[OpenMP] Extend CLANG target options with device offloading kind.

[gtbercea]

Summary

Pass the type of the device offloading when building the tool chain for a particular target architecture. This is required when supporting multiple tool chains that target a single device type. In our particular use case, the OpenMP and CUDA tool chains will use the same `addClangTargetOptions ` method. This enables the reuse of common options and ensures control over options only supported by a particular tool chain.

[jlebar]

lib/Driver/ToolChains.cpp

4902 ↗

(On Diff #87440)

Not sure this assertion message helps us much beyond what's already in the code.

4914 ↗

(On Diff #87440)

Are these changes related to this patch?

I have no problem cleaning up whitespace errors like these, but would prefer for them to be split out separately if possible.

4961 ↗

(On Diff #87440)

s/device/compilation/?

4961 ↗

(On Diff #87440)

An "otherwise" would probably be helpful in this comment.

4966 ↗

(On Diff #87440)

TODO is idiomatically one word

lib/Driver/Tools.cpp

12136 ↗

(On Diff #87440)

Why does JA.getOffloadingArch have the wrong value? Isn't the purpose of JA.getOffloadingArch to have this particular value? If it's broken, can we fix it instead of doing this hack?

[gtbercea]

Update patch to reflect latest source code changes.

[gtbercea]

lib/Driver/ToolChains.cpp

4902 ↗

(On Diff #87440)

Agreed. I've changed it to reflect that, currently, only OpenMP and CUDA offloading kinds are supported for this toolchain.

4914 ↗

(On Diff #87440)

Reverted them, since they only concern line length.

4961 ↗

(On Diff #87440)

Changed.

[gtbercea]

lib/Driver/Tools.cpp

12136 ↗

(On Diff #87440)

I have to investigate this. Thanks for pointing this out.

[Hahnfeld]

One minor drive-by comment. I think there is still one outstanding from Justin...

lib/Driver/ToolChains/Cuda.cpp

347

You can move the declaration down to the initialization

[hfinkel]

In general, this patch seems to be missing tests (unless it is actually NFC, or you can't write tests yet, which, in either case, need to be explained).

lib/Driver/ToolChains/Cuda.cpp

217

Do we have a test case for this?

412

Why is this a TODO? Is the necessary information not currently available in the offloading arguments, or are we just not grabbing it in this patch?

[gtbercea]

lib/Driver/ToolChains/Cuda.cpp

412

This is for a future patch in which we will grab the compute capability from a special flag. I believe there has already been some discussion on the dev mailing list as to what that flag should be but I don't think there was a general consensus as to how that flag should be named.

[hfinkel]

lib/Driver/ToolChains/Cuda.cpp

412

Which mailing list thread discussed this?

[gtbercea]

lib/Driver/ToolChains/Cuda.cpp

412

I believe it was openmp-dev. The proposal was initially posted by Gregory Rodgers.

[hfinkel]

lib/Driver/ToolChains/Cuda.cpp

412

Can you please find it and post the link from here: http://lists.llvm.org/pipermail/openmp-dev/

I do recall discussing this at some point somewhere, but I'm not finding it in my email.

[gtbercea]

lib/Driver/ToolChains/Cuda.cpp

412

I managed to find the e-mail at last. It looks like it was a side conversation which ended with Justin Lebar suggesting Greg to post the discussion on the llvm-dev list. That post never happened. :-(

[gtbercea]

Updated diff to address comments.

[gtbercea]

lib/Driver/ToolChains/Cuda.cpp

217

Just added a test case for this.

347

Done! Thanks for spotting that.

[hfinkel]

test/Driver/openmp-offload.c

614 ↗

(On Diff #104427)

How does this work on x86 where the host also uses -march?

[gtbercea]

test/Driver/openmp-offload.c

614 ↗

(On Diff #104427)

That's where the additional flag comes in. The new flag *should* be used instead.

[hfinkel]

test/Driver/openmp-offload.c

614 ↗

(On Diff #104427)

Ah, okay. Please just propose a patch which adds a new flag. This workaround is a bit strong for my tastes.

[gtbercea]

test/Driver/openmp-offload.c

614 ↗

(On Diff #104427)

Completely agree! It's not a solution I am happy with it either as temporary as it is. I will propose a patch with a new flag.

[gtbercea]

Split previous diff into a "device offloading kind" patch (show here) and a new patch D34784 which relies on a new compiler flag.

A TODO has been added to signal that the compute capability is to be handled in the new patch mentioned above.

