[Driver] Driver changes to support CUDA compilation on Windows.

[jlebar]

Summary

For the most part this is straightforward: Just add a CudaInstallation
object to the MSVC and MinGW toolchains.

CudaToolChain has to override computeMSVCVersion so that
Clang::constructJob passes the right version flag to cc1. We have to
modify IsWindowsMSVC and friends in Clang::constructJob to be true when
compiling CUDA device code on Windows for the same reason.

Depends on: D28319

[tra]

clang/lib/Driver/ToolChains.cpp

1819

I'd keep unix-like paths as defaults under else {}.
If you do want to enumerate supported OS variants explicitly, then me want to throw an error if particular host OS is not supported.

[jlebar]

Thank you for the reviews!

clang/lib/Driver/ToolChains.cpp

1819

Sure, done locally, will be reflected when I commit.

[chapuni]

It fails with LLVM_DEFAULT_TARGET_TRIPLE=i686-pc-win32, on linux.
http://bb.pgr.jp/builders/ninja-x64-msvc-RA-centos6/builds/33445
Seems it assumes Windows' layout even if the host is not Windows.

128838 stat64("/path/to/llvm-project/clang/test/Driver/Inputs/CUDA/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v8.0", 0xffc9bdc0) = -1 ENOENT (No such file or directory)

If you wouldn't consider such a configuration, I could add exclusion to lit.cfg.

[rnk]

I went ahead and landed rL291255 to try to fix forward. Take a look and let me know if that's not the right fix.

clang/include/clang/Driver/ToolChain.h

public:
   virtual ~ToolChain();
 
   // Accessors
 
   const Driver &getDriver() const { return D; }
   vfs::FileSystem &getVFS() const;
   const llvm::Triple &getTriple() const { return Triple; }
 
+  /// Get the toolchain's aux triple, if it has one.
+  ///
+  /// Exactly what the aux triple represents depends on the toolchain, but for
+  /// example when compiling CUDA code for the GPU, the triple might be NVPTX,
+  /// while the aux triple is the host (CPU) toolchain, e.g. x86-linux-gnu.
+  virtual const llvm::Triple *getAuxTriple() const { return nullptr; }
+
   llvm::Triple::ArchType getArch() const { return Triple.getArch(); }
   StringRef getArchName() const { return Triple.getArchName(); }
   StringRef getPlatform() const { return Triple.getVendorName(); }
   StringRef getOS() const { return Triple.getOSName(); }
 
   /// \brief Provide the default architecture name (as expected by -arch) for
   /// this toolchain.
   StringRef getDefaultUniversalArchName() const;

clang/lib/Driver/MSVCToolChain.cpp

#ifdef USE_WIN32
   #include <windows.h>
 #endif
 
 using namespace clang::driver;
 using namespace clang::driver::toolchains;
 using namespace clang;
 using namespace llvm::opt;
 
-MSVCToolChain::MSVCToolChain(const Driver &D, const llvm::Triple& Triple,
+MSVCToolChain::MSVCToolChain(const Driver &D, const llvm::Triple &Triple,
                              const ArgList &Args)
-  : ToolChain(D, Triple, Args) {
+    : ToolChain(D, Triple, Args), CudaInstallation(D, Triple, Args) {
   getProgramPaths().push_back(getDriver().getInstalledDir());
   if (getDriver().getInstalledDir() != getDriver().Dir)
     getProgramPaths().push_back(getDriver().Dir);
 }
 
 Tool *MSVCToolChain::buildLinker() const {
   return new tools::visualstudio::Linker(*this);
 }

 bool MSVCToolChain::isPIEDefault() const {
   return false;
 }
 
 bool MSVCToolChain::isPICDefaultForced() const {
   return getArch() == llvm::Triple::x86_64;
 }
 
+void MSVCToolChain::AddCudaIncludeArgs(const ArgList &DriverArgs,
+                                       ArgStringList &CC1Args) const {
+  CudaInstallation.AddCudaIncludeArgs(DriverArgs, CC1Args);
+}
+
+void MSVCToolChain::printVerboseInfo(raw_ostream &OS) const {
+  CudaInstallation.print(OS);
+}
+
 #ifdef USE_WIN32
 static bool readFullStringValue(HKEY hkey, const char *valueName,
                                 std::string &value) {
   std::wstring WideValueName;
   if (!llvm::ConvertUTF8toWide(valueName, WideValueName))
     return false;
 
   DWORD result = 0;

clang/lib/Driver/MinGWToolChain.cpp

for (StringRef CandidateArch : Archs) {
         Arch = CandidateArch;
         return;
       }
     }
   }
 }
 
