#include "private.h"
#include "xbt/time.h"
#include "mc/mc.h"
+#include "xbt/replay.h"
+#include <errno.h>
+#include "surf/surf.h"
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_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);
-
-static int match_recv(void* a, void* b) {
+
+static int match_recv(void* a, void* b, smx_action_t ignored) {
MPI_Request ref = (MPI_Request)a;
MPI_Request req = (MPI_Request)b;
+ XBT_DEBUG("Trying to match a recv of src %d against %d, tag %d against %d",ref->src,req->src, ref->tag, req->tag);
xbt_assert(ref, "Cannot match recv against null reference");
xbt_assert(req, "Cannot match recv against null request");
&& (ref->tag == MPI_ANY_TAG || req->tag == ref->tag);
}
-static int match_send(void* a, void* b) {
+static int match_send(void* a, void* b,smx_action_t ignored) {
MPI_Request ref = (MPI_Request)a;
MPI_Request req = (MPI_Request)b;
-
+ XBT_DEBUG("Trying to match a send of src %d against %d, tag %d against %d",ref->src,req->src, ref->tag, req->tag);
xbt_assert(ref, "Cannot match send against null reference");
xbt_assert(req, "Cannot match send against null request");
return (req->src == MPI_ANY_SOURCE || req->src == ref->src)
return request;
}
+void smpi_action_trace_run(char *path)
+{
+ char *name;
+ xbt_dynar_t todo;
+ xbt_dict_cursor_t cursor;
+
+ action_fp=NULL;
+ if (path) {
+ action_fp = fopen(path, "r");
+ xbt_assert(action_fp != NULL, "Cannot open %s: %s", path,
+ strerror(errno));
+ }
+
+ if (!xbt_dict_is_empty(action_queues)) {
+ XBT_WARN
+ ("Not all actions got consumed. If the simulation ended successfully (without deadlock), you may want to add new processes to your deployment file.");
+
+
+ xbt_dict_foreach(action_queues, cursor, name, todo) {
+ XBT_WARN("Still %lu actions for %s", xbt_dynar_length(todo), name);
+ }
+ }
+
+ if (path)
+ fclose(action_fp);
+ xbt_dict_free(&action_queues);
+ action_queues = xbt_dict_new_homogeneous(NULL);
+}
+
+static void smpi_mpi_request_free_voidp(void* request)
+{
+ MPI_Request req = request;
+ smpi_mpi_request_free(&req);
+}
+
/* MPI Low level calls */
MPI_Request smpi_mpi_send_init(void *buf, int count, MPI_Datatype datatype,
int dst, int tag, MPI_Comm comm)
"Cannot (re)start a non-finished communication");
if(request->flags & RECV) {
print_request("New recv", request);
+ if (request->size < xbt_cfg_get_int(_surf_cfg_set, "smpi/async_small_thres"))
+ mailbox = smpi_process_mailbox_small();
+ else
mailbox = smpi_process_mailbox();
+
// FIXME: SIMIX does not yet support non-contiguous datatypes
- request->action = SIMIX_req_comm_irecv(mailbox, request->buf, &request->size, &match_recv, request);
+ request->action = simcall_comm_irecv(mailbox, request->buf, &request->size, &match_recv, request);
} else {
print_request("New send", request);
- mailbox = smpi_process_remote_mailbox(
- smpi_group_index(smpi_comm_group(request->comm), request->dst));
- // FIXME: SIMIX does not yet support non-contiguous datatypes
- if (request->size < 64*1024 ) { // eager mode => detached send (FIXME: this limit should be configurable)
- void *oldbuf = request->buf;
- detached = 1;
- request->buf = malloc(request->size);
- memcpy(request->buf,oldbuf,request->size);
- XBT_DEBUG("Send request %p is detached; buf %p copied into %p",request,oldbuf,request->buf);
- } else {
- XBT_DEBUG("Send request %p is not detached (buf: %p)",request,request->buf);
+ if (request->size < xbt_cfg_get_int(_surf_cfg_set, "smpi/async_small_thres")) { // eager mode => detached send (FIXME: this limit should be configurable)
+ mailbox = smpi_process_remote_mailbox_small(
+ smpi_group_index(smpi_comm_group(request->comm), request->dst));
+ }else{
