Skip to content

[SYCL] Don't run SYCLOptimizeBarriersPass with O0 #20037

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 8 commits into from
Sep 16, 2025
Merged

[SYCL] Don't run SYCLOptimizeBarriersPass with O0 #20037

Show file tree
Hide file tree
Changes from 7 commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
b78605b
[SYCL] Don't run SYCLOptimizeBarriersPass with O0
MrSidims Sep 10, 2025
96a7a7a
format
MrSidims Sep 10, 2025
1aeedf7
add test
MrSidims Sep 10, 2025
68a84f0
adjust test
MrSidims Sep 10, 2025
73266d4
move the test
MrSidims Sep 11, 2025
71bdb3f
adjust test
MrSidims Sep 16, 2025
6347c9c
Merge remote-tracking branch 'origin/sycl' into run-barriers-with-o0
MrSidims Sep 16, 2025
3496abb
shorted test
MrSidims Sep 16, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions clang/lib/CodeGen/BackendUtil.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1093,8 +1093,9 @@ void EmitAssemblyHelper::RunOptimizationPipeline(
ThinOrFullLTOPhase) {
MPM.addPass(createModuleToFunctionPassAdaptor(
InferAddressSpacesPass(clang::targets::SPIR_GENERIC_AS)));
MPM.addPass(
createModuleToFunctionPassAdaptor(SYCLOptimizeBarriersPass()));
if (Level != OptimizationLevel::O0)
MPM.addPass(createModuleToFunctionPassAdaptor(
SYCLOptimizeBarriersPass()));
});
}

Expand Down
14 changes: 14 additions & 0 deletions clang/test/CodeGenSYCL/kernel-early-optimization-pipeline.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@
// CHECK: InferFunctionAttrsPass
// CHECK: AlwaysInlinerPass
// CHECK: ModuleInlinerWrapperPass
// CHECK: SYCLOptimizeBarriersPass
// CHECK: ConstantMergePass
// CHECK: SYCLMutatePrintfAddrspacePass
// CHECK: SYCLPropagateAspectsUsagePass
Expand All @@ -22,4 +23,17 @@
//
// RUN: %clang_cc1 -O2 -fsycl-is-device -triple spir64-unknown-unknown %s -mdebug-pass Structure -emit-llvm -fno-sycl-early-optimizations -o /dev/null 2>&1 | FileCheck %s --check-prefix=CHECK-NOEARLYOPT
// CHECK-NOEARLYOPT-NOT: ConstantMergePass1
// CHECK-NOEARLYOPT-NOT: SYCLOptimizeBarriersPass
// CHECK-NOEARLYOPT: SYCLMutatePrintfAddrspacePass

// RUN: %clang_cc1 -O0 -fsycl-is-device -triple spir64-unknown-unknown %s -mdebug-pass Structure -emit-llvm -o /dev/null 2>&1 | FileCheck %s --check-prefix=CHECK-O0
// CHECK-O0-NOT: SYCLOptimizeBarriersPass

template <typename name, typename Func>
void kernel(const Func &f) __attribute__((sycl_kernel)) {
f();
}

void bar() {
kernel<class MyKernel>([=]() {});
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

}
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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 .

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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.

Loading