Compiling GAMESS

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

I have been compiling the latest GAMESS with pgf77 7.0.4 and the include acml libraries(Both 64 an 32 bits). It compiles OK but some tests fail and give incorrect numerical results. The support pages on this site refer to version 6 of the compiler and Nov 2004 of GAMESS. I wonder whether somebody has some more recent advices about this compilation.

Thanks

2

Hi fjm,

We typically will update our community application test source for community before a major release. When testing GAMESS with 7.0, we used the April 2006 version since this was the latest at the time. We found no problem with this version. I’m in the process of obtaining the current version (May 2007) and will re-test.

Which tests are failing and how do they fail? Are you compiling in 32 or 64-bits? What steps did you perform to configure GAMESS?

Thanks,
Mat

3

I am testing the current version downloaded a week ago. It is ver. 24 MAR 2007 (R1). I am trying to compile both for 32 and 64( but just for 64 at the moment). I have used the default parameters for TARGET=Linux64 and exam files 06 25 36 and 38 have failed.
Switching from -lacml to -lblas for $LIBRARIES in lked seems to improve the things as just 36 and 38 fail.
As for exam 36 is concerned I think it is just a problem with the system’s shmmax parameter. I have to change it and chek. For the error with exam38 I have no clue.
As far as the 32 bit cross-compilation is concerned I have found that many exams don’t finish properly but I have still to check the reason.

Thanks,
F.J. Modrego

4

Thanks F.J. Exam36 is expected since it requires a different system configuration. I’ll look into exam38 and 32-bits once I obtain the source.

  • Mat
5

I have just modified the shmmax parameter and, as expected, exam36 does not fail. On the other hand I haver recompiled a 32 bit version using -lblas instead of -lacml and all 42 tests finish and are numerically correct.
Only the exam38 in the 64 bit version keeps giving incorrect numerical results.

Cheers,
F.J. Modreog

6

Hi F.J.

I was able to reproduce the wrong answers with exam38 and I have begun investigating the cause. Unfortunately, my time is currently limited but I’ll do my best to make some time for this.

Thanks,
Mat