+ XBT_DEBUG("Send request %p is not in the permanent receive mailbox (buf: %p)",request,request->buf);
+ mailbox = smpi_process_remote_mailbox(
+ smpi_group_index(smpi_comm_group(request->comm), request->dst));
+ }
+ if (request->size < 64*1024 ) { //(FIXME: this limit should be configurable)
+ void *oldbuf = request->buf;
+ detached = 1;
+ request->buf = malloc(request->size);
+ if (oldbuf)
+ memcpy(request->buf,oldbuf,request->size);
+ XBT_DEBUG("Send request %p is detached; buf %p copied into %p",request,oldbuf,request->buf);
+ }
+
+ request->action =
+ simcall_comm_isend(mailbox, request->size, -1.0,
+ request->buf, request->size,
+ &match_send,
+ &smpi_mpi_request_free_voidp, // how to free the userdata if a detached send fails
+ request,
+ // detach if msg size < eager/rdv switch limit
+ detached);
+
+ #ifdef HAVE_TRACING
+ /* FIXME: detached sends are not traceable (request->action == NULL) */
+ if (request->action)
+ simcall_set_category(request->action, TRACE_internal_smpi_get_category());
+ #endif
+
}
- request->action =
- SIMIX_req_comm_isend(mailbox, request->size, -1.0,
- request->buf, request->size,
- &match_send,free, // a simple free is used to free the buffer
- request,
- // detach if msg size < eager/rdv switch limit
- detached);
-#ifdef HAVE_TRACING
- SIMIX_req_set_category (request->action, TRACE_internal_smpi_get_category());
-#endif
- }
}
void smpi_mpi_startall(int count, MPI_Request * requests)
{
- int i;
+ int i;
for(i = 0; i < count; i++) {
smpi_mpi_start(requests[i]);
smpi_mpi_wait(&request, status);
}
+
+
void smpi_mpi_send(void *buf, int count, MPI_Datatype datatype, int dst,
int tag, MPI_Comm comm)
{
- MPI_Request request;
-
- request = smpi_mpi_isend(buf, count, datatype, dst, tag, comm);
- smpi_mpi_wait(&request, MPI_STATUS_IGNORE);
+ MPI_Request request;
+ request = smpi_mpi_isend(buf, count, datatype, dst, tag, comm);
+ smpi_mpi_wait(&request, MPI_STATUS_IGNORE);
}
void smpi_mpi_sendrecv(void *sendbuf, int sendcount, MPI_Datatype sendtype,
static void finish_wait(MPI_Request * request, MPI_Status * status)
{
MPI_Request req = *request;
+ // if we have a sender, we should use its data, and not the data from the receive
+ if((req->action)&&
+ (req->src==MPI_ANY_SOURCE || req->tag== MPI_ANY_TAG))
+ req = (MPI_Request)SIMIX_comm_get_src_data((*request)->action);
if(status != MPI_STATUS_IGNORE) {
status->MPI_SOURCE = req->src;
// right?
status->count = req->size;
}
+ req = *request;
+
print_request("Finishing", req);
if(req->flags & NON_PERSISTENT) {
smpi_mpi_request_free(request);
}
int smpi_mpi_test(MPI_Request * request, MPI_Status * status) {
-int flag;
-
- if ((*request)->action == NULL)
- flag = 1;
- else
- flag = SIMIX_req_comm_test((*request)->action);
- if(flag) {
- smpi_mpi_wait(request, status);
- }
- return flag;
+ int flag;
+ if ((*request)->action == NULL)
+ flag = 1;
+ else
+ flag = simcall_comm_test((*request)->action);
+ if(flag) {
+ finish_wait(request, status);
+ }
+ return flag;
}
int smpi_mpi_testany(int count, MPI_Request requests[], int *index,
}
}
if(size > 0) {
- i = SIMIX_req_comm_testany(comms);
+ i = simcall_comm_testany(comms);
// FIXME: MPI_UNDEFINED or does SIMIX have a return code?
if(i != MPI_UNDEFINED) {
*index = map[i];
return flag;
}
+
+int smpi_mpi_testall(int count, MPI_Request requests[],
+ MPI_Status status[])
+{
+ int flag=1;
+ int i;
+ for(i=0; i<count; i++)
+ if(requests[i]!= MPI_REQUEST_NULL)
+ if (smpi_mpi_test(&requests[i], &status[i])!=1)
+ flag=0;
+
+ return flag;
+}
+
+void smpi_mpi_probe(int source, int tag, MPI_Comm comm, MPI_Status* status){
+ int flag=0;
+ //FIXME find another wait to avoid busy waiting ?
