[Pw_forum] Using scalapack in PW/tests causes failures

[Jack Deslippe]

Hi,

I apologize ahead of time if this is a duplicate of a previous report (I
searched for recent matches and didn't find any).

I am trying to build ESPRESSO 5.0 for a couple different parallel machines
but am having trouble verifying the build.

In particular, for builds with scalapack, the PW/tests fail pretty
substantially when using more than 1 processor. Here is the output using
PGI Compilers + Cray LibSci using 4 processors.

hecking atom...passed
Checking atom-lsda...passed
Checking atom-pbe...discrepancy in number of scf iterations detected
Reference: 7, You got: 5
discrepancy in pressure detected
Reference: -14.46, You got: 94.93
Checking atom-sigmapbe...discrepancy in number of scf iterations detected
Reference: 11, You got: 15
discrepancy in pressure detected
Reference: -14.98, You got: 104.92

I get the same result using GNU Compilers + LibSci and using various levels
of optimization include -O0.  I also see the same problem using my own
scalapack build as well as using MKL scalapack on another machine. When
using only one processor, these tests pass as well as with 4 processors and
scalapack disabled.

It is possible I am missing something obvious (like these tests only being
intended to be run in serial), but I want to make sure the code is behaving
as expected.

Sincerely,

Jack Deslippe
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20120522/da829bb7/attachment.html>