Re: BUG and solution for ALTIX 4000 series(particularly at LRZ). On intel compiler for ALTIX causes wrong results with PM module in P-Gadget2.

From: Arman Khalatyan <arm2arm_at_googlemail.com>
Date: Thu, 22 Feb 2007 12:41:37 +0100

Yes nasa was replacing old CPUs to Newer what we have at LRZ, and partially
Old cpus where celled to UAM(Madrid).(Gustavo Yepes was helping me to test
3700 cpus at UAM)
I was not sure is that trouble on Columbia or no, I think this complex
brunching optimization in Intel causes trouble. Even -O1 gives wrong code.

Also one needs to keep in mind on Itanium-2 gcc or other compilers can
produce unpredictable results.

The conversions between types on Altix always painful.

> The code always crashed in the initialization of pm_nonperiodic.c. We
> traced it to a call of the erfc() function. For an argument larger than
> about ~30.0, this function would not return zero as it should (with
> underflow flagged), but instead it would simply crash the application.
> Replacing erfc() with a call of the GSL math library, i.e.
> gsl_sf_erfc(), solved this.
>
Always for gadget better to use -ansi.
I was testing GSL at lrz as well and simply compiled with -ansi -O2 -mp1
options. It warks for me :)

Arman.
Received on 2007-02-22 12:41:38