* Fix a number of environment variable naming issues in the cofs code. This

really needs to use the param registry code
* make mpirun use -f -n to start processes - seems to make ssh happier
* update oob test to go through the mca layer

This commit was SVN r663.

src/mca/lam/oob/cofs/src/oob_cofs_module.c

@@ -110,7 +110,7 @@ mca_oob_cofs_init(int *priority, bool *allow_multi_user_threads,
    * BWB - fix me, make register the "right" way...
    */
   /* find our vpid */
-  tmp = getenv("MCA_OOB_BASE_VPID");
+  tmp = getenv("MCA_common_lam_cofs_my_vpid");
   if (tmp == NULL) {
     printf("oob_cofs can not find vpid\n");
     return NULL;

src/mca/lam/oob/oob.h

@@ -101,5 +101,7 @@ typedef mca_oob_1_0_0_t mca_oob_t;
   /* oob v1.0 */ \
   "oob", 1, 0, 0
 
+extern mca_oob_t mca_oob;
+
 
 #endif

src/mca/lam/pcm/pcm.h

@@ -58,6 +58,8 @@
 #include "mca/mca.h"
 #include "lam/types.h"
 
+#include <sys/param.h>
+
 /*
  * "PCM" global types
  */
@@ -66,6 +68,7 @@
 #define LAM_PCM_PROC_OTHER  2
 
 struct mca_pcm_rte_node_t {
+  char name[MAXHOSTNAMELEN];
   int32_t node_num;
   int32_t num_procs;
 };

src/mca/lam/registry/cofs/src/registry_cofs_module.c

@@ -108,7 +108,7 @@ mca_registry_cofs_init(int *priority, bool *allow_multi_user_threads,
    * BWB - fix me, make register the "right" way...
    */
   /* find our vpid */
-  tmp = getenv("MCA_REGISTRY_BASE_VPID");
+  tmp = getenv("MCA_common_lam_cofs_my_vpid");
   if (tmp == NULL) {
     printf("registry_cofs can not find vpid\n");
     return NULL;

src/tools/mpirun/mpiboot

@@ -91,5 +91,4 @@ fi
 # Spawn and run
 #
 ######################################################################
-exec $@ &
+exec $@
-exit 0

src/tools/mpirun/mpirun

@@ -116,7 +116,7 @@ start_proc() {
     local argv="$*"
     local rmt_boot="${lam_mydir}/mpiboot"
 
-    local cmd="ssh ${rmt_hostname} ${rmt_boot} -myvpid ${rmt_vpid}"
+    local cmd="ssh ${rmt_hostname} -f -n ${rmt_boot} -myvpid ${rmt_vpid}"
     cmd="${cmd} -numprocs ${lam_numprocs} -jobhandle ${lam_jobhandle}"
     cmd="${cmd} -pwd ${lam_cwd}"
     if test ! -z "${MCA_common_lam_cofs_comm_dir}" ; then
@@ -193,7 +193,7 @@ launch_procs() {
 # Do the prep work
 #
 ######################################################################
-lam_jobhandle="pcm_cofs_job_handle_${$}_0"
+lam_jobhandle="pcm-cofs-job-handle-${$}-0"
 
 
 ######################################################################

test/unit/lam/oob_cofs/Makefile.am

@@ -5,15 +5,9 @@
 
 include $(top_srcdir)/config/Makefile.options
 
-noinst_PROGRAMS		= oob_cofs_test pcm_cofs_test
+noinst_PROGRAMS		= oob_cofs_test
 
 oob_cofs_test_SOURCES 	= oob_cofs_test.c
 oob_cofs_test_LDADD	= \
 			../../../../src/mca/lam/oob/cofs/libmca_lam_oob_cofs.la \
 			../../../../src/liblam.la
-
-pcm_cofs_test_SOURCES 	= pcm_cofs_test.c
-pcm_cofs_test_LDADD	= \
-			../../../../src/mca/lam/pcm/cofs/libmca_lam_pcm_cofs.la \
-			../../../../src/mca/lam/oob/cofs/libmca_lam_oob_cofs.la \
-			../../../../src/liblam.la

test/unit/lam/oob_cofs/oob_cofs_test.c

@@ -1,74 +1,92 @@
+#include "lam/runtime/runtime.h"
 #include "mca/lam/oob/oob.h"
-#include "mca/lam/oob/cofs/src/oob_cofs.h"
+#include "mca/lam/pcm/pcm.h"
+#include "mca/lam/base/base.h"
 
