Merge branch 'sycl' into pre-ci-ur-2249

Author: zhaomaosu

.github/workflows/scorecard.yml

@@ -57,6 +57,6 @@ jobs:
 
       # Upload the results to GitHub's code scanning dashboard.
       - name: "Upload to code-scanning"
-        uses: github/codeql-action/upload-sarif@e2b3eafc8d227b0241d48be5f425d47c2d750a13 # v3.26.10
+        uses: github/codeql-action/upload-sarif@662472033e021d55d94146f66f6058822b0b39fd # v3.27.0
         with:
           sarif_file: results.sarif

.github/workflows/sycl-macos-build-and-test.yml

@@ -24,7 +24,7 @@ permissions: read-all
 jobs:
   build:
     name: Build
-    runs-on: macos-12
+    runs-on: macos-13
     env:
       CCACHE_DIR: $GITHUB_WORKSPACE/build_cache_${{ inputs.build_cache_suffix }}
       CCACHE_MAXSIZE: ${{ inputs.build_cache_size }}

.github/workflows/sycl-nightly.yml

@@ -198,7 +198,7 @@ jobs:
           echo "TAG=$(date +'%Y-%m-%d')-${GITHUB_SHA::7}" >> "$GITHUB_OUTPUT"
         fi
     - name: Upload binaries
-      uses: softprops/action-gh-release@c062e08bd532815e2082a85e87e3ef29c3e6d191
+      uses: softprops/action-gh-release@e7a8f85e1c67a31e6ed99a94b41bd0b71bbee6b8
       with:
         files: |
           sycl_linux.tar.gz

clang/include/clang/Basic/LangOptions.def

@@ -326,6 +326,7 @@ LANGOPT(SYCLExperimentalRangeRounding, 1, 0, "Use experimental parallel for rang
 LANGOPT(SYCLEnableIntHeaderDiags, 1, 0, "Enable diagnostics that require the "
         "SYCL integration header")
 LANGOPT(SYCLIsNativeCPU      , 1, 0, "Generate code for SYCL Native CPU")
+LANGOPT(SYCLRTCMode, 1, 0, "Compile in RTC mode")
 
 LANGOPT(HIPUseNewLaunchAPI, 1, 0, "Use new kernel launching API for HIP")
 LANGOPT(OffloadUniformBlock, 1, 0, "Assume that kernels are launched with uniform block sizes (default true for CUDA/HIP and false otherwise)")

clang/include/clang/Driver/Driver.h

@@ -586,9 +586,9 @@ class Driver {
   /// @name Helper Methods
   /// @{
 
-  /// MakeSYCLDeviceTriple - Returns the SYCL device triple for the
+  /// getSYCLDeviceTriple - Returns the SYCL device triple for the
   /// specified subarch
-  llvm::Triple MakeSYCLDeviceTriple(StringRef TargetArch = "spir64") const;
+  llvm::Triple getSYCLDeviceTriple(StringRef TargetArch = "spir64") const;
 
   /// PrintActions - Print the list of actions.
   void PrintActions(const Compilation &C) const;

clang/include/clang/Driver/Options.td

@@ -6877,6 +6877,11 @@ defm sycl_esimd_force_stateless_mem : BoolFOption<"sycl-esimd-force-stateless-me
   NegFlag<SetFalse, [], [ClangOption, CLOption], "Do not enforce using "
           "stateless memory accesses.">,
   BothFlags<[], [ClangOption, CLOption, CC1Option], "">>;
+defm sycl_rtc_mode: BoolFOption<"sycl-rtc-mode",
+  LangOpts<"SYCLRTCMode">, DefaultFalse,
+  PosFlag<SetTrue, [], [ClangOption], "Enable">,
+  NegFlag<SetFalse, [], [ClangOption], "Disable">,
+  BothFlags<[HelpHidden], [ClangOption, CC1Option], " RTC mode in SYCL.">>;
 // TODO: Remove this option once ESIMD headers are updated to
 // guard vectors to be device only.
 def fno_sycl_esimd_build_host_code :  Flag<["-"], "fno-sycl-esimd-build-host-code">,

clang/include/clang/Driver/ToolChain.h

@@ -781,7 +781,7 @@ class ToolChain {
   virtual void AddHIPIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                                  llvm::opt::ArgStringList &CC1Args) const;
 
