Peformance of gfortran12 with pi

[tenomoto]

I have the following benchmark with my FreeBSD 13.1 box. I compiled OpenCoarrays 2.10.1 against openmpi-4.1.4 installed using pkg. I had to promote float to double to avoid Infinity. The result is impressive. CPU is two gen newer and gfortran and OpenCoarrays must have been improved over versions.

dx	gfortran 12.2.0
1e-7	0.161E-01
1e-8	0.317E+00
1e-9	0.174E+01

% sysctl hw.model hw.ncpu
hw.model: Intel(R) Core(TM) i5-6500 CPU @ 3.20GHz
hw.ncpu: 4
% caf -O3 -fdefault-real-8 pi.f90 -o pi
% cafrun -n 4 ./pi 1e-7
number of Fortran coarray images:  4
 approximating pi in     20000000  steps.
 pi:   3.1415926535897931        iterated pi:    3.1409395631965324     
pi error 0.653E-03
Elapsed wall clock time  0.161E-01 seconds, using  4 images.
% cafrun -n 4 ./pi 1e-8
number of Fortran coarray images:  4
 approximating pi in    200000000  steps.
 pi:   3.1415926535897931        iterated pi:    3.1413861282747124     
pi error 0.207E-03
Elapsed wall clock time  0.317E+00 seconds, using  4 images.
% cafrun -n 4 ./pi 1e-9
number of Fortran coarray images:  4
 approximating pi in   1999999999  steps.
 pi:   3.1415926535897931        iterated pi:    3.1415049838804494     
pi error 0.877E-04
Elapsed wall clock time  0.174E+01 seconds, using  4 images.