[hfinkel]

LGTM

[hfinkel]

In D29647#795271, @hfinkel wrote:

LGTM

When you commit this, please make sure to mention in the commit message that the test cases will be associated with follow-up commits.

[gtbercea]

.

include/clang/Driver/ToolChain.h

public:
   /// This routine is responsible for adding the necessary cc1 arguments to
   /// include headers from standard system header directories.
   virtual void
   AddClangSystemIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                             llvm::opt::ArgStringList &CC1Args) const;
 
   /// \brief Add options that need to be passed to cc1 for this target.
   virtual void addClangTargetOptions(const llvm::opt::ArgList &DriverArgs,
-                                     llvm::opt::ArgStringList &CC1Args) const;
+                                     llvm::opt::ArgStringList &CC1Args,
+                                     Action::OffloadKind DeviceOffloadKind) const;
 
   /// \brief Add warning options that need to be passed to cc1 for this target.
   virtual void addClangWarningOptions(llvm::opt::ArgStringList &CC1Args) const;
 
   // GetRuntimeLibType - Determine the runtime library type to use with the
   // given compilation arguments.
   virtual RuntimeLibType
   GetRuntimeLibType(const llvm::opt::ArgList &Args) const;

lib/Driver/ToolChain.cpp

std::string ToolChain::ComputeEffectiveClangTriple(const ArgList &Args,
   return ComputeLLVMTriple(Args, InputType);
 }
 
 void ToolChain::AddClangSystemIncludeArgs(const ArgList &DriverArgs,
                                           ArgStringList &CC1Args) const {
   // Each toolchain should provide the appropriate include flags.
 }
 
-void ToolChain::addClangTargetOptions(const ArgList &DriverArgs,
-                                      ArgStringList &CC1Args) const {
-}
+void ToolChain::addClangTargetOptions(
+    const ArgList &DriverArgs, ArgStringList &CC1Args,
+    Action::OffloadKind DeviceOffloadKind) const {}
 
 void ToolChain::addClangWarningOptions(ArgStringList &CC1Args) const {}
 
 void ToolChain::addProfileRTLibs(const llvm::opt::ArgList &Args,
                                  llvm::opt::ArgStringList &CmdArgs) const {
   if (!needsProfileRT(Args)) return;
 
   CmdArgs.push_back(getCompilerRTArgString(Args, "profile"));

lib/Driver/ToolChains/BareMetal.h

public:
   }
 
   const char *getDefaultLinker() const override { return "ld.lld"; }
 
   std::string getRuntimesDir() const;
   void AddClangSystemIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                                  llvm::opt::ArgStringList &CC1Args) const override;
   void addClangTargetOptions(const llvm::opt::ArgList &DriverArgs,
-                             llvm::opt::ArgStringList &CC1Args) const override;
+                             llvm::opt::ArgStringList &CC1Args,
+                             Action::OffloadKind DeviceOffloadKind) const override;
   std::string findLibCxxIncludePath(ToolChain::CXXStdlibType LibType) const;
   void AddClangCXXStdlibIncludeArgs(
       const llvm::opt::ArgList &DriverArgs,
       llvm::opt::ArgStringList &CC1Args) const override;
   void AddCXXStdlibLibArgs(const llvm::opt::ArgList &Args,
                            llvm::opt::ArgStringList &CmdArgs) const override;
   void AddLinkRuntimeLib(const llvm::opt::ArgList &Args,
                          llvm::opt::ArgStringList &CmdArgs) const;

lib/Driver/ToolChains/BareMetal.cpp

void BareMetal::AddClangSystemIncludeArgs(const ArgList &DriverArgs,
   if (!DriverArgs.hasArg(options::OPT_nostdlibinc)) {
     SmallString<128> Dir(getDriver().SysRoot);
     llvm::sys::path::append(Dir, "include");
     addSystemInclude(DriverArgs, CC1Args, Dir.str());
   }
 }
 
 void BareMetal::addClangTargetOptions(const ArgList &DriverArgs,
-                                      ArgStringList &CC1Args) const {
+                                      ArgStringList &CC1Args,
+                                      Action::OffloadKind) const {
   CC1Args.push_back("-nostdsysteminc");
 }
 
 std::string BareMetal::findLibCxxIncludePath(CXXStdlibType LibType) const {
   StringRef SysRoot = getDriver().SysRoot;
   if (SysRoot.empty())
     return "";
 

lib/Driver/ToolChains/Clang.cpp

bool AsynchronousUnwindTables =
                    options::OPT_fno_asynchronous_unwind_tables,
                    (getToolChain().IsUnwindTablesDefault() ||
                     getToolChain().getSanitizerArgs().needsUnwindTables()) &&
                        !KernelOrKext);
   if (Args.hasFlag(options::OPT_funwind_tables, options::OPT_fno_unwind_tables,
                    AsynchronousUnwindTables))
     CmdArgs.push_back("-munwind-tables");
 