-  /// Add arguments to use SYCL specific includes.
+  /// Add arguments to use system-specific SYCL includes.
   virtual void AddSYCLIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                                   llvm::opt::ArgStringList &CC1Args) const;
 

clang/lib/Driver/Compilation.cpp

@@ -344,7 +344,7 @@ static bool ActionFailed(const Action *A,
 
   // 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 pipeline on errors.
+  // 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;

clang/lib/Driver/Driver.cpp

@@ -834,6 +834,7 @@ static bool isValidSYCLTriple(llvm::Triple T) {
 }
 
 static const char *getDefaultSYCLArch(Compilation &C) {
+  // If -fsycl is supplied we will assume SPIR-V
   if (C.getDefaultToolChain().getTriple().getArch() == llvm::Triple::x86)
     return "spir";
   return "spir64";
@@ -847,7 +848,7 @@ static bool addSYCLDefaultTriple(Compilation &C,
   if (C.getInputArgs().hasArg(options::OPT_fsycl_force_target_EQ))
     return false;
   llvm::Triple DefaultTriple =
-      C.getDriver().MakeSYCLDeviceTriple(getDefaultSYCLArch(C));
+      C.getDriver().getSYCLDeviceTriple(getDefaultSYCLArch(C));
   for (const auto &SYCLTriple : SYCLTriples) {
     if (SYCLTriple == DefaultTriple)
       return false;
@@ -1079,22 +1080,21 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
   // We need to generate a SYCL toolchain if the user specified -fsycl.
   // If -fsycl is supplied without any of these we will assume SPIR-V.
   // Use of -fsycl-device-only overrides -fsycl.
-  bool HasValidSYCLRuntime =
-      C.getInputArgs().hasFlag(options::OPT_fsycl, options::OPT_fno_sycl,
-                               false) ||
-      C.getInputArgs().hasArg(options::OPT_fsycl_device_only);
+  bool IsSYCL = C.getInputArgs().hasFlag(options::OPT_fsycl,
+                                         options::OPT_fno_sycl, false) ||
+                C.getInputArgs().hasArg(options::OPT_fsycl_device_only);
 
   Arg *SYCLfpga = C.getInputArgs().getLastArg(options::OPT_fintelfpga);
 
   // Make -fintelfpga flag imply -fsycl.
-  if (SYCLfpga && !HasValidSYCLRuntime)
-    HasValidSYCLRuntime = true;
+  if (SYCLfpga && !IsSYCL)
+    IsSYCL = true;
 
   // A mechanism for retrieving SYCL-specific options, erroring out
   // if SYCL offloading wasn't enabled prior to that
   auto getArgRequiringSYCLRuntime = [&](OptSpecifier OptId) -> Arg * {
     Arg *SYCLArg = C.getInputArgs().getLastArg(OptId);
-    if (SYCLArg && !HasValidSYCLRuntime) {
+    if (SYCLArg && !IsSYCL) {
       Diag(clang::diag::err_drv_expecting_fsycl_with_sycl_opt)
           // Dropping the '=' symbol, which would otherwise pollute
           // the diagnostics for the most of options
@@ -1123,7 +1123,7 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
         << "-fsycl-host-compiler";
 
   auto argSYCLIncompatible = [&](OptSpecifier OptId) {
-    if (!HasValidSYCLRuntime)
+    if (!IsSYCL)
       return;
     if (Arg *IncompatArg = C.getInputArgs().getLastArg(OptId))
       Diag(clang::diag::err_drv_argument_not_allowed_with)
@@ -1182,7 +1182,7 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
       getArgRequiringSYCLRuntime(options::OPT_fsycl_force_target_EQ);
   if (SYCLForceTarget) {
     StringRef Val(SYCLForceTarget->getValue());
-    llvm::Triple TT(MakeSYCLDeviceTriple(Val));
+    llvm::Triple TT(getSYCLDeviceTriple(Val));
     if (!isValidSYCLTriple(TT))
       Diag(clang::diag::err_drv_invalid_sycl_target) << Val;
   }
@@ -1240,7 +1240,7 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
             continue;
           }
 
-          llvm::Triple DeviceTriple(MakeSYCLDeviceTriple(UserTargetName));
+          llvm::Triple DeviceTriple(getSYCLDeviceTriple(UserTargetName));
           if (!isValidSYCLTriple(DeviceTriple)) {
             Diag(clang::diag::err_drv_invalid_sycl_target) << Val;
             continue;
@@ -1271,7 +1271,7 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
           }
 
