[Openmp-commits] [PATCH] D60578: [OPENMP][NVPTX]Fix dynamic scheduling in L2+ SPMD parallel regions.

Jonas Hahnfeld via Phabricator via Openmp-commits openmp-commits at lists.llvm.org
Mon Apr 15 01:18:11 PDT 2019

Hahnfeld added a comment.

In D60578#1465958 <https://reviews.llvm.org/D60578#1465958>, @hfinkel wrote:

> What check command do you run to run these nvptx tests?

The target is called `check-libomptarget-nvptx` and it's not run by `check-openmp`. The reasoning is basically what I've described in my previous answers, (maybe) some more in the initial revision D51687 <https://reviews.llvm.org/D51687>.

>> In that case these tests need to be marked `UNSUPPORTED` for versions of Clang that will not pass them. There's infrastructure for that, but it's not applied in the current form of this patch.
> Okay. This patch review is not the right place to discuss the libomptarget support for old Clang versions. We should have a separate thread on this subject.

> In any case, let's move forward with adding this test in that directory, and then we'll address the infrastructure issue as follow-up work.

So put differently, you're proposing to land this in its current form (which will break for some users, including me) and wait for "somebody" to work on the infrastructure to fix things?

> I'm aware of several groups working on different libomptarget plugins and other related things (including my team), so I suspect that reality might not be as sad as you believe.

Working on internal testing (such as many have) is not the same as having this upstream. That's what I was referring to (sorry if that was ambiguous), I'm sure that many people are working on OpenMP offloading and the related runtime libraries.

> Nevertheless, what infrastructure do we actually want here? Should we have the ability to ask make check-openmp to take a list of targets to use to run all of the offload tests so that the user can specify the names of the targets that will actually work?

In my opinion we need exactly what we have with `check-libomptarget-nvptx`, maybe it needs to be generalized for future targets. In its easiest setup we would just have a target for each one of them, like the current name `check-libomptarget-nvptx` for Nvidia, `check-libomptarget-gcn` for AMD and so on.

  rOMP OpenMP



More information about the Openmp-commits mailing list