Comments / Profile of Lockal / Habr

User

Lockal Sep 4 2012 at 19:36

Или gcc -fopenmp -std=c99 -O2 pi.c -fdump-tree-optimized. Но там мало чего интересного. Содержимое блоков OpenMP вынесено в отдельные функции main._omp_fn.0 и main._omp_fn.1, а в main вызывается GOMP_parallel_start, которой передаётся указатель на функцию. Редукция чуть поинтересней.

Версия на чистом си

#include <stdio.h>
#include <math.h>
#include <unistd.h>
#include <sys/times.h>
#include <omp.h>

int main(int argc, char **argv)
{
    const unsigned long numSteps = 500000000; /* default # of rectangles */
    double PI25DT = 3.141592653589793238462643;
    double step, pi = 0, sum=0.0, x;

    #pragma omp parallel
    {
    #pragma omp master
        {
            int cntThreads = omp_get_num_threads();
            printf("OpenMP. number of threads = %d\n", cntThreads);
        }
    }

    clock_t clockStart, clockStop;
    struct tms tmsStart, tmsStop;
    step = 1./(double)numSteps;
    clockStart = times(&tmsStart);
    #pragma omp parallel for private (x), reduction (+:sum)
        for (int i = 0; i < numSteps; i++)
        {
            x = (i + .5) * step;
            sum = sum + 4.0 / (1. + x * x);
        }
    
    pi = sum * step;
    clockStop = times(&tmsStop);
    printf("The value of PI is %lf Error is %lf\n", pi, fabs(pi - PI25DT));
    double secs = (clockStop - clockStart)/(double)sysconf(_SC_CLK_TCK);
    printf("The time to calculate PI was %lf  seconds\n", secs);
    
    return 0;
}

Результат -fdump-tree-optimized

;; Function main._omp_fn.0 (main._omp_fn.0, funcdef_no=23, decl_uid=3556, cgraph_uid=23)

main._omp_fn.0 (void * .omp_data_i)
{
  int cntThreads;
  int D.3622;

<bb 2>:
  D.3622_1 = __builtin_omp_get_thread_num ();
  if (D.3622_1 == 0)
    goto <bb 4>;
  else
    goto <bb 3>;

<bb 3>:
  return;

<bb 4>:
  cntThreads_2 = __builtin_omp_get_num_threads ();
  __printf_chk (1, "OpenMP. number of threads = %d\n", cntThreads_2); [tail call]
  goto <bb 3>;

}



;; Function main._omp_fn.1 (main._omp_fn.1, funcdef_no=24, decl_uid=3560, cgraph_uid=27)

Removing basic block 10
Removing basic block 11
main._omp_fn.1 (struct .omp_data_s.1 * .omp_data_i)
{
  double pretmp.15;
  long unsigned int D.3620;
  long unsigned int D.3618;
  double D.3617;
  double D.3616;
  long unsigned int * {ref-all} D.3613;
  double D.3611;
  double D.3610;
  double D.3609;
  double x;
  double D.3606;
  double D.3605;
  int D.3604;
  int D.3602;
  int tt.5;
  int q.4;
  int D.3599;
  int D.3598;
  int i;
  double sum;

<bb 2>:
  D.3598_9 = __builtin_omp_get_num_threads ();
  D.3599_10 = __builtin_omp_get_thread_num ();
  q.4_11 = 500000000 / D.3598_9;
  tt.5_12 = 500000000 % D.3598_9;
  if (D.3599_10 < tt.5_12)
    goto <bb 9>;
  else
    goto <bb 3>;

<bb 3>:
  # q.4_4 = PHI <q.4_14(9), q.4_11(2)>
  # tt.5_5 = PHI <0(9), tt.5_12(2)>
  D.3602_15 = q.4_4 * D.3599_10;
  i_16 = D.3602_15 + tt.5_5;
  D.3604_17 = i_16 + q.4_4;
  if (i_16 >= D.3604_17)
    goto <bb 6>;
  else
    goto <bb 4>;

