[clang] [Clang] Fix sema checks thinking kernels aren't kernels (PR #104460)

[via cfe-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-amdgpu

Author: Joseph Huber (jhuber6)

<details>
<summary>Changes</summary>

Summary:
Currently we have some sema checks to make sure users don't apply
kernel-only attributes to non-kernel functions. However, this currently
did not correctly check for bare NVPTX / AMDGPU kernel attributes,
making it impossible to use them at all w/o CUDA enabled. This patch
fixes that by checking for the calling convention / attributes directly.


---
Full diff: https://github.com/llvm/llvm-project/pull/104460.diff


2 Files Affected:

- (modified) clang/lib/Sema/SemaDeclAttr.cpp (+5-2) 
- (modified) clang/test/CodeGenCXX/amdgpu-kernel-arg-pointer-type.cpp (+5) 


``````````diff
diff --git a/clang/lib/Sema/SemaDeclAttr.cpp b/clang/lib/Sema/SemaDeclAttr.cpp
index 3b5e984f4ee773..96c3759de7edc1 100644
--- a/clang/lib/Sema/SemaDeclAttr.cpp
+++ b/clang/lib/Sema/SemaDeclAttr.cpp
@@ -7147,7 +7147,9 @@ void Sema::ProcessDeclAttributeList(
   // good to have a way to specify "these attributes must appear as a group",
   // for these. Additionally, it would be good to have a way to specify "these
   // attribute must never appear as a group" for attributes like cold and hot.
-  if (!D->hasAttr<OpenCLKernelAttr>()) {
+  const FunctionType *FnTy = D->getFunctionType();
+  if (!D->hasAttr<OpenCLKernelAttr>() && FnTy &&
+      FnTy->getCallConv() != CallingConv::CC_AMDGPUKernelCall) {
     // These attributes cannot be applied to a non-kernel function.
     if (const auto *A = D->getAttr<ReqdWorkGroupSizeAttr>()) {
       // FIXME: This emits a different error message than
@@ -7163,7 +7165,8 @@ void Sema::ProcessDeclAttributeList(
     } else if (const auto *A = D->getAttr<OpenCLIntelReqdSubGroupSizeAttr>()) {
       Diag(D->getLocation(), diag::err_opencl_kernel_attr) << A;
       D->setInvalidDecl();
-    } else if (!D->hasAttr<CUDAGlobalAttr>()) {
+    } else if (!D->hasAttr<CUDAGlobalAttr>() &&
+               !D->hasAttr<NVPTXKernelAttr>()) {
       if (const auto *A = D->getAttr<AMDGPUFlatWorkGroupSizeAttr>()) {
         Diag(D->getLocation(), diag::err_attribute_wrong_decl_type)
             << A << A->isRegularKeywordAttribute() << ExpectedKernelFunction;
diff --git a/clang/test/CodeGenCXX/amdgpu-kernel-arg-pointer-type.cpp b/clang/test/CodeGenCXX/amdgpu-kernel-arg-pointer-type.cpp
index c8fb12315d0eec..8df1d75e9ec8fa 100644
--- a/clang/test/CodeGenCXX/amdgpu-kernel-arg-pointer-type.cpp
+++ b/clang/test/CodeGenCXX/amdgpu-kernel-arg-pointer-type.cpp
@@ -6,6 +6,10 @@
 // The original test passes the result through opt O2, but that seems to introduce invalid
 // addrspace casts which are not being fixed as part of the present change.
 
+// COMMON: define{{.*}} amdgpu_kernel void @_Z6kernelv() #[[ATTR:[0-9]+]]
+__attribute__((amdgpu_kernel, amdgpu_flat_work_group_size(1, 256))) void
+    kernel() {}
+
 // COMMON-LABEL: define{{.*}} amdgpu_kernel void @_Z7kernel1Pi(ptr {{.*}} %x)
 // CHECK-NOT: ={{.*}} addrspacecast [[TYPE:.*]] addrspace(1)* %{{.*}} to ptr
 __attribute__((amdgpu_kernel)) void kernel1(int *x) {
@@ -81,3 +85,4 @@ __attribute__((amdgpu_kernel)) void kernel8(struct SS a) {
   *a.x += 3.f;
 }
 
+// COMMON: attributes #[[ATTR]] = { {{.*}}"amdgpu-flat-work-group-size"="1,256"{{.*}} }

``````````

</details>


https://github.com/llvm/llvm-project/pull/104460