Speex dev - Mar 2010 - Symbian/S60 encoding failures when building -O2/-O1 optimization and FIXED_POINT

Hello -

We are building Speex 1.2rc1 on Symbian/S60 using the GNU 
csl-arm-2005Q1C tool chain provided in the S60 SDK.  We build with the 
supplied config.h and define FIXED_POINT to 1.

We found that the encoded audio is "garbled" when we build with GCC
-O1
or -O2 optimization level with fixed point enabled.  But if fixed point 
is enabled and no GCC optimization level is specified, the encoded audio 
is correct.  The encoded audio is also correct with fixed point disabled 
and the library built at -O1 or -O2 optimization level.  So it seems 
that the GCC optimization is somehow causing the fixed point encoder 
build to fail.

I'm wondering if anyone else has run into this, and/or if there are 
suggestions to help pinpoint the failure.

Regards,
Brian

Seems like it could be an optimizer bug because many people have 
compiled the fixed-point port on all kinds of platforms without problem. 
If you want to narrow down the problem, you might want to see which file 
causes the corruption when compiled with optimization on. Then you could 
even check per function and per optimization. Or you could just try 
another version of gcc.

	Jean-Marc

On 2010-03-05 19:09, Brian Friedkin wrote:
> Hello -
>
> We are building Speex 1.2rc1 on Symbian/S60 using the GNU
> csl-arm-2005Q1C tool chain provided in the S60 SDK.  We build with the
> supplied config.h and define FIXED_POINT to 1.
>
> We found that the encoded audio is "garbled" when we build with
GCC -O1
> or -O2 optimization level with fixed point enabled.  But if fixed point
> is enabled and no GCC optimization level is specified, the encoded audio
> is correct.  The encoded audio is also correct with fixed point disabled
> and the library built at -O1 or -O2 optimization level.  So it seems
> that the GCC optimization is somehow causing the fixed point encoder
> build to fail.
>
> I'm wondering if anyone else has run into this, and/or if there are
> suggestions to help pinpoint the failure.
>
> Regards,
> Brian
> _______________________________________________
> Speex-dev mailing list
> Speex-dev at xiph.org
> http://lists.xiph.org/mailman/listinfo/speex-dev
>
>