[Pw_forum] Broken PP routines in espresso-3.2.2?

[Tyler Grassman]

Axel,

Thank you very much for your helpful response.  I went ahead and recompiled
using ifort 9.1.040 (FFLAGS = -O3 -xW -static) and acml 3.6 (ifort
compiled), and it worked.  It's a tiny bit slower than the PGI code, (2.5%
on ni.scf.in from example08), so that's good.  Thanks again!

Regards,
Tyler


> -----Original Message-----
> From: Axel Kohlmeyer [mailto:akohlmey at cmm.chem.upenn.edu] 
> Sent: Wednesday, August 15, 2007 11:29 AM
> To: Tyler Grassman
> Cc: pw_forum at pwscf.org
> Subject: Re: [Pw_forum] Broken PP routines in espresso-3.2.2?
> 
> 
> On Wed, 15 Aug 2007, Tyler Grassman wrote:
> 
> 
> hello tyler,
> 
> 
> TG> My system and compilation setup is as follows...
> TG>   
> TG> OS/Architecture = CentOS 4.3 (AMD64) on dual Opteron 2214 
> TG> (dual-core) F90 = pgf90 (v.6.2-5) CC = gcc
> TG> LAPACK/BLAS = acml (v.3.6)
> TG> FFTW = internal
> TG> 
> TG> I can give my optimizations and such if anyone wants 
> them, but I've 
> TG> also tried turning off all optimizations (i.e. setting 
> both CFLAGS 
> TG> and FFLAGS to -O0), and I get the same exact crashing, so I'm 
> TG> guessing that's not the issue.  I've also tried compiling 
> with pgf90 
> TG> v.5.2-4, but pw.x itself segfaults, and I didn't think it 
> was worth 
> TG> trying to figure that one out...
> 
> correct. see below.
> 
> TG> But I'm not so sure that it's necessarily a compiler issue, since 
> TG> I've had plenty of suggess with pgf90 6.2-5 with pw.x, as well as 
> TG> other codes (such as VASP and CPMD).
> 
> the pgi f90 compilers all have a number of (serious) issues with 
> current standard complying fortran 90/95 code. btw, there are 
> also some known issues with CPMD but they are mostly in harmless or 
> rarely used places (e.g. the wavefunction guess, ultra-soft support).
> 
> if you have access to the intel compilers (make sure you have 
> updated versions there as well), it should work ok. or you 
> may want to try g95 ( from www.g95.org, it is straightforward 
> to try it out from 
> the precompiled snapshots and to get rid of it, too).
> 
> TG> So, has anyone else run into these problems?  Or does anyone have 
> TG> any ideas that I can do to try to remedy the situation?
> 
> the i/o code based on iotk is a bit sensitive to compiler bugs, 
> so switching compilers would be the best strategy to get a 
> hint where the problem is coming from. sometimes code can be 
> replaced with an equivalent less demanding implementation, 
> sometimes there are features used that only work with certain 
> compilers.
> 
> cheers,
>    axel.
> 
> TG> 
> TG> Thank you,
> TG> 
> TG> Tyler Grassman
> TG> Depts. of Materials Science, Chemistry/Biochemistry University of 
> TG> California, San Diego 9500 Gilman Dr. #0358
> TG> La Jolla, CA 92093-0358
> TG> 
> TG> _______________________________________________
> TG> Pw_forum mailing list
> TG> Pw_forum at pwscf.org 
> TG> http://www.democritos.it/mailman/listinfo/pw_forum
> TG> 
> 
> -- 
> ==============================================================
> =========
> Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   
> http://www.cmm.upenn.edu
>    Center for Molecular Modeling   --   University of Pennsylvania
> Department of Chemistry, 231 S.34th Street, Philadelphia, PA 
> 19104-6323
> tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 
> 1-215-898-5425 
> ==============================================================
> =========
> If you make something idiot-proof, the universe creates a 
> better idiot.
> 
>

Tyler Grassman
Depts. of Materials Science, Chemistry/Biochemistry University of 
California, San Diego 9500 Gilman Dr. #0358
La Jolla, CA 92093-0358