[Openmp-commits] [PATCH] D82718: [OpenMP] Use primary context in CUDA plugin

Shilei Tian via Phabricator via Openmp-commits openmp-commits at lists.llvm.org
Tue Jun 30 19:32:05 PDT 2020

tianshilei1992 added a comment.

In D82718#2123820 <https://reviews.llvm.org/D82718#2123820>, @ye-luo wrote:

> In D82718#2123747 <https://reviews.llvm.org/D82718#2123747>, @tianshilei1992 wrote:
> > In D82718#2123675 <https://reviews.llvm.org/D82718#2123675>, @protze.joachim wrote:
> >
> > > This patch drops the `CU_CTX_SCHED_BLOCKING_SYNC` property currently selected for the context. Is this intended? Should we add another function call to request this behavior for the primary context?
> >
> >
> > That is good point. We depend on the synchronous behavior in some cases in the RTL.
> Are you sure this is the right flag you need?

>From CUDA documentation, it says:

> CU_CTX_SCHED_BLOCKING_SYNC: Instruct CUDA to block the CPU thread on a synchronization primitive when waiting for the GPU to finish work.

I assume it affects all synchronization operations, not only the context synchronization. If it's true, then it is definitely what we need and are currently using in the RTL.



More information about the Openmp-commits mailing list