[SYCL] Don't run SYCLOptimizeBarriersPass with O0

[MrSidims]

No description provided.

[MrSidims]

Not entirely sure, how to add a test for this change - with O0 SYCL code won't be having memory semantic constants be propagated in the first place.

[maarquitos14]

Not entirely sure, how to add a test for this change - with O0 SYCL code won't be having memory semantic constants be propagated in the first place.

Maybe with some kind of log to make sure the barrier does not run?

[Fznamznon]

Not entirely sure, how to add a test for this change - with O0 SYCL code won't be having memory semantic constants be propagated in the first place.

Why do we disable it then?

[MrSidims]

Not entirely sure, how to add a test for this change - with O0 SYCL code won't be having memory semantic constants be propagated in the first place.

Why do we disable it then?

Because of the OpenMP compiler, that inserts barriers implicitly (with the constants) and OpenMP offloading reuses SYCL offloading model, making this pass running.

[Fznamznon]

Not entirely sure, how to add a test for this change - with O0 SYCL code won't be having memory semantic constants be propagated in the first place.

Why do we disable it then?

Because of the OpenMP compiler, that inserts barriers implicitly (with the constants) and OpenMP offloading reuses SYCL offloading model, making this pass running.

Can we use an OpenMP test?

[MrSidims]

Not entirely sure, how to add a test for this change - with O0 SYCL code won't be having memory semantic constants be propagated in the first place.

Why do we disable it then?

Because of the OpenMP compiler, that inserts barriers implicitly (with the constants) and OpenMP offloading reuses SYCL offloading model, making this pass running.

Can we use an OpenMP test?

AFAIK OpenMP solution is a part of our downstream, so adding a test here won't help much :) The only other solution I see is to add an LLVM IR test with opt -O0 passes as a RUN command, but not sure if it's 'LLVM way' of creating unit tests.

[MrSidims]

Not entirely sure, how to add a test for this change - with O0 SYCL code won't be having memory semantic constants be propagated in the first place.

Why do we disable it then?

Because of the OpenMP compiler, that inserts barriers implicitly (with the constants) and OpenMP offloading reuses SYCL offloading model, making this pass running.

Can we use an OpenMP test?

AFAIK OpenMP solution is a part of our downstream, so adding a test here won't help much :) The only other solution I see is to add an LLVM IR test with opt -O0 passes as a RUN command, but not sure if it's 'LLVM way' of creating unit tests.

Actually, I have an idea for SYCL'ish test, let me try it.

[github-actions]

@intel/llvm-gatekeepers please consider merging

[MrSidims]

Apparently non-empty source code is required to make the pass run. @Fznamznon please comment if you are okay with this change or if I should separate the barrier case from this test.

[Fznamznon]

I think it is fine in this file, I don't really get why it happens.

[MrSidims]

Same failures are observed in #20082

[Fznamznon]

I find it really weird that the pass manager doesn't report this pass if the source is empty, however I'm not an expert in this. Can we simplify the code and explain why it is required, it is dummy anyway

Suggested change

	template <typename name, typename Func>
	void kernel(const Func &f) __attribute__((sycl_kernel)) {
	f();
	}
	void bar() {
	kernel<class MyKernel>([=]() {});
	}
	// Some passes don't run on empty code
	__attribute__((sycl_device)) void bar() {
	}

If a kernel is required please use https://github.com/intel/llvm/blob/sycl/sycl/doc/developer/ContributeToDPCPP.md#dpc-clang-fe-tests .

[MrSidims]

Applied. So passes registered via registerOptimizerLastEPCallback are not being run fir the empty test. Passes registered like this are being run at the end of a normal pipeline via invokeOptimizerLastEPCallbacks and ... I haven't yet finished with the debugging, but it seems like it's module-size sensitive.

[Fznamznon]

I think it is fine in this file, I don't really get why it happens.

[MrSidims]

@intel/llvm-gatekeepers please help with merge