MPI_Recv期间冻结MPI程序

Question

I'm a beginner in MPI parallel programming. I've written this little piece of code to draw the Mandelbrot fracta. The idea is that the first slave will calculate the first half, stick it in a pointer, and send it to master node who is waiting for receiving pointer. The same thing happens for the second node. Finally, the master node should have the result in 2 different variables and will write them in one file.

......
    if((itertab=malloc((sizeof(int)*sizebuffre))) == NULL) { 
        printf("ERREUR , errno : %d (%s) .\n",errno,strerror(errno)); 
        return EXIT_FAILURE; 
    } 
    int rank, size,start,end;

    MPI_Init (&argc, &argv); /* starts MPI */ 
    MPI_Comm_rank (MPI_COMM_WORLD, &rank); /* get current process id */ 
    MPI_Comm_size (MPI_COMM_WORLD, &size); /* get number of processes */
    MPI_Status st;

    /*allocation du tableau de pixel*/ 
    if (rank==1) { 
        xpixel = 0; 
        end = (nbpixelx/MACHINE_NUM); 
        Calcule(xpixel,end); 
        printf("rank 1 start : %d, end : %d\n",xpixel,end); 
        MPI_Send(&itertab,sizebuffre,MPI_INT,0,5,MPI_COMM_WORLD); 
        free(itertab); 
        printf("work done : i am rank 1 \n"); 
    } 
    if (rank==2) {
        xpixel = (nbpixelx/MACHINE_NUM); 
        end = nbpixelx; 
        Calcule(xpixel,end); 
        printf("rank 2 start : %d, end : %d\n",xpixel,end); 
        MPI_Send(&itertab,sizebuffre,MPI_INT,0,6,MPI_COMM_WORLD); 
        printf("work done : i am rank 2 \n"); 
        free(itertab); 
    }

    if (rank==0) { 
        if((itertabA=malloc((sizeof(int)*sizebuffre))) == NULL) { 
            printf("ERREUR d'allocation de itertabA, errno : %d (%s) .\n",errno,strerror(errno)); 
            return EXIT_FAILURE; 
        } 
        if((itertabB=malloc((sizeof(int)*sizebuffre))) == NULL) { 
            printf("ERREUR d'allocation de itertabB, errno : %d (%s) .\n",errno,strerror(errno)); 
            return EXIT_FAILURE; 
        }
        printf("test before reciving result from first slave\n");
        MPI_Recv(itertabA,sizebuffre,MPI_INT,1,5,MPI_COMM_WORLD,&st); 
        printf("result recived  rank 1 \n"); 
        MPI_Recv(itertabB,sizebuffre,MPI_INT,2,6,MPI_COMM_WORLD,&st); 
        printf("result recived rank 2 \n");



    }

    MPI_Finalize(); 
    return EXIT_SUCCESS; 
}

The problem is that my code freezes in the line where the master receives the result from first slave, but I don't know why?

I tried to debug the result. I added some printf to see where it freezes. This is the result:

test before reciving result from first slave
test in calcule function
trairment xpixel 0
trairment xpixel 1
trairment xpixel 2
...snip...
trairment xpixel 399
test after the end off calculating loop
rank 1 start : 0, end : 400
^C

Answer 1

Your MPI code is not working properly, because you are passing the wrong argument to MPI_Send . Your variable itertab is already a pointer to your data buffer, you thus don't need to de-reference it again.

Instead of:

MPI_Send(&itertab,sizebuffre,MPI_INT,0,5,MPI_COMM_WORLD);

do:

MPI_Send(itertab,sizebuffre,MPI_INT,0,5,MPI_COMM_WORLD);

Another issue is that you are accessing non allocated memory, both in your Calcule function, and in the output loop. In the Calcule function, you are writing into itertab[xpixel*nbpixely+ypixel]=iter . This will fail for process 2 , since it allocates only its local part of the itertab buffer. You'll need to subtract an offset for xpixel .

In the output loop, you are reading itertabB with the global index. Here, you should also subtract an offset for xpixel , like so:

fprintf(file,"%f %f %d\n", x, y,itertabB[(xpixel-(nbpixelx/MACHINE_NUM))*nbpixely+ypixel]);