diff --git a/clang/lib/Driver/ToolChains/Cuda.cpp b/clang/lib/Driver/ToolChains/Cuda.cpp
index fdfcea852b4f..327cb5183f83 100644
--- a/clang/lib/Driver/ToolChains/Cuda.cpp
+++ b/clang/lib/Driver/ToolChains/Cuda.cpp
@@ -611,9 +611,12 @@ void NVPTX::Linker::ConstructJob(Compilation &C, const JobAction &JA,
     CmdArgs.push_back(Args.MakeArgString(
         "--pxtas-path=" + Args.getLastArgValue(options::OPT_ptxas_path_EQ)));
 
-  if (Args.hasArg(options::OPT_cuda_path_EQ))
-    CmdArgs.push_back(Args.MakeArgString(
-        "--cuda-path=" + Args.getLastArgValue(options::OPT_cuda_path_EQ)));
+  if (Args.hasArg(options::OPT_cuda_path_EQ) || TC.CudaInstallation.isValid()) {
+    StringRef CudaPath = Args.getLastArgValue(
+        options::OPT_cuda_path_EQ,
+        llvm::sys::path::parent_path(TC.CudaInstallation.getBinPath()));
+    CmdArgs.push_back(Args.MakeArgString("--cuda-path=" + CudaPath));
+  }
 
   // Add paths specified in LIBRARY_PATH environment variable as -L options.
   addDirectoryList(Args, CmdArgs, "-L", "LIBRARY_PATH");
diff --git a/clang/test/Driver/cuda-cross-compiling.c b/clang/test/Driver/cuda-cross-compiling.c
index 1df231ecb447..6717ae82d0e7 100644
--- a/clang/test/Driver/cuda-cross-compiling.c
+++ b/clang/test/Driver/cuda-cross-compiling.c
@@ -104,3 +104,11 @@
 // RUN:   -nogpulib -nogpuinc -### %s 2>&1 | FileCheck -check-prefix=STARTUP %s
 
 // STARTUP: clang-nvlink-wrapper{{.*}}"-lc" "-lm" "{{.*}}crt1.o"
+
+//
+// Test cuda path handling
+//
+// RUN: %clang -target nvptx64-nvidia-cuda -march=sm_52 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
+// RUN:   -nogpulib -nogpuinc -### %s 2>&1 | FileCheck -check-prefix=PATH %s
+
+// PATH: clang-nvlink-wrapper{{.*}}"--cuda-path={{.*}}/Inputs/CUDA/usr/local/cuda"
diff --git a/clang/tools/clang-nvlink-wrapper/ClangNVLinkWrapper.cpp b/clang/tools/clang-nvlink-wrapper/ClangNVLinkWrapper.cpp
index 4b63971214f8..58eb671c6198 100644
--- a/clang/tools/clang-nvlink-wrapper/ClangNVLinkWrapper.cpp
+++ b/clang/tools/clang-nvlink-wrapper/ClangNVLinkWrapper.cpp
@@ -286,12 +286,16 @@ struct Symbol {
 };
 
 Expected<StringRef> runPTXAs(StringRef File, const ArgList &Args) {
-  std::string CudaPath = Args.getLastArgValue(OPT_cuda_path_EQ).str();
-  std::string GivenPath = Args.getLastArgValue(OPT_ptxas_path_EQ).str();
-  Expected<std::string> PTXAsPath =
-      findProgram(Args, "ptxas", {CudaPath + "/bin", GivenPath});
+  SmallVector<StringRef, 1> SearchPaths;
+  if (Arg *A = Args.getLastArg(OPT_cuda_path_EQ))
+    SearchPaths.push_back(Args.MakeArgString(A->getValue() + Twine("/bin")));
+  if (Arg *A = Args.getLastArg(OPT_ptxas_path_EQ))
+    SearchPaths.push_back(Args.MakeArgString(A->getValue()));
+
+  Expected<std::string> PTXAsPath = findProgram(Args, "ptxas", SearchPaths);
   if (!PTXAsPath)
     return PTXAsPath.takeError();
+
   if (!Args.hasArg(OPT_arch))
     return createStringError(
         "must pass in an explicit nvptx64 gpu architecture to 'ptxas'");
@@ -691,9 +695,11 @@ Error runNVLink(ArrayRef<StringRef> Files, const ArgList &Args) {
   if (Args.hasArg(OPT_lto_emit_asm) || Args.hasArg(OPT_lto_emit_llvm))
     return Error::success();
 
-  std::string CudaPath = Args.getLastArgValue(OPT_cuda_path_EQ).str();
-  Expected<std::string> NVLinkPath =
-      findProgram(Args, "nvlink", {CudaPath + "/bin"});
+  SmallVector<StringRef, 1> SearchPaths;
+  if (Arg *A = Args.getLastArg(OPT_cuda_path_EQ))
+    SearchPaths.push_back(Args.MakeArgString(A->getValue() + Twine("/bin")));
+
+  Expected<std::string> NVLinkPath = findProgram(Args, "nvlink", SearchPaths);
   if (!NVLinkPath)
     return NVLinkPath.takeError();