 MinGW::MinGW(const Driver &D, const llvm::Triple &Triple, const ArgList &Args)
-    : ToolChain(D, Triple, Args) {
+    : ToolChain(D, Triple, Args), CudaInstallation(D, Triple, Args) {
   getProgramPaths().push_back(getDriver().getInstalledDir());
 
 // In Windows there aren't any standard install locations, we search
 // for gcc on the PATH. In Linux the base is always /usr.
 #ifdef LLVM_ON_WIN32
   if (getDriver().SysRoot.size())
     Base = getDriver().SysRoot;
   else if (llvm::ErrorOr<std::string> GPPName =

 bool MinGW::isPICDefaultForced() const {
   return getArch() == llvm::Triple::x86_64;
 }
 
 bool MinGW::UseSEHExceptions() const {
   return getArch() == llvm::Triple::x86_64;
 }
 
+void MinGW::AddCudaIncludeArgs(const ArgList &DriverArgs,
+                               ArgStringList &CC1Args) const {
+  CudaInstallation.AddCudaIncludeArgs(DriverArgs, CC1Args);
+}
+
+void MinGW::printVerboseInfo(raw_ostream &OS) const {
+  CudaInstallation.print(OS);
+}
+
 // Include directories for various hosts:
 
 // Windows, mingw.org
 // c:\mingw\lib\gcc\mingw32\4.8.1\include\c++
 // c:\mingw\lib\gcc\mingw32\4.8.1\include\c++\mingw32
 // c:\mingw\lib\gcc\mingw32\4.8.1\include\c++\backward
 // c:\mingw\lib\gcc\mingw32\4.8.1\include
 // c:\mingw\include

clang/lib/Driver/ToolChains.h

public:
 
   void
   AddClangSystemIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                             llvm::opt::ArgStringList &CC1Args) const override;
   void AddClangCXXStdlibIncludeArgs(
       const llvm::opt::ArgList &DriverArgs,
       llvm::opt::ArgStringList &CC1Args) const override;
 
+  void AddCudaIncludeArgs(const llvm::opt::ArgList &DriverArgs,
+                          llvm::opt::ArgStringList &CC1Args) const override;
+
+  void printVerboseInfo(raw_ostream &OS) const override;
+
 protected:
   Tool *getTool(Action::ActionClass AC) const override;
   Tool *buildLinker() const override;
   Tool *buildAssembler() const override;
 
 private:
+  CudaInstallationDetector CudaInstallation;
+
   std::string Base;
   std::string GccLibDir;
   std::string Ver;
   std::string Arch;
   mutable std::unique_ptr<tools::gcc::Preprocessor> Preprocessor;
   mutable std::unique_ptr<tools::gcc::Compiler> Compiler;
   void findGccLibDir();
 };
protected:
   Tool *buildLinker() const override;
 };
 
 class LLVM_LIBRARY_VISIBILITY CudaToolChain : public ToolChain {
 public:
   CudaToolChain(const Driver &D, const llvm::Triple &Triple,
                 const ToolChain &HostTC, const llvm::opt::ArgList &Args);
 
+  virtual const llvm::Triple *getAuxTriple() const override {
+    return &HostTC.getTriple();
+  }
+
   llvm::opt::DerivedArgList *
   TranslateArgs(const llvm::opt::DerivedArgList &Args, StringRef BoundArch,
                 Action::OffloadKind DeviceOffloadKind) const override;
   void addClangTargetOptions(const llvm::opt::ArgList &DriverArgs,
                              llvm::opt::ArgStringList &CC1Args) const override;
 
   // Never try to use the integrated assembler with CUDA; always fork out to
   // ptxas.
public:
   void AddClangCXXStdlibIncludeArgs(
       const llvm::opt::ArgList &Args,
       llvm::opt::ArgStringList &CC1Args) const override;
   void AddIAMCUIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                            llvm::opt::ArgStringList &CC1Args) const override;
 
   SanitizerMask getSupportedSanitizers() const override;
 
+  VersionTuple
+  computeMSVCVersion(const Driver *D,
+                     const llvm::opt::ArgList &Args) const override;
+
   const ToolChain &HostTC;
   CudaInstallationDetector CudaInstallation;
 
 protected:
   Tool *buildAssembler() const override;  // ptxas
   Tool *buildLinker() const override;     // fatbinary (ok, not really a linker)
 };
 
public:
 
   void
   AddClangSystemIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                             llvm::opt::ArgStringList &CC1Args) const override;
   void AddClangCXXStdlibIncludeArgs(
       const llvm::opt::ArgList &DriverArgs,
       llvm::opt::ArgStringList &CC1Args) const override;
 