<bb 4>:
  pretmp.15_56 = .omp_data_i_21(D)->step;

<bb 5>:
  # sum_1 = PHI <0.0(4), sum_27(5)>
  # i_3 = PHI <i_16(4), i_28(5)>
  D.3605_19 = (double) i_3;
  D.3606_20 = D.3605_19 + 5.0e-1;
  x_23 = D.3606_20 * pretmp.15_56;
  D.3609_24 = x_23 * x_23;
  D.3610_25 = D.3609_24 + 1.0e+0;
  D.3611_26 = 4.0e+0 / D.3610_25;
  sum_27 = D.3611_26 + sum_1;
  i_28 = i_3 + 1;
  if (i_28 != D.3604_17)
    goto <bb 5>;
  else
    goto <bb 6>;

<bb 6>:
  # sum_2 = PHI <0.0(3), sum_27(5)>
  D.3613_30 = &.omp_data_i_21(D)->sum;
  D.3620_31 = MEM[(long unsigned int * {ref-all}).omp_data_i_21(D) + 8B];

<bb 7>:
  # D.3620_6 = PHI <D.3620_31(6), D.3620_36(7)>
  D.3616_33 = VIEW_CONVERT_EXPR<double>(D.3620_6);
  D.3617_34 = D.3616_33 + sum_2;
  D.3618_35 = VIEW_CONVERT_EXPR<long unsigned int>(D.3617_34);
  D.3620_36 = __sync_val_compare_and_swap_8 (D.3613_30, D.3620_6, D.3618_35);
  if (D.3620_6 != D.3620_36)
    goto <bb 7>;
  else
    goto <bb 8>;

<bb 8>:
  return;

<bb 9>:
  q.4_14 = q.4_11 + 1;
  goto <bb 3>;

}



;; Function main (main, funcdef_no=22, decl_uid=3527, cgraph_uid=22) (executed once)

main (int argc, char * * argv)
{
  double secs;
  struct tms tmsStop;
  struct tms tmsStart;
  clock_t clockStop;
  clock_t clockStart;
  double sum;
  double pi;
  double step;
  double D.3553;
  long int D.3552;
  double D.3551;
  long int D.3550;
  double D.3549;
  double D.3548;
  struct .omp_data_s.1 .omp_data_o.2;

<bb 2>:
  __builtin_GOMP_parallel_start (main._omp_fn.0, 0B, 0);
  main._omp_fn.0 (0B);
  __builtin_GOMP_parallel_end ();
  clockStart_6 = times (&tmsStart);
  .omp_data_o.2.sum = 0.0;
  .omp_data_o.2.step = 2.00000000000000012456318291555971283779413738557195756584e-9;
  __builtin_GOMP_parallel_start (main._omp_fn.1, &.omp_data_o.2, 0);
  main._omp_fn.1 (&.omp_data_o.2);
  __builtin_GOMP_parallel_end ();
  sum_7 = .omp_data_o.2.sum;
  step_8 = .omp_data_o.2.step;
  pi_9 = sum_7 * step_8;
  clockStop_10 = times (&tmsStop);
  D.3548_11 = pi_9 - 3.141592653589793115997963468544185161590576171875e+0;
  D.3549_12 = ABS_EXPR <D.3548_11>;
  __printf_chk (1, "The value of PI is %lf Error is %lf\n", pi_9, D.3549_12);
  D.3550_13 = clockStop_10 - clockStart_6;
  D.3551_14 = (double) D.3550_13;
  D.3552_15 = sysconf (2);
  D.3553_16 = (double) D.3552_15;
  secs_17 = D.3551_14 / D.3553_16;
  __printf_chk (1, "The time to calculate PI was %lf  seconds\n", secs_17);
  tmsStart ={v} {CLOBBER};
  tmsStop ={v} {CLOBBER};
  return 0;

}