PAPI_lowlevel_flops.c

#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <sys/time.h>
#include <omp.h>
#include "papi.h"

void
LU (double * mat,int num);

#define SIZE 11
double mat[(1 << SIZE) * (1 << SIZE)];

int
main ()
{
  srand (time (NULL));
  int i;
  int EventSet = PAPI_NULL;

  // 諸々の初期化
  if (PAPI_library_init(PAPI_VER_CURRENT) != PAPI_VER_CURRENT) {
    fprintf(stderr, "PAPI library init error!\n");
    exit(1);
  }
  if (PAPI_thread_init((unsigned long (*)(void))(omp_get_num_threads)) != PAPI_OK) {
    fprintf(stderr, "PAPI thread init error.\n");
    exit(1);
  }

  if (PAPI_num_counters() < 2) {
    fprintf(stderr, "No hardware counters here, or PAPI not supported.\n");
    exit(1);
  }
  // 浮動小数点演算の回数を取得する
  PAPI_create_eventset( &EventSet );
  PAPI_add_event(EventSet, PAPI_FP_OPS);

  puts ("Size\tReal_Time\tProcess_Time\tFLPops\tMFlops");

  int nthreads;
#pragma omp parallel
  {
    // 0番スレッドのときはスレッド数を取得する
    if (omp_get_thread_num() == 0)
      nthreads = omp_get_num_threads();
  }

  for (i = 1;i <= SIZE; ++i)
    {
      float rtime,ptime,mflops;
      long long flpops = 0;
      int ret;
      const int num = 1 << i;
      int j;
      // 適当に行列に要素を入れる
      for (j = 0;j < num * num; ++j)
        {
          mat[j] = rand () / 1000.0;
        }

      long long values[nthreads];
      for (j = 0 ; j < nthreads;j++) {
        values[j] = 0;
      }

      struct timeval start,end;
      long long v_start,v_end;
      // 開始時間の取得
      v_start = PAPI_get_virt_usec();
      gettimeofday(&start, NULL);
      // 命令数のカウントの開始
      if ((ret = PAPI_start (EventSet)) != PAPI_OK) {
        fprintf(stderr, "PAPI failed to start counters: %s\n", PAPI_strerror(ret));
        exit(1);
      }
#pragma omp parallel
      {
        LU (mat,num);
        // 命令数のカウントの終了
        if ((ret = PAPI_stop (EventSet,values + omp_get_thread_num ())) != PAPI_OK) {
          fprintf(stderr, "PAPI failed to read counters: %s\n", PAPI_strerror(ret));
          exit(1);
        }
      }
      // 終了時間の取得
      gettimeofday(&end, NULL);
      v_end = PAPI_get_virt_usec();
      rtime = (end.tv_sec - start.tv_sec) + 0.000001 * (end.tv_usec - start.tv_usec);
      ptime = (v_end - v_start) * 1.0e-6;

      // 各スレッドでの実行命令数を足す
      flpops = 0;
      for (j = 0 ; j < nthreads;j++) {
        flpops += values[j];
      }
      // flopsの計算
      mflops = flpops/ptime*1.0e-6;

      printf ("%d\t%f\t%f\t%lld\t%f\n",num,rtime,ptime,flpops,mflops);
    }

}