[Driver][SYCL] Add initial SYCL offload compilation support (#117268)

Introduces the SYCL based toolchain and initial toolchain construction
when using the '-fsycl' option. This option will enable SYCL based
offloading, creating a SPIR-V based IR file packaged into the compiled
host object.

This includes early support for creating the host/device object using
the new offloading model. The device object is created using the
spir64-unknown-unknown target triple.

New/Updated Options:
 -fsycl  Enables SYCL offloading for host and device
 -fsycl-device-only
         Enables device only compilation for SYCL
 -fsycl-host-only
         Enables host only compilation for SYCL

RFC Reference:
https://discourse.llvm.org/t/rfc-sycl-driver-enhancements/74092

This is a reland of: https://github.com/llvm/llvm-project/pull/107493

16 files changed, 424 insertions, 22 deletions

diff --git a/clang/lib/Driver/Action.cpp b/clang/lib/Driver/Action.cpp
index 849bf60..23dbceb 100644
--- a/clang/lib/Driver/Action.cpp
+++ b/clang/lib/Driver/Action.cpp
@@ -111,6 +111,8 @@ std::string Action::getOffloadingKindPrefix() const {
     return "device-openmp";
   case OFK_HIP:
     return "device-hip";
+  case OFK_SYCL:
+    return "device-sycl";
 
     // TODO: Add other programming models here.
   }
@@ -128,6 +130,8 @@ std::string Action::getOffloadingKindPrefix() const {
     Res += "-hip";
   if (ActiveOffloadKindMask & OFK_OpenMP)
     Res += "-openmp";
+  if (ActiveOffloadKindMask & OFK_SYCL)
+    Res += "-sycl";
 
   // TODO: Add other programming models here.
 
@@ -164,6 +168,8 @@ StringRef Action::GetOffloadKindName(OffloadKind Kind) {
     return "openmp";
   case OFK_HIP:
     return "hip";
+  case OFK_SYCL:
+    return "sycl";
 
     // TODO: Add other programming models here.
   }
@@ -320,7 +326,7 @@ void OffloadAction::DeviceDependences::add(Action &A, const ToolChain &TC,
   DeviceBoundArchs.push_back(BoundArch);
 
   // Add each active offloading kind from a mask.
-  for (OffloadKind OKind : {OFK_OpenMP, OFK_Cuda, OFK_HIP})
+  for (OffloadKind OKind : {OFK_OpenMP, OFK_Cuda, OFK_HIP, OFK_SYCL})
     if (OKind & OffloadKindMask)
       DeviceOffloadKinds.push_back(OKind);
 }
diff --git a/clang/lib/Driver/CMakeLists.txt b/clang/lib/Driver/CMakeLists.txt
index 57d04c3..5bdb661 100644
--- a/clang/lib/Driver/CMakeLists.txt
+++ b/clang/lib/Driver/CMakeLists.txt
@@ -78,6 +78,7 @@ add_clang_library(clangDriver
   ToolChains/Solaris.cpp
   ToolChains/SPIRV.cpp
   ToolChains/SPIRVOpenMP.cpp
+  ToolChains/SYCL.cpp
   ToolChains/TCE.cpp
   ToolChains/UEFI.cpp
   ToolChains/VEToolchain.cpp
diff --git a/clang/lib/Driver/Compilation.cpp b/clang/lib/Driver/Compilation.cpp
index 4d40805..a39952e 100644
--- a/clang/lib/Driver/Compilation.cpp
+++ b/clang/lib/Driver/Compilation.cpp
@@ -214,10 +214,11 @@ static bool ActionFailed(const Action *A,
   if (FailingCommands.empty())
     return false;
 
-  // CUDA/HIP can have the same input source code compiled multiple times so do
-  // not compiled again if there are already failures. It is OK to abort the
-  // CUDA pipeline on errors.
-  if (A->isOffloading(Action::OFK_Cuda) || A->isOffloading(Action::OFK_HIP))
+  // CUDA/HIP/SYCL can have the same input source code compiled multiple times
+  // so do not compile again if there are already failures. It is OK to abort
+  // the CUDA/HIP/SYCL pipeline on errors.
+  if (A->isOffloading(Action::OFK_Cuda) || A->isOffloading(Action::OFK_HIP) ||
+      A->isOffloading(Action::OFK_SYCL))
     return true;
 
   for (const auto &CI : FailingCommands)
diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
index bc5ce9f..36d6c93 100644
--- a/clang/lib/Driver/Driver.cpp
+++ b/clang/lib/Driver/Driver.cpp
@@ -44,6 +44,7 @@
 #include "ToolChains/RISCVToolchain.h"
 #include "ToolChains/SPIRV.h"
 #include "ToolChains/SPIRVOpenMP.h"
+#include "ToolChains/SYCL.h"
 #include "ToolChains/Solaris.h"
 #include "ToolChains/TCE.h"
 #include "ToolChains/UEFI.h"
@@ -781,6 +782,35 @@ Driver::OpenMPRuntimeKind Driver::getOpenMPRuntime(const ArgList &Args) const {
   return RT;
 }
 