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <assert.h>
 
 int
 main(int argc, char* argv[])
 {
   int ret;
-  char *tmp;
+  mca_pcm_proc_t *procs;
-  struct mca_oob_1_0_0_t *init_ret;
+  size_t nprocs;
-  int target_vpid, source_vpid, source_tag;
+  mca_pcm_proc_t *me;
-  size_t source_len;
+  int left_vpid, right_vpid, me_vpid;
-  char *source_data;
+  int count = 2;
-  char buffer[2048];
+  lam_job_handle_t job;
-  int msg_count = 0;
+  int data = 0xDEADBEEF;
-  int priority = 0;
+  int tag = MCA_OOB_ANY_TAG;
-  bool allow_threads = false;
+  bool threads, hidden;
-  bool have_hidden_threads = false;
 
+  printf("hello, world!\n");
+  ret = lam_init(argc, argv);
+  assert(ret == LAM_SUCCESS);
 
-  if (argc != 3) {
+  ret = mca_base_open();
-    printf("usage: %s my_vpid target_vpid\n", argv[0]);
+  assert(ret == LAM_SUCCESS);
-    exit(1);
+
+  ret = lam_rte_init(&threads, &hidden);
+  assert(ret == LAM_SUCCESS);
+
+  ret = mca_pcm.pcm_proc_startup();
+  assert(ret == LAM_SUCCESS);
+
+  ret = mca_pcm.pcm_proc_get_peers(&procs, &nprocs);
+  assert(ret == LAM_SUCCESS);
+
+  job = mca_pcm.pcm_handle_get();
+  assert(job != NULL);
+
+  me = mca_pcm.pcm_proc_get_me();
+  assert(me != NULL);
+
+  /* time to play the ring game! */
+  me_vpid = me->vpid;
+  printf("Hello, World.  I am vpid %d\n", me_vpid);
+
+  left_vpid = me_vpid == 0 ? nprocs - 1 : me_vpid - 1;
+  right_vpid = (me_vpid + 1) % nprocs;
+
+  if (me_vpid == 0) {
+    printf("vpid %d sending to vpid %d\n", me_vpid, right_vpid);
+    ret = mca_oob.oob_send(job, right_vpid, 0, &data, sizeof(int));
+    assert(ret == LAM_SUCCESS);
+    count--;
   }
 
-  tmp = malloc(strlen("MCA_OOB_BASE_VPID") + strlen(argv[1]) + 2);
+  while (count > 0) {
-  sprintf(tmp, "MCA_OOB_BASE_VPID=%s", argv[1]);
+    int *data_ptr;
-  putenv(tmp);
+    size_t data_ptr_len; 
+    printf("vpid %d recving from vpid %d\n", me_vpid, left_vpid);
+    ret = mca_oob.oob_recv(job, left_vpid, &tag, &data_ptr, &data_ptr_len);
+    assert(ret == LAM_SUCCESS);
+    assert(data_ptr_len == sizeof(int));
+    assert(*data_ptr == data);
 
-  target_vpid = atoi(argv[2]);
+    printf("vpid %d sending to vpid %d\n", me_vpid, right_vpid);
+    ret = mca_oob.oob_send(job, right_vpid, 0, &data, sizeof(int));
+    assert(ret == LAM_SUCCESS);
 
-  ret = mca_oob_cofs_open();
+    count--;
-  if (ret != LAM_SUCCESS) {
-    printf("mca_oob_cofs_open returned %d\n", ret);
-    exit(1);
   }
 
