On Wed, Oct 09, 2019 at 11:46:37AM +0200, Thomas Schwinge wrote: > On 2019-10-09T08:04:29+0200, Bernhard Reutner-Fischer <rep.dot....@gmail.com> > wrote: > > On 8 October 2019 14:12:47 CEST, Jakub Jelinek <ja...@redhat.com> wrote: > >>On Tue, Oct 08, 2019 at 02:04:17PM +0200, Tobias Burnus wrote: > >>> Seemingly, 'target simd' was forgotten – which yielded the error: > >>> "Unexpected !$OMP TARGET SIMD statement" > > Shouldn't this get a PR opened, and the fix eventually backported also to > release branches?
Not sure about that. target simd is an OpenMP 4.5 feature, and we've never claimed OpenMP 4.5 is fully implemented in GCC 9 or earlier, so I think it is fine if it is just one of several missing 4.5 features implemented in GCC 10. If we've accepted target simd, fine, it would be a bug fix, but my understanding is that because of this bug we've not really supported it. The actual problem is that the https://gcc.gnu.org/ml/gcc-patches/2016-05/msg01125.html patch wasn't accompanied by sufficient test coverage, some of it has been later covered, like taskloop/taskloop simd, but most of it has not, because I eventually got busy with non-OpenMP stuff that year. In particular, we don't have any testsuite coverage for target enter data target exit data target parallel target parallel do target parallel do simd (for target simd we now have some) just from the above patch. The plan was basically to port the C/C++ testcases from the OpenMP 4.5 submission, so https://gcc.gnu.org/ml/gcc-patches/2015-10/msg01291.html https://gcc.gnu.org/ml/gcc-patches/2015-11/msg00477.html https://gcc.gnu.org/ml/gcc-patches/2015-11/msg01835.html to Fortran, but that never happened, like finishing the 4.5 support. Jakub