           // Make sure we don't have a duplicate triple.
-          std::string NormalizedName = MakeSYCLDeviceTriple(Val).normalize();
+          std::string NormalizedName = getSYCLDeviceTriple(Val).normalize();
           auto Duplicate = FoundNormalizedTriples.find(NormalizedName);
           if (Duplicate != FoundNormalizedTriples.end()) {
             Diag(clang::diag::warn_drv_sycl_offload_target_duplicate)
@@ -1303,9 +1303,8 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
   // Create a toolchain for each valid triple.
   // We do not support SYCL offloading if any of the inputs is a
   // .cu (for CUDA type) or .hip (for HIP type) file.
-  else if (HasValidSYCLRuntime &&
-           C.getInputArgs().hasArg(options::OPT_offload_arch_EQ) && !IsHIP &&
-           !IsCuda) {
+  else if (IsSYCL && C.getInputArgs().hasArg(options::OPT_offload_arch_EQ) &&
+           !IsHIP && !IsCuda) {
     // SYCL offloading to AOT Targets with '--offload-arch'
     // is currently enabled only with '--offload-new-driver' option.
     // Emit a diagnostic if '--offload-arch' is invoked without
@@ -1352,7 +1351,7 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
                      getProcessorFromTargetID(*AMDTriple, Arch)))) {
         DerivedArchs[AMDTriple->getTriple()].insert(Arch);
       } else if (IsSYCLSupportedIntelCPUArch(StringToOffloadArchSYCL(Arch))) {
-        DerivedArchs[MakeSYCLDeviceTriple("spir64_x86_64").getTriple()].insert(
+        DerivedArchs[getSYCLDeviceTriple("spir64_x86_64").getTriple()].insert(
             Arch);
       } else if (IsSYCLSupportedIntelGPUArch(StringToOffloadArchSYCL(Arch))) {
         StringRef IntelGPUArch;
@@ -1363,7 +1362,7 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
         // offloading to Intel GPUs and the corresponding '-device' value passed
         // to OCLOC.
         IntelGPUArch = mapIntelGPUArchName(Arch).data();
-        DerivedArchs[MakeSYCLDeviceTriple("spir64_gen").getTriple()].insert(
+        DerivedArchs[getSYCLDeviceTriple("spir64_gen").getTriple()].insert(
             IntelGPUArch);
       } else {
         Diag(clang::diag::err_drv_invalid_sycl_target) << Arch;
@@ -1381,7 +1380,7 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
       SYCLTriples.insert(TripleAndArchs.first());
 
     for (const auto &Val : SYCLTriples) {
-      llvm::Triple SYCLTargetTriple(MakeSYCLDeviceTriple(Val.getKey()));
+      llvm::Triple SYCLTargetTriple(getSYCLDeviceTriple(Val.getKey()));
       std::string NormalizedName = SYCLTargetTriple.normalize();
 
       // Make sure we don't have a duplicate triple.
@@ -1403,12 +1402,12 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
   } else {
     // If -fsycl is supplied without -fsycl-targets we will assume SPIR-V.
     // For -fsycl-device-only, we also setup the implied triple as needed.
-    if (HasValidSYCLRuntime) {
+    if (IsSYCL) {
       StringRef SYCLTargetArch = getDefaultSYCLArch(C);
       if (SYCLfpga)
         // Triple for -fintelfpga is spir64_fpga.
         SYCLTargetArch = "spir64_fpga";
-      UniqueSYCLTriplesVec.push_back(MakeSYCLDeviceTriple(SYCLTargetArch));
+      UniqueSYCLTriplesVec.push_back(getSYCLDeviceTriple(SYCLTargetArch));
       addSYCLDefaultTriple(C, UniqueSYCLTriplesVec);
     }
   }
@@ -2529,7 +2528,7 @@ void Driver::PrintHelp(bool ShowHidden) const {
                       VisibilityMask);
 }
 
