This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: Gfortran OpenMP SINGLE vs MASTER
- From: Matt Fago <fago at earthlink dot net>
- To: Steve Kargl <sgk at troutmask dot apl dot washington dot edu>
- Cc: fortran at gcc dot gnu dot org
- Date: Wed, 20 Jan 2010 10:32:21 -0700
- Subject: Re: Gfortran OpenMP SINGLE vs MASTER
- References: <20910604.1262191283299.JavaMail.root@elwamui-polski.atl.sa.earthlink.net> <20091230171725.GA49296@troutmask.apl.washington.edu>
On Dec 30, 2009, at 10:17 AM, Steve Kargl wrote:
> On Wed, Dec 30, 2009 at 09:41:23AM -0700, Matt Fago wrote:
>>
>> I seem to have run into an issue with OMP SINGLE on gfortran 4.3.0. For
>> the code snippet:
>>
>> !$OMP SINGLE
>> write(*,*) " -- work --", omp_get_thread_num()
>> ! Do things that must only be done once
>> !$OMP END SINGLE
>>
>> about half-way through the program execution I start getting output from
>> each thread. Perhaps memory corruption (stomping on libgomp), but valgrind
>> does not show anything. If I change it to MASTER everything works ok.
>> Intel Fortran and FSF gcc 4.4.0 also works ok (neither would be surprising
>> if it were a memory issue).
>
> (snip)
>
>> I can also certainly open a bug (I didn't find a previous report), but
>> thought it best to ask first, due to the issue with Redhat and the
>> impending 4.5.
>
> A gfortran bug in a version older than 4.4.x that is
> not present in either 4.4.x or trunk(aka 4.5) is unlikely
> to draw any attention. This is simply a man power issue.
> There are too many bugs (against 4.4 and 4.5) and too few
> worker bees.
>
> If it's possible, it would advantageous to switch the
> latest 4.4.x release.
To close the issue, this is a bug with all versions of FSF gcc 4.3 on x86_64 linux.
It is fixed in 4.4.x, so I have moved to that. I will test 4.5 RC when available.
Would the test case be useful so this issue doesn't crop up again?
Thanks,
Matt