-  init_ret = mca_oob_cofs_init(&priority, &allow_threads, &have_hidden_threads);
+
-  if (init_ret == NULL) {
+  if (me_vpid == 0) {
-    printf("mca_oob_cofs_init returned NULL\n");
+    int *data_ptr;
-    exit(1);
+    size_t data_ptr_len;
-  } else {
+    printf("vpid %d recving from vpid %d\n", me_vpid, left_vpid);
-    printf("mca_oob_cofs_query said \"go\" with priority %d\n", priority);
+    ret = mca_oob.oob_recv(job, left_vpid, &tag, &data_ptr, &data_ptr_len);
+    assert(ret == LAM_SUCCESS);
+    assert(data_ptr_len == sizeof(int));
+    assert(*data_ptr == data);
   }
 
-  printf("#\n# Sending Messages\n#\n\n");
+  ret = lam_rte_finalize();
-  for (msg_count = 0 ; msg_count < 20 ; ++msg_count) {
+  assert(ret == LAM_SUCCESS);
-    sprintf(buffer, "%s's message number %d\n", argv[1], msg_count);
-    printf("%d %d: %s\n", target_vpid, 1, buffer);
-    ret = mca_oob_cofs_send("foobar", target_vpid, 1, buffer, strlen(buffer) + 1);
-    if (ret != LAM_SUCCESS) {
-      printf("mca_oob_cofs_send failed on msg_count %d\n", msg_count);
-      exit(1);
-    }
-  }
 
-  printf("#\n# Receiving Messages\n#\n\n");
-  for (msg_count = 0 ; msg_count < 20 ; ++msg_count) {
-    source_tag = 1;
-    ret = mca_oob_cofs_recv("foobar", target_vpid, &source_tag, (void**) &source_data, &source_len);
-    if (ret != LAM_SUCCESS) {
-      printf("mca_oob_cofs_recv failed on msg_count %d, %d\n", msg_count, ret);
-      exit(1);
-    }
-    printf("%d %d: %s\n", source_vpid, source_tag, source_data);
-    free(source_data);
-  }
-
-  printf("#\n# Finished\n#\n\n");
   return 0;
 }

test/unit/lam/oob_cofs/pcm_cofs_test.c

@@ -1,66 +0,0 @@
-#include "lam_config.h"
-
-#include "lam/constants.h"
-#include "mca/lam/oob/oob.h"
-#include "mca/lam/oob/cofs/src/oob_cofs.h"
-#include "mca/lam/pcm/pcm.h"
-#include "mca/lam/pcm/cofs/src/pcm_cofs.h"
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-
-
-int
-main(int argc, char *argv[])
-{
-  int ret;
-  struct mca_pcm_1_0_0_t* modret;
-  mca_pcm_proc_t *procs;
-  size_t nprocs;
-  mca_pcm_proc_t *my_proc;
-  int priority = 0;
-  bool allow_threads = false;
-  bool have_hidden_threads = false;
-
-  modret = mca_pcm_cofs_init(&priority, &allow_threads, &have_hidden_threads);
-  if (modret == NULL) {
-    printf("failed to init PCM module.  Aborting.\n");
-    exit(1);
-  }
-
-  /*
-   * Do all the stuff that a pcm module user would do
-   */
-  ret = mca_pcm_cofs_proc_startup();
-  if (ret != MPI_SUCCESS) {
-    printf("Failed in cofs_startup() with retcode: %d\n", ret);
-    exit(1);
-  }
-
-  ret = mca_pcm_cofs_proc_get_peers(&procs, &nprocs);
-  if (ret != MPI_SUCCESS) {
-    printf("Failed in cofs_proc_get_peers() with retcode %d\n", ret);
-    exit(1);
-  }
-
-  my_proc = mca_pcm_cofs_proc_get_me();
-  if (my_proc == NULL) {
-    printf("Failed to get my proc entry\n");
-    exit(1);
-  }
-
-  printf("Hello, World, I am vpid %d of job %s\n", my_proc->vpid, my_proc->job_handle);
-
-  /*
-   * from here, we can bring up the OOB interface (if it isn't already) and be
-   * ready to run...
-   */
-
-  /*
-   * Clean up after ourselves - normall that mca interface would do that for us...
-   */
-  mca_pcm_cofs_close();
-
-  return 0;
-}