+static llvm::Triple getSYCLDeviceTriple(StringRef TargetArch) {
+  SmallVector<StringRef, 5> SYCLAlias = {"spir", "spir64", "spirv", "spirv32",
+                                         "spirv64"};
+  if (llvm::is_contained(SYCLAlias, TargetArch)) {
+    llvm::Triple TargetTriple;
+    TargetTriple.setArchName(TargetArch);
+    TargetTriple.setVendor(llvm::Triple::UnknownVendor);
+    TargetTriple.setOS(llvm::Triple::UnknownOS);
+    return TargetTriple;
+  }
+  return llvm::Triple(TargetArch);
+}
+
+static bool addSYCLDefaultTriple(Compilation &C,
+                                 SmallVectorImpl<llvm::Triple> &SYCLTriples) {
+  // Check current set of triples to see if the default has already been set.
+  for (const auto &SYCLTriple : SYCLTriples) {
+    if (SYCLTriple.getSubArch() == llvm::Triple::NoSubArch &&
+        SYCLTriple.isSPIROrSPIRV())
+      return false;
+  }
+  // Add the default triple as it was not found.
+  llvm::Triple DefaultTriple = getSYCLDeviceTriple(
+      C.getDefaultToolChain().getTriple().isArch32Bit() ? "spirv32"
+                                                        : "spirv64");
+  SYCLTriples.insert(SYCLTriples.begin(), DefaultTriple);
+  return true;
+}
+
 void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
                                               InputList &Inputs) {
 
@@ -842,7 +872,6 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
       return;
     auto *HIPTC = &getOffloadingDeviceToolChain(C.getInputArgs(), *HIPTriple,
                                                 *HostTC, OFK);
-    assert(HIPTC && "Could not create offloading device tool chain.");
     C.addOffloadDeviceToolChain(HIPTC, OFK);
   }
 
@@ -997,6 +1026,38 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
     return;
   }
 
+  // We need to generate a SYCL toolchain if the user specified -fsycl.
+  bool IsSYCL = C.getInputArgs().hasFlag(options::OPT_fsycl,
+                                         options::OPT_fno_sycl, false);
+
+  auto argSYCLIncompatible = [&](OptSpecifier OptId) {
+    if (!IsSYCL)
+      return;
+    if (Arg *IncompatArg = C.getInputArgs().getLastArg(OptId))
+      Diag(clang::diag::err_drv_argument_not_allowed_with)
+          << IncompatArg->getSpelling() << "-fsycl";
+  };
+  // -static-libstdc++ is not compatible with -fsycl.
+  argSYCLIncompatible(options::OPT_static_libstdcxx);
+  // -ffreestanding cannot be used with -fsycl
+  argSYCLIncompatible(options::OPT_ffreestanding);
+
+  llvm::SmallVector<llvm::Triple, 4> UniqueSYCLTriplesVec;
+
+  if (IsSYCL) {
+    addSYCLDefaultTriple(C, UniqueSYCLTriplesVec);
+
+    // We'll need to use the SYCL and host triples as the key into
+    // getOffloadingDeviceToolChain, because the device toolchains we're
+    // going to create will depend on both.
+    const ToolChain *HostTC = C.getSingleOffloadToolChain<Action::OFK_Host>();
+    for (const auto &TargetTriple : UniqueSYCLTriplesVec) {
+      auto SYCLTC = &getOffloadingDeviceToolChain(
+          C.getInputArgs(), TargetTriple, *HostTC, Action::OFK_SYCL);
+      C.addOffloadDeviceToolChain(SYCLTC, Action::OFK_SYCL);
+    }
+  }
+
   //
   // TODO: Add support for other offloading programming models here.
   //
@@ -4234,6 +4295,7 @@ void Driver::BuildActions(Compilation &C, DerivedArgList &Args,
 
   bool UseNewOffloadingDriver =
       C.isOffloadingHostKind(Action::OFK_OpenMP) ||
+      C.isOffloadingHostKind(Action::OFK_SYCL) ||
       Args.hasFlag(options::OPT_foffload_via_llvm,
                    options::OPT_fno_offload_via_llvm, false) ||
       Args.hasFlag(options::OPT_offload_new_driver,
@@ -4651,6 +4713,8 @@ Driver::getOffloadArchs(Compilation &C, const llvm::opt::DerivedArgList &Args,
       Archs.insert(OffloadArchToString(OffloadArch::HIPDefault));
     else if (Kind == Action::OFK_OpenMP)
       Archs.insert(StringRef());
+    else if (Kind == Action::OFK_SYCL)
+      Archs.insert(StringRef());
   } else {
     Args.ClaimAllArgs(options::OPT_offload_arch_EQ);
     Args.ClaimAllArgs(options::OPT_no_offload_arch_EQ);
@@ -4675,7 +4739,7 @@ Action *Driver::BuildOffloadingActions(Compilation &C,
   OffloadAction::DeviceDependences DDeps;
 
   const Action::OffloadKind OffloadKinds[] = {
-      Action::OFK_OpenMP, Action::OFK_Cuda, Action::OFK_HIP};
+      Action::OFK_OpenMP, Action::OFK_Cuda, Action::OFK_HIP, Action::OFK_SYCL};
 
   for (Action::OffloadKind Kind : OffloadKinds) {
     SmallVector<const ToolChain *, 2> ToolChains;
@@ -4712,6 +4776,15 @@ Action *Driver::BuildOffloadingActions(Compilation &C,
     if (DeviceActions.empty())
       return HostAction;
 
