Skip to content

[CIR][HIP] Use GlobalView to access stub #1957

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 4 commits into from
Nov 4, 2025
Merged

[CIR][HIP] Use GlobalView to access stub #1957

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
53e4509
Fix stub code generation
koparasy Oct 24, 2025
2b45b50
Fix tests
koparasy Oct 25, 2025
d308b09
Revert change
koparasy Oct 25, 2025
3e6110c
LLVM Code generation is not supported due to missing ABI
koparasy Oct 30, 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
17 changes: 12 additions & 5 deletions clang/lib/CIR/CodeGen/CIRGenCUDANV.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -265,12 +265,20 @@ void CIRGenNVCUDARuntime::emitDeviceStubBodyNew(CIRGenFunction &cgf,

void CIRGenNVCUDARuntime::emitDeviceStub(CIRGenFunction &cgf, cir::FuncOp fn,
FunctionArgList &args) {

if (auto globalOp =
llvm::dyn_cast<cir::GlobalOp>(KernelHandles[fn.getSymName()])) {
auto symbol = mlir::FlatSymbolRefAttr::get(fn.getSymNameAttr());
// Set the initializer for the global
cgm.setInitializer(globalOp, symbol);
auto &builder = cgm.getBuilder();
auto fnPtrTy = globalOp.getSymType();
auto sym = mlir::FlatSymbolRefAttr::get(fn.getSymNameAttr());
auto gv = cir::GlobalViewAttr::get(fnPtrTy, sym);

globalOp->setAttr("initial_value", gv);
globalOp->removeAttr("sym_visibility");
globalOp->setAttr("alignment", builder.getI64IntegerAttr(
cgm.getPointerAlign().getQuantity()));
}

// CUDA 9.0 changed the way to launch kernels.
if (CudaFeatureEnabled(cgm.getTarget().getSDKVersion(),
CudaFeature::CUDA_USES_NEW_LAUNCH) ||
Expand Down Expand Up @@ -322,12 +330,11 @@ mlir::Operation *CIRGenNVCUDARuntime::getKernelHandle(cir::FuncOp fn,
cgm, fn->getLoc(), globalName,
builder.getPointerTo(fn.getFunctionType()), true,
cir::AddressSpace::Default,
/*insertPoint=*/nullptr, fn.getLinkage());
/*insertPoint=*/nullptr);
});

globalOp->setAttr("alignment", builder.getI64IntegerAttr(
cgm.getPointerAlign().getQuantity()));
globalOp->setAttr("visibility", fn->getAttr("sym_visibility"));

// Store references
KernelHandles[fn.getSymName()] = globalOp;
Expand Down
40 changes: 34 additions & 6 deletions clang/test/CIR/CodeGen/HIP/simple.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,10 +5,15 @@
// RUN: -emit-cir %s -o %t.cir
// RUN: FileCheck --check-prefix=CIR-HOST --input-file=%t.cir %s

// RUN: %clang_cc1 -triple=amdgcn-amd-amdhsa -x hip \
// RUN: %clang_cc1 -triple=amdgcn-amd-amdhsa -x hip -fclangir \
// RUN: -fcuda-is-device -fhip-new-launch-api \
// RUN: -emit-cir %s -o %t.cir
// RUN: FileCheck --check-prefix=CIR-DEVICE --input-file=%t.cir %s
//
// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir \
// RUN: -x hip -emit-llvm -fhip-new-launch-api \
// RUN: %s -o %t.ll
// RUN: FileCheck --check-prefix=LLVM-HOST --input-file=%t.ll %s

// Attribute for global_fn
// CIR-HOST: [[Kernel:#[a-zA-Z_0-9]+]] = {{.*}}#cir.cu.kernel_name<_Z9global_fni>{{.*}}
Expand All @@ -25,14 +30,21 @@ __device__ void device_fn(int* a, double b, float c) {}
__global__ void global_fn(int a) {}
// CIR-DEVICE: @_Z9global_fni

// CIR-HOST: cir.alloca {{.*}}"kernel_args"
// CIR-HOST: @_Z24__device_stub__global_fni{{.*}}extra([[Kernel]])
// CIR-HOST: %[[#CIRKernelArgs:]] = cir.alloca {{.*}}"kernel_args"
// CIR-HOST: %[[#Decayed:]] = cir.cast array_to_ptrdecay %[[#CIRKernelArgs]]
// CIR-HOST: cir.call @__hipPopCallConfiguration

// Host access the global stub instead of the functiond evice stub.
// The stub has the mangled name of the function
// CIR-HOST: cir.get_global @_Z9global_fni
// CIR-HOST: cir.get_global @_Z9global_fni : !cir.ptr<!cir.ptr<!cir.func<(!s32i)>>>
// CIR-HOST: cir.call @hipLaunchKernel

// LLVM-HOST: void @_Z24__device_stub__global_fni
// LLVM-HOST: %[[#KernelArgs:]] = alloca [1 x ptr], i64 1, align 16
// LLVM-HOST: %[[#GEP1:]] = getelementptr ptr, ptr %[[#KernelArgs]], i32 0
// LLVM-HOST: %[[#GEP2:]] = getelementptr [1 x ptr], ptr %[[#KernelArgs]], i32 0, i64 0
// LLVM-HOST: call i32 @__hipPopCallConfiguration
// LLVM-HOST: call i32 @hipLaunchKernel(ptr @_Z9global_fni


int main() {
global_fn<<<1, 1>>>(1);
}
Expand All @@ -49,4 +61,20 @@ int main() {
// CIR-HOST: cir.call @_Z24__device_stub__global_fni([[Arg]])
// CIR-HOST: }

// LLVM-HOST: define dso_local i32 @main
// LLVM-HOST: alloca %struct.dim3
// LLVM-HOST: alloca %struct.dim3
// LLVM-HOST: call void @_ZN4dim3C1Ejjj
// LLVM-HOST: call void @_ZN4dim3C1Ejjj
// LLVM-HOST: %[[#ConfigOK:]] = call i32 @__hipPushCallConfiguration
// LLVM-HOST: %[[#ConfigCond:]] = icmp ne i32 %[[#ConfigOK]], 0
// LLVM-HOST: br i1 %[[#ConfigCond]], label %[[#Good:]], label %[[#Bad:]]
// LLVM-HOST: [[#Good]]:
// LLVM-HOST: br label %[[#End:]]
// LLVM-HOST: [[#Bad]]:
// LLVM-HOST: call void @_Z24__device_stub__global_fni(i32 1)
// LLVM-HOST: br label %[[#End:]]
// LLVM-HOST: [[#End]]:
// LLVM-HOST: %[[#]] = load i32
// LLVM-HOST: ret i32

Loading