-llvm::Triple Driver::MakeSYCLDeviceTriple(StringRef TargetArch) const {
+llvm::Triple Driver::getSYCLDeviceTriple(StringRef TargetArch) const {
   SmallVector<StringRef, 5> SYCLAlias = {
       "spir",       "spir64",  "spir64_fpga", "spir64_x86_64",
       "spir64_gen", "spirv32", "spirv64",     "nvptx64"};
@@ -2557,13 +2556,13 @@ void Driver::PrintSYCLToolHelp(const Compilation &C) const {
     StringRef AV(A->getValue());
     llvm::Triple T;
     if (AV == "gen" || AV == "all")
-      HelpArgs.push_back(std::make_tuple(MakeSYCLDeviceTriple("spir64_gen"),
+      HelpArgs.push_back(std::make_tuple(getSYCLDeviceTriple("spir64_gen"),
                                          "ocloc", "--help", ""));
     if (AV == "fpga" || AV == "all")
-      HelpArgs.push_back(std::make_tuple(MakeSYCLDeviceTriple("spir64_fpga"),
+      HelpArgs.push_back(std::make_tuple(getSYCLDeviceTriple("spir64_fpga"),
                                          "aoc", "-help", "-sycl"));
     if (AV == "x86_64" || AV == "all")
-      HelpArgs.push_back(std::make_tuple(MakeSYCLDeviceTriple("spir64_x86_64"),
+      HelpArgs.push_back(std::make_tuple(getSYCLDeviceTriple("spir64_x86_64"),
                                          "opencl-aot", "--help", ""));
     if (HelpArgs.empty()) {
       C.getDriver().Diag(diag::err_drv_unsupported_option_argument)
@@ -3614,7 +3613,7 @@ static bool hasSYCLDefaultSection(Compilation &C, const StringRef &File) {
   if (!(IsArchive || isObjectFile(File.str())))
     return false;
 
-  llvm::Triple TT(C.getDriver().MakeSYCLDeviceTriple(getDefaultSYCLArch(C)));
+  llvm::Triple TT(C.getDriver().getSYCLDeviceTriple(getDefaultSYCLArch(C)));
   // Checking uses -check-section option with the input file, no output
   // file and the target triple being looked for.
   const char *Targets =
@@ -3821,7 +3820,7 @@ bool Driver::checkForSYCLDefaultDevice(Compilation &C,
   // or if -fsycl-targets isn't passed (that implies default device)
   if (const Arg *A = Args.getLastArg(options::OPT_fsycl_targets_EQ)) {
     for (const char *Val : A->getValues()) {
-      llvm::Triple TT(C.getDriver().MakeSYCLDeviceTriple(Val));
+      llvm::Triple TT(C.getDriver().getSYCLDeviceTriple(Val));
       if ((TT.isSPIROrSPIRV()) && TT.getSubArch() == llvm::Triple::NoSubArch)
         // Default triple found
         return false;
@@ -6303,7 +6302,7 @@ class OffloadingActionBuilder final {
           // There are a few different variants for FPGA, if we see one, just
           // use the default FPGA triple to reduce possible match confusion.
           if (Arch.compare(0, 4, "fpga") == 0)
-            Arch = C.getDriver().MakeSYCLDeviceTriple("spir64_fpga").str();
+            Arch = C.getDriver().getSYCLDeviceTriple("spir64_fpga").str();
 
           if (std::find(UniqueSections.begin(), UniqueSections.end(), Arch) ==
               UniqueSections.end())
@@ -6460,8 +6459,9 @@ class OffloadingActionBuilder final {
                 // Unrecognized, we have already diagnosed this earlier; skip.
                 continue;
               // Add the proper -device value to the list.
-              GpuArchList.emplace_back(C.getDriver().MakeSYCLDeviceTriple(
-                                       "spir64_gen"), ValidDevice->data());
+              GpuArchList.emplace_back(
+                  C.getDriver().getSYCLDeviceTriple("spir64_gen"),
+                  ValidDevice->data());
               UserTargetName = "spir64_gen";
             } else if (auto ValidDevice =
                            gen::isGPUTarget<gen::NvidiaGPU>(Val)) {
@@ -6470,7 +6470,7 @@ class OffloadingActionBuilder final {
                 continue;
               // Add the proper -device value to the list.
               GpuArchList.emplace_back(
-                  C.getDriver().MakeSYCLDeviceTriple("nvptx64-nvidia-cuda"),
+                  C.getDriver().getSYCLDeviceTriple("nvptx64-nvidia-cuda"),
                   ValidDevice->data());
               UserTargetName = "nvptx64-nvidia-cuda";
             } else if (auto ValidDevice = gen::isGPUTarget<gen::AmdGPU>(Val)) {
@@ -6479,7 +6479,7 @@ class OffloadingActionBuilder final {
                 continue;
               // Add the proper -device value to the list.
               GpuArchList.emplace_back(
-                  C.getDriver().MakeSYCLDeviceTriple("amdgcn-amd-amdhsa"),
+                  C.getDriver().getSYCLDeviceTriple("amdgcn-amd-amdhsa"),
                   ValidDevice->data());
               UserTargetName = "amdgcn-amd-amdhsa";
             } else if (Val == "native_cpu") {
@@ -6490,7 +6490,7 @@ class OffloadingActionBuilder final {
               continue;
             }
 
