[clang] Add missing intrinsics to cuda headers (PR #143664)

Artem Belevich via cfe-commits cfe-commits at lists.llvm.org
Fri Jun 13 12:53:24 PDT 2025


================
@@ -479,7 +479,291 @@ inline __device__ unsigned __funnelshift_rc(unsigned low32, unsigned high32,
   return ret;
 }
 
-#endif // !defined(__CUDA_ARCH__) || __CUDA_ARCH__ >= 320
+#pragma push_macro("__INTRINSIC_LOAD")
+#define __INTRINSIC_LOAD(__FnName, __AsmOp, __DeclType, __TmpType, __AsmType, \
+                         __Clobber)                                           \
+  inline __device__ __DeclType __FnName(const __DeclType *ptr) {              \
----------------
Artem-B wrote:

`ptr`, and local variables should probably be prefixed, too. Nothing stops the users from passing `-Dptr=1`

https://github.com/llvm/llvm-project/pull/143664


More information about the cfe-commits mailing list