Skip to content

[SYCL][NATIVECPU][PERF] always inline kernels with -O3 #17360

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

Open
wants to merge 6 commits into
base: sycl
Choose a base branch
from
Open

[SYCL][NATIVECPU][PERF] always inline kernels with -O3 #17360

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
1f09921
[SYCL][NATIVECPU] always inline kernels with -O3
uwedolinsky Mar 5, 2025
d9129a0
[SYCL][NATIVECPU] fix lit test to account for more properties occurin…
uwedolinsky Mar 10, 2025
74539a6
Merge remote-tracking branch 'origin/sycl' into uwe/alwaysinlinekernels
uwedolinsky Sep 22, 2025
4fc7da6
[NATIVECPU] use LinkOnceODRLinkage on kernels
uwedolinsky Sep 23, 2025
5f46321
[NATIVECPU] pass OptimizationLevel
uwedolinsky Sep 25, 2025
e6b90e1
[NATIVECPU] file mode fix
uwedolinsky Sep 25, 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
4 changes: 4 additions & 0 deletions llvm/include/llvm/SYCLLowerIR/PrepareSYCLNativeCPU.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,14 +16,18 @@

#include "llvm/IR/Module.h"
#include "llvm/IR/PassManager.h"
#include "llvm/Passes/OptimizationLevel.h"

namespace llvm {

class ModulePass;

class PrepareSYCLNativeCPUPass
: public PassInfoMixin<PrepareSYCLNativeCPUPass> {
const OptimizationLevel OptLevel;

public:
PrepareSYCLNativeCPUPass(OptimizationLevel OL) : OptLevel(OL) {}
PreservedAnalyses run(Module &M, ModuleAnalysisManager &MAM);
};

Expand Down
2 changes: 1 addition & 1 deletion llvm/lib/SYCLNativeCPUUtils/PipelineSYCLNativeCPU.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -109,7 +109,7 @@ void llvm::sycl::utils::addSYCLNativeCPUBackendPasses(
MPM.addPass(compiler::utils::ReplaceLocalModuleScopeVariablesPass());
MPM.addPass(AlwaysInlinerPass());
#endif
MPM.addPass(PrepareSYCLNativeCPUPass());
MPM.addPass(PrepareSYCLNativeCPUPass(OptLevel));
#ifdef NATIVECPU_USE_OCK
MPM.addPass(compiler::utils::DefineMuxBuiltinsPass());
#endif
Expand Down
6 changes: 6 additions & 0 deletions llvm/lib/SYCLNativeCPUUtils/PrepareSYCLNativeCPU.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -403,6 +403,12 @@ PreservedAnalyses PrepareSYCLNativeCPUPass::run(Module &M,
for (auto &OldF : OldKernels) {
auto *NewF =
cloneFunctionAndAddParam(OldF, StatePtrType, CurrentStatePointerTLS);
NewF->setLinkage(llvm::GlobalValue::LinkageTypes::LinkOnceODRLinkage);
if (OptLevel == OptimizationLevel::O3 &&
!NewF->hasFnAttribute(Attribute::NoInline)) {
if (!NewF->hasFnAttribute(Attribute::AlwaysInline))
NewF->addFnAttr(Attribute::AlwaysInline);
}
NewF->takeName(OldF);
OldF->replaceAllUsesWith(NewF);
OldF->eraseFromParent();
Expand Down
7 changes: 7 additions & 0 deletions sycl/test/check_device_code/native_cpu/kernelhandler-scalar.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
// RUN: %clangxx -fsycl-device-only -fno-sycl-libspirv -Wno-unsafe-libspirv-not-linked -fsycl-targets=native_cpu -Xclang -sycl-std=2020 -mllvm -sycl-opt -S -emit-llvm -o %t_temp.ll %s
// RUN: %clangxx -mllvm -sycl-native-cpu-backend -S -emit-llvm -o - %t_temp.ll | FileCheck %s
// RUN: %clangxx -mllvm -sycl-native-cpu-backend -O3 -S -emit-llvm -o - %t_temp.ll | FileCheck %s --check-prefix=CHECK-O3
#include <sycl/sycl.hpp>

#include <iostream>
Expand Down Expand Up @@ -51,3 +52,9 @@ int main() {
// CHECK-DAG: @_ZTS6init_aIjE.NativeCPUKernel(ptr {{.*}}%0, ptr {{.*}}%1, i32 {{.*}}%2, ptr {{.*}}%3){{.*}}
// CHECK-DAG: @_ZTS6init_aIfE.NativeCPUKernel(ptr {{.*}}%0, ptr {{.*}}%1, float {{.*}}%2, ptr {{.*}}%3){{.*}}
// CHECK-DAG: @_ZTS6init_aIdE.NativeCPUKernel(ptr {{.*}}%0, ptr {{.*}}%1, double {{.*}}%2, ptr {{.*}}%3){{.*}}

// CHECK-O3-NOT: @{{.*}}.NativeCPUKernel
// CHECK-O3-DAG: define void @_ZTS6init_aIiE.SYCLNCPU(ptr {{.*}}%0, ptr addrspace(1) {{.*}}%1) {{.*}} #{{.*}} {
// CHECK-O3-DAG: define void @_ZTS6init_aIjE.SYCLNCPU(ptr {{.*}}%0, ptr addrspace(1) {{.*}}%1) {{.*}} #{{.*}} {
// CHECK-O3-DAG: define void @_ZTS6init_aIfE.SYCLNCPU(ptr {{.*}}%0, ptr addrspace(1) {{.*}}%1) {{.*}} #{{.*}} {
// CHECK-O3-DAG: define void @_ZTS6init_aIdE.SYCLNCPU(ptr {{.*}}%0, ptr addrspace(1) {{.*}}%1) {{.*}} #{{.*}} {
2 changes: 1 addition & 1 deletion sycl/test/check_device_code/native_cpu/local_module_scope.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
// find the original function after this pass is run

// CHECK: %localVarTypes = type { ptr addrspace(1) }
// CHECK: define void @_ZTS4TestILi1ELi4EiE.NativeCPUKernel{{.*}} #[[ATTR:[0-9]*]]
// CHECK: define linkonce_odr void @_ZTS4TestILi1ELi4EiE.NativeCPUKernel{{.*}} #[[ATTR:[0-9]*]]
// CHECK: alloca %localVarTypes
// CHECK: attributes #[[ATTR]] = {{.*}} "mux-orig-fn"="_ZTS4TestILi1ELi4EiE"

Expand Down
2 changes: 1 addition & 1 deletion sycl/test/check_device_code/native_cpu/native_cpu_builtins.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,7 @@ int main() {
// CHECK: call{{.*}}__dpcpp_nativecpu_get_global_id(i32 0, ptr addrspace(1) %2)
// CHECK-NOT: @llvm.threadlocal

// CHECK-TL: define void @_ZTSN4sycl3_V16detail19__pf_kernel_wrapperI5Test1EE.NativeCPUKernel({{.*}}
// CHECK-TL: define linkonce_odr void @_ZTSN4sycl3_V16detail19__pf_kernel_wrapperI5Test1EE.NativeCPUKernel({{.*}}
// CHECK-TL-NEXT:entry:
// CHECK-TL-NEXT: %[[VAL1:.*]] = call ptr addrspace(1) @llvm.threadlocal.address.p1(ptr addrspace(1) @_ZL28nativecpu_thread_local_state)
// CHECK-TL-NEXT: %[[VAL2:.*]] = load ptr addrspace(1), ptr addrspace(1) %[[VAL1]], align 8
Expand Down
Loading