Introduce a new parameter "use_alltoallv" that decides whether to use

MPI_Alltoallv or to use MPI_Irecv, MPI_Isend, and MPI_Wait to
communicate.  The latter is not tested, but may be faster if only a
few of all processors are involved in the communication.

Disable some self tests: Allow processors who do not participate in
the communication (who specify a zero slab size) to not know the
geometry of the source and destination slabs.


git-svn-id: http://svn.cactuscode.org/arrangements/CactusNumerical/Slab/trunk@50 2e825fa2-fb71-486d-8b7f-a5ff3f0f6cb8

1 files changed, 4 insertions, 0 deletions

diff --git a/param.ccl b/param.ccl
index 3be94cc..e3d2ae2 100644
--- a/param.ccl
+++ b/param.ccl
@@ -4,3 +4,7 @@
 BOOLEAN timer_output "Print slabbing timings at shutdown time" STEERABLE=always
 {
 } "no"
+
+BOOLEAN use_alltoallv "Use MPI_Alltoallv for communication?" STEERABLE=always
+{
+} "yes"