+ // the issue here is that we have to wait on a nonexistent comm
+ while(flag==0){
+ smpi_mpi_iprobe(source, tag, comm, &flag, status);
+ XBT_DEBUG("Busy Waiting on probing : %d", flag);
+ if(!flag) {
+ simcall_process_sleep(0.0001);
+ }
+ }
+}
+
+void smpi_mpi_iprobe(int source, int tag, MPI_Comm comm, int* flag, MPI_Status* status){
+ MPI_Request request =build_request(NULL, 0, MPI_CHAR, source, smpi_comm_rank(comm), tag,
+ comm, NON_PERSISTENT | RECV);
+ // behave like a receive, but don't do it
+ smx_rdv_t mailbox;
+
+ print_request("New iprobe", request);
+ // We have to test both mailboxes as we don't know if we will receive one one or another
+ if (xbt_cfg_get_int(_surf_cfg_set, "smpi/async_small_thres")>0){
+ mailbox = smpi_process_mailbox_small();
+ XBT_DEBUG("trying to probe the perm recv mailbox");
+ request->action = simcall_comm_iprobe(mailbox, request->src, request->tag, &match_recv, (void*)request);
+ }
+ if (request->action==NULL){
+ mailbox = smpi_process_mailbox();
+ XBT_DEBUG("trying to probe the other mailbox");
+ request->action = simcall_comm_iprobe(mailbox, request->src, request->tag, &match_recv, (void*)request);
+ }
+
+ if(request->action){
+ MPI_Request req = (MPI_Request)SIMIX_comm_get_src_data(request->action);
+ *flag=true;
+ if(status != MPI_STATUS_IGNORE) {
+ status->MPI_SOURCE = req->src;
+ status->MPI_TAG = req->tag;
+ status->MPI_ERROR = MPI_SUCCESS;
+ status->count = req->size;
+ }
+ }
+ else *flag=false;
+ smpi_mpi_request_free(&request);
+
+ return;
+}
+
void smpi_mpi_wait(MPI_Request * request, MPI_Status * status)
{
print_request("Waiting", *request);
if ((*request)->action != NULL) { // this is not a detached send
- SIMIX_req_comm_wait((*request)->action, -1.0);
+ simcall_comm_wait((*request)->action, -1.0);
finish_wait(request, status);
}
// FIXME for a detached send, finish_wait is not called:
XBT_DEBUG("Wait for one of");
for(i = 0; i < count; i++) {
if((requests[i] != MPI_REQUEST_NULL) && (requests[i]->action != NULL)) {
- print_request(" ", requests[i]);
+ print_request("Waiting any ", requests[i]);
xbt_dynar_push(comms, &requests[i]->action);
map[size] = i;
size++;
}
}
if(size > 0) {
- i = SIMIX_req_comm_waitany(comms);
+ i = simcall_comm_waitany(comms);
+
// FIXME: MPI_UNDEFINED or does SIMIX have a return code?
if (i != MPI_UNDEFINED) {
index = map[i];
finish_wait(&requests[index], status);
+
}
}
xbt_free(map);
// FIXME: check for errors
smpi_datatype_extent(datatype, &lb, &dataext);
// Local copy from root
- smpi_datatype_copy(sendbuf, count, datatype, recvbuf, count, datatype);
+ if (sendbuf && recvbuf)
+ smpi_datatype_copy(sendbuf, count, datatype, recvbuf, count, datatype);
// Receive buffers from senders
//TODO: make a MPI_barrier here ?
requests = xbt_new(MPI_Request, size - 1);
smpi_mpi_startall(size - 1, requests);
for(src = 0; src < size - 1; src++) {
index = smpi_mpi_waitany(size - 1, requests, MPI_STATUS_IGNORE);
+ XBT_DEBUG("finished waiting any request with index %d", index);
if(index == MPI_UNDEFINED) {
break;
}
- smpi_op_apply(op, tmpbufs[index], recvbuf, &count, &datatype);
+ if(op) /* op can be MPI_OP_NULL that does nothing */
+ smpi_op_apply(op, tmpbufs[index], recvbuf, &count, &datatype);
}
for(index = 0; index < size - 1; index++) {
xbt_free(tmpbufs[index]);
