X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/a6bb32631746cb8f7744196398e11d9d499f7d8b..1372e35c79dba117ae1b592394821a3f716a3f13:/src/smpi/smpi_base.c

diff --git a/src/smpi/smpi_base.c b/src/smpi/smpi_base.c
index 7f0fd9d1a2..eb9436f598 100644
--- a/src/smpi/smpi_base.c
+++ b/src/smpi/smpi_base.c
@@ -1,253 +1,210 @@
 #include "private.h"
+#include "xbt/time.h"
 
-SMPI_MPI_Global_t smpi_mpi_global = NULL;
+XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_base, smpi,
+                                "Logging specific to SMPI (base)");
+XBT_LOG_EXTERNAL_CATEGORY(smpi_base);
+XBT_LOG_EXTERNAL_CATEGORY(smpi_bench);
+XBT_LOG_EXTERNAL_CATEGORY(smpi_kernel);
+XBT_LOG_EXTERNAL_CATEGORY(smpi_mpi);
+XBT_LOG_EXTERNAL_CATEGORY(smpi_receiver);
+XBT_LOG_EXTERNAL_CATEGORY(smpi_sender);
+XBT_LOG_EXTERNAL_CATEGORY(smpi_util);
 
-void smpi_mpi_land_func(void *x, void *y, void *z)
+smpi_mpi_global_t smpi_mpi_global = NULL;
+
+void smpi_mpi_land_func(void *a, void *b, int *length,
+                        MPI_Datatype * datatype);
+
+void smpi_mpi_land_func(void *a, void *b, int *length,
+                        MPI_Datatype * datatype)
 {
-	*(int *)z = *(int *)x && *(int *)y;
+  int i;
+  if (*datatype == smpi_mpi_global->mpi_int) {
+    int *x = a, *y = b;
+    for (i = 0; i < *length; i++) {
+      y[i] = x[i] && y[i];
+    }
+  }
 }
 
-void smpi_mpi_sum_func(void *x, void *y, void *z)
+void smpi_mpi_sum_func(void *a, void *b, int *length,
+                       MPI_Datatype * datatype);
+
+void smpi_mpi_sum_func(void *a, void *b, int *length, MPI_Datatype * datatype)
 {
-	*(int *)z = *(int *)x + *(int *)y;
+  int i;
+  if (*datatype == smpi_mpi_global->mpi_int) {
+    int *x = a, *y = b;
+    for (i = 0; i < *length; i++) {
+      y[i] = x[i] + y[i];
+    }
+  }
 }
 
-int inline smpi_mpi_comm_size(smpi_mpi_communicator_t comm)
+int smpi_mpi_comm_rank(smpi_mpi_communicator_t comm)
 {
-	return comm->size;
+  return comm->index_to_rank_map[smpi_host_index()];
 }
 
-// FIXME: smarter algorithm?
-int smpi_mpi_comm_rank(smpi_mpi_communicator_t comm, smx_host_t host)
+void smpi_process_init()
 {
-	int i;
+  smx_host_t host;
+  int i;
+  smpi_host_data_t hdata;
 
-	for(i = comm->size - 1; i > 0 && host != comm->hosts[i]; i--);
+  smpi_global->running_hosts_count++;
 
-	return i;
-}
+  // initialize some local variables
+  host = SIMIX_host_self();
 
-int inline smpi_mpi_comm_rank_self(smpi_mpi_communicator_t comm)
-{
-	return smpi_mpi_comm_rank(comm, SIMIX_host_self());
-}
+  hdata = xbt_new(s_smpi_host_data_t, 1);
+  SIMIX_host_set_data(host, hdata);
+  SIMIX_process_set_data(SIMIX_process_self(),hdata);
 