-  getToolChain().addClangTargetOptions(Args, CmdArgs);
+  getToolChain().addClangTargetOptions(Args, CmdArgs,
+                                       JA.getOffloadingDeviceKind());
 
   if (Arg *A = Args.getLastArg(options::OPT_flimited_precision_EQ)) {
     CmdArgs.push_back("-mlimit-float-precision");
     CmdArgs.push_back(A->getValue());
   }
 
   // FIXME: Handle -mtune=.
   (void)Args.hasArg(options::OPT_mtune_EQ);

lib/Driver/ToolChains/Cuda.h

public:
   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;
+                             llvm::opt::ArgStringList &CC1Args,
+                             Action::OffloadKind DeviceOffloadKind) const override;
 
   // Never try to use the integrated assembler with CUDA; always fork out to
   // ptxas.
   bool useIntegratedAs() const override { return false; }
   bool isCrossCompiling() const override { return true; }
   bool isPICDefault() const override { return false; }
   bool isPIEDefault() const override { return false; }
   bool isPICDefaultForced() const override { return false; }

lib/Driver/ToolChains/Cuda.cpp

void NVPTX::Assembler::ConstructJob(Compilation &C, const JobAction &JA,
                                     const ArgList &Args,
                                     const char *LinkingOutput) const {
   const auto &TC =
       static_cast<const toolchains::CudaToolChain &>(getToolChain());
   assert(TC.getTriple().isNVPTX() && "Wrong platform");
 
   // Obtain architecture from the action.
   CudaArch gpu_arch = StringToCudaArch(JA.getOffloadingArch());
   assert(gpu_arch != CudaArch::UNKNOWN &&
          "Device action expected to have an architecture.");
 
   // Check that our installation's ptxas supports gpu_arch.
   if (!Args.hasArg(options::OPT_no_cuda_version_check)) {
     TC.CudaInstallation.CheckCudaVersionSupportsArch(gpu_arch);
   }
 
   ArgStringList CmdArgs;
CudaToolChain::CudaToolChain(const Driver &D, const llvm::Triple &Triple,
     : ToolChain(D, Triple, Args), HostTC(HostTC),
       CudaInstallation(D, HostTC.getTriple(), Args) {
   if (CudaInstallation.isValid())
     getProgramPaths().push_back(CudaInstallation.getBinPath());
 }
 
 void CudaToolChain::addClangTargetOptions(
     const llvm::opt::ArgList &DriverArgs,
-    llvm::opt::ArgStringList &CC1Args) const {
-  HostTC.addClangTargetOptions(DriverArgs, CC1Args);
+    llvm::opt::ArgStringList &CC1Args,
+    Action::OffloadKind DeviceOffloadingKind) const {
+  HostTC.addClangTargetOptions(DriverArgs, CC1Args, DeviceOffloadingKind);
 
+  StringRef GpuArch = DriverArgs.getLastArgValue(options::OPT_march_EQ);
+  assert(!GpuArch.empty() && "Must have an explicit GPU arch.");
+  assert((DeviceOffloadingKind == Action::OFK_OpenMP ||
+          DeviceOffloadingKind == Action::OFK_Cuda) &&
+         "Only OpenMP or CUDA offloading kinds are supported for NVIDIA GPUs.");
+
+  if (DeviceOffloadingKind == Action::OFK_Cuda) {
     CC1Args.push_back("-fcuda-is-device");
 
     if (DriverArgs.hasFlag(options::OPT_fcuda_flush_denormals_to_zero,
                            options::OPT_fno_cuda_flush_denormals_to_zero, false))
       CC1Args.push_back("-fcuda-flush-denormals-to-zero");
 
     if (DriverArgs.hasFlag(options::OPT_fcuda_approx_transcendentals,
                            options::OPT_fno_cuda_approx_transcendentals, false))
       CC1Args.push_back("-fcuda-approx-transcendentals");
 
     if (DriverArgs.hasArg(options::OPT_nocudalib))
       return;
+  }
 
-  StringRef GpuArch = DriverArgs.getLastArgValue(options::OPT_march_EQ);
-  assert(!GpuArch.empty() && "Must have an explicit GPU arch.");
   std::string LibDeviceFile = CudaInstallation.getLibDeviceFile(GpuArch);
 
   if (LibDeviceFile.empty()) {
     getDriver().Diag(diag::err_drv_no_cuda_libdevice) << GpuArch;
     return;
   }
 
