Re: [patch] Add support for #pragma GCC unroll

Re: [patch] Add support for #pragma GCC unroll

[Bernhard Reutner-Fischer]

On 20 November 2017 at 12:26, Eric Botcazou <ebotcazou@adacore.com> wrote:
>> If anybody finds the time to push the corresponding Fortran changes then I'd
>> be grateful. I won't have time for this until end of stage 1...
>>
>> https://gcc.gnu.org/ml/fortran/2015-02/msg00014.html
>
> OK, I'm going to merge it in the main patch.

[CCing fortran@]
Thanks alot in advance!
>
> --
> Eric Botcazou

Re: [patch] Add support for #pragma GCC unroll

[Steve Kargl]

On Mon, Nov 20, 2017 at 12:57:47PM +0100, Bernhard Reutner-Fischer wrote:
> On 20 November 2017 at 12:26, Eric Botcazou <ebotcazou@adacore.com> wrote:
> >> If anybody finds the time to push the corresponding Fortran changes then I'd
> >> be grateful. I won't have time for this until end of stage 1...
> >>
> >> https://gcc.gnu.org/ml/fortran/2015-02/msg00014.html
> >
> > OK, I'm going to merge it in the main patch.
> 
> [CCing fortran@]
> Thanks alot in advance!

The URL points to a nearly 3 year old patch.  I noticed
that there is no documentation of the new Fortran directive.
Section 7.2 of gfortran.info should be updated.  In particular,
does '!GCC$ UNROLL 4' affect only the immediately following
DO-LOOP or all DO-LOOPs that follow the directive until another
'GCC$ UNROLL ...' is found?  How does this new directive interface
with OpenMP and OpenACC?

-- 
Steve

Re: [patch] Add support for #pragma GCC unroll

[Bernhard Reutner-Fischer]

On 20 November 2017 at 15:45, Steve Kargl
<sgk@troutmask.apl.washington.edu> wrote:
> On Mon, Nov 20, 2017 at 12:57:47PM +0100, Bernhard Reutner-Fischer wrote:
>> On 20 November 2017 at 12:26, Eric Botcazou <ebotcazou@adacore.com> wrote:
>> >> If anybody finds the time to push the corresponding Fortran changes then I'd
>> >> be grateful. I won't have time for this until end of stage 1...
>> >>
>> >> https://gcc.gnu.org/ml/fortran/2015-02/msg00014.html
>> >
>> > OK, I'm going to merge it in the main patch.
>>
>> [CCing fortran@]
>> Thanks alot in advance!
>
> The URL points to a nearly 3 year old patch.  I noticed
> that there is no documentation of the new Fortran directive.
> Section 7.2 of gfortran.info should be updated.  In particular,
> does '!GCC$ UNROLL 4' affect only the immediately following
> DO-LOOP or all DO-LOOPs that follow the directive until another
> 'GCC$ UNROLL ...' is found?  How does this new directive interface
> with OpenMP and OpenACC?

The documentation for the directive is missing indeed. We can fix this
during stage3.

Currently the directive works on the whole function (see
gfc_cfun_has_unroll()) and instructs the loop-optimizers to run on
that function.
The loop-optimizers will discover the ANNOTATE_EXPR and act accordingly.
Richard B. already noted that the RTL unroller might do more than
intended, see https://gcc.gnu.org/ml/gcc-patches/2017-11/msg01468.html
I expect updates to the C and C++ in this area to be reflected to Fortran too.

The interaction with OpenMP and OpenACC in the Fortran FE is the same
as in the other frontends, obviously.
Eric's current respin of Mikes patch is here, FYI:
https://gcc.gnu.org/ml/gcc-patches/2017-11/msg01452.html

thanks,

Re: [patch] Add support for #pragma GCC unroll

[Eric Botcazou]

> The documentation for the directive is missing indeed. We can fix this
> during stage3.

Someone who speaks Fortran will have to write it down...

> Currently the directive works on the whole function (see
> gfc_cfun_has_unroll()) and instructs the loop-optimizers to run on
> that function.

gfc_cfun_has_unroll is superfluous and has already been dropped because the 
flag will be set by the middle-end, but this doesn't change the behavior.

> The loop-optimizers will discover the ANNOTATE_EXPR and act accordingly.
> Richard B. already noted that the RTL unroller might do more than
> intended, see https://gcc.gnu.org/ml/gcc-patches/2017-11/msg01468.html
> I expect updates to the C and C++ in this area to be reflected to Fortran
> too.

Yes, it's a generic issue.

-- 
Eric Botcazou