-void smpi_mpi_init()
-{
-	smx_process_t process;
-	smx_host_t host;
-	smx_host_t *hosts;
-	int size;
-
-	SIMIX_mutex_lock(smpi_global->running_hosts_count_mutex);
-	smpi_global->running_hosts_count++;
-	SIMIX_mutex_unlock(smpi_global->running_hosts_count_mutex);
-
-	// initialize some local variables
-	process = SIMIX_process_self();
-	host    = SIMIX_host_self();
-	hosts   = SIMIX_host_get_table();
-	size    = SIMIX_host_get_number();
-
-	// node 0 sets the globals
-	if (host == hosts[0]) {
-
-		smpi_mpi_global                                = xbt_new(s_SMPI_MPI_Global_t, 1);
-
-		// global communicator
-		smpi_mpi_global->mpi_comm_world                = xbt_new(s_smpi_mpi_communicator_t, 1);
-		smpi_mpi_global->mpi_comm_world->size          = size;
-		smpi_mpi_global->mpi_comm_world->hosts         = hosts;
-		smpi_mpi_global->mpi_comm_world->processes     = xbt_new(smx_process_t, size);
-		smpi_mpi_global->mpi_comm_world->processes[0]  = process;
-		smpi_mpi_global->mpi_comm_world->barrier_count = 0;
-		smpi_mpi_global->mpi_comm_world->barrier_mutex = SIMIX_mutex_init();
-		smpi_mpi_global->mpi_comm_world->barrier_cond  = SIMIX_cond_init();
-
-		// mpi datatypes
-		smpi_mpi_global->mpi_byte                      = xbt_new(s_smpi_mpi_datatype_t, 1);
-		smpi_mpi_global->mpi_byte->size                = (size_t)1;
-		smpi_mpi_global->mpi_int                       = xbt_new(s_smpi_mpi_datatype_t, 1);
-		smpi_mpi_global->mpi_int->size                 = sizeof(int);
-		smpi_mpi_global->mpi_double                    = xbt_new(s_smpi_mpi_datatype_t, 1);
-		smpi_mpi_global->mpi_double->size              = sizeof(double);
-
-		// mpi operations
-		smpi_mpi_global->mpi_land                      = xbt_new(s_smpi_mpi_op_t, 1);
-		smpi_mpi_global->mpi_land->func                = smpi_mpi_land_func;
-		smpi_mpi_global->mpi_sum                       = xbt_new(s_smpi_mpi_op_t, 1);
-		smpi_mpi_global->mpi_sum->func                 = smpi_mpi_sum_func;
-
-		// signal all nodes to perform initialization
-		SIMIX_mutex_lock(smpi_global->start_stop_mutex);
-		smpi_global->root_ready = 1;
-		SIMIX_cond_broadcast(smpi_global->start_stop_cond);
-		SIMIX_mutex_unlock(smpi_global->start_stop_mutex);
-
-	} else {
-
-		// make sure root is done before own initialization
-		SIMIX_mutex_lock(smpi_global->start_stop_mutex);
-		if (!smpi_global->root_ready) {
-			SIMIX_cond_wait(smpi_global->start_stop_cond, smpi_global->start_stop_mutex);
-		}
-		SIMIX_mutex_unlock(smpi_global->start_stop_mutex);
-
-		smpi_mpi_global->mpi_comm_world->processes[smpi_mpi_comm_rank_self(smpi_mpi_global->mpi_comm_world)] = process;
-	}
-
-	// wait for all nodes to signal initializatin complete
-	SIMIX_mutex_lock(smpi_global->start_stop_mutex);
-	smpi_global->ready_process_count++;
-	if (smpi_global->ready_process_count < 3 * size) {
-		SIMIX_cond_wait(smpi_global->start_stop_cond, smpi_global->start_stop_mutex);
-	} else {
-		SIMIX_cond_broadcast(smpi_global->start_stop_cond);
-	}
-	SIMIX_mutex_unlock(smpi_global->start_stop_mutex);
-
-	return;
+  for (i = 0; i < smpi_global->host_count && host != smpi_global->hosts[i]; i++);
+
+  hdata->index = i;
+  hdata->mutex = SIMIX_mutex_init();
+  hdata->cond = SIMIX_cond_init();
+  hdata->finalize = 0;
+
+  hdata->pending_recv_request_queue = xbt_fifo_new();
+
+  hdata->main = SIMIX_process_self();
+  hdata->sender = SIMIX_process_create("smpi_sender",
+          smpi_sender, hdata,
+          SIMIX_host_get_name(SIMIX_host_self()), 0, NULL,
+          /*props */ NULL);
+  hdata->receiver = SIMIX_process_create("smpi_receiver",
+          smpi_receiver, hdata,
+          SIMIX_host_get_name(SIMIX_host_self()), 0, NULL,
+          /*props */ NULL);
+
+  smpi_global->main_processes[hdata->index] = SIMIX_process_self();
+  return;
 }
 
-void smpi_mpi_finalize()
+void smpi_process_finalize()
 {
-	int i;
+  int i;
+  smpi_host_data_t hdata =  SIMIX_host_get_data(SIMIX_host_self());
 
-	SIMIX_mutex_lock(smpi_global->running_hosts_count_mutex);
-	i = --smpi_global->running_hosts_count;
-	SIMIX_mutex_unlock(smpi_global->running_hosts_count_mutex);
+  i = --smpi_global->running_hosts_count;
 
-	SIMIX_mutex_lock(smpi_global->start_stop_mutex);
-	smpi_global->ready_process_count--;
-	SIMIX_mutex_unlock(smpi_global->start_stop_mutex);
+  hdata->finalize = 2; /* Tell sender and receiver to quit */
+  SIMIX_process_resume(hdata->sender);
+  SIMIX_process_resume(hdata->receiver);
+  while (hdata->finalize>0) { /* wait until it's done */
+	  SIMIX_cond_wait(hdata->cond,hdata->mutex);
+  }
 
