Coarrays from Fortran 2008/2018 are native Fortran high-level abstractions that are supported by a range of libraries, including MPI.
By using htop or other CPU monitor, you can see that multiple CPU cores are used.
Compute value of Pi iteratively:
cafrun coarray/coarray_piYou can optionally specify the resolution of Pi, say 1e-8:
cafrun coarray/coarray_pi 1e-8Comparing gfortran and ifort coarray performance (computation time in seconds on i7-4650, 4 threads). -O3 was used for both compilers.
Notice that ifort is over 5x faster than gfortran.
YES this was using pi2008.f90 for both, to ensure that Fortran 2018 co_sum() didn't have a disadvantage over the explicit Fortran 2008 loop.
The performance of co_sum was essentially the same in pi.f90 as in pi2008.f90.
| dx | gfortran 7.2.0 | ifort 18.1 |
|---|---|---|
| 1e-7 | 0.254 | 0.049 |
| 1e-8 | 2.72 | 0.489 |
| 1e-9 | 26.0 | 4.88 |