In D145823#4195082, @jdoerfert wrote:

In D145823#4194245, @ddpagan wrote:

In D145823#4192464, @ABataev wrote:

Also, can any clause affect the behavior?

If I understand your question, I don't believe that any of the allowed clauses affect the behavior in a way that would invalidate transforming the combined loops in such a manner.

At least the way we tried to design them, they should not. That said, we need to ensure the clauses apply properly, e.g., the reduction is applied to all loop-associated directives, but that should happen automagically, I think. The AMDGPU test case shows at least the right reduction.

In D145823#4192464, @ABataev wrote:

Also, can any clause affect the behavior?

In D145823#4192463, @ABataev wrote:

Can we add also simd for such loops?

Adding context.

How is bind(thread) not working? Runtime issue?

Thanks for the review, Alexey. Suggested changes have been made.

Corrected clang formatting issues, except those relating to updated tests. Because of known clang-format deficiencies, making these changes causes them to fail.

Per Alexey's suggestion, 'inoutset' now accepted only with OpenMP5.1. Also updated affected dependency type tests to expect and to test 'inoutset' only when version is at least 5.1.

Is it a new mode from OpenMP 5.2? Or 5.1? Can you add a check for OpenMP version, if so?

Fixed format problem for test/OpenMP/allocate_codegen_attr.cpp
Successfully rebased, built, and tested.

Alexey, you weren't missing anything. This was an unnecessary declaration of createOMPAlignedAlloc OMPIRBuilder.h that I had forgotten to delete.

Thanks for catching that, Alexey. As suggested, we now keep the allocator Expr * and reemit it later when needed.

Fixed build problem (found during build of flang). Added align clause to simple clause check.

Successfully rebased, built, and tested.

Thanks for reviewing the code, Aaron.

Thanks for the review, Aaron! I neglected to submit my responses that I made inline, but you can still see them.

Thanks for the review, Alexey.