-	if (0 >= i) {
+  SIMIX_mutex_destroy(hdata->mutex);
+  SIMIX_cond_destroy(hdata->cond);
+  xbt_fifo_free(hdata->pending_recv_request_queue);
 
-		// wake up senders/receivers
-		for (i = 0; i < smpi_mpi_global->mpi_comm_world->size; i++) {
-			if (SIMIX_process_is_suspended(smpi_global->sender_processes[i])) {
-				SIMIX_process_resume(smpi_global->sender_processes[i]);
-			}
-			if (SIMIX_process_is_suspended(smpi_global->receiver_processes[i])) {
-				SIMIX_process_resume(smpi_global->receiver_processes[i]);
-			}
-		}
 
-		SIMIX_mutex_destroy(smpi_mpi_global->mpi_comm_world->barrier_mutex);
-		SIMIX_cond_destroy(smpi_mpi_global->mpi_comm_world->barrier_cond);
-		xbt_free(smpi_mpi_global->mpi_comm_world->processes);
-		xbt_free(smpi_mpi_global->mpi_comm_world);
+  if (0 >= i) {
 
-		xbt_free(smpi_mpi_global->mpi_byte);
-		xbt_free(smpi_mpi_global->mpi_int);
-		xbt_free(smpi_mpi_global->mpi_double);
+    // wake up senders/receivers
+	  /* MQ: (FIXME) Don't do so: it breaks since some hosts are already gone
+    for (i = 0; i < smpi_global->host_count; i++) {
+      smpi_host_data_t remote_hdata =  SIMIX_process_get_data(smpi_global->main_processes[i]);
 
-		xbt_free(smpi_mpi_global->mpi_land);
-		xbt_free(smpi_mpi_global->mpi_sum);
+      if (SIMIX_process_is_suspended(remote_hdata->sender))
+        SIMIX_process_resume(remote_hdata->sender);
 
-		xbt_free(smpi_mpi_global);
+      if (SIMIX_process_is_suspended(remote_hdata->receiver))
+        SIMIX_process_resume(remote_hdata->receiver);
+    }*/
 
-	}
+    SIMIX_mutex_destroy(smpi_mpi_global->mpi_comm_world->barrier_mutex);
+    SIMIX_cond_destroy(smpi_mpi_global->mpi_comm_world->barrier_cond);
+    xbt_free(smpi_mpi_global->mpi_comm_world);
 
-}
+    xbt_free(smpi_mpi_global->mpi_byte);
+    xbt_free(smpi_mpi_global->mpi_int);
+    xbt_free(smpi_mpi_global->mpi_double);
 
-void smpi_barrier(smpi_mpi_communicator_t comm)
-{
-
-	SIMIX_mutex_lock(comm->barrier_mutex);
-	if(++comm->barrier_count < comm->size) {
-		SIMIX_cond_wait(comm->barrier_cond, comm->barrier_mutex);
-	} else {
-		comm->barrier_count = 0;
-		SIMIX_cond_broadcast(comm->barrier_cond);
-	}
-	SIMIX_mutex_unlock(comm->barrier_mutex);
+    xbt_free(smpi_mpi_global->mpi_land);
+    xbt_free(smpi_mpi_global->mpi_sum);
 
-	return;
+    xbt_free(smpi_mpi_global);
+  }
 }
 
-int smpi_create_request(void *buf, int count, smpi_mpi_datatype_t datatype,
-	int src, int dst, int tag, smpi_mpi_communicator_t comm, smpi_mpi_request_t *request)
+int smpi_mpi_barrier(smpi_mpi_communicator_t comm)
 {
-	int retval = MPI_SUCCESS;
-
-	*request = NULL;
-
-	if (0 > count) {
-		retval = MPI_ERR_COUNT;
-	} else if (NULL == buf) {
-		retval = MPI_ERR_INTERN;
-	} else if (NULL == datatype) {
-		retval = MPI_ERR_TYPE;
-	} else if (NULL == comm) {
-		retval = MPI_ERR_COMM;
-	} else if (MPI_ANY_SOURCE != src && (0 > src || comm->size <= src)) {
-		retval = MPI_ERR_RANK;
-	} else if (0 > dst || comm->size <= dst) {
-		retval = MPI_ERR_RANK;
-	} else if (0 > tag) {
-		retval = MPI_ERR_TAG;
-	} else {
-		*request               = xbt_mallocator_get(smpi_global->request_mallocator);
-		(*request)->comm       = comm;
-		(*request)->src        = src;
-		(*request)->dst        = dst;
-		(*request)->tag        = tag;
-		(*request)->buf        = buf;
-		(*request)->count      = count;
-		(*request)->datatype   = datatype;
-	}
-	return retval;
+
+  SIMIX_mutex_lock(comm->barrier_mutex);
+  ++comm->barrier_count;
+  if (comm->barrier_count > comm->size) {       // only happens on second barrier...
+    comm->barrier_count = 0;
+  } else if (comm->barrier_count == comm->size) {
+    SIMIX_cond_broadcast(comm->barrier_cond);
+  }
+  while (comm->barrier_count < comm->size) {
+    SIMIX_cond_wait(comm->barrier_cond, comm->barrier_mutex);
+  }
+  SIMIX_mutex_unlock(comm->barrier_mutex);
+
+  return MPI_SUCCESS;
 }
 
