[clang] [nvptx-arch] allowed load `nvcuda.dll` on Windows (PR #135300)

[via cfe-commits]

https://github.com/SquallATF created https://github.com/llvm/llvm-project/pull/135300

Make nvptx-arch work on windows

>From a4e22b9f5da3578fc2fb02d55e144509759ee705 Mon Sep 17 00:00:00 2001
From: Peiyuan Song <squallatf at gmail.com>
Date: Fri, 11 Apr 2025 09:51:44 +0800
Subject: [PATCH] [nvptx-arch] allowed load `nvcuda.dll` on Windows

---
 clang/tools/offload-arch/NVPTXArch.cpp | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/clang/tools/offload-arch/NVPTXArch.cpp b/clang/tools/offload-arch/NVPTXArch.cpp
index c7b7fcdf80500..dc06f836898ca 100644
--- a/clang/tools/offload-arch/NVPTXArch.cpp
+++ b/clang/tools/offload-arch/NVPTXArch.cpp
@@ -39,7 +39,11 @@ CUresult (*cuGetErrorString)(CUresult, const char **);
 CUresult (*cuDeviceGet)(CUdevice *, int);
 CUresult (*cuDeviceGetAttribute)(int *, CUdevice_attribute, CUdevice);
 
+#if defined(_WIN32)
+constexpr const char *DynamicCudaPath = "nvcuda.dll";
+#else
 constexpr const char *DynamicCudaPath = "libcuda.so.1";
+#endif
 
 llvm::Error loadCUDA() {
   std::string ErrMsg;