checkgpus/main.cpp at main · dylanjude/checkgpus · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
#include <cstdio>
#include <cstdlib>
#include <string>
#include "mpi.h"
#include "cuda_runtime.h"
#include "helper.h"

void bwtest(int size0, int size1);
void datatest(int size0, int size1);

void setdev(){

  int myid, numprocs, i, j;
  double t_start = 0.0, t_end = 0.0, t = 0.0;
  char cpu_name[MPI_MAX_PROCESSOR_NAME];
  char gpu_name[MPI_MAX_PROCESSOR_NAME];
  char line[128];
  int namelen;

  MPI_Comm_size(MPI_COMM_WORLD, &numprocs);
  MPI_Comm_rank(MPI_COMM_WORLD, &myid);
  MPI_Get_processor_name(cpu_name, &namelen);

  char *s_buf;
  char *r_buf;

  int device, ngpu;
  HANDLE_ERROR(cudaGetDeviceCount(&ngpu));
  device = myid % ngpu;
  HANDLE_ERROR( cudaSetDevice( device) );

  cudaDeviceProp prop;
  HANDLE_ERROR( cudaGetDeviceProperties(&prop, device) );
  sprintf(gpu_name,"%s_%d[%X]", prop.name,device,((unsigned long long*)prop.uuid.bytes)[0]);

  MPI_Status stat;

  if(myid == 0){
    printf("\n----------- Info about matching ranks to gpus and hosts ------------\n");
  }

  for(int p=0; p<numprocs; p++){
    if(myid == p){
      sprintf(line,"Rank %3d on %30s from %20s\n", myid, gpu_name, cpu_name);
      if(p>0){
        MPI_Send(line, 128, MPI_CHAR, 0, 200+p, MPI_COMM_WORLD);
      }
    }
    if(myid == 0){
      if(p>0){
        MPI_Recv(line, 128, MPI_CHAR, p, 200+p, MPI_COMM_WORLD,&stat);
      }
      printf(line);
    }
    MPI_Barrier(MPI_COMM_WORLD);
  }
  if(myid == 0){
    printf("------------------------------------------------\n");
  }
  fflush(stdout);
  MPI_Barrier(MPI_COMM_WORLD);
}


int main(int argc, char* argv[]){

  MPI_Init(&argc, &argv);

  setdev();

  // bwtest(262144,4194304);
  bwtest(4096,4194304);

  datatest(4096,4194304);

  MPI_Finalize();

  return 0;
}