Changeset 15559

Timestamp:

07/23/13 12:06:26 (12 years ago)

Author:

Mathieu Morlighem

Message:

BUG: on fe233, pfopen crashed because the same file cannot be created simultaneously by all processor

Location:

issm/trunk-jpl/src/c

Files:

• classes/FemModel.cpp (modified) (2 diffs)
• modules/OutputResultsx/OutputResultsx.cpp (modified) (1 diff)
• shared/io/Disk/diskio.h (modified) (1 diff)
• shared/io/Disk/pfopen.cpp (modified) (2 diffs)

issm/trunk-jpl/src/c/classes/FemModel.cpp

@@ -188 +188 @@
 
         /*Open input file on cpu 0: */
-        if(my_rank==0) IOMODEL = pfopen(inputfilename ,"rb");
+        if(my_rank==0) IOMODEL = pfopen0(inputfilename ,"rb");
 
         /*Initialize internal data: */
@@ -1465 +1465 @@
         /*Get response*/
         J=0;
+        //_printf_(true,"list of misfits: ");
         for(int i=0;i<num_responses;i++){
                 this->Responsex(&Jplus,EnumToStringx(responses[i]),i);
+                //_printf_(true," %12.7g ",Jplus);
                 J+=Jplus;
         }
+        //_printf_(true," \n");
 
         /*Assign output pointers: */

issm/trunk-jpl/src/c/modules/OutputResultsx/OutputResultsx.cpp

@@ -61 +61 @@
                 if(io_gather){
                         /*Just open the file for output on cpu 0. We are gathering the data on cpu 0 from all other cpus: */
-                        if(my_rank==0) fid=pfopen(outputfilename ,"wb");
+                        if(my_rank==0) fid=pfopen0(outputfilename ,"wb");
                 }
                 else{

issm/trunk-jpl/src/c/shared/io/Disk/diskio.h

@@ -9 +9 @@
 
 FILE* pfopen(char* filename,const char* format);
+FILE* pfopen0(char* filename,const char* format);
 void  pfclose(FILE* fid,char* filename);
 void WriteLockFile(char* filename);

issm/trunk-jpl/src/c/shared/io/Disk/pfopen.cpp

@@ -4 +4 @@
 
 #ifdef HAVE_CONFIG_H
-        #include <config.h>
+#include <config.h>
 #else
 #error "Cannot compile with HAVE_CONFIG_H symbol! run configure first!"
@@ -10 +10 @@
 
 #include <stdio.h>
-#include "../../shared.h"
+#include "../Print/Print.h"
+#include "../Comm/Comm.h"
+#include "../../Exceptions/exceptions.h"
 
+FILE* pfopen0(char* filename,const char* format){
+
+        FILE* fid=NULL;
+
+        /*recover my_rank:*/
+        int my_rank  = IssmComm::GetRank();
+        if(my_rank) _error_("This function should only be called by cpu 0");
+
+        /*Open handle to data on disk*/
+        fid = fopen(filename,format);
+        if(fid==NULL) _error_("could not open file " << filename << " for binary reading or writing");
+
+        return fid;
+}
 FILE* pfopen(char* filename,const char* format){
 
         FILE* fid=NULL;
 
-        /*Open handle to data on disk: */
-        fid=fopen(filename,format);
-        if(fid==NULL) _error_("could not open file " << filename << " for binary reading or writing"); 
+        /*recover my_rank:*/
+        int my_rank  = IssmComm::GetRank();
+        int num_proc = IssmComm::GetSize();
+
+        /*Open handle to data on disk (one by one to avoid errors)*/
+        for(int i=0;i<num_proc;i++){
+                if(my_rank==i) fid = fopen(filename,format);
+#ifdef _HAVE_MPI_
+                MPI_Barrier(IssmComm::GetComm());
+#endif
+        }
+        if(fid==NULL) _error_("could not open file " << filename << " for binary reading or writing");
 
         return fid;