fmtconvert_mips.c |  334 +++++++++++++++++++++++
>   13 files changed, 1169 insertions(+), 0 deletions(-)
>   create mode 100644 libavcodec/mips/dsputil_mips.c
>   create mode 100644 libavcodec/mips/fft_init_table.c
>   create mode 100644 libavcodec/mips/fft_mips.c
>   create mode 100644 libavcodec/mips/fft_table.h
>   create mode 100644 libavcodec/mips/fmtconvert_mips.c
>

> diff --git a/libavcodec/fft.c b/libavcodec/fft.c
> index 6b93a5c..39c8972 100644
> --- a/libavcodec/fft.c
> +++ b/libavcodec/fft.c
> @@ -31,6 +31,7 @@
>   #include "libavutil/mathematics.h"
>   #include "fft.h"
>   #include "fft-internal.h"
> +#include "mips/fft_table.h"

This include should not be needed (and probably isn't). MIPS-specific
code in common file is bad.

Otherwise, patch OK.

-Vitor