add some debug code.

This commit was SVN r18096.
2008-04-07 21:03:23 +00:00 · 2008-04-07 21:03:23 +00:00 · fa696734d5
--- a/ompi/mca/coll/sm2/coll_sm2_allreduce.c
+++ b/ompi/mca/coll/sm2/coll_sm2_allreduce.c
@ -1460,31 +1460,31 @@ int mca_coll_sm2_allreduce_intra(void *sbuf, void *rbuf, int count,
    /* local variables */
    int rc;

-#if 0 /* just for some testing */    
+#if 0
    if( 0 != (op->o_flags & OMPI_OP_FLAGS_COMMUTE)) {
+#endif
        /* Commutative Operation */
        rc= mca_coll_sm2_allreduce_intra_recursive_doubling(sbuf, rbuf, count,
                dtype, op, comm, module);
        if( OMPI_SUCCESS != rc ) {
            goto Error;
        }
+#if 0
    } else {
-#endif /* testing */
        /* Non-Commutative Operation */
+#endif
+#if 0
        rc= mca_coll_sm2_allreduce_intra_fanin_fanout_pipeline(
                sbuf, rbuf, count,dtype, op, comm, module);
        if( OMPI_SUCCESS != rc ) {
            goto Error;
        }
-#if 0
        /* Non-Commutative Operation */
        rc= mca_coll_sm2_allreduce_intra_fanin_fanout(sbuf, rbuf, count,
                dtype, op, comm, module);
        if( OMPI_SUCCESS != rc ) {
            goto Error;
        }
-#endif
-#if 0 /* just for some testing */
    }
 #endif

--- a/ompi/mca/coll/sm2/coll_sm2_barrier.c
+++ b/ompi/mca/coll/sm2/coll_sm2_barrier.c
@ -24,6 +24,9 @@
 #include "ompi/mca/coll/coll.h"
 #include "opal/sys/atomic.h"
 #include "coll_sm2.h"
+/* debug */
+extern int debug_print;
+/* end debug */

 /**
 * Shared memory barrier.
@ -117,6 +120,15 @@ int mca_coll_sm2_nbbarrier_intra(struct ompi_communicator_t *comm,
                /* if parent has not checked in - set parameters for async
                 *   completion, incomplet barrier flag, and bail
                 */
+/* debug */
+if( debug_print ) {
+fprintf(stderr," A-I  rank %d parent %d -tag %lld sm_address->flag %lld \n",
+ ompi_comm_rank(comm),
+ sm_module->sm_buffer_mgmt_barrier_tree.parent_rank,
+-tag,sm_address->flag);
+}
+/* end debug */
+                /* child not arrived, just break out */
                request->sm2_barrier_phase=NB_BARRIER_FAN_OUT;
                return OMPI_SUCCESS;
            }
@ -302,6 +314,12 @@ int mca_coll_sm2_nbbarrier_intra_progress(struct ompi_communicator_t *comm,
            /* if parent has not checked in - set parameters for async
             *   completion, incomplet barrier flag, and bail
             */
+/* debug */
+if( debug_print ) {
+fprintf(stderr," I rank %d -tag %lld sm_address->flag %lld \n",
+ ompi_comm_rank(comm),-tag,sm_address->flag);
+}
+/* end debug */
            request->sm2_barrier_phase=NB_BARRIER_FAN_OUT;
            return OMPI_SUCCESS;
        }
@ -310,6 +328,13 @@ int mca_coll_sm2_nbbarrier_intra_progress(struct ompi_communicator_t *comm,
         * set my completion flag
         */
        request->sm2_barrier_phase=NB_BARRIER_DONE;
+/* debug */
+if( debug_print ) {
+fprintf(stderr," rank %d tag %lld done \n",
+  ompi_comm_rank(comm),
+  tag);
+}
+/* end debug */
    } else if( INTERIOR_NODE == sm_module->sm_buffer_mgmt_barrier_tree.my_node_type ) {
        phase=request->sm2_barrier_phase;
        if( NB_BARRIER_FAN_OUT == phase ) {
@ -335,6 +360,14 @@ int mca_coll_sm2_nbbarrier_intra_progress(struct ompi_communicator_t *comm,
                /* child arrived */
                cnt++;
            } else {
+/* debug */
+if( debug_print ) {
+fprintf(stderr," II rank %d child %d tag %lld sm_address->flag %lld \n",
+ ompi_comm_rank(comm),
+ sm_module->sm_buffer_mgmt_barrier_tree.children_ranks[child],
+tag,sm_address->flag);
+}
+/* end debug */
                /* child not arrived, just break out */
                break;
            }
@ -375,6 +408,14 @@ int mca_coll_sm2_nbbarrier_intra_progress(struct ompi_communicator_t *comm,
            /* if parent has not checked in - set parameters for async
             *   completion, incomplet barrier flag, and bail
             */
+/* debug */
+if( debug_print ) {
+fprintf(stderr," III rank %d parent %d -tag %lld sm_address->flag %lld \n",
+ ompi_comm_rank(comm),
+ sm_module->sm_buffer_mgmt_barrier_tree.parent_rank,
+-tag,sm_address->flag);
+}
+/* end debug */
            request->sm2_barrier_phase=NB_BARRIER_FAN_OUT;
            return OMPI_SUCCESS;
        }
@ -389,6 +430,13 @@ int mca_coll_sm2_nbbarrier_intra_progress(struct ompi_communicator_t *comm,
         * set my completion flag
         */
        request->sm2_barrier_phase=NB_BARRIER_DONE;
+/* debug */
+if( debug_print ) {
+fprintf(stderr," rank %d tag %lld done \n",
+  ompi_comm_rank(comm),
+  tag);
+}
+/* end debug */
    } else {
        /* root node */
        phase=request->sm2_barrier_phase;
@ -413,6 +461,14 @@ int mca_coll_sm2_nbbarrier_intra_progress(struct ompi_communicator_t *comm,
                /* child arrived */
                cnt++;
            } else {
+/* debug */
+if( debug_print ) {
+fprintf(stderr," IV rank %d parent %d tag %lld sm_address->flag %lld \n",
+ ompi_comm_rank(comm),
+ sm_module->sm_buffer_mgmt_barrier_tree.children_ranks[child],
+ tag,sm_address->flag);
+}
+/* end debug */
                /* child not arrived, just break out */
                break;
            }
@ -438,6 +494,13 @@ int mca_coll_sm2_nbbarrier_intra_progress(struct ompi_communicator_t *comm,
         * set my completion flag
         */
        request->sm2_barrier_phase=NB_BARRIER_DONE;
+/* debug */
+if( debug_print ) {
+fprintf(stderr," rank %d tag %lld done \n",
+  ompi_comm_rank(comm),
+  tag);
+}
+/* end debug */
    }
    