   CC1Args.push_back("-mlink-cuda-bitcode");
CudaToolChain::TranslateArgs(const llvm::opt::DerivedArgList &Args,
                              Action::OffloadKind DeviceOffloadKind) const {
   DerivedArgList *DAL =
       HostTC.TranslateArgs(Args, BoundArch, DeviceOffloadKind);
   if (!DAL)
     DAL = new DerivedArgList(Args.getBaseArgs());
 
   const OptTable &Opts = getDriver().getOpts();
 
+  // For OpenMP device offloading, append derived arguments. Make sure
+  // flags are not duplicated.
+  // TODO: Append the compute capability.
+  if (DeviceOffloadKind == Action::OFK_OpenMP) {
+    for (Arg *A : Args){
+      bool IsDuplicate = false;
+      for (Arg *DALArg : *DAL){
+        if (A == DALArg) {
+          IsDuplicate = true;
+          break;
+        }
+      }
+      if (!IsDuplicate)
+        DAL->append(A);
+    }
+    return DAL;
+  }
+
   for (Arg *A : Args) {
     if (A->getOption().matches(options::OPT_Xarch__)) {
       // Skip this argument unless the architecture matches BoundArch
       if (BoundArch.empty() || A->getValue(0) != BoundArch)
         continue;
 
       unsigned Index = Args.getBaseArgs().MakeIndex(A->getValue(1));
       unsigned Prev = Index;

lib/Driver/ToolChains/Darwin.h

protected:
   }
 
   /// Return true if c++17 aligned allocation/deallocation functions are not
   /// implemented in the c++ standard library of the deployment target we are
   /// targeting.
   bool isAlignedAllocationUnavailable() const;
 
   void addClangTargetOptions(const llvm::opt::ArgList &DriverArgs,
-                             llvm::opt::ArgStringList &CC1Args) const override;
+                             llvm::opt::ArgStringList &CC1Args,
+                             Action::OffloadKind DeviceOffloadKind) const override;
 
   StringRef getPlatformFamily() const;
   static StringRef getSDKName(StringRef isysroot);
   StringRef getOSLibraryNameSuffix() const;
 
 public:
   /// }
   /// @name ToolChain Implementation

lib/Driver/ToolChains/Darwin.cpp

bool Darwin::isAlignedAllocationUnavailable() const {
   case WatchOS:
   case WatchOSSimulator: // Earlier than 4.0.
     return TargetVersion < VersionTuple(4U, 0U, 0U);
   }
   llvm_unreachable("Unsupported platform");
 }
 
 void Darwin::addClangTargetOptions(const llvm::opt::ArgList &DriverArgs,
-                                   llvm::opt::ArgStringList &CC1Args) const {
+                                   llvm::opt::ArgStringList &CC1Args,
+                                   Action::OffloadKind DeviceOffloadKind) const {
   if (isAlignedAllocationUnavailable())
     CC1Args.push_back("-faligned-alloc-unavailable");
 }
 
 DerivedArgList *
 Darwin::TranslateArgs(const DerivedArgList &Args, StringRef BoundArch,
                       Action::OffloadKind DeviceOffloadKind) const {
   // First get the generic Apple args, before moving onto Darwin-specific ones.

lib/Driver/ToolChains/Fuchsia.h

public:
   SanitizerMask getSupportedSanitizers() const override;
 
   RuntimeLibType
   GetRuntimeLibType(const llvm::opt::ArgList &Args) const override;
   CXXStdlibType
   GetCXXStdlibType(const llvm::opt::ArgList &Args) const override;
 
   void addClangTargetOptions(const llvm::opt::ArgList &DriverArgs,
-                             llvm::opt::ArgStringList &CC1Args) const override;
+                             llvm::opt::ArgStringList &CC1Args,
+                             Action::OffloadKind DeviceOffloadKind) const override;
   void
   AddClangSystemIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                             llvm::opt::ArgStringList &CC1Args) const override;
   std::string findLibCxxIncludePath() const override;
   void AddCXXStdlibLibArgs(const llvm::opt::ArgList &Args,
                            llvm::opt::ArgStringList &CmdArgs) const override;
 
