XBT_LOG_EXTERNAL_CATEGORY(smpi_bench);
XBT_LOG_EXTERNAL_CATEGORY(smpi_kernel);
XBT_LOG_EXTERNAL_CATEGORY(smpi_mpi);
+XBT_LOG_EXTERNAL_CATEGORY(smpi_mpi_dt);
+XBT_LOG_EXTERNAL_CATEGORY(smpi_coll);
XBT_LOG_EXTERNAL_CATEGORY(smpi_receiver);
XBT_LOG_EXTERNAL_CATEGORY(smpi_sender);
XBT_LOG_EXTERNAL_CATEGORY(smpi_util);
smpi_mpi_global_t smpi_mpi_global = NULL;
+
/**
* Operations of MPI_OP : implemented=land,sum,min,max
**/
void smpi_mpi_sum_func(void *a, void *b, int *length, MPI_Datatype * datatype)
{
int i;
- if (*datatype == smpi_mpi_global->mpi_byte) {
+ if ((*datatype == smpi_mpi_global->mpi_byte) || (*datatype == smpi_mpi_global->mpi_char)) {
char *x = a, *y = b;
for (i = 0; i < *length; i++) {
y[i] = x[i] + y[i];
}
}
}
+/**
+ *i multiply two vectors element-wise
+ *
+ * @param a the first vectors
+ * @param b the second vectors
+ * @return the second vector is modified and contains the element-wise products
+ **/
+void smpi_mpi_prod_func(void *a, void *b, int *length,
+ MPI_Datatype * datatype);
+void smpi_mpi_prod_func(void *a, void *b, int *length, MPI_Datatype * datatype)
+{
+ int i;
+ if ((*datatype == smpi_mpi_global->mpi_byte) || (*datatype == smpi_mpi_global->mpi_char)) {
+ char *x = a, *y = b;
+ for (i = 0; i < *length; i++) {
+ y[i] = x[i] * y[i];
+ }
+ } else if (*datatype == smpi_mpi_global->mpi_int) {
+ int *x = a, *y = b;
+ for (i = 0; i < *length; i++) {
+ y[i] = x[i] * y[i];
+ }
+ } else if (*datatype == smpi_mpi_global->mpi_float) {
+ float *x = a, *y = b;
+ for (i = 0; i < *length; i++) {
+ y[i] = x[i] * y[i];
+ }
+ } else if (*datatype == smpi_mpi_global->mpi_double) {
+ double *x = a, *y = b;
+ for (i = 0; i < *length; i++) {
+ y[i] = x[i] * y[i];
+ }
+ }
+}
/**
* compute the min of two vectors element-wise
**/
void smpi_mpi_min_func(void *a, void *b, int *length, MPI_Datatype * datatype)
{
int i;
- if (*datatype == smpi_mpi_global->mpi_byte) {
+ if ((*datatype == smpi_mpi_global->mpi_byte) || (*datatype == smpi_mpi_global->mpi_char)) {
char *x = a, *y = b;
for (i = 0; i < *length; i++) {
y[i] = x[i] < y[i] ? x[i] : y[i];
void smpi_mpi_max_func(void *a, void *b, int *length, MPI_Datatype * datatype)
{
int i;
- if (*datatype == smpi_mpi_global->mpi_byte) {
+ if ((*datatype == smpi_mpi_global->mpi_byte) || (*datatype == smpi_mpi_global->mpi_char)) {
char *x = a, *y = b;
for (i = 0; i < *length; i++) {
y[i] = x[i] > y[i] ? x[i] : y[i];
/* get rank from command line, and remove it from argv */
pdata->index = atoi((*argv)[1]);
- DEBUG1("I'm rank %d", pdata->index);
+ DEBUG1("I'm rank <%d>", pdata->index);
if (*argc > 2) {
memmove((*argv)[1], (*argv)[2], sizeof(char *) * (*argc - 2));
(*argv)[(*argc) - 1] = NULL;
xbt_free(pdata);
}
-int smpi_mpi_barrier(smpi_mpi_communicator_t comm)
+
+/*int smpi_mpi_barrier(smpi_mpi_communicator_t comm)
{
SIMIX_mutex_lock(comm->barrier_mutex);
return MPI_SUCCESS;
}
+*/
int smpi_mpi_isend(smpi_mpi_request_t request)
{
return retval;
}
+void print_req( smpi_mpi_request_t r );
+void print_req( smpi_mpi_request_t r ) {
+ fprintf(stderr,"***req %p-> src=%d dst=%d tag=%d completed=0x%x consumed=0x%x\n",r,r->src,r->dst,r->tag,r->completed,r->consumed);
+}
+
+
+/**
+ * wait and friends ...