 DONE:
--- a/ompi/mca/coll/sm2/coll_sm2_component.c
+++ b/ompi/mca/coll/sm2/coll_sm2_component.c
@ -36,6 +36,18 @@
 #include "ompi/mca/coll/base/base.h"
 #include "orte/mca/rml/rml.h"

+/* debug */
+#include <signal.h>
+
+extern int debug_print;
+extern int my_debug_rank;
+extern void debug_module(void);
+
+void dbg_handler(int my_signal) {
+ debug_print=1;
+ debug_module();
+}
+/* end debug */

 /*
 * Public string showing the coll ompi_sm V2 component version number
@ -117,6 +129,11 @@ mca_coll_sm2_component_t mca_coll_sm2_component = {
 */
 static int sm2_open(void)
 {
+/* debug */
+    int retVal;
+    struct sigaction new_sigact;
+/* end debug */
+
    /* local variables */
    mca_coll_sm2_component_t *cs = &mca_coll_sm2_component;

@ -176,6 +193,13 @@ static int sm2_open(void)
    cs->n_poll_loops=
        mca_coll_sm2_param_register_int("n_poll_loops",4);

+/* debug */
+    new_sigact.sa_handler=dbg_handler;
+    sigemptyset(&(new_sigact.sa_mask));
+
+    retVal=sigaction(SIGUSR2,&new_sigact,NULL);
+/* end debug */
+
    return OMPI_SUCCESS;
 }

--- a/ompi/mca/coll/sm2/coll_sm2_module.c
+++ b/ompi/mca/coll/sm2/coll_sm2_module.c
@ -44,6 +44,51 @@
 static int sm2_module_enable(struct mca_coll_base_module_1_1_0_t *module,
        struct ompi_communicator_t *comm);

+/* debug */
+extern int debug_print;
+extern int my_debug_rank;
+extern int my_debug_comm_size;
+extern void debug_module(void);
+static mca_coll_sm2_module_t *module_dbg;
+void debug_module(void) {
+ int i,j,k;
+ char *ptr;
+ mca_coll_sm2_nb_request_process_shared_mem_t * ctl_ptr;
+   /* control regions */
+   if ( 0 == my_debug_rank ) {
+       for( i=0 ; i < 2 ; i++ ) {
+           for( j=0 ; j < 2 ; j++ ) {
+              fprintf(stderr," bank %d index %d \n", i,j);
+                  for( k=0 ; k < my_debug_comm_size ; k++ ) {
+                      ctl_ptr=module_dbg->barrier_request[i].barrier_base_address[j];
+                      ctl_ptr=(mca_coll_sm2_nb_request_process_shared_mem_t *) (
+                          (char *)ctl_ptr+k*module_dbg->sm2_size_management_region_per_proc
+                      );
+                      fprintf(stderr," bank %d index %d flag %lld \n",
+                         i,j,ctl_ptr->flag);
+                  }
+           }
+       }
+    }
+    /* data regions */
+    fprintf(stderr," my_debug_rank %d current index %d freed index %d \n",
+         my_debug_rank,
+         module_dbg->sm2_allocated_buffer_index,module_dbg->sm2_freed_buffer_index);
+    if( 0 == my_debug_rank ) {
+        for( i=0 ; i < module_dbg->sm2_module_num_buffers ; i++ ) {
+            for( j=0 ; j < my_debug_comm_size ; j++ ) {
+                fprintf(stderr," buffer index %d tag %lld \n",
+                   i,
+                   module_dbg->sm_buffer_descriptor[i].proc_memory[j].control_region->flag);
+            }
+        }
+    }
+
+    fflush(stderr);
+ 
+}
+/* end debug */
+
 /*
 * Local functions
 */
@ -883,6 +928,9 @@ mca_coll_sm2_comm_query(struct ompi_communicator_t *comm, int *priority)
    /* touch pages to apply memory affinity - Note: do we really need this or will
     * the algorithms do this */

+/* debug */
+module_dbg=&(sm_module->super);
+/* end debug */

    /* return */
    return &(sm_module->super);