test_omp_write.c

/* -----------------------------------------------------------------------------
MIT License

Copyright (c) 2022 CSC HPC

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
------------------------------------------------------------------------------ */

#include "shmem_tests.h"
#include <omp.h>
#ifdef USE_INTRINSICS
#include <immintrin.h>
#endif

/* Borrowed from util-linux-2.13-pre7/schedutils/taskset.c */
static char *cpuset_to_cstr(cpu_set_t *mask, char *str)
{
  char *ptr = str;
  int i, j, entry_made = 0;
  for (i = 0; i < CPU_SETSIZE; i++) {
    if (CPU_ISSET(i, mask)) {
      int run = 0; 
      entry_made = 1; 
      for (j = i + 1; j < CPU_SETSIZE; j++) {
        if (CPU_ISSET(j, mask)) run++;
        else break;
      }
      if (!run)
        sprintf(ptr, "%d,", i);
      else if (run == 1) {
        sprintf(ptr, "%d,%d,", i, i + 1);
        i++; 
      } else {
        sprintf(ptr, "%d-%d,", i, i + run);
        i += run;
      }
      while (*ptr != 0) ptr++;
    }
  }
  ptr -= entry_made;
  *ptr = 0;
  return(str);
}

double mysecond()
{
        struct timeval tp;
        struct timezone tzp;
        int i;

        i = gettimeofday(&tp,&tzp); 
        return ( (double) tp.tv_sec + (double) tp.tv_usec * 1.e-6 );
}

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

  double times[NTIMES];
  double avgtime = 0;
  double mintime = FLT_MAX;
  double maxtime = 0;

  int nthreads;
#pragma omp parallel
  nthreads = omp_get_num_threads();

  if (nthreads != 2) {
    printf("Only two threads can be used\n");
    return -1;
  }

  cpu_set_t coremask;
  char clbuf[7 * CPU_SETSIZE];
  sched_getaffinity(0, sizeof(coremask), &coremask);
  memset(clbuf, 0, sizeof(clbuf));
  cpuset_to_cstr(&coremask, clbuf);
  printf("Original affinity: %s\n", clbuf);
  fflush(stdout);

  double *a, *b;

  printf("Array size: %d MB\n", BUF_SIZE / 1000000);
#ifdef USE_AVX      
  printf("Using omp SIMD (note: compiler might still optimize for memcpy)\n");
#elif defined USE_INTRINSICS
  #ifdef USE_STREAMING
  printf("Using intrinsics with non-temporal stores\n");
  #else
  printf("Using intrinsics\n");
  #endif
#else
  printf("Using memcpy\n");
#endif
  
  
#pragma omp parallel
  {
  int tid = omp_get_thread_num();
  // child
  if (1 == tid) {

    char clbuf[7 * CPU_SETSIZE];
    int ncpus = CPU_COUNT(&coremask);
    CPU_ZERO(&coremask);
    CPU_SET(ncpus-1, &coremask);
    sched_setaffinity(0, sizeof(coremask), &coremask); 
    memset(clbuf, 0, sizeof(clbuf));
    cpuset_to_cstr(&coremask, clbuf);
    printf("Affinity in child: %s\n", clbuf);
    fflush(stdout);

    // a = (double *) malloc(BUF_SIZE * sizeof(double));
    posix_memalign((void *)&a, 32, BUF_SIZE * sizeof(double));
    // invalidate data
    for (int i=0; i < BUF_SIZE; i++) {   
       a[i] = i;
    }

    // synchronize with parent
    #pragma omp barrier

    // write data

    for (int k=0; k < NTIMES; k++)
      {
        times[k] = mysecond();
        // write data
#ifdef USE_AVX
      #pragma omp simd aligned(a, b : 32)
      for (int i=0; i < BUF_SIZE; i++)
        b[i] = a[i];
#elif defined USE_INTRINSICS
      for (int i=0; i < BUF_SIZE; i += 4) {
        __m256d src = _mm256_load_pd(&a[i]);
        #ifdef USE_STREAMING
        _mm256_stream_pd(&b[i], src);
        #else
        _mm256_store_pd(&b[i], src);
        #endif
      }
#else
        memcpy(b, a, sizeof(double) * BUF_SIZE);
#endif
        times[k] = mysecond() - times[k];
      }

    for (int k=1; k < NTIMES; k++) /* note -- skip first iteration */
      {
        avgtime = avgtime + times[k];
        mintime = MIN(mintime, times[k]);
        maxtime = MAX(maxtime, times[k]);
      }
    
    size_t bytes = BW_CONVENTION * sizeof(double) * BUF_SIZE;
  
    printf("Function    Best Rate MB/s  Avg time     Min time     Max time\n");
    avgtime = avgtime / (double)(NTIMES-1);

    printf("%s%12.1f  %11.6f  %11.6f  %11.6f\n", "omp write",
           1.0E-06 * bytes/mintime,
           avgtime,
           mintime,
           maxtime);

    #pragma omp barrier

  // parent
  } else {

    char clbuf[7 * CPU_SETSIZE];
    cpu_set_t coremask;
    CPU_ZERO(&coremask);
    CPU_SET(0, &coremask);
    sched_setaffinity(0, sizeof(coremask), &coremask); 
    memset(clbuf, 0, sizeof(clbuf));
    cpuset_to_cstr(&coremask, clbuf);
    printf("Affinity in parent: %s\n", clbuf);
    fflush(stdout);

    // b = (double *) malloc(BUF_SIZE * sizeof(double));
    posix_memalign((void *)&b, 32, BUF_SIZE * sizeof(double));
    
    // invalidate data
    for (int i=0; i < BUF_SIZE; i++) {
       b[i] = -1;
    }

    // synchronize with child
    #pragma omp barrier

    // synchronize after write
    #pragma omp barrier
    // Use Kahan's algorithm for summation
    double checksum = b[0];
    double tmp_c = 0.0;
    for (size_t i=1; i < BUF_SIZE; i++) {
      double y = b[i] - tmp_c;
      double t = checksum + y;
      tmp_c = (t - checksum) - y;
      checksum = t;
    }
    checksum /= (double) BUF_SIZE;
    
    printf(HLINE);
    printf("check: %f %f\n", checksum, 0.5*(BUF_SIZE-1));

  }
  } // end omp parallel
}