Merging

--HG-- branch : week-of-code
clairekope · Jul 16, 2009 · 0c9d7e1 · 0c9d7e1
2 parents 23209ff + 1405dce
commit 0c9d7e1
Show file tree

Hide file tree

Showing 12 changed files with 349 additions and 156 deletions.
diff --git a/input/restart.sh b/input/restart.sh
@@ -0,0 +1,69 @@
+#!/bin/sh
+
+# Initial parameter file.  Will change after restarts
+AMRFILE=nBody.RunParamFile
+
+# Set to 1 if the initial run is a restart
+INIT_RESTART=0
+
+# Redirection of output (doesn't work ...)
+OUTPUT=""
+#OUTPUT=">& estd.out"
+
+# MPI call
+MPI=mpirun
+
+# Number of processors
+NPROCS=1
+
+# Enzo executable
+EXE=/Users/jwise/codes/EnzoWOC/week-of-code/src/enzo/enzo.exe
+
+# These shouldn't be changed as they're (for now) hard-coded into enzo
+RESTART_FILE=RestartParamFile
+FINISH_FILE=RunFinished
+
+########################################################################
+########################################################################
+
+# Remove restarting and stopping flag files
+if [ -e $RESTART_FILE ]; then
+    rm -f $RESTART_FILE
+fi
+if [ -e $FINISH_FILE ]; then
+    rm -f $FINISH_FILE
+fi
+
+if [ $INIT_RESTART -eq 1 ]; then
+    RESTART_FLAG="-r"
+else
+    RESTART_FLAG=""
+fi
+
+# Loop until the run is finished.  Restarting when found the restart flag.
+FIRST_TIME=1
+while [ ! -f $FINISH_FILE ]; do
+    echo "$MPI -np $NPROCS $EXE -d $RESTART_FLAG $AMRFILE $OUTPUT"
+    $MPI -np $NPROCS $EXE -d $RESTART_FLAG $AMRFILE $OUTPUT
+
+    # Exit if an error code is returned
+    if [ "$?" -ne "0" ]; then
+	exit 1;
+    fi
+
+    # Get restart parameter file from RESTART_FILE
+    if [ -f $RESTART_FILE ]; then
+
+	# If this isn't the first loop, erase the previous restart dump
+	DIR=${AMRFILE%%/*}
+	if [ -d $DIR -a $FIRST_TIME -eq 0 ]; then
+	    rm -r $DIR
+	fi
+
+	AMRFILE=`cat $RESTART_FILE`
+	RESTART_FLAG="-r"
+	FIRST_TIME=0
+
+    fi
+
+done
diff --git a/src/enzo/CheckForOutput.C b/src/enzo/CheckForOutput.C
@@ -64,11 +64,8 @@ int CheckForOutput(HierarchyEntry *TopGrid, TopGridData &MetaData,
     MetaData.TimeLastDataDump += MetaData.dtDataDump;
 
     //#ifdef USE_HDF5_GROUPS
-    if (Group_WriteAllData(MetaData.DataDumpName, MetaData.DataDumpNumber++,
-		     TopGrid, MetaData, Exterior) == FAIL) {
-	fprintf(stderr, "Error in Group_WriteAllData.\n");
-	ENZO_FAIL("");
-    }
+    Group_WriteAllData(MetaData.DataDumpName, MetaData.DataDumpNumber++,
+		       TopGrid, MetaData, Exterior);
 // #else
 //     if (WriteAllData(MetaData.DataDumpName, MetaData.DataDumpNumber++,
 // 		     TopGrid, MetaData, Exterior) == FAIL) {
@@ -88,11 +85,8 @@ int CheckForOutput(HierarchyEntry *TopGrid, TopGridData &MetaData,
     MetaData.CycleLastDataDump += MetaData.CycleSkipDataDump;
 
     //#ifdef USE_HDF5_GROUPS
-    if (Group_WriteAllData(MetaData.DataDumpName, MetaData.DataDumpNumber++,
-		     TopGrid, MetaData, Exterior) == FAIL) {
-	fprintf(stderr, "Error in Group_WriteAllData.\n");
-	ENZO_FAIL("");
-    }
+    Group_WriteAllData(MetaData.DataDumpName, MetaData.DataDumpNumber++,
+		       TopGrid, MetaData, Exterior);
 // #else
 //     if (WriteAllData(MetaData.DataDumpName, MetaData.DataDumpNumber++,
 // 		     TopGrid, MetaData, Exterior) == FAIL) {
@@ -123,11 +117,8 @@ int CheckForOutput(HierarchyEntry *TopGrid, TopGridData &MetaData,
     MetaData.CycleLastDataDump = MetaData.CycleNumber;
     if (debug) printf("CPUtime-based output!\n");
 //#ifdef USE_HDF5_GROUPS
-    if (Group_WriteAllData(MetaData.DataDumpName, MetaData.DataDumpNumber++,
-		     TopGrid, MetaData, Exterior) == FAIL) {
-	fprintf(stderr, "Error in Group_WriteAllData.\n");
-	ENZO_FAIL("");
-    }
+    Group_WriteAllData(MetaData.DataDumpName, MetaData.DataDumpNumber++,
+		       TopGrid, MetaData, Exterior);
 // #else
 //     if (WriteAllData(MetaData.DataDumpName, MetaData.DataDumpNumber++,
 // 		     TopGrid, MetaData, Exterior) == FAIL) {
@@ -138,6 +129,48 @@ int CheckForOutput(HierarchyEntry *TopGrid, TopGridData &MetaData,
     WroteData = TRUE;
   } // ENDIF
 