-int smpi_isend(smpi_mpi_request_t request)
+int smpi_mpi_isend(smpi_mpi_request_t request)
 {
-	int retval = MPI_SUCCESS;
-	int rank   = smpi_mpi_comm_rank_self(smpi_mpi_global->mpi_comm_world);
+	smpi_host_data_t hdata =  SIMIX_host_get_data(SIMIX_host_self());
+  int retval = MPI_SUCCESS;
 
-	if (NULL != request) {
-		SIMIX_mutex_lock(smpi_global->pending_send_request_queues_mutexes[rank]);
-		xbt_fifo_push(smpi_global->pending_send_request_queues[rank], request);
-		SIMIX_mutex_unlock(smpi_global->pending_send_request_queues_mutexes[rank]);
-	}
+  if (NULL == request) {
+    retval = MPI_ERR_INTERN;
+  } else {
+    xbt_fifo_push(smpi_global->pending_send_request_queues[hdata->index], request);
 
-	if (SIMIX_process_is_suspended(smpi_global->sender_processes[rank])) {
-		SIMIX_process_resume(smpi_global->sender_processes[rank]);
-	}
+    if (SIMIX_process_is_suspended(hdata->sender)) {
+      SIMIX_process_resume(hdata->sender);
+    }
+  }
 
-	return retval;
+  return retval;
 }
 
-int smpi_irecv(smpi_mpi_request_t request)
+int smpi_mpi_irecv(smpi_mpi_request_t request)
 {
-	int retval = MPI_SUCCESS;
-	int rank = smpi_mpi_comm_rank_self(smpi_mpi_global->mpi_comm_world);
+  int retval = MPI_SUCCESS;
+  smpi_host_data_t hdata =  SIMIX_host_get_data(SIMIX_host_self());
 
-	if (NULL != request) {
-		SIMIX_mutex_lock(smpi_global->pending_recv_request_queues_mutexes[rank]);
-		xbt_fifo_push(smpi_global->pending_recv_request_queues[rank], request);
-		SIMIX_mutex_unlock(smpi_global->pending_recv_request_queues_mutexes[rank]);
-	}
+  if (NULL == request) {
+    retval = MPI_ERR_INTERN;
+  } else {
+    xbt_fifo_push(hdata->pending_recv_request_queue, request);
 
-	if (SIMIX_process_is_suspended(smpi_global->receiver_processes[rank])) {
-		SIMIX_process_resume(smpi_global->receiver_processes[rank]);
-	}
+    if (SIMIX_process_is_suspended(hdata->receiver)) {
+      SIMIX_process_resume(hdata->receiver);
+    }
+  }
 
-	return retval;
+  return retval;
 }
 
-int smpi_wait(smpi_mpi_request_t request, smpi_mpi_status_t *status)
+int smpi_mpi_wait(smpi_mpi_request_t request, smpi_mpi_status_t * status)
 {
-	if (NULL != request) {
-		SIMIX_mutex_lock(request->mutex);
-		if (!request->completed) {
-			SIMIX_cond_wait(request->cond, request->mutex);
-		}
-		if (NULL != status) {
-			status->MPI_SOURCE = request->src;
-		}
-		SIMIX_mutex_unlock(request->mutex);
-	}
-
-	return MPI_SUCCESS;
+  int retval = MPI_SUCCESS;
+
+  if (NULL == request) {
+    retval = MPI_ERR_INTERN;
+  } else {
+    SIMIX_mutex_lock(request->mutex);
+    while (!request->completed) {
+      SIMIX_cond_wait(request->cond, request->mutex);
+    }
+    if (NULL != status) {
+      status->MPI_SOURCE = request->src;
+      status->MPI_TAG = request->tag;
+      status->MPI_ERROR = MPI_SUCCESS;
+    }
+    SIMIX_mutex_unlock(request->mutex);
+  }
+
+  return retval;
 }