diff --git a/clang/lib/Driver/ToolChains/Cuda.cpp b/clang/lib/Driver/ToolChains/Cuda.cpp
index 1462576ca870e..8a9d0caaccf30 100644
--- a/clang/lib/Driver/ToolChains/Cuda.cpp
+++ b/clang/lib/Driver/ToolChains/Cuda.cpp
@@ -738,9 +738,22 @@ NVPTXToolChain::TranslateArgs(const llvm::opt::DerivedArgList &Args,
     if (!llvm::is_contained(*DAL, A))
       DAL->append(A);
 
-  if (!DAL->hasArg(options::OPT_march_EQ))
+  if (!DAL->hasArg(options::OPT_march_EQ)) {
     DAL->AddJoinedArg(nullptr, Opts.getOption(options::OPT_march_EQ),
                       CudaArchToString(CudaArch::CudaDefault));
+  } else if (DAL->getLastArgValue(options::OPT_march_EQ) == "native") {
+    auto GPUsOrErr = getSystemGPUArchs(Args);
+    if (!GPUsOrErr) {
+      getDriver().Diag(diag::err_drv_undetermined_gpu_arch)
+          << getArchName() << llvm::toString(GPUsOrErr.takeError()) << "-march";
+    } else {
+      if (GPUsOrErr->size() > 1)
+        getDriver().Diag(diag::warn_drv_multi_gpu_arch)
+            << getArchName() << llvm::join(*GPUsOrErr, ", ") << "-march";
+      DAL->AddJoinedArg(nullptr, Opts.getOption(options::OPT_march_EQ),
+                        Args.MakeArgString(GPUsOrErr->front()));
+    }
+  }
 
   return DAL;
 }
@@ -783,6 +796,31 @@ void NVPTXToolChain::adjustDebugInfoKind(
   }
 }
 
+Expected<SmallVector<std::string>>
+NVPTXToolChain::getSystemGPUArchs(const ArgList &Args) const {
+  // Detect NVIDIA GPUs availible on the system.
+  std::string Program;
+  if (Arg *A = Args.getLastArg(options::OPT_nvptx_arch_tool_EQ))
+    Program = A->getValue();
+  else
+    Program = GetProgramPath("nvptx-arch");
+
+  auto StdoutOrErr = executeToolChainProgram(Program);
+  if (!StdoutOrErr)
+    return StdoutOrErr.takeError();
+
+  SmallVector<std::string, 1> GPUArchs;
+  for (StringRef Arch : llvm::split((*StdoutOrErr)->getBuffer(), "\n"))
+    if (!Arch.empty())
+      GPUArchs.push_back(Arch.str());
+
+  if (GPUArchs.empty())
+    return llvm::createStringError(std::error_code(),
+                                   "No NVIDIA GPU detected in the system");
+
+  return std::move(GPUArchs);
+}
+
 /// CUDA toolchain.  Our assembler is ptxas, and our "linker" is fatbinary,
 /// which isn't properly a linker but nonetheless performs the step of stitching
 /// together object files from the assembler into a single blob.
@@ -948,31 +986,6 @@ CudaToolChain::TranslateArgs(const llvm::opt::DerivedArgList &Args,
   return DAL;
 }
 
-Expected<SmallVector<std::string>>
-CudaToolChain::getSystemGPUArchs(const ArgList &Args) const {
-  // Detect NVIDIA GPUs availible on the system.
-  std::string Program;
-  if (Arg *A = Args.getLastArg(options::OPT_nvptx_arch_tool_EQ))
-    Program = A->getValue();
-  else
-    Program = GetProgramPath("nvptx-arch");
-
-  auto StdoutOrErr = executeToolChainProgram(Program);
-  if (!StdoutOrErr)
-    return StdoutOrErr.takeError();
-
-  SmallVector<std::string, 1> GPUArchs;
-  for (StringRef Arch : llvm::split((*StdoutOrErr)->getBuffer(), "\n"))
-    if (!Arch.empty())
-      GPUArchs.push_back(Arch.str());
-
-  if (GPUArchs.empty())
-    return llvm::createStringError(std::error_code(),
-                                   "No NVIDIA GPU detected in the system");
-
-  return std::move(GPUArchs);
-}
-
 Tool *NVPTXToolChain::buildAssembler() const {
   return new tools::NVPTX::Assembler(*this);
 }
diff --git a/clang/lib/Driver/ToolChains/Cuda.h b/clang/lib/Driver/ToolChains/Cuda.h
index 8a053f3393e12..43c17ba7c0ba0 100644
--- a/clang/lib/Driver/ToolChains/Cuda.h
+++ b/clang/lib/Driver/ToolChains/Cuda.h
@@ -168,6 +168,11 @@ class LLVM_LIBRARY_VISIBILITY NVPTXToolChain : public ToolChain {
   unsigned GetDefaultDwarfVersion() const override { return 2; }
   unsigned getMaxDwarfVersion() const override { return 2; }
 
