"Woodcrest" Xeon 3.0ghz optimal -march and options???

"Woodcrest" Xeon 3.0ghz optimal -march and options???

[Conor Robinson]

Hi,

I'm running a mac pro with darwin 8.8.1 osx 10.4.8.  3.0ghz Woodcrest
dual core xeon
chip with 1.33ghz bus.

I have gcc 4.0.3 from hpc and 4.1.1 from apple.

What I would like to know is, what is my -(m)arch and besides -03, how
can I invoke auto loop vectorization -sse3 etc, or can I since I don't
see my arch.  I have intels specific gnu friendly compilers, however
they are not that friendly and would like to stick with gcc.

Any comments or ideas would be greatly appreciated.

Thank you,
Conor

Re: "Woodcrest" Xeon 3.0ghz optimal -march and options???

[Tim Prince]

Conor Robinson wrote:
> Hi,
> 
> I'm running a mac pro with darwin 8.8.1 osx 10.4.8.  3.0ghz Woodcrest
> dual core xeon
> chip with 1.33ghz bus.
> 
> I have gcc 4.0.3 from hpc and 4.1.1 from apple.
> 
> What I would like to know is, what is my -(m)arch and besides -03, how
> can I invoke auto loop vectorization -sse3 etc, or can I since I don't
> see my arch.  I have intels specific gnu friendly compilers, however
> they are not that friendly and would like to stick with gcc.
> 

If you are interested in vectorization, you should find the Intel 
compilers more friendly than the gnu compilers you mention.
  -free-vectorize is the option for gnu auto-vectorization.  It works 
well in gfortran 4.1.1 for simple cases.
For gcc, if you don't want to upgrade to a version with better default 
optimization for Intel, -march=nocona (64-bit) or prescott (32-bit) 
should work well most of the time.  The 64-bit gnu compilers default to 
-mfpmath=sse; specifying that option may help if you use 32-bit compilers.