+    // FIXME: Do not collapse the host side for Darwin targets with SYCL offload
+    // compilations. The toolchain is not properly initialized for the target.
+    if (isa<CompileJobAction>(HostAction) && Kind == Action::OFK_SYCL &&
+        HostAction->getType() != types::TY_Nothing &&
+        C.getSingleOffloadToolChain<Action::OFK_Host>()
+            ->getTriple()
+            .isOSDarwin())
+      HostAction->setCannotBeCollapsedWithNextDependentAction();
+
     auto PL = types::getCompilationPhases(*this, Args, InputType);
 
     for (phases::ID Phase : PL) {
@@ -4720,6 +4793,11 @@ Action *Driver::BuildOffloadingActions(Compilation &C,
         break;
       }
 
+      // Assemble actions are not used for the SYCL device side.  Both compile
+      // and backend actions are used to generate IR and textual IR if needed.
+      if (Kind == Action::OFK_SYCL && Phase == phases::Assemble)
+        continue;
+
       auto TCAndArch = TCAndArchs.begin();
       for (Action *&A : DeviceActions) {
         if (A->getType() == types::TY_Nothing)
@@ -4958,6 +5036,7 @@ Action *Driver::ConstructPhaseAction(
       return C.MakeAction<BackendJobAction>(Input, Output);
     }
     if (Args.hasArg(options::OPT_emit_llvm) ||
+        TargetDeviceOffloadKind == Action::OFK_SYCL ||
         (((Input->getOffloadingToolChain() &&
            Input->getOffloadingToolChain()->getTriple().isAMDGPU()) ||
           TargetDeviceOffloadKind == Action::OFK_HIP) &&
@@ -6644,11 +6723,16 @@ const ToolChain &Driver::getOffloadingDeviceToolChain(
                                                            HostTC, Args);
       break;
     }
+    case Action::OFK_SYCL:
+      if (Target.isSPIROrSPIRV())
+        TC = std::make_unique<toolchains::SYCLToolChain>(*this, Target, HostTC,
+                                                         Args);
+      break;
     default:
       break;
     }
   }
-
+  assert(TC && "Could not create offloading device tool chain.");
   return *TC;
 }
 
diff --git a/clang/lib/Driver/ToolChain.cpp b/clang/lib/Driver/ToolChain.cpp
index 9f174fb..2b4df64 100644
--- a/clang/lib/Driver/ToolChain.cpp
+++ b/clang/lib/Driver/ToolChain.cpp
@@ -1485,6 +1485,9 @@ void ToolChain::AddCudaIncludeArgs(const ArgList &DriverArgs,
 void ToolChain::AddHIPIncludeArgs(const ArgList &DriverArgs,
                                   ArgStringList &CC1Args) const {}
 
+void ToolChain::addSYCLIncludeArgs(const ArgList &DriverArgs,
+                                   ArgStringList &CC1Args) const {}
+
 llvm::SmallVector<ToolChain::BitCodeLibraryInfo, 12>
 ToolChain::getDeviceLibs(const ArgList &DriverArgs) const {
   return {};
diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index daf863c..a000237 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -24,6 +24,7 @@
 #include "Hexagon.h"
 #include "MSP430.h"
 #include "PS4CPU.h"
+#include "SYCL.h"
 #include "clang/Basic/CLWarnings.h"
 #include "clang/Basic/CharInfo.h"
 #include "clang/Basic/CodeGenOptions.h"
@@ -122,6 +123,13 @@ forAllAssociatedToolChains(Compilation &C, const JobAction &JA,
   } else if (JA.isDeviceOffloading(Action::OFK_OpenMP))
     Work(*C.getSingleOffloadToolChain<Action::OFK_Host>());
 
+  if (JA.isHostOffloading(Action::OFK_SYCL)) {
+    auto TCs = C.getOffloadToolChains<Action::OFK_SYCL>();
+    for (auto II = TCs.first, IE = TCs.second; II != IE; ++II)
+      Work(*II->second);
+  } else if (JA.isDeviceOffloading(Action::OFK_SYCL))
+    Work(*C.getSingleOffloadToolChain<Action::OFK_Host>());
+
   //
   // TODO: Add support for other offloading programming models here.
   //
@@ -1070,14 +1078,16 @@ void Clang::AddPreprocessingOptions(Compilation &C, const JobAction &JA,
   Args.AddLastArg(CmdArgs, options::OPT_MP);
   Args.AddLastArg(CmdArgs, options::OPT_MV);
 
-  // Add offload include arguments specific for CUDA/HIP.  This must happen
+  // Add offload include arguments specific for CUDA/HIP/SYCL. This must happen
   // before we -I or -include anything else, because we must pick up the
-  // CUDA/HIP headers from the particular CUDA/ROCm installation, rather than
-  // from e.g. /usr/local/include.
+  // CUDA/HIP/SYCL headers from the particular CUDA/ROCm/SYCL installation,
+  // rather than from e.g. /usr/local/include.
   if (JA.isOffloading(Action::OFK_Cuda))
     getToolChain().AddCudaIncludeArgs(Args, CmdArgs);
   if (JA.isOffloading(Action::OFK_HIP))
     getToolChain().AddHIPIncludeArgs(Args, CmdArgs);
+  if (JA.isOffloading(Action::OFK_SYCL))
+    getToolChain().addSYCLIncludeArgs(Args, CmdArgs);
 
