I have this code for testing MPI_Irecv and MPI_Isend
if(rank==1){
int cc;
MPI_Request request;
MPI_Status status;
int flag;
do{
MPI_Irecv(&cc, 1, MPI_INT, MPI_ANY_SOURCE, MPI_ANY_TAG, allcomm, &request);
//MPI_Wait(&request, &status);
MPI_Test(&request, &flag, &status);
if (flag != 0) {
printf("recv : %d, slave : %d\n", cc, status.MPI_SOURCE);
}
} while(flag != 0);
}
if(rank==0){
int cc=0;
MPI_Request request;
for(int i=1;i<10;i++){
cc+=1;
MPI_Isend(&cc, 1, MPI_INT, 1, 0, allcomm, &request);
}
}
If I execute this code, sometimes it prints cc from 1 to 9 and sometimes it prints nothing.
It makes sense, since the process 0 may be executed after process 1 finish.
However, if I let process 1 be a infinity loop like this:
if(rank==1){
int cc;
MPI_Request request;
MPI_Status status;
int flag;
do{
MPI_Irecv(&cc, 1, MPI_INT, MPI_ANY_SOURCE, MPI_ANY_TAG, allcomm, &request);
//MPI_Wait(&request, &status);
MPI_Test(&request, &flag, &status);
if (flag != 0) {
printf("recv : %d, slave : %d\n", cc, status.MPI_SOURCE);
}
} while(true);
}
I think process 1 will print cc from 1 to 9 every time, but it won't.
If I insert MPI_Wait() in process 1, it seems process 1 will print cc from 1 to 9 every time. But it will also wait forever if there is nothing sent by other process and I want non-blocking communication actually.
Reply to comments:
1.About version: mpiexec --version mpiexec (OpenRTE) 2.1.1
2.I don't get the point what 'MPI progress thread' means.
Actually I want each process can communicate with each other without blocking and get the information in the buffer each time when receive.
3.executable code following
#include <stdio.h>
#include <stdlib.h>
#include <mpi.h>
#include <unistd.h>
int main(void)
{
MPI_Comm allcomm;
int i;
int size, rank, irank;
int tablesize, localtablesize;
int *table,*table2;
allcomm = MPI_COMM_WORLD;
MPI_Init(NULL, NULL);
MPI_Comm_size(allcomm, &size);
MPI_Comm_rank(allcomm, &rank);
if(rank==1){
int cc;
MPI_Request request;
MPI_Status status;
int flag;
do{
MPI_Irecv(&cc, 1, MPI_INT, MPI_ANY_SOURCE, MPI_ANY_TAG, allcomm, &request);
//MPI_Wait(&request, &status);
MPI_Test(&request, &flag, &status);
if (flag != 0) {
printf("recv : %d, slave : %d\n", cc, status.MPI_SOURCE);
}
} //while(flag!=0);
while(true);
}
if(rank==0){
int cc=0;
MPI_Request request;
for(int i=1;i<10;i++){
cc+=1;
MPI_Isend(&cc, 1, MPI_INT, 1, 0, allcomm, &request);
}
}
MPI_Finalize();
}
Your program is incorrect.
For example, if rank 1
is faster than rank 0
, the first MPI_Test()
will not match any message and then go straight to MPI_Finalize()
Bottom line, this program might print between 0 and 10 lines. This program will only print 10 lines if rank 1
is somehow lagging behind rank 0
, but since both ranks are not synchronized, this is quite unlikely to happen.
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.