jdoerfert added a comment. In D101976#2743470 <https://reviews.llvm.org/D101976#2743470>, @JonChesterfield wrote:
> In D101976#2742919 <https://reviews.llvm.org/D101976#2742919>, @jdoerfert > wrote: > >> So what do you wnat me to change for this patch now? > > Equivalent change to amdgpu target_impl to the nvptx target_impl, which looks > like syncthreads should call a new barrier. > > Iiuc this has run successfully even without that change so hopefully that's > sufficient that we won't regress on amdgpu. I'd like to get miniqmc running > locally to verify as well, but we may not be able to wait for that. So we don't need changes? I'm not sure what the problem here is. ================ Comment at: openmp/libomptarget/deviceRTLs/common/src/omptarget.cu:195 + +EXTERN __attribute__((weak)) +int32_t __kmpc_target_init(ident_t *Ident, bool IsSPMD, ---------------- JonChesterfield wrote: > why are these weak? So we do not perform IPO but will inline them. If we perform IPO we specialize the arguments even though we still want to do potentially change the mode from non-SPMD to SPMD. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D101976/new/ https://reviews.llvm.org/D101976 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