   // If we are offloading to a target via OpenMP we need to include the
   // openmp_wrappers folder which contains alternative system headers.
@@ -5037,17 +5047,21 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   // second input. Module precompilation accepts a list of header files to
   // include as part of the module. API extraction accepts a list of header
   // files whose API information is emitted in the output. All other jobs are
-  // expected to have exactly one input.
+  // expected to have exactly one input. SYCL compilation only expects a
+  // single input.
   bool IsCuda = JA.isOffloading(Action::OFK_Cuda);
   bool IsCudaDevice = JA.isDeviceOffloading(Action::OFK_Cuda);
   bool IsHIP = JA.isOffloading(Action::OFK_HIP);
   bool IsHIPDevice = JA.isDeviceOffloading(Action::OFK_HIP);
+  bool IsSYCL = JA.isOffloading(Action::OFK_SYCL);
+  bool IsSYCLDevice = JA.isDeviceOffloading(Action::OFK_SYCL);
   bool IsOpenMPDevice = JA.isDeviceOffloading(Action::OFK_OpenMP);
   bool IsExtractAPI = isa<ExtractAPIJobAction>(JA);
   bool IsDeviceOffloadAction = !(JA.isDeviceOffloading(Action::OFK_None) ||
                                  JA.isDeviceOffloading(Action::OFK_Host));
   bool IsHostOffloadingAction =
       JA.isHostOffloading(Action::OFK_OpenMP) ||
+      JA.isHostOffloading(Action::OFK_SYCL) ||
       (JA.isHostOffloading(C.getActiveOffloadKinds()) &&
        Args.hasFlag(options::OPT_offload_new_driver,
                     options::OPT_no_offload_new_driver, false));
@@ -5097,10 +5111,10 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   bool IsWindowsMSVC = RawTriple.isWindowsMSVCEnvironment();
   bool IsIAMCU = RawTriple.isOSIAMCU();
 
-  // Adjust IsWindowsXYZ for CUDA/HIP compilations.  Even when compiling in
+  // Adjust IsWindowsXYZ for CUDA/HIP/SYCL compilations.  Even when compiling in
   // device mode (i.e., getToolchain().getTriple() is NVPTX/AMDGCN, not
   // Windows), we need to pass Windows-specific flags to cc1.
-  if (IsCuda || IsHIP)
+  if (IsCuda || IsHIP || IsSYCL)
     IsWindowsMSVC |= AuxTriple && AuxTriple->isWindowsMSVCEnvironment();
 
   // C++ is not supported for IAMCU.
@@ -5184,11 +5198,33 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   if (const Arg *PF = Args.getLastArg(options::OPT_mprintf_kind_EQ))
     PF->claim();
 
-  if (Args.hasFlag(options::OPT_fsycl, options::OPT_fno_sycl, false)) {
-    CmdArgs.push_back("-fsycl-is-device");
+  if (IsSYCL) {
+    if (IsSYCLDevice) {
+      // Host triple is needed when doing SYCL device compilations.
+      llvm::Triple AuxT = C.getDefaultToolChain().getTriple();
+      std::string NormalizedTriple = AuxT.normalize();
+      CmdArgs.push_back("-aux-triple");
+      CmdArgs.push_back(Args.MakeArgString(NormalizedTriple));
 
-    if (Arg *A = Args.getLastArg(options::OPT_sycl_std_EQ)) {
-      A->render(Args, CmdArgs);
+      // We want to compile sycl kernels.
+      CmdArgs.push_back("-fsycl-is-device");
+
+      // Set O2 optimization level by default
+      if (!Args.getLastArg(options::OPT_O_Group))
+        CmdArgs.push_back("-O2");
+    } else {
+      // Add any options that are needed specific to SYCL offload while
+      // performing the host side compilation.
+
+      // Let the front-end host compilation flow know about SYCL offload
+      // compilation.
+      CmdArgs.push_back("-fsycl-is-host");
+    }
+
+    // Set options for both host and device.
+    Arg *SYCLStdArg = Args.getLastArg(options::OPT_sycl_std_EQ);
+    if (SYCLStdArg) {
+      SYCLStdArg->render(Args, CmdArgs);
     } else {
       // Ensure the default version in SYCL mode is 2020.
       CmdArgs.push_back("-sycl-std=2020");
@@ -6135,7 +6171,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   // Prepare `-aux-target-cpu` and `-aux-target-feature` unless
   // `--gpu-use-aux-triple-only` is specified.
   if (!Args.getLastArg(options::OPT_gpu_use_aux_triple_only) &&
-      (IsCudaDevice || IsHIPDevice)) {
+      (IsCudaDevice || IsHIPDevice || IsSYCLDevice)) {
     const ArgList &HostArgs =
         C.getArgsForToolChain(nullptr, StringRef(), Action::OFK_None);
     std::string HostCPU =
diff --git a/clang/lib/Driver/ToolChains/Darwin.cpp b/clang/lib/Driver/ToolChains/Darwin.cpp
index 4105d38..56b6dd7 100644
--- a/clang/lib/Driver/ToolChains/Darwin.cpp
+++ b/clang/lib/Driver/ToolChains/Darwin.cpp
@@ -969,7 +969,8 @@ MachO::MachO(const Driver &D, const llvm::Triple &Triple, const ArgList &Args)
 /// Darwin - Darwin tool chain for i386 and x86_64.
 Darwin::Darwin(const Driver &D, const llvm::Triple &Triple, const ArgList &Args)
     : MachO(D, Triple, Args), TargetInitialized(false),
-      CudaInstallation(D, Triple, Args), RocmInstallation(D, Triple, Args) {}
+      CudaInstallation(D, Triple, Args), RocmInstallation(D, Triple, Args),
+      SYCLInstallation(D, Triple, Args) {}
 