+  void AddCudaIncludeArgs(const llvm::opt::ArgList &DriverArgs,
+                          llvm::opt::ArgStringList &CC1Args) const override;
+
   bool getWindowsSDKDir(std::string &path, int &major,
                         std::string &windowsSDKIncludeVersion,
                         std::string &windowsSDKLibVersion) const;
   bool getWindowsSDKLibraryPath(std::string &path) const;
   /// \brief Check if Universal CRT should be used if available
   bool useUniversalCRT(std::string &visualStudioDir) const;
   bool getUniversalCRTSdkDir(std::string &path, std::string &ucrtVersion) const;
   bool getUniversalCRTLibraryPath(std::string &path) const;
   bool getVisualStudioInstallDir(std::string &path) const;
   bool getVisualStudioBinariesFolder(const char *clangProgramPath,
                                      std::string &path) const;
   VersionTuple
   computeMSVCVersion(const Driver *D,
                      const llvm::opt::ArgList &Args) const override;
 
   std::string ComputeEffectiveClangTriple(const llvm::opt::ArgList &Args,
                                           types::ID InputType) const override;
   SanitizerMask getSupportedSanitizers() const override;
 
+  void printVerboseInfo(raw_ostream &OS) const override;
+
 protected:
   void AddSystemIncludeWithSubfolder(const llvm::opt::ArgList &DriverArgs,
                                      llvm::opt::ArgStringList &CC1Args,
                                      const std::string &folder,
                                      const Twine &subfolder1,
                                      const Twine &subfolder2 = "",
                                      const Twine &subfolder3 = "") const;
 
   Tool *buildLinker() const override;
   Tool *buildAssembler() const override;
 private:
   VersionTuple getMSVCVersionFromTriple() const;
   VersionTuple getMSVCVersionFromExe() const;
+
+  CudaInstallationDetector CudaInstallation;
 };
 
 class LLVM_LIBRARY_VISIBILITY CrossWindowsToolChain : public Generic_GCC {
 public:
   CrossWindowsToolChain(const Driver &D, const llvm::Triple &T,
                         const llvm::opt::ArgList &Args);
 
   bool IsIntegratedAssemblerDefault() const override { return true; }

clang/lib/Driver/ToolChains.cpp

 }
 
 CudaInstallationDetector::CudaInstallationDetector(
     const Driver &D, const llvm::Triple &HostTriple,
     const llvm::opt::ArgList &Args)
     : D(D) {
   SmallVector<std::string, 4> CudaPathCandidates;
 
+  // In decreasing order so we prefer newer versions to older versions.
+  std::initializer_list<const char *> Versions = {"8.0", "7.5", "7.0"};
+
   if (Args.hasArg(options::OPT_cuda_path_EQ))
     CudaPathCandidates.push_back(
         Args.getLastArgValue(options::OPT_cuda_path_EQ));
-  else {
+  else if (HostTriple.isOSLinux() || HostTriple.isMacOSX()) {
     CudaPathCandidates.push_back(D.SysRoot + "/usr/local/cuda");
-    CudaPathCandidates.push_back(D.SysRoot + "/usr/local/cuda-8.0");
-    CudaPathCandidates.push_back(D.SysRoot + "/usr/local/cuda-7.5");
-    CudaPathCandidates.push_back(D.SysRoot + "/usr/local/cuda-7.0");
+    for (const char *Ver : Versions)
+      CudaPathCandidates.push_back(D.SysRoot + "/usr/local/cuda-" + Ver);
+  } else if (HostTriple.isOSWindows()) {
+    for (const char *Ver : Versions)
+      CudaPathCandidates.push_back(
+          D.SysRoot + "/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v" +
+          Ver);
   }
 
   for (const auto &CudaPath : CudaPathCandidates) {
     if (CudaPath.empty() || !D.getVFS().exists(CudaPath))
       continue;
 
     InstallPath = CudaPath;
     BinPath = CudaPath + "/bin";
SanitizerMask CudaToolChain::getSupportedSanitizers() const {
   // sanitization of device code is actually supported at this time.
   //
   // This behavior is necessary because the host and device toolchains
   // invocations often share the command line, so the device toolchain must
   // tolerate flags meant only for the host toolchain.
   return HostTC.getSupportedSanitizers();
 }
 
