Package: release.debian.org
Severity: normal
User: release.debian....@packages.debian.org
Usertags: unblock
X-Debbugs-Cc: nwc...@packages.debian.org
Control: affects -1 + src:nwchem
Please unblock package nwchem
[ Reason ]
nwchem-mpich had a problem running over multiple nodes with mpich,
discussed at https://github.com/nwchemgit/nwchem/issues/633
It was a problem in mpich which got fixed in mpich 4.0.3, but
currently we have mpich 4.0.2. The workaround for nwchem is to set
the environment variable ARMCI_USE_WIN_ALLOCATE=0
The need for this workaround is documented in a README.Debian, added
in nwchem/7.0.2-4
[ Impact ]
Without this patch, users may be unaware of the environment variable
setting required to run nwchem over multiple nodes (with mpich)
[ Tests ]
This is a documentation update, no source changes, debci tests
continue to pass (or fail on some arches as before, no regression)
[ Risks ]
Documentation update only. No source change. Negligible rish.
[ Checklist ]
[x ] all changes are documented in the d/changelog
[x ] I reviewed all changes and I approve them
[x ] attach debdiff against the package in testing
unblock nwchem/7.0.2-4
diff -Nru nwchem-7.0.2/debian/changelog nwchem-7.0.2/debian/changelog
--- nwchem-7.0.2/debian/changelog 2022-03-10 17:20:23.000000000 +0100
+++ nwchem-7.0.2/debian/changelog 2023-03-19 15:01:42.000000000 +0100
@@ -1,3 +1,13 @@
+nwchem (7.0.2-4) unstable; urgency=medium
+
+ * Team upload.
+ * create nwchem-mpich.README.Debian to document the need to use
+ ARMCI_USE_WIN_ALLOCATE=0 when running nwchem with MPICH 4.0.2
+ (binary nwchem.mpich, fixed in mpich 4.0.3).
+ See upstream Issue#633.
+
+ -- Drew Parsons <dpars...@debian.org> Sun, 19 Mar 2023 15:01:42 +0100
+
nwchem (7.0.2-3) unstable; urgency=medium
* Team upload.
diff -Nru nwchem-7.0.2/debian/nwchem-mpich.README.Debian
nwchem-7.0.2/debian/nwchem-mpich.README.Debian
--- nwchem-7.0.2/debian/nwchem-mpich.README.Debian 1970-01-01
01:00:00.000000000 +0100
+++ nwchem-7.0.2/debian/nwchem-mpich.README.Debian 2023-03-19
15:01:42.000000000 +0100
@@ -0,0 +1,38 @@
+Running NWChem with MPICH
+-------------------------
+
+tldr:
+ set ARMCI_USE_WIN_ALLOCATE=0 when running nwchem.mpich
+
+
+When nwchem.mpich is run over multiple processes, it may give an error
+e.g. on 2 processes
+
+ iter energy gnorm gmax time
+ ----- ------------------- --------- --------- --------
+ 1 -75.9473154351 8.06D-01 3.50D-01 0.2
+ ga_iter_lsolve: convergence stagnant ... aborting solve
+ Increased level shift to 2.00
+ ga_iter_lsolve: convergence stagnant ... aborting solve
+
+
+or on 3 processes
+
+ Symmetry analysis of molecular orbitals - initial
+ -------------------------------------------------
+
+ sym_movecs_adapt: orbital 10 negative proj.
+ 1.00D+00 -2.08D-05 -1.04D-05 -5.55D-17
+ ------------------------------------------------------------------------
+ sym_movecs_adapt: negative proj 0
+ ...
+ For further details see manual section: No section for this category
+[0] ARMCI Error: 0:sym_movecs_adapt: negative proj:
+Abort(-1) on node 0 (rank 0 in comm 496): application called
MPI_Abort(comm=0x84000002, -1) - process 0
+
+
+When this happens, the fix is to set the environment variable
+
+ ARMCI_USE_WIN_ALLOCATE=0
+
+For more discussion, see https://github.com/nwchemgit/nwchem/issues/633
