[Openmp-commits] [PATCH] D64217: [OpenMP][NFCI] Cleanup the target state queue implementation
Johannes Doerfert via Phabricator via Openmp-commits
openmp-commits at lists.llvm.org
Fri Jul 5 13:37:12 PDT 2019
jdoerfert marked 4 inline comments as done.
jdoerfert added inline comments.
Comment at: openmp/libomptarget/deviceRTLs/nvptx/src/state-queue.h:1
-//===--------- statequeue.h - NVPTX OpenMP GPU State Queue ------- CUDA -*-===//
+//===--- state-queue.h --- OpenMP target state queue ------------*- C++ -*-===//
> jdoerfert wrote:
> > ABataev wrote:
> > > I think it is better to mark it as Cuda, not C/C++, since it contains some Cuda specific constructs.
> > I though I removed all of them, if I haven't we should. The idea is that this "core logic" code should _not_ be CUDA code but something else, e.g., C++ seems natural.
> Nope, `INLINE` macro is expanded into cuda specific attributes.
It is a macro and macros are not specific to CUDA. To what it expands if this file is included from the CUDA code right now is not relevant. This is a first step and we will have to move these files into the common folder (see the RFC). The macro definition will be in the device specific parts and they can then use whatever device (language) specific extensions they want without making these part specific to that device (language) extension. That is, after all, the main motivation behind these patches.
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
More information about the Openmp-commits