+  /// Uses nvptx-arch tool to get arch of the system GPU. Will return error
+  /// if unable to find one.
+  virtual Expected<SmallVector<std::string>>
+  getSystemGPUArchs(const llvm::opt::ArgList &Args) const override;
+
   CudaInstallationDetector CudaInstallation;
 
 protected:
@@ -223,11 +228,6 @@ class LLVM_LIBRARY_VISIBILITY CudaToolChain : public NVPTXToolChain {
 
   const ToolChain &HostTC;
 
-  /// Uses nvptx-arch tool to get arch of the system GPU. Will return error
-  /// if unable to find one.
-  virtual Expected<SmallVector<std::string>>
-  getSystemGPUArchs(const llvm::opt::ArgList &Args) const override;
-
 protected:
   Tool *buildAssembler() const override; // ptxas
   Tool *buildLinker() const override;    // fatbinary (ok, not really a linker)
diff --git a/clang/test/Driver/Inputs/nvptx-arch/nvptx_arch_sm_89_sm_80 b/clang/test/Driver/Inputs/nvptx-arch/nvptx_arch_sm_89_sm_80
new file mode 100644
index 0000000000000..c9df565b9f8d7
--- /dev/null
+++ b/clang/test/Driver/Inputs/nvptx-arch/nvptx_arch_sm_89_sm_80
@@ -0,0 +1,4 @@
+#!/bin/sh
+echo sm_89
+echo sm_80
+exit 0
diff --git a/clang/test/Driver/nvptx-cuda-system-arch.c b/clang/test/Driver/nvptx-cuda-system-arch.c
index 037215fd52a88..81cad37150669 100644
--- a/clang/test/Driver/nvptx-cuda-system-arch.c
+++ b/clang/test/Driver/nvptx-cuda-system-arch.c
@@ -6,9 +6,11 @@
 // RUN: mkdir -p %t
 // RUN: cp %S/Inputs/nvptx-arch/nvptx_arch_fail %t/
 // RUN: cp %S/Inputs/nvptx-arch/nvptx_arch_sm_70 %t/
+// RUN: cp %S/Inputs/nvptx-arch/nvptx_arch_sm_89_sm_80 %t/
 // RUN: echo '#!/bin/sh' > %t/nvptx_arch_empty
 // RUN: chmod +x %t/nvptx_arch_fail
 // RUN: chmod +x %t/nvptx_arch_sm_70
+// RUN: chmod +x %t/nvptx_arch_sm_89_sm_80
 // RUN: chmod +x %t/nvptx_arch_empty
 
 // case when nvptx-arch returns nothing or fails
@@ -31,3 +33,14 @@
 // RUN:   %clang -### --target=x86_64-unknown-linux-gnu -nogpulib --offload-arch=native --offload-new-driver --nvptx-arch-tool=%t/nvptx_arch_sm_70 -x cuda --cuda-path=%S/Inputs/CUDA_102/usr/local/cuda %s 2>&1 \
 // RUN:   | FileCheck %s --check-prefix=ARCH-sm_70
 // ARCH-sm_70: "-cc1" "-triple" "nvptx64-nvidia-cuda"{{.*}}"-target-cpu" "sm_70"
+
+// case when nvptx-arch is used via '-march=native'
+// RUN:   %clang -### --target=nvptx64-nvidia-cuda -nogpulib -march=native --nvptx-arch-tool=%t/nvptx_arch_sm_70 \
+// RUN:     --cuda-path=%S/Inputs/CUDA_102/usr/local/cuda %s 2>&1 | FileCheck %s --check-prefix=MARCH-sm_70
+// MARCH-sm_70: "-cc1" "-triple" "nvptx64-nvidia-cuda"{{.*}}"-target-cpu" "sm_70"
+
+// case when nvptx-arch is used via '-march=native'
+// RUN:   %clang -### --target=nvptx64-nvidia-cuda -nogpulib -march=native --nvptx-arch-tool=%t/nvptx_arch_sm_89_sm_80 \
+// RUN:     --cuda-path=%S/Inputs/CUDA_102/usr/local/cuda %s 2>&1 | FileCheck %s --check-prefix=MARCH-sm_89
+// MARCH-sm_89: warning: multiple nvptx64 architectures are detected: sm_89, sm_80; only the first one is used for '-march' [-Wmulti-gpu]
+// MARCH-sm_89: "-cc1" "-triple" "nvptx64-nvidia-cuda"{{.*}}"-target-cpu" "sm_89"