a Mac(port) problem ?

[pagani laurent]

Hi,

Not sure this is totally relevent to MacPorts Users who usually discuss install problems, but I have a weird one here and you might help!

Testing OpenMP on my Mac, I find that the output is slightly random when running in parallel mode while it is correct in sequential mode.
the same test on an Ubuntu machine did not reveal the same problem.

I use Macport gcc7.3 :

OpenMP>gcc --version
gcc (MacPorts gcc7 7.3.0_0) 7.3.0

I compile with or without the -fopenmp option and in parallel mode I get this :

OpenMP>gcc -fopenmp -o essai_par 2boucles.c
OpenMP>essai_par
a[0,1] 2, a[99,199] 20000, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998
OpenMP>essai_par
a[0,1] 2, a[99,199] 20000, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998
OpenMP>essai_par
a[0,1] 2, a[99,199] 20000, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998
OpenMP>essai_par
a[0,1] 2, a[99,199] 20000, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998
OpenMP>essai_par
a[0,1] 2, a[99,199] 20000, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998
OpenMP>essai_par
a[0,1] 0, a[99,199] 20000, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998
OpenMP>essai_par
a[0,1] 2, a[99,199] 0, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998

On the 6th call, I got zero instead of 2, on the 7th call, I got zero instead of 20000. No need to say this is totally random. I use the #pragma parallel private
I also tried with the #pragma parallel default(shared) and I got this :

OpenMP>gcc -fopenmp -o essai_pardef 2boucles.c
OpenMP>essai_pardef
a[0,1] 2, a[99,199] 0, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998
OpenMP>essai_pardef
a[0,1] 0, a[99,199] 0, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998
OpenMP>essai_pardef
a[0,1] 2, a[99,199] 20000, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998
OpenMP>essai_pardef
a[0,1] 2, a[99,199] 20000, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998
OpenMP>essai_pardef
a[0,1] 2, a[99,199] 20000, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998

got the first two wrong, then a few correct ones.

I returned to gcc5.5, just in case 7.3 would be faulty :

OpenMP>gcc --version
gcc (MacPorts gcc5 5.5.0_0) 5.5.0
OpenMP>gcc -fopenmp -o essai_pardef 2boucles.c
OpenMP>essai_pardef
a[0,1] 1, a[99,199] 0, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998
OpenMP>essai_pardef
a[0,1] 2, a[99,199] 20000, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998
OpenMP>essai_pardef
a[0,1] 2, a[99,199] 0, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998
OpenMP>essai_pardef
a[0,1] 2, a[99,199] 0, a[999,999] 1000000
b[1] 2, b[100] 200, b[999] 1998

and it looks even worse.

Is it a problem with the Mac (MBP, OS X 10.12.6) or a problem with the MP gcc ?

I attach the test program if one wants to test it on another Mac machine.

Laurent



"S'il n'y a pas de solution, c'est qu'il n'y a pas de problème" (devise Shadok)

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.macports.org/pipermail/macports-users/attachments/20180702/2ae117e9/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 2boucles.c
Type: application/octet-stream
Size: 817 bytes
Desc: not available
URL: <http://lists.macports.org/pipermail/macports-users/attachments/20180702/2ae117e9/attachment.obj>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.macports.org/pipermail/macports-users/attachments/20180702/2ae117e9/attachment-0001.html>