MPI-CUDA-C++ Programing using MPI ,CUDA calls and C++

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello

I wrote a simple program in C++ using file extension as .cu and used MPI libraries.
My program is as follows;
#include
#include
#include
#include<unistd.h>
#include<cuda.h>
#include"mpi.h"

using namespace std;

int main(int argc, char *argv)
{
int MyRank, NumberOfProcessors, Root = 0;

    MPI::Init(argc, argv);
    NumberOfProcessors = MPI::COMM_WORLD.Get_size();
    MyRank = MPI::COMM_WORLD.Get_rank();

    if(MyRank == Root)
            cout<<"My Rank is ..";

    MPI::Finalize();

    return(0);

}

When i am compiling i am getting following error.

I am compiling with the following command
nvcc -I/usr/local/mpich2-1.0.7/include -L/usr/local/mpich2-1.0.7/lib -lmpich -DMPICH_IGNORE_CXX_SEEK --host-compilation=C++ -o Sample_Program Sample_Programm_C++.cu

And the error is
/tmp/tmpxft_00001609_00000000-11_Sample_Programm_C++.o: In function main': tmpxft_00001609_00000000-10_Sample_Programm_C++.ii:(.text+0x12f): undefined reference to MPI::Init(int&, char**&)’
tmpxft_00001609_00000000-10_Sample_Programm_C++.ii:(.text+0x134): undefined reference to MPI::COMM_WORLD' tmpxft_00001609_00000000-10_Sample_Programm_C++.ii:(.text+0x141): undefined reference to MPI::COMM_WORLD’
tmpxft_00001609_00000000-10_Sample_Programm_C++.ii:(.text+0x165): undefined reference to `MPI::Finalize()’

Can anyone tell me what is the problem??

2

Please use code tags next time!

Did u used the commands right, i mean, did you compiled this with NVCC, cause there’s AFAIK no need!

3

Thank you for your quick reply

I am compiling using the following command

nvcc -I/usr/local/mpich2-1.0.7/include -L/usr/local/mpich2-1.0.7/lib -lmpich -DMPICH_IGNORE_CXX_SEEK --host-compilation=C++ -o Sample_Program Sample_Programm_C++.cu

Please see the command and tell me if any changes hev to be made.

4

Why don’t you use the default makefile delivered by NVidia SDK?

There you can include the namespaces/libraries!

5

Your need to link the C++ support library. Add -lmpichcxx. And read the MPICH documentation next time. This has absolutely nothing to do with CUDA.

6

Sorry for my ignorance. I dont know about NVIDIA SDK.

Till now i was writing programs in cuda, calling MPI libraries. Which is the mixed programing.

Then i thought can we write C++ also. They say that CUDA is extension of C. Will it support C++ also??

Thats why i wrote this simple program to see whether we can compile it or not.

I have used same flags expect the --host-compilation=C++ to compile the program CUDA which calls the MPI .

Then i surfed the net and got that we have to include the flag to compile C++.

I have got the above mentioned error.

Now can you guide me??

7

Thank you very much.
Its working after including that library.