[cfe-dev] Clang/LLVM OpenMP program not spawning threads

andreybokhanko at gmail.com andreybokhanko at gmail.com
Tue Jul 28 15:32:34 PDT 2015


Jack,

Here is what 3.7 release notes (http://llvm.org/viewvc/llvm-project/cfe/branches/release_37/docs/ReleaseNotes.rst?r1=242225&r2=243164) say at the moment:

"OpenMP 3.1 is fully supported, but disabled by default. To enable it, please use ``-fopenmp=libomp`` command line option."

IMHO, this is the best description we can provide. If you believe it can be improved, please submit your changes.

As for enabling libomp by default, Hans is right -- it's definitely too late for 3.7.

Yours,
Andrey

Отправлено с iPad

> 28 июля 2015 г., в 20:44, Jack Howarth <howarth.mailing.lists at gmail.com> написал(а):
> 
> Hans,
>     If we miss the 3.7.0 release, the fact that the LLVM OpenMP
> library is defaulted off in the 3.7.0 release should be clearly stated
> on the first mention of OpenMP library support in the release notes.
> We should also apologize for this suboptimal situation and promise to
> remedy it in the 3.7.1 release. Frankly admitting this defect in the
> 3.7.0 release might at least mitigate the amount of needless traffic
> on the mailing lists from users befuddled by this unexpected
> *feature*.
>            Jack
> 
>> On Tue, Jul 28, 2015 at 12:59 PM, Hans Wennborg <hans at chromium.org> wrote:
>>> On Tue, Jul 28, 2015 at 7:11 AM, Hal Finkel <hfinkel at anl.gov> wrote:
>>> ----- Original Message -----
>>>> From: "David Chisnall" <David.Chisnall at cl.cam.ac.uk>
>>>> To: "Anton Korobeynikov" <anton at korobeynikov.info>
>>>> Cc: "Jaime Arteaga" <jandres742 at gmail.com>, "cfe-dev" <cfe-dev at cs.uiuc.edu>
>>>> Sent: Tuesday, July 28, 2015 6:18:51 AM
>>>> Subject: Re: [cfe-dev] Clang/LLVM OpenMP program not spawning threads
>>>> 
>>>> On 28 Jul 2015, at 12:13, Anton Korobeynikov
>>>> <anton at korobeynikov.info> wrote:
>>>>> 
>>>>> Probably for 3.7 we'd need to issue a warning / error for -fopenmp
>>>>> w/o libomp.
>>>> 
>>>> And make everyone using OpenMP modify their build systems for a
>>>> single release of clang, when we could just make -fopenmp do the
>>>> right thing?
>>> 
>>> I agree that this would be quite unfortunate. It took some time to have a CMake build system for the runtime library that everyone has been satisfied with. I'm under the impression that we could change the default at this point, but I'm not sure what that would require in terms of the release cycle. I've cc'd Chandler, as he's been driving the project-level QoI requirements for this, and Hans for the obvious reason ;)
>> 
>> As for the release cycle, I'd still be willing to take a merge for
>> this, but not for much longer. I'm planning to cut RC2 this week (late
>> Wednesday, or Thursday) and after that it's bug fixes only. The
>> question of merging is still very hypothetical though, as there hasn't
>> been any change on trunk that I'm aware of.
>> 
>> - Hans
>> 
>> _______________________________________________
>> cfe-dev mailing list
>> cfe-dev at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
> 
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev




More information about the cfe-dev mailing list