Possible bug in OpenMP implementation

[eZWALT]

Possible bug in OpenMP implementation

[Lancern]

The second line of the CIR error output gives the position of the error:

clangir/clang/lib/CIR/CodeGen/CIRGenerator.cpp

Lines 162 to 165 in b26f21d

	// For OpenMP emit declare reduction functions, if required.
	if (astCtx->getLangOpts().OpenMP) {
	llvm_unreachable("NYI");
	}

Basically this code checks whether we have -fopenmp flag in the command line. If the flag is present, then we crash with llvm_unreachable. So if you are implementing OpenMP stuff, you will need to replace this llvm_unreachable with your implementation code.

Note that this code only gets executed when we meet a TagDecl (which represents a struct / class / enum / union) on the clang AST. The inclusing of omp.h introduces a bunch of TagDecls so you get the error once you include omp.h.

If you are not sure what to do at here, a good reference is always the original clang CodeGen module. The corresponding part in the clang CodeGen is:

clangir/clang/lib/CodeGen/ModuleBuilder.cpp

Lines 258 to 268 in b26f21d

	if (Ctx->getLangOpts().OpenMP) {
	for (Decl *Member : D->decls()) {
	if (auto *DRD = dyn_cast<OMPDeclareReductionDecl>(Member)) {
	if (Ctx->DeclMustBeEmitted(DRD))
	Builder->EmitGlobal(DRD);
	} else if (auto *DMD = dyn_cast<OMPDeclareMapperDecl>(Member)) {
	if (Ctx->DeclMustBeEmitted(DMD))
	Builder->EmitGlobal(DMD);
	}
	}
	}

You can start by understanding what the original clang CodeGen is doing and port it to CIRGen accordingly.

[bcardosolopes]

Thanks for the detailed explanation @Lancern.

@eZWALT another possibility is to mock the relevant constructs out of #include <omp.h> and use them directly into your testcase, so you don't need to handle all the things not yet supported from that header. It's also possible that under clang/test/OpenMP you will find those mocked versions to copy from.

[eZWALT]

Great explanation @Lancern , thanks for all the time invested, after implementing the task construct I will be checking this out!