Sun Studio 12 comparison?

[jdrodrig]

I just downloaded the SunStudio 12 for Linux today, and I was wondering if any of you had performed comparative testings of Sun's f95 vs. Intel Fortran 9.x o 10.

[jdrodrig]

As a reference, I performed some test on some simple code dealing with numerical optimization taking as initial condition values on a grid that I split using OpenMP ...

Using a 4xOpteron275..
ifort xyz.f90 -O3 -xW -axW -ip -openmp
performs close to 33% faster than the code generated with the compiler on SunStudio 12
f95 xyz.f90 -fast -openmp -m32 -xipo=2

[Ron_Green]

Interesting. Can you share with us the version of ifort you used? I would be curious: in 10.0 we added -xO option to generate SSE3 for Opteron. Not sure if that would make a difference or not. I can't imagine it hurting performance.

ron

[TimP]

The relative performance quoted above is consistent with expectation. Presumably, -xO would be an advantage only where there is vectorizable complex arithmetic and the CPU supports SSE3.

[jdrodrig]

I am still using the ifort 9.1.043, I will try to upgrade later this week.

To be fair, concerning the role of SSE3, I am relatively new to Sun's f95, I just read about the flag -xvector=simd. I will try later today, but I would be surprised if the vectorization was not included in the -xarch=sse3a that is part of the -fast macro.

[zzli]

Try f95 xyz.f90 -fast -xtarget=amd64 -xipo=2 instead.