do not attempt reductions in local mode unless auto_res[n] is setHEADsvnmaster

currently SphericalSurface_SetupRes will attempt a reduction in local mode even
if all auto_res values are false, ie if there are no spheres whose resolution
is to be set automatically. This causes MPI hangs if different processors own
different numbers of components, as produced by the recursive load balancer and
carpet::same_number_of_components_on_each_process = false (true by default but
might change).

This patch ameliorates the situation by only attempting a reduction if
auto_res[n] is set, thus at least allowing SphericalSurface to be used when all
ntheta, nphi are giving explicitly. It also adds a comment warning about
auto_res.



git-svn-id: http://svn.cactuscode.org/arrangements/CactusNumerical/SphericalSurface/trunk@51 40f6ab95-0e4f-0410-8daa-ee8d7420be1d

2 files changed, 5 insertions, 0 deletions

diff --git a/param.ccl b/param.ccl
index 4217adf..231cb0b 100644
--- a/param.ccl
+++ b/param.ccl
@@ -19,6 +19,8 @@ CCTK_STRING name[42] "User friendly name of spherical surface" STEERABLE=always
 } ""
 
 
+# Warning: this will MPI hang when the number of components per refinement
+# level differs between processors
 BOOLEAN auto_res[42] "Automatically determine resolution according to radius and Cartesian resolution"
 {
 } no
diff --git a/src/setup.cc b/src/setup.cc
index db6a92f..76bf945 100644
--- a/src/setup.cc
+++ b/src/setup.cc
@@ -336,6 +336,9 @@ extern "C" void SphericalSurface_SetupRes (CCTK_ARGUMENTS)
     // can decide which reflevel to use
      
     // set resolution according to Radius and Cartesian resolution
+
+    if (!auto_res[n])
+        continue;
       
     CCTK_REAL my_radius;