 types::ID MachO::LookupTypeForExtension(StringRef Ext) const {
   types::ID Ty = ToolChain::LookupTypeForExtension(Ext);
@@ -1028,6 +1029,11 @@ void Darwin::AddHIPIncludeArgs(const ArgList &DriverArgs,
   RocmInstallation->AddHIPIncludeArgs(DriverArgs, CC1Args);
 }
 
+void Darwin::addSYCLIncludeArgs(const ArgList &DriverArgs,
+                                ArgStringList &CC1Args) const {
+  SYCLInstallation->addSYCLIncludeArgs(DriverArgs, CC1Args);
+}
+
 // This is just a MachO name translation routine and there's no
 // way to join this into ARMTargetParser without breaking all
 // other assumptions. Maybe MachO should consider standardising
diff --git a/clang/lib/Driver/ToolChains/Darwin.h b/clang/lib/Driver/ToolChains/Darwin.h
index 2e55b496..5bc1858 100644
--- a/clang/lib/Driver/ToolChains/Darwin.h
+++ b/clang/lib/Driver/ToolChains/Darwin.h
@@ -12,6 +12,7 @@
 #include "Cuda.h"
 #include "LazyDetector.h"
 #include "ROCm.h"
+#include "SYCL.h"
 #include "clang/Basic/DarwinSDKInfo.h"
 #include "clang/Basic/LangOptions.h"
 #include "clang/Driver/Tool.h"
@@ -331,6 +332,7 @@ public:
 
   LazyDetector<CudaInstallationDetector> CudaInstallation;
   LazyDetector<RocmInstallationDetector> RocmInstallation;
+  LazyDetector<SYCLInstallationDetector> SYCLInstallation;
 
 private:
   void AddDeploymentTarget(llvm::opt::DerivedArgList &Args) const;
@@ -563,6 +565,8 @@ public:
                           llvm::opt::ArgStringList &CC1Args) const override;
   void AddHIPIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                          llvm::opt::ArgStringList &CC1Args) const override;
+  void addSYCLIncludeArgs(const llvm::opt::ArgList &DriverArgs,
+                          llvm::opt::ArgStringList &CC1Args) const override;
 
   bool UseObjCMixedDispatch() const override {
     // This is only used with the non-fragile ABI and non-legacy dispatch.
diff --git a/clang/lib/Driver/ToolChains/Gnu.cpp b/clang/lib/Driver/ToolChains/Gnu.cpp
index 8397f11..e5db1b2 100644
--- a/clang/lib/Driver/ToolChains/Gnu.cpp
+++ b/clang/lib/Driver/ToolChains/Gnu.cpp
@@ -3058,7 +3058,8 @@ bool Generic_GCC::GCCInstallationDetector::ScanGentooGccConfig(
 Generic_GCC::Generic_GCC(const Driver &D, const llvm::Triple &Triple,
                          const ArgList &Args)
     : ToolChain(D, Triple, Args), GCCInstallation(D),
-      CudaInstallation(D, Triple, Args), RocmInstallation(D, Triple, Args) {
+      CudaInstallation(D, Triple, Args), RocmInstallation(D, Triple, Args),
+      SYCLInstallation(D, Triple, Args) {
   getProgramPaths().push_back(getDriver().Dir);
 }
 
@@ -3274,6 +3275,11 @@ void Generic_GCC::AddClangCXXStdlibIncludeArgs(const ArgList &DriverArgs,
   }
 }
 
+void Generic_GCC::addSYCLIncludeArgs(const ArgList &DriverArgs,
+                                     ArgStringList &CC1Args) const {
+  SYCLInstallation->addSYCLIncludeArgs(DriverArgs, CC1Args);
+}
+
 void
 Generic_GCC::addLibCxxIncludePaths(const llvm::opt::ArgList &DriverArgs,
                                    llvm::opt::ArgStringList &CC1Args) const {
diff --git a/clang/lib/Driver/ToolChains/Gnu.h b/clang/lib/Driver/ToolChains/Gnu.h
index 0b664a1..3b8df71 100644
--- a/clang/lib/Driver/ToolChains/Gnu.h
+++ b/clang/lib/Driver/ToolChains/Gnu.h
@@ -12,6 +12,7 @@
 #include "Cuda.h"
 #include "LazyDetector.h"
 #include "ROCm.h"
+#include "SYCL.h"
 #include "clang/Driver/Tool.h"
 #include "clang/Driver/ToolChain.h"
 #include <set>
@@ -288,6 +289,7 @@ protected:
   GCCInstallationDetector GCCInstallation;
   LazyDetector<CudaInstallationDetector> CudaInstallation;
   LazyDetector<RocmInstallationDetector> RocmInstallation;
+  LazyDetector<SYCLInstallationDetector> SYCLInstallation;
 
 public:
   Generic_GCC(const Driver &D, const llvm::Triple &Triple,
@@ -336,6 +338,9 @@ protected:
       const llvm::opt::ArgList &DriverArgs,
       llvm::opt::ArgStringList &CC1Args) const override;
 
