This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH][libstdc++-v3 parallel mode] PR 33893 work correctly even if omp_get_dynamic()
- From: Paolo Carlini <pcarlini at suse dot de>
- To: Johannes Singler <singler at ira dot uka dot de>
- Cc: libstdc++ <libstdc++ at gcc dot gnu dot org>, gcc-patches at gcc dot gnu dot org
- Date: Wed, 21 Nov 2007 23:03:11 +0100
- Subject: Re: [PATCH][libstdc++-v3 parallel mode] PR 33893 work correctly even if omp_get_dynamic()
- References: <4744137B.2050701@ira.uka.de>
Hi Johannes,
>> Johannes Singler wrote:
>>
>>> I'm aware of that, and got a pointer to a detailed instruction on how to
>>> do that by Benjamin Kosnik only recently. I already tried to format the
>>> changed parts according to the guidelines.
>>>
>> Sorry, frankly this is not true: there are obviously many lines wider
>> than 80 columns (remember, by the way, that it means 80 overall, CR
>> included, therefore from 0 to 78 of actual code)
>>
>
> Well, I meant the code parts that were actually changed. Anyway, now I
> reformatted all concerned files. I will process the other files later.
> New patch attached.
>
Thanks for your effort, I appreciate that.
> Major revision of the parallel mode algorithms, to work correctly even
> if omp_get_dynamic(), PR 33893.
> Also, the quicksort variants work correctly if !omp_get_nested().
> However, the speedup is limited to 2 in this case. Getting rid of this
> limitation is very elaborate since OpenMP 2.5 does not support barriers
> among a subset of a team. Maybe things get better with OpenMP 3.0, I
> still have to look into this.
>
> Because the process included a lot of formatting cleanup, I'm also
> attaching a version of the patch ignoring white space, for better
> readability.
>
> Please approve.
>
A few days passed since you posted the first version of the patch, which
already contained the substantive changes, and nobody else commented
besides me (Benjamin, in particular), therefore I assume there no other
comments or remarks. Please commit, patch is OK.
Thanks.
Paolo.