[cfe-commits] [PATCH] First OpenMP patch

[Mahesha HS]

Sorry for the spam to llvm-commit list (to some extent).

Somehow, CFE COMMIT list was got replaced by LLVM COMMIT list during
mail exchange.

--
mahesha


On Fri, Nov 23, 2012 at 9:35 AM, Mahesha HS <mahesha.llvm at gmail.com> wrote:
> On Fri, Nov 23, 2012 at 1:53 AM, Hal Finkel <hfinkel at anl.gov> wrote:
>>
>> ----- Original Message -----
>>> From: "Mahesha HS" <mahesha.llvm at gmail.com>
>>> To: "Hal Finkel" <hfinkel at anl.gov>, "Dmitri Gribenko" <gribozavr at gmail.com>
>>> Cc: "llvm-commits" <llvm-commits at cs.uiuc.edu>, "Joey Gouly" <joey.gouly at arm.com>, "Benjamin Kramer"
>>> <benny.kra at gmail.com>
>>> Sent: Wednesday, November 21, 2012 11:52:39 PM
>>> Subject: Re: [cfe-commits] [PATCH] First OpenMP patch
>>>
>>> That is fine. I will experiment with GCC for different flags like -W,
>>> -Wall etc. And make sure that the implementation is in compatible
>>> with
>>> GCC (when it is not really a bug-to-bug compatible). I think, this
>>> additional implementation will be on top of what the current patch
>>> contains. This, I will send as an another patch.
>>
>> Makes sense.
>>
>>>
>>> Do you guys agree to the internal requirement for two lang options -
>>> OpenMP and NoOpenMP - to keep track of the options -fopenmp and
>>> -fno-openmp? If so, I will change the "comment and test case" as
>>> mentioned by Joey and re-submit the patch.
>>
>> If the only purpose of this is to affect the warnings, then Dmitri's suggestion of using something like -Wsource-uses-openmp is (in my opinion) better.
>
> I agree. As of now, it affects only warnings though I do not know
> about future requirements. So I will remove "NoOpenMP" for the time
> being.
>
> --
> mahesha
>
>
>>
>> Thanks again,
>> Hal
>>
>>>
>>> --
>>> mahesha
>>>
>>>
>>>
>>>
>>>
>>> On Thu, Nov 22, 2012 at 2:09 AM, Hal Finkel <hfinkel at anl.gov> wrote:
>>> > ----- Original Message -----
>>> >> From: "Dmitri Gribenko" <gribozavr at gmail.com>
>>> >> To: "Mahesha HS" <mahesha.llvm at gmail.com>
>>> >> Cc: "Benjamin Kramer" <benny.kra at gmail.com>, "llvm cfe"
>>> >> <cfe-commits at cs.uiuc.edu>
>>> >> Sent: Wednesday, November 21, 2012 11:06:50 AM
>>> >> Subject: Re: [cfe-commits] [PATCH] First OpenMP patch
>>> >>
>>> >> On Wed, Nov 21, 2012 at 6:02 PM, Mahesha HS
>>> >> <mahesha.llvm at gmail.com>
>>> >> wrote:
>>> >> > Though -fopenmp and -fno-openmp are mutually exclusive options,
>>> >> > underlying functionality, that we intended to implement differ a
>>> >> > little. In a nutshell, we, sometime back discussed in CFE DEV
>>> >> > list
>>> >> > to
>>> >> > implement OpenMP as follows. We thought that passing neither
>>> >> > -fopenmp
>>> >> > nor -fno-openmp  is not same as passing only -fno-openmp.
>>> >> >  Following
>>> >> > is the intended implementation. This forced me to add two lang
>>> >> > options
>>> >> > as I could not handle it using only one lang option. If this was
>>> >> > not
>>> >> > there, I guess, all your comments would not have araised.
>>> >> >
>>> >> > if (source contains OpenMP statements) {
>>> >> >      if (neither -fopenmp nor -fno-openmp present) {
>>> >> >          Throw warning.
>>> >> >      }
>>> >> >      else if (only -fopenmp present) {
>>> >> >          Process OpenMP
>>> >> >      }
>>> >> >      else if (only -fno-openmp) {
>>> >> >         Silently ignore OpenMP. Do not throw warning.
>>> >> >      }
>>> >> >     else if (both -fopenmp and -fno-openmp are present) {
>>> >> >         Decide based on which one appears later in the command
>>> >> >         line.
>>> >> >     }
>>> >> > }
>>> >>
>>> >> I think the implementation should use the general diagnostics
>>> >> infrastructure.  For example, by default add something like
>>> >> -Wsource-uses-openmp and disable it in case of -fno-openmp.
>>> >>
>>> >> $ g++ -W -Wall /tmp/zzz.cc
>>> >> /tmp/zzz.cc:3:0: warning: ignoring #pragma omp barrier
>>> >> [-Wunknown-pragmas]
>>> >>
>>> >> By the way, the algorithm above is not consistent with gcc, which
>>> >> ignores the pragma when no -f flag is passed.
>>> >
>>> > I think that we'd like to act the way that gcc does, except that we
>>> > don't want to warn about unused OpenMP pragmas when -fno-openmp is
>>> > given even if -Wall is also specified.
>>> >
>>> >  -Hal
>>> >
>>> >>
>>> >> Dmitri
>>> >>
>>> >> --
>>> >> main(i,j){for(i=2;;i++){for(j=2;j<i;j++){if(!(i%j)){j=0;break;}}if
>>> >> (j){printf("%d\n",i);}}} /*Dmitri Gribenko <gribozavr at gmail.com>*/
>>> >> _______________________________________________
>>> >> cfe-commits mailing list
>>> >> cfe-commits at cs.uiuc.edu
>>> >> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>>> >>
>>> >
>>> > --
>>> > Hal Finkel
>>> > Postdoctoral Appointee
>>> > Leadership Computing Facility
>>> > Argonne National Laboratory
>>>
>>>
>>>
>>> --
>>> mahesha
>>>
>>
>> --
>> Hal Finkel
>> Postdoctoral Appointee
>> Leadership Computing Facility
>> Argonne National Laboratory
>
>
>
> --
> mahesha



-- 
mahesha