+  void addSYCLIncludeArgs(const llvm::opt::ArgList &DriverArgs,
+                          llvm::opt::ArgStringList &CC1Args) const override;
+
   virtual void
   addLibCxxIncludePaths(const llvm::opt::ArgList &DriverArgs,
                         llvm::opt::ArgStringList &CC1Args) const;
diff --git a/clang/lib/Driver/ToolChains/Linux.cpp b/clang/lib/Driver/ToolChains/Linux.cpp
index c91b55b..1c56355 100644
--- a/clang/lib/Driver/ToolChains/Linux.cpp
+++ b/clang/lib/Driver/ToolChains/Linux.cpp
@@ -777,6 +777,11 @@ void Linux::AddIAMCUIncludeArgs(const ArgList &DriverArgs,
   }
 }
 
+void Linux::addSYCLIncludeArgs(const ArgList &DriverArgs,
+                               ArgStringList &CC1Args) const {
+  SYCLInstallation->addSYCLIncludeArgs(DriverArgs, CC1Args);
+}
+
 bool Linux::isPIEDefault(const llvm::opt::ArgList &Args) const {
   return CLANG_DEFAULT_PIE_ON_LINUX || getTriple().isAndroid() ||
          getTriple().isMusl() || getSanitizerArgs(Args).requiresPIE();
diff --git a/clang/lib/Driver/ToolChains/Linux.h b/clang/lib/Driver/ToolChains/Linux.h
index 2d9e674..2eb2d05 100644
--- a/clang/lib/Driver/ToolChains/Linux.h
+++ b/clang/lib/Driver/ToolChains/Linux.h
@@ -41,6 +41,8 @@ public:
                             llvm::opt::ArgStringList &CmdArgs) const override;
   void AddIAMCUIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                            llvm::opt::ArgStringList &CC1Args) const override;
+  void addSYCLIncludeArgs(const llvm::opt::ArgList &DriverArgs,
+                          llvm::opt::ArgStringList &CC1Args) const override;
   RuntimeLibType GetDefaultRuntimeLibType() const override;
   unsigned GetDefaultDwarfVersion() const override;
   CXXStdlibType GetDefaultCXXStdlibType() const override;
diff --git a/clang/lib/Driver/ToolChains/MSVC.cpp b/clang/lib/Driver/ToolChains/MSVC.cpp
index 752c2e2..bae41fc 100644
--- a/clang/lib/Driver/ToolChains/MSVC.cpp
+++ b/clang/lib/Driver/ToolChains/MSVC.cpp
@@ -430,7 +430,7 @@ void visualstudio::Linker::ConstructJob(Compilation &C, const JobAction &JA,
 MSVCToolChain::MSVCToolChain(const Driver &D, const llvm::Triple &Triple,
                              const ArgList &Args)
     : ToolChain(D, Triple, Args), CudaInstallation(D, Triple, Args),
-      RocmInstallation(D, Triple, Args) {
+      RocmInstallation(D, Triple, Args), SYCLInstallation(D, Triple, Args) {
   getProgramPaths().push_back(getDriver().Dir);
 
   std::optional<llvm::StringRef> VCToolsDir, VCToolsVersion;
@@ -509,6 +509,11 @@ void MSVCToolChain::AddHIPIncludeArgs(const ArgList &DriverArgs,
   RocmInstallation->AddHIPIncludeArgs(DriverArgs, CC1Args);
 }
 
+void MSVCToolChain::addSYCLIncludeArgs(const ArgList &DriverArgs,
+                                       ArgStringList &CC1Args) const {
+  SYCLInstallation->addSYCLIncludeArgs(DriverArgs, CC1Args);
+}
+
 void MSVCToolChain::AddHIPRuntimeLibArgs(const ArgList &Args,
                                          ArgStringList &CmdArgs) const {
   CmdArgs.append({Args.MakeArgString(StringRef("-libpath:") +
diff --git a/clang/lib/Driver/ToolChains/MSVC.h b/clang/lib/Driver/ToolChains/MSVC.h
index 3950a8e..b35390c 100644
--- a/clang/lib/Driver/ToolChains/MSVC.h
+++ b/clang/lib/Driver/ToolChains/MSVC.h
@@ -12,6 +12,7 @@
 #include "AMDGPU.h"
 #include "Cuda.h"
 #include "LazyDetector.h"
+#include "SYCL.h"
 #include "clang/Driver/Compilation.h"
 #include "clang/Driver/Tool.h"
 #include "clang/Driver/ToolChain.h"
@@ -100,6 +101,9 @@ public:
   void AddHIPRuntimeLibArgs(const llvm::opt::ArgList &Args,
                             llvm::opt::ArgStringList &CmdArgs) const override;
 