+  /* Check for output: restart-based. */
+
+  char *param;
+  FILE *pfptr;
+
+  if ((CurrentCPUTime >= MetaData.dtRestartDump && 
+       MetaData.dtRestartDump > 0 ) ||
+      (MetaData.CycleNumber - MetaData.CycleLastRestartDump >= 
+       MetaData.CycleSkipRestartDump &&
+       MetaData.CycleSkipRestartDump > 0)) {
+
+    MetaData.CycleLastRestartDump = MetaData.CycleNumber;
+
+    if (debug) printf("Writing restart dump.\n");
+    Group_WriteAllData(MetaData.RestartDumpName, MetaData.RestartDumpNumber++,
+		       TopGrid, MetaData, Exterior);
+
+    /* On the root processor, write the restart parameter filename to
+       a file that will be read by a (batch) script to restart enzo.
+       We cannot call another MPI application from here. */
+
+    if (MyProcessorNumber == ROOT_PROCESSOR) {
+      param = new char[512];
+      if (MetaData.RestartDumpDir != NULL)
+	sprintf(param, "%s%"CYCLE_TAG_FORMAT""ISYM"/%s%"CYCLE_TAG_FORMAT""ISYM,
+		MetaData.RestartDumpDir, MetaData.RestartDumpNumber-1,
+		MetaData.RestartDumpName, MetaData.RestartDumpNumber-1);
+      else
+	sprintf(param, "%s%"CYCLE_TAG_FORMAT""ISYM,
+		MetaData.RestartDumpName, MetaData.RestartDumpNumber-1);
+
+      if ((pfptr = fopen("RestartParamFile", "w")) == NULL)
+	ENZO_FAIL("Error opening RestartParamFile");
+      fprintf(pfptr, "%s", param);
+      fclose(pfptr);
+
+      delete [] param;
+    } // ENDIF ROOT_PROCESSOR
+
+    WroteData = TRUE;
+  }
+
   /* Check for output: redshift-based. */
 
   if (ComovingCoordinates)
@@ -155,10 +188,7 @@ int CheckForOutput(HierarchyEntry *TopGrid, TopGridData &MetaData,
 	  }
 
 	  //#ifdef USE_HDF5_GROUPS
-	  if (Group_WriteAllData(Name, Number, TopGrid, MetaData, Exterior) == FAIL) {
-	    fprintf(stderr, "Error in Group_WriteAllData.\n");
-	    ENZO_FAIL("");
-	  }
+	  Group_WriteAllData(Name, Number, TopGrid, MetaData, Exterior);
 // #else
 // 	  if (WriteAllData(Name, Number, TopGrid, MetaData, Exterior) == FAIL) {
 // 	    fprintf(stderr, "Error in WriteAllData.\n");

diff --git a/src/enzo/CheckForResubmit.C b/src/enzo/CheckForResubmit.C
@@ -31,7 +31,7 @@ int CheckForResubmit(TopGridData &MetaData, int &Stop)
   if (MetaData.ResubmitOn == FALSE)
     return SUCCESS;
 
-  char *cmd = new char[256];
+  char *cmd = new char[512];
   double CurrentCPUTime = ReturnWallTime() - MetaData.StartCPUTime;
 
   if (CurrentCPUTime + MetaData.LastCycleCPUTime > MetaData.StopCPUTime) {
@@ -40,19 +40,21 @@ int CheckForResubmit(TopGridData &MetaData, int &Stop)
 	     "Executing resubmission script, %s\n", MetaData.ResubmitCommand);
     if (MyProcessorNumber == ROOT_PROCESSOR) {
       if (MetaData.DataDumpDir != NULL)
-	sprintf(cmd, "%s/%s %d %s%4.4d/%s%4.4d", MetaData.GlobalDir,
-		MetaData.ResubmitCommand, NumberOfProcessors, 
+	sprintf(cmd, "%s/%s %"ISYM" %s%"CYCLE_TAG_FORMAT""ISYM"/%s%"CYCLE_TAG_FORMAT""ISYM,
+		MetaData.GlobalDir, MetaData.ResubmitCommand, NumberOfProcessors, 
 		MetaData.DataDumpDir, MetaData.DataDumpNumber-1, 
 		MetaData.DataDumpName, MetaData.DataDumpNumber-1);
       else
-	sprintf(cmd, "%s/%s %d %s%4.4d", MetaData.GlobalDir, 
-		MetaData.ResubmitCommand, NumberOfProcessors, 
+	sprintf(cmd, "%s/%s %"ISYM" %s%"CYCLE_TAG_FORMAT""ISYM, 
+		MetaData.GlobalDir, MetaData.ResubmitCommand, NumberOfProcessors, 
 		MetaData.DataDumpName, MetaData.DataDumpNumber-1);
       printf("command: %s\n", cmd);
       system(cmd);
     }
     Stop = TRUE;
   } // ENDIF
 
+  delete [] cmd;
+
   return SUCCESS;
 }