I also forgot to mention, for both of these we'd need some way to test for
PPC type builds...
> I've been wanting to use -mfpmath=sse as it be beneficial (I've
> benchmarked it myself) however, it would limit us to the PIII ('99) and
> the Athlon 4 (2001). While for the Athlon that was 6 years ago, I'm not
> convinced we don't have a significant userbase still running us on
> machines pre-sse. (coincidentally this was one of the unanswered
> questions that inspired Case Metrics... so we should have a better idea
> after this release).
>
> I can see your point with the -march option, however my only concern is
> with 64-bit builds. I don't currently know of a way to test for 64/32-bit
> in Scons and specifying -march=pentium would mean we always compile 32-bit
> (Being a 64-bit user I don't much like the idea of that :)).
>
> So I guess I'd summarize my opinion as:
> -mfpmath=sse: Not yet
> -march=pentium: Not until we find a way to make sure this doesn't affect
> 64-bit builds
>
> -John
>
>> Albert Santoni wrote:
>>> Ok, couple of problems:
>>>
>>> 1) We already have -O3 in two other places on unix (always turned on).
>>> We really do want these turned on by default on unix, so I've removed
>>> the -O3 from the other spots and made your optimizations block enabled
>>> by default on unix.
>>
>> It is worthwhile to specify whether the CPU is a pentium or not:
>> i.e.: -march=pentium
>> since this enables more efficient float <=> int conversion. The
>> performance penalty of the stock GCC conversion is rather large.
>>
>> if you can figure out during configuration whether the architecture is
>> i686 and supports mmx/sse, the following will gain you even more float
>> performance:
>> -mmmx -msse -mfpmath=sse
>>
>> Some more general GCC optimization flags:
>> -fomit-frame-pointer -ffast-math -funroll-loops
>>
>> Especially the inclusion of floating point math optimization will give
>> you a lot of improvement. And who uses mixxx on a 486 anyway?
>>
>> Greets,
>>
>> Pieter
>>
>
>
>
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2005.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Mixxx-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mixxx-devel
