[Users] Meeting Minutes for 2014-10-27

[Roland Haas]

Present: Frank, Peter, Steve, Matt, Roland, Ian, Erik, Yosef

* intel compiler miscompiles NewRad
** Peter tried intel 15
** Ian or Roland to try with intel 15 on datura
** Steve found and fixed some issues in compiler test

* svn checkouts out of private repositories requiring login: ask Bruno
for patch, we initially introduced non-interactive flag to cope with svn
asking permission to accept new ssl certificates and seeming to stall
since the user did not see the prompt

* OpenBLAS as replacement for LAPACK and BLAS
** desire to build OpenBLAS rather than use a system installed reference
LAPACK which is very slow
** difficult to detect exactly what the system installed LAPACK library
is since some optimized versions keep parts of the reference library intact
** issue only arises on user workstations/laptops and possibly private
clusters where simfactory does not explicitly provide a "good" LAPACK
library installation
** balance between compilation speed and execution speed (and reputation
of Cactus being slow when using PETSc due to slow LAPACK)
** decision: come up with way to auto-detect in the next 6 months then
decide what to do
** decision: keep OpenBLAS in thornlist but keep it commented out for now

Upcoming release:
* change group separator in output file names from "::" to "-" after the
release
* example to solve Poisson equation using CactusPetsc
** works with unigrid only (PUGH and Carpet), no mesh refinement, no
multi-patch
** currently has no test case but may not need one given that it itself
is an example
** needs documentation
** in test there are NaN values on the outer edges of the residual grid
function that should be set to something harmless instead
* testing:
** Erik ran tests using simfactory's distribute script
** no major issues so far
** send out call to mailing list for user to send in option lists for
cluster they use but that are not in simfactory, in particular European
ones (Ian and Roland have access to the RZG machine in Munich but they
are Max-Planck machines so not public, have yet to hand in a PRACE
allocation request for the XSEDE equivalent)

Yours,
Roland





-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: OpenPGP digital signature
Url : http://lists.einsteintoolkit.org/pipermail/users/attachments/20141029/75786cc9/attachment.bin