+ **/
int smpi_mpi_wait(smpi_mpi_request_t request, smpi_mpi_status_t * status)
{
int retval = MPI_SUCCESS;
if (NULL == request) {
retval = MPI_ERR_INTERN;
} else {
+
+ DEBUG3("entered smpi_mpi_wait() for req_src=%d,req_dst=%d,req_tag=%d",
+ request->src,request->dst,request->tag);
SIMIX_mutex_lock(request->mutex);
+//#define DEBUG_STEPH
+#ifdef DEBUG_STEPH
+ print_req( request ); //@@
+#endif
while (!request->completed) {
SIMIX_cond_wait(request->cond, request->mutex);
}
return retval;
}
+/**
+ * waitall
+ **/
int smpi_mpi_waitall(int count, smpi_mpi_request_t requests[],
- smpi_mpi_status_t status[])
+ smpi_mpi_status_t status[])
{
- int cpt;
- int index;
- int retval;
- smpi_mpi_status_t stat;
-
- for (cpt = 0; cpt < count; cpt++) {
- retval = smpi_mpi_waitany(count, requests, &index, &stat);
- if (retval != MPI_SUCCESS)
- return retval;
- memcpy(&(status[index]), &stat, sizeof(stat));
- }
- return MPI_SUCCESS;
+ int cpt;
+ int index;
+ int retval;
+ smpi_mpi_status_t stat;
+
+ for (cpt = 0; cpt < count; cpt++) {
+ retval = smpi_mpi_waitany(count, requests, &index, &stat);
+ if (retval != MPI_SUCCESS)
+ return retval;
+ if (MPI_STATUS_IGNORE != status)
+ memcpy(&(status[index]), &stat, sizeof(stat));
+ }
+ return MPI_SUCCESS;
}
+/**
+ * waitany
+ **/
int smpi_mpi_waitany(int count, smpi_mpi_request_t * requests, int *index,
smpi_mpi_status_t * status)
{
int cpt;
+ DEBUG0("entering smpi_wait_any() ...");
*index = MPI_UNDEFINED;
if (NULL == requests) {
return MPI_ERR_INTERN;
}
/* First check if one of them is already done */
for (cpt = 0; cpt < count; cpt++) {
+ DEBUG2(" exam req[%d] of msg from <%d>",cpt,requests[cpt]->src);
if (requests[cpt]->completed && !requests[cpt]->consumed) { /* got ya */
+ DEBUG2("smpi_wait_any() found match req[%d] of msg from <%d>",cpt,requests[cpt]->src);
*index = cpt;
goto found_request;
}
/* FIXME: should use a SIMIX_cond_waitany, when implemented. For now, block on the first one */
while (1) {
for (cpt = 0; cpt < count; cpt++) {
+
+#ifdef DEBUG_STEPH
+ print_req( requests[cpt] );
+#endif
if (!requests[cpt]->completed) { /* this one is not done, wait on it */
+ DEBUG3("smpi_waitany() blocked waiting a msg <%d> -> <%d>, tag=%d",requests[cpt]->src,requests[cpt]->dst,requests[cpt]->tag);
while (!requests[cpt]->completed)
SIMIX_cond_wait(requests[cpt]->cond, requests[cpt]->mutex);
}
found_request:
+#ifdef DEBUG_STEPH
+ print_req( requests[cpt] );
+#endif
requests[*index]->consumed = 1;
-
+#ifdef DEBUG_STEPH
+ print_req( requests[cpt] );
+#endif
+ DEBUG2("smpi_waitany() request %p unblocked ... mark *req[%d]->consumed",requests[*index],cpt);
if (NULL != status) {
- status->MPI_SOURCE = requests[*index]->src;
- status->MPI_TAG = requests[*index]->tag;
- status->MPI_ERROR = MPI_SUCCESS;
+ status->MPI_SOURCE = requests[*index]->src;
+ status->MPI_TAG = requests[*index]->tag;
+ status->MPI_ERROR = MPI_SUCCESS;
}
return MPI_SUCCESS;