+VersionTuple CudaToolChain::computeMSVCVersion(const Driver *D,
+                                               const ArgList &Args) const {
+  return HostTC.computeMSVCVersion(D, Args);
+}
+
 /// XCore tool chain
 XCoreToolChain::XCoreToolChain(const Driver &D, const llvm::Triple &Triple,
                                const ArgList &Args)
     : ToolChain(D, Triple, Args) {
   // ProgramPaths are found via 'PATH' environment variable.
 }
 
 Tool *XCoreToolChain::buildAssembler() const {

clang/lib/Driver/Tools.cpp

void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   const llvm::Triple &Triple = getToolChain().getEffectiveTriple();
   const std::string &TripleStr = Triple.getTriple();
 
   bool KernelOrKext =
       Args.hasArg(options::OPT_mkernel, options::OPT_fapple_kext);
   const Driver &D = getToolChain().getDriver();
   ArgStringList CmdArgs;
 
-  bool IsWindowsGNU = getToolChain().getTriple().isWindowsGNUEnvironment();
-  bool IsWindowsCygnus =
-      getToolChain().getTriple().isWindowsCygwinEnvironment();
-  bool IsWindowsMSVC = getToolChain().getTriple().isWindowsMSVCEnvironment();
-  bool IsPS4CPU = getToolChain().getTriple().isPS4CPU();
-  bool IsIAMCU = getToolChain().getTriple().isOSIAMCU();
-
   // Check number of inputs for sanity. We need at least one input.
   assert(Inputs.size() >= 1 && "Must have at least one input.");
   const InputInfo &Input = Inputs[0];
   // CUDA compilation may have multiple inputs (source file + results of
   // device-side compilations). OpenMP device jobs also take the host IR as a
   // second input. All other jobs are expected to have exactly one
   // input.
   bool IsCuda = JA.isOffloading(Action::OFK_Cuda);
   bool IsOpenMPDevice = JA.isDeviceOffloading(Action::OFK_OpenMP);
   assert((IsCuda || (IsOpenMPDevice && Inputs.size() == 2) ||
           Inputs.size() == 1) &&
          "Unable to handle multiple inputs.");
 
+  bool IsWindowsGNU = getToolChain().getTriple().isWindowsGNUEnvironment();
+  bool IsWindowsCygnus =
+      getToolChain().getTriple().isWindowsCygwinEnvironment();
+  bool IsWindowsMSVC = getToolChain().getTriple().isWindowsMSVCEnvironment();
+  bool IsPS4CPU = getToolChain().getTriple().isPS4CPU();
+  bool IsIAMCU = getToolChain().getTriple().isOSIAMCU();
+
+  // Adjust IsWindowsXYZ for CUDA compilations.  Even when compiling in device
+  // mode (i.e., getToolchain().getTriple() is NVPTX, not Windows), we need to
+  // pass Windows-specific flags to cc1.
+  if (IsCuda) {
+    const llvm::Triple *AuxTriple = getToolChain().getAuxTriple();
+    IsWindowsMSVC |= AuxTriple && AuxTriple->isWindowsMSVCEnvironment();
+    IsWindowsGNU |= AuxTriple && AuxTriple->isWindowsGNUEnvironment();
+    IsWindowsCygnus |= AuxTriple && AuxTriple->isWindowsCygwinEnvironment();
+  }
+
   // C++ is not supported for IAMCU.
   if (IsIAMCU && types::isCXX(Input.getType()))
     D.Diag(diag::err_drv_clang_unsupported) << "C++ for IAMCU";
 
   // Invoke ourselves in -cc1 mode.
   //
   // FIXME: Implement custom jobs for internal actions.
   CmdArgs.push_back("-cc1");

clang/test/Driver/cuda-windows.cu

+// REQUIRES: clang-driver
+// REQUIRES: x86-registered-target
+// REQUIRES: nvptx-registered-target
+//
+// RUN: %clang -v --target=i386-pc-windows-msvc \
+// RUN:   --sysroot=%S/Inputs/CUDA-windows 2>&1 %s -### | FileCheck %s
+// RUN: %clang -v --target=i386-pc-windows-mingw32 \
+// RUN:   --sysroot=%S/Inputs/CUDA-windows 2>&1 %s -### | FileCheck %s
+
+// CHECK: Found CUDA installation: {{.*}}/Inputs/CUDA-windows/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v8.0
+// CHECK: "-cc1" "-triple" "nvptx-nvidia-cuda"
+// CHECK-SAME: "-fms-extensions"
+// CHECK-SAME: "-fms-compatibility"
+// CHECK-SAME: "-fms-compatibility-version=