   const char *getDefaultLinker() const override {

lib/Driver/ToolChains/Fuchsia.cpp

if (Value != "libc++")
       getDriver().Diag(diag::err_drv_invalid_stdlib_name)
         << A->getAsString(Args);
   }
 
   return ToolChain::CST_Libcxx;
 }
 
 void Fuchsia::addClangTargetOptions(const ArgList &DriverArgs,
-                                    ArgStringList &CC1Args) const {
+                                    ArgStringList &CC1Args,
+                                    Action::OffloadKind) const {
   if (DriverArgs.hasFlag(options::OPT_fuse_init_array,
                          options::OPT_fno_use_init_array, true))
     CC1Args.push_back("-fuse-init-array");
 }
 
 void Fuchsia::AddClangSystemIncludeArgs(const ArgList &DriverArgs,
                                         ArgStringList &CC1Args) const {
   const Driver &D = getDriver();

lib/Driver/ToolChains/Gnu.h

class LLVM_LIBRARY_VISIBILITY Generic_ELF : public Generic_GCC {
   virtual void anchor();
 
 public:
   Generic_ELF(const Driver &D, const llvm::Triple &Triple,
               const llvm::opt::ArgList &Args)
       : Generic_GCC(D, Triple, Args) {}
 
   void addClangTargetOptions(const llvm::opt::ArgList &DriverArgs,
-                             llvm::opt::ArgStringList &CC1Args) const override;
+                             llvm::opt::ArgStringList &CC1Args,
+                             Action::OffloadKind DeviceOffloadKind) const override;
 };
 
 } // end namespace toolchains
 } // end namespace driver
 } // end namespace clang
 
 #endif // LLVM_CLANG_LIB_DRIVER_TOOLCHAINS_GNU_H

lib/Driver/ToolChains/Gnu.cpp

if (DeviceOffloadKind == Action::OFK_OpenMP) {
     return DAL;
   }
   return nullptr;
 }
 
 void Generic_ELF::anchor() {}
 
 void Generic_ELF::addClangTargetOptions(const ArgList &DriverArgs,
-                                        ArgStringList &CC1Args) const {
+                                        ArgStringList &CC1Args,
+                                        Action::OffloadKind) const {
   const Generic_GCC::GCCVersion &V = GCCInstallation.getVersion();
   bool UseInitArrayDefault =
       getTriple().getArch() == llvm::Triple::aarch64 ||
       getTriple().getArch() == llvm::Triple::aarch64_be ||
       (getTriple().getOS() == llvm::Triple::Linux &&
        (!V.isOlderThan(4, 7, 0) || getTriple().isAndroid())) ||
       getTriple().getOS() == llvm::Triple::NaCl ||
       (getTriple().getVendor() == llvm::Triple::MipsTechnologies &&
        !getTriple().hasEnvironment());
 
   if (DriverArgs.hasFlag(options::OPT_fuse_init_array,
                          options::OPT_fno_use_init_array, UseInitArrayDefault))
     CC1Args.push_back("-fuse-init-array");
 }

lib/Driver/ToolChains/Hexagon.h

protected:
   unsigned getOptimizationLevel(const llvm::opt::ArgList &DriverArgs) const;
 
 public:
   HexagonToolChain(const Driver &D, const llvm::Triple &Triple,
                    const llvm::opt::ArgList &Args);
   ~HexagonToolChain() override;
 
   void addClangTargetOptions(const llvm::opt::ArgList &DriverArgs,
-                             llvm::opt::ArgStringList &CC1Args) const override;
+                             llvm::opt::ArgStringList &CC1Args,
+                             Action::OffloadKind DeviceOffloadKind) const override;
   void
   AddClangSystemIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                             llvm::opt::ArgStringList &CC1Args) const override;
   void addLibStdCxxIncludePaths(
       const llvm::opt::ArgList &DriverArgs,
       llvm::opt::ArgStringList &CC1Args) const override;
   CXXStdlibType GetCXXStdlibType(const llvm::opt::ArgList &Args) const override;
 

lib/Driver/ToolChains/Hexagon.cpp

unsigned HexagonToolChain::getOptimizationLevel(
 
   unsigned OptLevel;
   if (S.getAsInteger(10, OptLevel))
     return 0;
   return OptLevel;
 }
 
