This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PING] [PING] Make the OpenACC C++ acc_on_device wrapper "always inline"
- From: Thomas Schwinge <thomas at codesourcery dot com>
- To: Jakub Jelinek <jakub at redhat dot com>
- Cc: Nathan Sidwell <nathan at acm dot org>, GCC Patches <gcc-patches at gcc dot gnu dot org>
- Date: Tue, 6 Jun 2017 13:16:03 +0200
- Subject: Re: [PING] [PING] Make the OpenACC C++ acc_on_device wrapper "always inline"
- Authentication-results: sourceware.org; auth=none
- References: <56327E51.1090507@acm.org> <568AB69F.4020502@acm.org> <568D3764.1050104@acm.org> <5632B856.4050509@acm.org> <87h90ba8u8.fsf@euler.schwinge.homeip.net> <87wp8yv7xq.fsf@hertz.schwinge.homeip.net> <871sqx1v43.fsf@euler.schwinge.homeip.net> <20170606065821.GJ2154@tucnak>
Hi Jakub!
On Tue, 6 Jun 2017 08:58:21 +0200, Jakub Jelinek <jakub@redhat.com> wrote:
> On Tue, Jun 06, 2017 at 08:35:40AM +0200, Thomas Schwinge wrote:
> > > > commit 9cc3a384c17e9f692f7864c604d2e2f9fbf0bac9
> > > > Author: Thomas Schwinge <thomas@codesourcery.com>
> > > > Date: Tue May 23 13:21:14 2017 +0200
> > > >
> > > > Make the OpenACC C++ acc_on_device wrapper "always inline"
> > > >
> > > > libgomp/
> > > > * openacc.h [__cplusplus] (acc_on_device): Mark as "always
> > > > inline".
> > > > * testsuite/libgomp.oacc-c-c++-common/acc-on-device-2.c: Remove
> > > > file; test cases already present...
> > > > * testsuite/libgomp.oacc-c-c++-common/acc_on_device-1.c: ... in
> > > > this file. Update.
> > > > * testsuite/libgomp.oacc-c-c++-common/acc-on-device.c: Remove
> > > > file; test cases now present...
> > > > * testsuite/libgomp.oacc-c-c++-common/acc_on_device-2.c: ... in
> > > > this new file.
> > > > * testsuite/libgomp.oacc-c-c++-common/parallel-dims.c: Update.
>
> I don't like this very much.
Thanks for having a look. Would you please clarify whether "this"
applies to my "always inline" changes and testing additions that you
quoted, or rather to the C++ "acc_on_device" wrapper function as it is
currently present?
> Can't you instead just turn the builtin into BT_FN_INT_VAR and diagnose
> during folding if it has no or 2+ arguments or if the argument is not type
> compatible with int?
Thanks for the suggestion, I'll look into that!
In terms of incremental progress, do you oppose that I commit my existing
patch now, and then rework the builtin in a later patch?
Grüße
Thomas