+  void addSYCLIncludeArgs(const llvm::opt::ArgList &DriverArgs,
+                          llvm::opt::ArgStringList &CC1Args) const override;
+
   bool getWindowsSDKLibraryPath(
       const llvm::opt::ArgList &Args, std::string &path) const;
   bool getUniversalCRTLibraryPath(const llvm::opt::ArgList &Args,
@@ -138,6 +142,7 @@ private:
   llvm::ToolsetLayout VSLayout = llvm::ToolsetLayout::OlderVS;
   LazyDetector<CudaInstallationDetector> CudaInstallation;
   LazyDetector<RocmInstallationDetector> RocmInstallation;
+  LazyDetector<SYCLInstallationDetector> SYCLInstallation;
 };
 
 } // end namespace toolchains
diff --git a/clang/lib/Driver/ToolChains/SYCL.cpp b/clang/lib/Driver/ToolChains/SYCL.cpp
new file mode 100644
index 0000000..e42b65c
--- /dev/null
+++ b/clang/lib/Driver/ToolChains/SYCL.cpp
@@ -0,0 +1,153 @@
+//===--- SYCL.cpp - SYCL Tool and ToolChain Implementations -----*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+#include "SYCL.h"
+#include "CommonArgs.h"
+#include "llvm/Support/Path.h"
+
+using namespace clang::driver;
+using namespace clang::driver::toolchains;
+using namespace clang::driver::tools;
+using namespace clang;
+using namespace llvm::opt;
+
+SYCLInstallationDetector::SYCLInstallationDetector(
+    const Driver &D, const llvm::Triple &HostTriple,
+    const llvm::opt::ArgList &Args)
+    : D(D) {}
+
+void SYCLInstallationDetector::addSYCLIncludeArgs(
+    const ArgList &DriverArgs, ArgStringList &CC1Args) const {
+  if (DriverArgs.hasArg(clang::driver::options::OPT_nobuiltininc))
+    return;
+
+  // Add the SYCL header search locations in the specified order.
+  // FIXME: Add the header file locations once the SYCL library and headers
+  //        are properly established within the build.
+}
+
+// Unsupported options for SYCL device compilation.
+static ArrayRef<OptSpecifier> getUnsupportedOpts() {
+  return {
+      options::OPT_fsanitize_EQ,      // -fsanitize
+      options::OPT_fcf_protection_EQ, // -fcf-protection
+      options::OPT_fprofile_generate,
+      options::OPT_fprofile_generate_EQ,
+      options::OPT_fno_profile_generate, // -f[no-]profile-generate
+      options::OPT_ftest_coverage,
+      options::OPT_fno_test_coverage, // -f[no-]test-coverage
+      options::OPT_fcoverage_mapping,
+      options::OPT_fno_coverage_mapping, // -f[no-]coverage-mapping
+      options::OPT_coverage,             // --coverage
+      options::OPT_fprofile_instr_generate,
+      options::OPT_fprofile_instr_generate_EQ,
+      options::OPT_fno_profile_instr_generate, // -f[no-]profile-instr-generate
+      options::OPT_fprofile_arcs,
+      options::OPT_fno_profile_arcs, // -f[no-]profile-arcs
+      options::OPT_fcreate_profile,  // -fcreate-profile
+      options::OPT_fprofile_instr_use,
+      options::OPT_fprofile_instr_use_EQ,       // -fprofile-instr-use
+      options::OPT_forder_file_instrumentation, // -forder-file-instrumentation
+      options::OPT_fcs_profile_generate,        // -fcs-profile-generate
+      options::OPT_fcs_profile_generate_EQ};
+}
+
+SYCLToolChain::SYCLToolChain(const Driver &D, const llvm::Triple &Triple,
+                             const ToolChain &HostTC, const ArgList &Args)
+    : ToolChain(D, Triple, Args), HostTC(HostTC),
+      SYCLInstallation(D, Triple, Args) {
+  // Lookup binaries into the driver directory, this is used to discover any
+  // dependent SYCL offload compilation tools.
+  getProgramPaths().push_back(getDriver().Dir);
+
+  // Diagnose unsupported options only once.
+  for (OptSpecifier Opt : getUnsupportedOpts()) {
+    if (const Arg *A = Args.getLastArg(Opt)) {
+      D.Diag(clang::diag::warn_drv_unsupported_option_for_target)
+          << A->getAsString(Args) << getTriple().str();
+    }
+  }
+}
+
+void SYCLToolChain::addClangTargetOptions(
+    const llvm::opt::ArgList &DriverArgs, llvm::opt::ArgStringList &CC1Args,
+    Action::OffloadKind DeviceOffloadingKind) const {
+  HostTC.addClangTargetOptions(DriverArgs, CC1Args, DeviceOffloadingKind);
+}
+
+llvm::opt::DerivedArgList *
+SYCLToolChain::TranslateArgs(const llvm::opt::DerivedArgList &Args,
+                             StringRef BoundArch,
+                             Action::OffloadKind DeviceOffloadKind) const {
+  DerivedArgList *DAL =
+      HostTC.TranslateArgs(Args, BoundArch, DeviceOffloadKind);
+
+  bool IsNewDAL = false;
+  if (!DAL) {
+    DAL = new DerivedArgList(Args.getBaseArgs());
