Does OpenMP need -Ofast to run properly? - c

Question

Hello i am trying to learn openMP and i am confused by the results, i have pi.c
#include <stdio.h>
#include <omp.h>
#define NUM_THREADS 6
static long num_steps = 1000000000;
double step;
int main(){
int i, nthreads; double pi, sum[NUM_THREADS];
step = 1.0/(double)num_steps;
double start_time, run_time;
omp_set_num_threads(NUM_THREADS);
start_time = omp_get_wtime();
#pragma omp parallel
{
int i, id, nthrds; double x;
id = omp_get_thread_num();
nthrds = omp_get_num_threads();
if(id == 0) nthreads = nthrds;
for(i = id, sum[id] = 0.0; i < num_steps; i+=nthrds){
x = (i+0.5)*step;
sum[id] += 4.0 / (1.0+x*x);
}
}
for(i = 0, pi = 0.0; i < nthreads; i++){
pi += step * sum[i];
}
run_time = omp_get_wtime();
printf("[PI %f TIME %.4fs ON %d THREADS]\n", pi, (run_time - start_time), nthreads);
}
and when i complile with gcc -fopenmp -Wall -Wextra pi.c i get these results:
[PI 3.141593 TIME 3.8663s ON 1 THREADS]
[PI 3.141593 TIME 7.9291s ON 2 THREADS]
[PI 3.141593 TIME 8.4961s ON 3 THREADS]
[PI 3.141593 TIME 10.8343s ON 4 THREADS]
[PI 3.141593 TIME 9.7167s ON 5 THREADS]
[PI 3.141593 TIME 10.0182s ON 6 THREADS]
but when i compile with gcc -fopenmp -Ofast -Wall -Wextra pi.c i get the results i expected:
[PI 3.141593 TIME 1.8380s ON 1 THREADS]
[PI 3.141593 TIME 0.7553s ON 2 THREADS]
[PI 3.141593 TIME 0.5525s ON 3 THREADS]
[PI 3.141593 TIME 0.3930s ON 4 THREADS]
[PI 3.141593 TIME 0.3694s ON 5 THREADS]
[PI 3.141593 TIME 0.3287s ON 6 THREADS]
-O2,-O3 behave similarly to -Ofast and -O1 has results similar to without compiler optimizations, with more threads giving worse results.

Answer 1

Short answer: No, you don't need Ofast to run omp properly.
If you do man gcc, you can see
-Ofast
Disregard strict standards compliance. -Ofast enables all -O3 optimizations.
It also enables optimizations that are not valid for all standard-compliant
programs. It turns on -ffast-math and the Fortran-specific
-fno-protect-parens and -fstack-arrays.
So basically, Ofast turns on O3 optimisations + other optimisations and hence faster.
If you check -ffast-math (set by Ofast) in manual, you can see:
-ffast-math
This option causes the preprocessor macro "__FAST_MATH__" to be defined.
This option is not turned on by any -O option besides -Ofast since it can
result in incorrect output for programs that depend on an exact implementation
of IEEE or ISO rules/specifications for math functions. It may, however, yield
faster code for programs that do not require the guarantees of these
specifications.
Key point here is Ofast disregard strict standards compliance. and it can result in incorrect output for programs that depend on an exact implementation of IEEE or ISO rules/specifications for math functions (which is also mentioned in the comments)
So to summarize, Ofast can give you faster results if you don't care about the standard compliance or catastrophic cancellation (as mentioned by zwol in comments) or there could be others that I don't know..