[MITgcm-devel] changing -Ofast in linux_amd64_open64

[Oliver Jahn]

Hi Jean-Michel,

I have no idea.  I'll run a test to find out.

Cheers,
Oliver

On 2012-04-01 17:46, Jean-Michel Campin wrote:
> Hi Oliver,
>
> I am going to change linux_amd64_open64 optfile, to remove the "-ipa" option
> (replacing "-Ofast" with "-O3 -OPT:Ofast -ffast-math"):
> On acresgrid (login), when I use "-ipa", the little changes I made last Wed.
> in eesupp/src/print.F are causing some differences for advect_xz&  the 5
> global_ocean.cs32x15 tests.
> see in: http://mitgcm.org/testing/results/2012_03/ when comparing:
>       tr_acesgrid-op64_20120330_0/summary.txt
> with tr_acesgrid-op64_20120328_0/summary.txt
>
> It seems strange to me and annoying regarding the robustness of the results.
> Also without "-ipa", I don't get a compiler internal error in
> tutorial_global_oce_lat_lon.
>
> Do you have an idea how the "-ipa" option impact the performance of your runs ?
>
> Cheers,
> Jean-Michel
>
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel