Project

General

Profile

Bug #1360

AMPI megampi test fails on mpi-crayxc and darwin builds

Added by Sam White over 2 years ago. Updated about 2 years ago.

Status:
Merged
Priority:
Normal
Category:
AMPI
Target version:
Start date:
01/15/2017
Due date:
% Done:

0%


Description

AMPI's megampi test passes on +p1 +vp1 and +p +vp2 but fails on +p2 +vp2. The same test passes all runs on Blue Waters.

History

#1 Updated by Sam White over 2 years ago

For some reason I can't get an interactive job on Edison now, and I get the same exact error message when I try to submit a job via sbatch:

$ salloc -N 1 -p debug -t 00:30:00 -L SCRATCH

salloc: error: Job submit/allocate failed: Unspecified error

#2 Updated by Sam White over 2 years ago

Also seeing a failure on netlrts-darwin-x86_64 on Wit in megampi with +p1 +vp1. Doesn't reproduce on my Mac, but is consistent in autobuild.

Edit: The darwin bug doesn't reproduce on Wit either, except in Autobuild....? Not sure what I'm doing different from autobuild...
http://charm.cs.uiuc.edu/autobuild/cur/

#3 Updated by Sam White over 2 years ago

  • Subject changed from AMPI megampi test fails on mpi-crayxc to AMPI megampi test fails on mpi-crayxc and darwin builds

#4 Updated by Sam White over 2 years ago

  • Assignee changed from Sam White to Matthias Diener

#5 Updated by Matthias Diener over 2 years ago

megampi fails with OPTS="-optimize" (even on my macbook), compiling without optimization works fine.

#6 Updated by Matthias Diener over 2 years ago

Seems to be a clang stack alignment issue. Adding explicit alignment

int rank __attribute__ ((aligned (16)));
to line 10 in test.C is a workaround.

#7 Updated by Sam White over 2 years ago

  • Status changed from New to In Progress

#8 Updated by Sam White over 2 years ago

  • Status changed from In Progress to Merged
  • translation missing: en.field_closed_date set to 2017-02-14 21:28:53.052235

The real issue was that we were returning the address of a variable on the stack when getting builtin keyvals like AMPI_MY_PE. The above patch was amended to fix the real issue.

Also available in: Atom PDF