From 6cacd197c40ad1694ad71045fa0724fd0dc783e2 Mon Sep 17 00:00:00 2001
From: rhaas <rhaas@ff385933-4943-42dc-877b-ffc776028de6>
Date: Tue, 27 Mar 2012 03:16:10 +0000
Subject: reduce nan mask in level mode

The previous local reduction did not quite work if a process has more than one
component of the grid, and deadlocks if processes have different numbers of
components.


git-svn-id: http://svn.cactuscode.org/arrangements/CactusUtils/NaNChecker/trunk@101 ff385933-4943-42dc-877b-ffc776028de6
---
 schedule.ccl               | 24 +++++++++++++++++++-----
 src/NaNCheck.c             | 36 ++++++++++++++++++++++++++++--------
 test/nancount.par          | 18 ++++++++++++++++++
 test/nancount/NaNsFound.xg |  2 ++
 4 files changed, 67 insertions(+), 13 deletions(-)
 create mode 100644 test/nancount.par
 create mode 100644 test/nancount/NaNsFound.xg

diff --git a/schedule.ccl b/schedule.ccl
index 050cb58..97237f7 100644
--- a/schedule.ccl
+++ b/schedule.ccl
@@ -17,22 +17,36 @@ if (*check_vars && check_every > 0)
     OPTIONS: global
   } "Reset the NaNChecker::NaNsFound counter"
 
-  schedule NaNChecker_NaNCheck as zzz_NaNChecker_NaNCheck at POSTSTEP
+  schedule NaNChecker_NaNCheck_Prepare IN NaNChecker_NaNCheck
   {
     LANG:    C
-    OPTIONS: local
+    OPTIONS: level
+  } "Prepare data structures to check for NaNs"
+
+  schedule GROUP NaNChecker_NaNCheck as zzz_NaNChecker_NaNCheck at POSTSTEP
+  {
   } "Check for NaNs and count them in NaNChecker::NaNsFound"
 
+  schedule NaNChecker_NaNCheck_Check IN NaNChecker_NaNCheck AFTER NaNChecker_NaNCheck_Prepare
+  {
+    LANG:    C
+    OPTIONS: local
+  } "Check for NaNs"
+
+  schedule NaNChecker_NaNCheck_Finish IN NaNChecker_NaNCheck AFTER NaNChecker_NaNCheck_Check
+  {
+    LANG:    C
+    OPTIONS: level
+  } "Count NaNs in NaNChecker::NaNsFound"
+
   schedule NaNChecker_TakeAction at POSTSTEP after zzz_NaNChecker_NaNCheck
   {
     LANG:    C
     OPTIONS: global loop-level
   } "Output NaNChecker::NaNmask and take action according to NaNChecker::action_if_found"
 
-  schedule NaNChecker_NaNCheck as zzz_NaNChecker_NaNCheck at POST_RECOVER_VARIABLES
+  schedule GROUP NaNChecker_NaNCheck as zzz_NaNChecker_NaNCheck at POST_RECOVER_VARIABLES
   {
-    LANG:    C
-    OPTIONS: local
   } "Check for NaNs and count them in NaNChecker::NaNsFound"
 
   schedule NaNChecker_TakeAction at POST_RECOVER_VARIABLES after zzz_NaNChecker_NaNCheck
diff --git a/src/NaNCheck.c b/src/NaNCheck.c
index dfa59be..129a2e5 100644
--- a/src/NaNCheck.c
+++ b/src/NaNCheck.c
@@ -127,12 +127,12 @@ void NaNChecker_ResetCounter (CCTK_ARGUMENTS)
    @vio        in
    @endvar
 @@*/
-void NaNChecker_NaNCheck (CCTK_ARGUMENTS)
+static t_nanchecker_info info;
+void NaNChecker_NaNCheck_Prepare (CCTK_ARGUMENTS)
 {
   DECLARE_CCTK_ARGUMENTS;
   DECLARE_CCTK_PARAMETERS;
   int i, nelems;
-  t_nanchecker_info info;
 
 
   if (cctk_iteration < check_after || cctk_iteration % check_every)
@@ -179,17 +179,37 @@ void NaNChecker_NaNCheck (CCTK_ARGUMENTS)
     info.check_for = CHECK_FOR_BOTH;
   }
 #endif
+}
+void NaNChecker_NaNCheck_Check (CCTK_ARGUMENTS)
+{
+  DECLARE_CCTK_ARGUMENTS;
+  DECLARE_CCTK_PARAMETERS;
+
+  if (cctk_iteration < check_after || cctk_iteration % check_every || 
+      (info.NaNmask && cctk_iteration == last_iteration_output))
+  {
+    return;
+  }
 
   CCTK_TraverseString (check_vars, CheckForNaN, &info, CCTK_GROUP_OR_VAR);
+}
+void NaNChecker_NaNCheck_Finish (CCTK_ARGUMENTS)
+{
+  DECLARE_CCTK_ARGUMENTS;
+  DECLARE_CCTK_PARAMETERS;
 
-  /* reduce the NaN counter globally and accumulate in NaNChecker::NaNsFound */
+  if (cctk_iteration < check_after || cctk_iteration % check_every || 
+      (info.NaNmask && cctk_iteration == last_iteration_output))
   {
-    int sum_handle = CCTK_ReductionHandle ("sum");
-    CCTK_INT count = 0;
-    CCTK_ReduceLocalScalar (cctkGH, -1, sum_handle, &info.count, &count,
-                            CCTK_VARIABLE_INT);
-    *NaNsFound += count;
+    return;
   }
+
+  /* reduce the NaN counter globally and accumulate in NaNChecker::NaNsFound */
+  int sum_handle = CCTK_ReductionHandle ("sum");
+  CCTK_INT count = 0;
+  CCTK_ReduceLocalScalar (cctkGH, -1, sum_handle, &info.count, &count,
+                          CCTK_VARIABLE_INT);
+  *NaNsFound += count;
 }
 
 
diff --git a/test/nancount.par b/test/nancount.par
new file mode 100644
index 0000000..b3a54e8
--- /dev/null
+++ b/test/nancount.par
@@ -0,0 +1,18 @@
+ActiveThorns = "boundary idscalarwavec wavetoyc time CartGrid3D CoordBase SymBase ioutil iobasic localreduce PUGH PUGHReduce PUGHSlab nanchecker"
+ 
+idscalarwave::radius = inf
+idscalarwave::sigma = inf
+
+cactus::cctk_itlast = 0
+driver::ghost_size = 0
+
+IO::out_fileinfo  = "none"
+IO::parfile_write = "no"
+IO::out_dir = $parfile
+
+NaNChecker::check_vars = "all"
+NaNChecker::check_every = 1
+
+IOBasic::outScalar_vars = "NaNChecker::NaNsFound"
+IOBasic::outScalar_reductions = "maximum"
+IOBasic::outScalar_every = 1
diff --git a/test/nancount/NaNsFound.xg b/test/nancount/NaNsFound.xg
new file mode 100644
index 0000000..305ce1c
--- /dev/null
+++ b/test/nancount/NaNsFound.xg
@@ -0,0 +1,2 @@
+"NaNsFound v time
+0.0000000000000	1000
-- 
cgit v1.2.3