mpich ch_gm device and totalview

Hi,
	Attached is a patch to make the ch_gm device work properly with
the TotalView multi-process debugger. Beware, there seems to be a bug in
the gm driver that if you are stopped at a breakpoint in the debugger and
you exit the program, it will hang the computer. I'm told that there will
be a fix in a future version of gm.

diff -u -r clean/mpich-1.2..2/mpid/ch_gm/adi2init.c mpich-1.2..2/mpid/ch_gm/adi2init.c
--- clean/mpich-1.2..2/mpid/ch_gm/adi2init.c	Wed Apr 19 03:10:54 2000
+++ mpich-1.2..2/mpid/ch_gm/adi2init.c	Tue Aug 15 14:01:02 2000
@@ -12,6 +12,8 @@
 #include "../util/cmnargs.h"
 #include "../util/queue.h"
 #include "reqalloc.h"
+#include "gmpi.h"
+#include <unistd.h>
 
 /* Home for these globals */
 int MPID_MyWorldSize, MPID_MyWorldRank;
@@ -55,8 +57,12 @@
 void *config;
 char ***argv;
 {
-    int i, np;
+    int i, ii, np, err, msgrp = 0, j, jj;
     MPID_Device *dev;
+    int mypid;
+    struct MPIR_COMMUNICATOR *comm = 0;
+    struct gm_global *gm_intermediate, *g;
+    MPI_Status status;    
 
 #ifdef SMP_PLUG
     /* here we bypass the default fake config function. Instead, we call 
@@ -162,7 +168,52 @@
 
     MPID_devset->req_pending = 0;
     *error_code = MPI_SUCCESS;
+
+    /* Initialize the array on node with MPI process 0 that contains
+       pid information for all processes. The index of the array is the same
+       as the mpi rank of the process. We associate this with a node by
+       matching it to the values in gmpi.node_name[].
+    */
+
+    mypid = (int ) getpid();
+    printf ("PID is, %d, index is %d\n",mypid,MPID_MyWorldRank);
+    printf("Getting PIDs\n");
+
+    printf("host name is, %s\n", gmpi.node_names[MPID_MyWorldRank]);
+    if (MPID_MyWorldRank == 0){
+      gmpi.mpi_pid[MPID_MyWorldRank] = mypid;
+      for (ii=1; ii<np; ii++) {
+	printf("recv started\n");
+	MPID_RecvContig(comm, &j, sizeof(int), ii, 0, 0,
+			&status ,&err);
+	gmpi.mpi_pid[ii] = j;
+	printf("PID, %d\n", j);
+      }
+    }
+    else
+      {
+	MPID_SendContig(comm, &mypid, sizeof(int), MPID_MyWorldRank, 
+			0, 0, 0, msgrp, &err);
+      }
+}
+
+/* Created a routine for remote PID's for TotalView debugger */
+
+int MPID_getpid(int i, char **host, char **exename){
+    int pid;
+
+    if (i == 0)
+      pid = (int) getpid();
+    else
+      pid = gmpi.mpi_pid[i];
+
+    *host = gmpi.node_names[i];
+    printf ("PID, %d on node, %s, index %d\n", pid, host,i);
+    printf ("I'm pid %d\n", getpid());
+    return(pid);
+
 }
+
 
 /* Barry Smith suggests that this indicate who is aborting the program.
    There should probably be a separate argument for whether it is a 
diff -u -r clean/mpich-1.2..2/mpid/ch_gm/chconfig.h mpich-1.2..2/mpid/ch_gm/chconfig.h
--- clean/mpich-1.2..2/mpid/ch_gm/chconfig.h	Wed Mar  8 19:49:24 2000
+++ mpich-1.2..2/mpid/ch_gm/chconfig.h	Wed Aug  2 13:11:51 2000
@@ -15,7 +15,8 @@
 #endif
 
 
-/* #define MPID_HAS_PROC_INFO  */
+#define MPID_HAS_PROC_INFO 
+
 /* This is needed if you want TotalView to acquire all of the processes
  * automagically.
  * If it is defined you must also define 


diff -u -r clean/mpich-1.2..2/mpid/ch_gm/gmpi.h mpich-1.2..2/mpid/ch_gm/gmpi.h
--- clean/mpich-1.2..2/mpid/ch_gm/gmpi.h	Wed Apr 19 03:10:54 2000
+++ mpich-1.2..2/mpid/ch_gm/gmpi.h	Wed Aug  2 13:11:51 2000
@@ -126,6 +126,7 @@
 
   unsigned int node_ids[GMPI_MAX_NUMNODES];/* node ids */
   char *node_names[GMPI_MAX_NUMNODES];/* names of machines */
+  unsigned int mpi_pid[GMPI_MAX_NUMNODES];/*PID of remote processes */
   unsigned int port_ids[GMPI_MAX_NUMNODES];/* port ids */
   unsigned int board_ids[GMPI_MAX_NUMNODES];/* board/unit number */
 #ifdef GMPI_CREDITS

-- 
Doug Johnson
Systems Developer/Engineer
The Ohio Supercomputer Center
[email protected]