 void HexagonToolChain::addClangTargetOptions(const ArgList &DriverArgs,
-                                             ArgStringList &CC1Args) const {
+                                             ArgStringList &CC1Args,
+                                             Action::OffloadKind) const {
   if (DriverArgs.hasArg(options::OPT_ffp_contract))
     return;
   unsigned OptLevel = getOptimizationLevel(DriverArgs);
   if (OptLevel >= 3)
     CC1Args.push_back("-ffp-contract=fast");
 }
 
 void HexagonToolChain::AddClangSystemIncludeArgs(const ArgList &DriverArgs,

lib/Driver/ToolChains/WebAssembly.h

private:
   bool isPIEDefault() const override;
   bool isPICDefaultForced() const override;
   bool IsIntegratedAssemblerDefault() const override;
   bool hasBlocksRuntime() const override;
   bool SupportsObjCGC() const override;
   bool SupportsProfiling() const override;
   bool HasNativeLLVMSupport() const override;
   void addClangTargetOptions(const llvm::opt::ArgList &DriverArgs,
-                             llvm::opt::ArgStringList &CC1Args) const override;
+                             llvm::opt::ArgStringList &CC1Args,
+                             Action::OffloadKind DeviceOffloadKind) const override;
   RuntimeLibType GetDefaultRuntimeLibType() const override;
   CXXStdlibType GetCXXStdlibType(const llvm::opt::ArgList &Args) const override;
   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;

lib/Driver/ToolChains/WebAssembly.cpp

 bool WebAssembly::hasBlocksRuntime() const { return false; }
 
 // TODO: Support profiling.
 bool WebAssembly::SupportsProfiling() const { return false; }
 
 bool WebAssembly::HasNativeLLVMSupport() const { return true; }
 
 void WebAssembly::addClangTargetOptions(const ArgList &DriverArgs,
-                                        ArgStringList &CC1Args) const {
+                                        ArgStringList &CC1Args,
+                                        Action::OffloadKind) const {
   if (DriverArgs.hasFlag(clang::driver::options::OPT_fuse_init_array,
                          options::OPT_fno_use_init_array, true))
     CC1Args.push_back("-fuse-init-array");
 }
 
 ToolChain::RuntimeLibType WebAssembly::GetDefaultRuntimeLibType() const {
   return ToolChain::RLT_CompilerRT;
 }

lib/Driver/ToolChains/XCore.h

public:
   bool isPIEDefault() const override;
   bool isPICDefaultForced() const override;
   bool SupportsProfiling() const override;
   bool hasBlocksRuntime() const override;
   void
   AddClangSystemIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                             llvm::opt::ArgStringList &CC1Args) const override;
   void addClangTargetOptions(const llvm::opt::ArgList &DriverArgs,
-                             llvm::opt::ArgStringList &CC1Args) const override;
+                             llvm::opt::ArgStringList &CC1Args,
+                             Action::OffloadKind DeviceOffloadKind) const override;
   void AddClangCXXStdlibIncludeArgs(
       const llvm::opt::ArgList &DriverArgs,
       llvm::opt::ArgStringList &CC1Args) const override;
   void AddCXXStdlibLibArgs(const llvm::opt::ArgList &Args,
                            llvm::opt::ArgStringList &CmdArgs) const override;
 };
 
 } // end namespace toolchains
 } // end namespace driver
 } // end namespace clang
 
 #endif // LLVM_CLANG_LIB_DRIVER_TOOLCHAINS_XCORE_H

lib/Driver/ToolChains/XCore.cpp

if (const char *cl_include_dir = getenv("XCC_C_INCLUDE_PATH")) {
     const char EnvPathSeparatorStr[] = {llvm::sys::EnvPathSeparator, '\0'};
     StringRef(cl_include_dir).split(Dirs, StringRef(EnvPathSeparatorStr));
     ArrayRef<StringRef> DirVec(Dirs);
     addSystemIncludes(DriverArgs, CC1Args, DirVec);
   }
 }
 
 void XCoreToolChain::addClangTargetOptions(const ArgList &DriverArgs,
-                                           ArgStringList &CC1Args) const {
+                                           ArgStringList &CC1Args,
+                                           Action::OffloadKind) const {
   CC1Args.push_back("-nostdsysteminc");
 }
 
 void XCoreToolChain::AddClangCXXStdlibIncludeArgs(
     const ArgList &DriverArgs, ArgStringList &CC1Args) const {
   if (DriverArgs.hasArg(clang::driver::options::OPT_nostdinc) ||
       DriverArgs.hasArg(options::OPT_nostdlibinc) ||
       DriverArgs.hasArg(options::OPT_nostdincxx))