[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Jakub's Recommendations for ia32 Support



Bill Nottingham wrote:
> Warren Togami (wtogami redhat com) said: 
>> This is a silly notion.   Atom requires code to be built entirely  
>> differently from all modern processors in order to get maximum  
>> performance.  Code must be optimized for in-line performance.  Atom  
>> target support hasn't even hit gcc-4.4 yet.
> 
> Interestingly, using the parts of OpenBench that can actually be
> cajoled to build and run correctly (why are benchmark suites such
> pain?)... the best option for atom is -march=i686 -mtune=generic.
> Tuning for i586 (the previous in-order processor) doesn't actually
> help.

-march=i686 is better than i586, but is it the best.
I though -march=prescott was optimal for atom?

> 
> Also, -march=i686 was a win over -march=i586, in general, in testing
> across Atom, Core2Duo, and a Athlon64, although not a particularly
> significant one.


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]