+    IsNewDAL = true;
+  }
+
+  for (Arg *A : Args) {
+    // Filter out any options we do not want to pass along to the device
+    // compilation.
+    auto Opt(A->getOption());
+    bool Unsupported = false;
+    for (OptSpecifier UnsupportedOpt : getUnsupportedOpts()) {
+      if (Opt.matches(UnsupportedOpt)) {
+        if (Opt.getID() == options::OPT_fsanitize_EQ &&
+            A->getValues().size() == 1) {
+          std::string SanitizeVal = A->getValue();
+          if (SanitizeVal == "address") {
+            if (IsNewDAL)
+              DAL->append(A);
+            continue;
+          }
+        }
+        if (!IsNewDAL)
+          DAL->eraseArg(Opt.getID());
+        Unsupported = true;
+      }
+    }
+    if (Unsupported)
+      continue;
+    if (IsNewDAL)
+      DAL->append(A);
+  }
+
+  const OptTable &Opts = getDriver().getOpts();
+  if (!BoundArch.empty()) {
+    DAL->eraseArg(options::OPT_march_EQ);
+    DAL->AddJoinedArg(nullptr, Opts.getOption(options::OPT_march_EQ),
+                      BoundArch);
+  }
+  return DAL;
+}
+
+void SYCLToolChain::addClangWarningOptions(ArgStringList &CC1Args) const {
+  HostTC.addClangWarningOptions(CC1Args);
+}
+
+ToolChain::CXXStdlibType
+SYCLToolChain::GetCXXStdlibType(const ArgList &Args) const {
+  return HostTC.GetCXXStdlibType(Args);
+}
+
+void SYCLToolChain::addSYCLIncludeArgs(const ArgList &DriverArgs,
+                                       ArgStringList &CC1Args) const {
+  SYCLInstallation.addSYCLIncludeArgs(DriverArgs, CC1Args);
+}
+
+void SYCLToolChain::AddClangSystemIncludeArgs(const ArgList &DriverArgs,
+                                              ArgStringList &CC1Args) const {
+  HostTC.AddClangSystemIncludeArgs(DriverArgs, CC1Args);
+}
+
+void SYCLToolChain::AddClangCXXStdlibIncludeArgs(const ArgList &Args,
+                                                 ArgStringList &CC1Args) const {
+  HostTC.AddClangCXXStdlibIncludeArgs(Args, CC1Args);
+}
diff --git a/clang/lib/Driver/ToolChains/SYCL.h b/clang/lib/Driver/ToolChains/SYCL.h
new file mode 100644
index 0000000..9af2fd0
--- /dev/null
+++ b/clang/lib/Driver/ToolChains/SYCL.h
@@ -0,0 +1,80 @@
+//===--- SYCL.h - SYCL ToolChain Implementations ----------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CLANG_LIB_DRIVER_TOOLCHAINS_SYCL_H
+#define LLVM_CLANG_LIB_DRIVER_TOOLCHAINS_SYCL_H
+
+#include "clang/Driver/Tool.h"
+#include "clang/Driver/ToolChain.h"
+
+namespace clang {
+namespace driver {
+
+class SYCLInstallationDetector {
+public:
+  SYCLInstallationDetector(const Driver &D, const llvm::Triple &HostTriple,
+                           const llvm::opt::ArgList &Args);
+
+  void addSYCLIncludeArgs(const llvm::opt::ArgList &DriverArgs,
+                          llvm::opt::ArgStringList &CC1Args) const;
+
+private:
+  const Driver &D;
+};
+
+namespace toolchains {
+
+class LLVM_LIBRARY_VISIBILITY SYCLToolChain : public ToolChain {
+public:
+  SYCLToolChain(const Driver &D, const llvm::Triple &Triple,
+                const ToolChain &HostTC, const llvm::opt::ArgList &Args);
+
+  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,
+                        Action::OffloadKind DeviceOffloadKind) const override;
+
+  bool useIntegratedAs() const override { return true; }
+  bool isPICDefault() const override { return false; }
+  llvm::codegenoptions::DebugInfoFormat getDefaultDebugFormat() const override {
+    return this->HostTC.getDefaultDebugFormat();
+  }
+  bool isPIEDefault(const llvm::opt::ArgList &Args) const override {
+    return false;
+  }
+  bool isPICDefaultForced() const override { return false; }
+
+  void addClangWarningOptions(llvm::opt::ArgStringList &CC1Args) const override;
+  CXXStdlibType GetCXXStdlibType(const llvm::opt::ArgList &Args) const override;
+  void addSYCLIncludeArgs(const llvm::opt::ArgList &DriverArgs,
+                          llvm::opt::ArgStringList &CC1Args) const override;
+  void
+  AddClangSystemIncludeArgs(const llvm::opt::ArgList &DriverArgs,
+                            llvm::opt::ArgStringList &CC1Args) const override;
+  void AddClangCXXStdlibIncludeArgs(
+      const llvm::opt::ArgList &Args,
+      llvm::opt::ArgStringList &CC1Args) const override;
+
+private:
+  const ToolChain &HostTC;
+  SYCLInstallationDetector SYCLInstallation;
+};
+
+} // end namespace toolchains
+
+} // end namespace driver
+} // end namespace clang
+
+#endif // LLVM_CLANG_LIB_DRIVER_TOOLCHAINS_SYCL_H