[CUDA][HIP] Fix host/device attribute of builtin #138162
Open
+44
−0
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
There was a problem hiding this comment.
Pull Request Overview
This PR fixes an error in CUDA/HIP compilation by ensuring that overloaded builtin functions are marked as implicit, allowing them to be callable on both host and device sides.
- Added a call to set the implicit attribute on the overload declaration
- Ensures that device functions can properly call the overloaded builtins
Files not reviewed (1)
- clang/test/SemaCUDA/overloaded-builtin.cu: Language not supported
Comments suppressed due to low confidence (1)
clang/lib/Sema/SemaExpr.cpp:6361
- Marking the overloaded builtin function as implicit fixes device function call errors. Please ensure that this change does not interfere with any subsequent attribute processing and document the rationale if necessary.
OverloadDecl->setImplicit(true);
|
@llvm/pr-subscribers-clang Author: Yaxun (Sam) Liu (yxsamliu) ChangesWhen a builtin function with generic pointer parameter is passed a pointer with address space, clang creates an overloaded builtin function but does not make it implicit. This causes error when the builtin is called by device functions since CUDA/HIP relies on the implicit attribute to treat a builtin function as callable on both host and device sides. Fixed by making the created overloaded builtin functions implicit. Full diff: https://github.com/llvm/llvm-project/pull/138162.diff 2 Files Affected:
diff --git a/clang/lib/Sema/SemaExpr.cpp b/clang/lib/Sema/SemaExpr.cpp
index 0cd86dc54b0ab..d9eccb31e6d1e 100644
--- a/clang/lib/Sema/SemaExpr.cpp
+++ b/clang/lib/Sema/SemaExpr.cpp
@@ -6358,6 +6358,7 @@ static FunctionDecl *rewriteBuiltinFunctionDecl(Sema *Sema, ASTContext &Context,
}
OverloadDecl->setParams(Params);
Sema->mergeDeclAttributes(OverloadDecl, FDecl);
+ OverloadDecl->setImplicit(true);
return OverloadDecl;
}
diff --git a/clang/test/SemaCUDA/overloaded-builtin.cu b/clang/test/SemaCUDA/overloaded-builtin.cu
new file mode 100644
index 0000000000000..719bfea4aef2f
--- /dev/null
+++ b/clang/test/SemaCUDA/overloaded-builtin.cu
@@ -0,0 +1,23 @@
+// expected-no-diagnostics
+
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -aux-triple amdgcn-amd-amdhsa -fsyntax-only -verify -xhip %s
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -fsyntax-only -fcuda-is-device -verify -xhip %s
+
+#include "Inputs/cuda.h"
+
+__global__ void kernel() {
+ __attribute__((address_space(0))) void *mem_ptr;
+ (void)__builtin_amdgcn_is_shared(mem_ptr);
+}
+
+template<typename T>
+__global__ void template_kernel(T *p) {
+ __attribute__((address_space(0))) void *mem_ptr;
+ (void)__builtin_amdgcn_is_shared(mem_ptr);
+}
+
+int main() {
+ int *p;
+ kernel<<<1,1>>>();
+ template_kernel<<<1,1>>>(p);
+}
|
Artem-B
reviewed
May 1, 2025
When a builtin function with generic pointer parameter is passed a pointer with address space, clang creates an overloaded builtin function but does copy the host/device attribute. This causes error when the builtin is called by device functions since CUDA/HIP relies on the host/device attribute to treat a builtin function as callable on both host and device sides. Fixed by copying the host/device attribute of the original builtin function to the created overloaded builtin function.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When a builtin function is passed a pointer with a different
address space, clang creates an overloaded
builtin function but does not copy the host/device attribute. This causes
error when the builtin is called by device functions
since CUDA/HIP relies on the host/device attribute to treat
a builtin function as callable on both host and device
sides.
Fixed by copying the host/device attribute of the original
builtin function to the created overloaded builtin function.