On Fri, Mar 04, 2016 at 05:04:31PM +0100, Jakub Jelinek wrote:
> On Fri, Mar 04, 2016 at 05:01:34PM +0100, Martin Jambor wrote:
> > Not in the limited runs that I experimented with so far, but I
> > certainly kept this possibility in mind too. If so, I would either
> > set it back before invoking dg-finish or dismiss the whole idea.
> >
> > > > However, the C++ and Fortran cases use gfortran-dg-runtest to cycle
> > > > through a set of torture options and I have not yet discovered the
> > > > right magic variable to set (for example, adding -Wno-hsa to
> > > > DG_TORTURE_OPTIONS elements does not work).
> > > >
> > > > I'm afraid I have spent way too much time on this already, so unless
> > > > someone has any ideas, I'd suggest that we use the (already approved)
> > > > name-changing gomp patch as it is. Or at least for C++ and Fortran.
> > >
> > > Do you have URL for what you refer to?
> > >
> >
> > Sure, the patch has been posted here:
> >
> > https://gcc.gnu.org/ml/gcc-patches/2016-03/msg00071.html
>
> For the g*.dg/gomp/, if you'd only move -Wno-hsa into the last argument
> next to -fopenmp, how many tests would be affected?
Out of 287 files that have dg-options with them in the gomp
directories, only 9 generate hsa warnings:
c-c++-common/gomp/clauses-1.c:/* { dg-options "-fopenmp" } */
c-c++-common/gomp/if-1.c:/* { dg-options "-fopenmp" } */
c-c++-common/gomp/pr61486-2.c:/* { dg-options "-fopenmp" } */
c-c++-common/gomp/target-teams-1.c:/* { dg-options "-fopenmp
-fdump-tree-gimple" } */
g++.dg/gomp/target-teams-1.C:// { dg-options "-fopenmp -fdump-tree-gimple" }
gcc.dg/gomp/pr68128-2.c:/* { dg-options "-O2 -fopenmp -fdump-tree-omplower" }
*/
gfortran.dg/gomp/target1.f90:! { dg-options "-fopenmp" }
gfortran.dg/gomp/target2.f90:! { dg-options "-fopenmp -ffree-line-length-160"
}
gfortran.dg/gomp/target3.f90:! { dg-options "-fopenmp" }
> If not really many,
> perhaps those could be changed to use dg-additional-options instead of
> dg-options.
I do not know what -ffree-line-length-160 is, but probably all of
them, even though putting -O2 in gcc.dg/gomp/pr68128-2.c to
"additional" flags feels just wrong.
However, the real question is: Would such a solution really be much
better than the first version of the patch
(https://gcc.gnu.org/ml/gcc-patches/2016-02/msg01813.html)? After
all, in comparison it would only avoid touching two tests and it will
not avoid issues with tests added in future if they use dg-options.
Martin