-            llvm::Triple TT(C.getDriver().MakeSYCLDeviceTriple(Val));
+            llvm::Triple TT(C.getDriver().getSYCLDeviceTriple(Val));
             std::string NormalizedName = TT.normalize();
 
             // Make sure we don't have a duplicate triple.
@@ -6503,7 +6503,7 @@ class OffloadingActionBuilder final {
             FoundNormalizedTriples[NormalizedName] = Val;
 
             SYCLTripleList.push_back(
-                C.getDriver().MakeSYCLDeviceTriple(UserTargetName));
+                C.getDriver().getSYCLDeviceTriple(UserTargetName));
             if (TT.getSubArch() == llvm::Triple::SPIRSubArch_fpga)
               SYCLfpgaTriple = true;
             // For user specified spir64_gen, add an empty device value as a
@@ -6567,7 +6567,7 @@ class OffloadingActionBuilder final {
         // -fsycl -fintelfpga implies spir64_fpga
         const char *SYCLTargetArch =
             SYCLfpga ? "spir64_fpga" : getDefaultSYCLArch(C);
-        llvm::Triple TT = C.getDriver().MakeSYCLDeviceTriple(SYCLTargetArch);
+        llvm::Triple TT = C.getDriver().getSYCLDeviceTriple(SYCLTargetArch);
         auto TCIt = llvm::find_if(
             ToolChains, [&](auto &TC) { return TT == TC->getTriple(); });
         assert(TCIt != ToolChains.end() &&
@@ -8372,14 +8372,14 @@ Action *Driver::ConstructPhaseAction(
       return C.MakeAction<BackendJobAction>(Input, Output);
     }
     if (Args.hasArg(options::OPT_emit_llvm) ||
-        ((TargetDeviceOffloadKind == Action::OFK_SYCL &&
-          C.getDriver().getUseNewOffloadingDriver()) ||
-         (((Input->getOffloadingToolChain() &&
-            Input->getOffloadingToolChain()->getTriple().isAMDGPU()) ||
-           TargetDeviceOffloadKind == Action::OFK_HIP) &&
-          (Args.hasFlag(options::OPT_fgpu_rdc, options::OPT_fno_gpu_rdc,
-                        false) ||
-           TargetDeviceOffloadKind == Action::OFK_OpenMP)))) {
+        (TargetDeviceOffloadKind == Action::OFK_SYCL &&
+         C.getDriver().getUseNewOffloadingDriver()) ||
+        (((Input->getOffloadingToolChain() &&
+           Input->getOffloadingToolChain()->getTriple().isAMDGPU()) ||
+          TargetDeviceOffloadKind == Action::OFK_HIP) &&
+         (Args.hasFlag(options::OPT_fgpu_rdc, options::OPT_fno_gpu_rdc,
+                       false) ||
+          TargetDeviceOffloadKind == Action::OFK_OpenMP))) {
       types::ID Output =
           Args.hasArg(options::OPT_S) &&
                   (TargetDeviceOffloadKind == Action::OFK_None ||

clang/lib/Driver/ToolChain.cpp

@@ -1660,7 +1660,7 @@ llvm::opt::DerivedArgList *ToolChain::TranslateOffloadTargetArgs(
         A->getOption().matches(options::OPT_Xsycl_frontend);
       if (A->getOption().matches(options::OPT_Xsycl_frontend_EQ)) {
         // Passing device args: -Xsycl-target-frontend=<triple> -opt=val.
-        if (getDriver().MakeSYCLDeviceTriple(A->getValue(0)) == getTriple())
+        if (getDriver().getSYCLDeviceTriple(A->getValue(0)) == getTriple())
           Index = Args.getBaseArgs().MakeIndex(A->getValue(1));
         else
           continue;