SunilKuravinakop wrote: > #if dispatch transformed to > #pragma omp taskwait depend(out:x) > #pragma omp dispatch nocontext(c2) > foo(); --> with traits call to foo_variant_dispatch(i,j) > you should have 2 captured regions - one for taskwait and one for > dispatch. It does not mean, >you need to transform the AST node, there should > be single AST node with 2 captured regions >(one caotures another), just like > it is done for combined constructs. > > > >I do not have this code anymore in SemaOpenMP.cpp. I am having code for > > >adding AnnotateAttr. In CodeGen this exact comment has been taken care of. > > I'm saying how it should be implemented. Otherwise, it may crash in some > cases or work incorrectly.
Do you want me to move the changes from CodeGen back into SemaOpenMP.cpp and avoid AnnotateAttr? ( In one of your feedbacks you had indicated that I should move the helper codes to CodeGen ). With changes that I have now, I was looking at: 1. With current changes there was a crash occurring with constant values e.g. nocontext(true) This was solved easily. 2. The changes for handling template was difficult. I could not get the variant method in LLVM IR (-emit-llvm). Probably I am missing something in VisitCXXMethodDecl(). https://github.com/llvm/llvm-project/pull/131838 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
