Merge remote-tracking branch 'origin/sycl' into llvmspirv_pulldown

Author: jsji

.github/workflows/sycl-linux-build.yml

@@ -300,7 +300,7 @@ jobs:
         testing_mode: build-only
         target_devices: all
         binaries_artifact: ${{ inputs.e2e_binaries_artifact }}
-        cxx_compiler: $GITHUB_WORKSPACE/toolchain/bin/clang++
+        sycl_compiler: $GITHUB_WORKSPACE/toolchain/bin/clang++
         extra_lit_opts: --param sycl_build_targets="spir;nvidia;amd"
 
     - name: Remove E2E tests before spirv-backend run
@@ -315,5 +315,5 @@ jobs:
         testing_mode: build-only
         target_devices: all
         binaries_artifact: ${{ inputs.e2e_binaries_artifact }}_spirv_backend
-        cxx_compiler: $GITHUB_WORKSPACE/toolchain/bin/clang++
+        sycl_compiler: $GITHUB_WORKSPACE/toolchain/bin/clang++
         extra_lit_opts: --param spirv-backend=True

.github/workflows/sycl-nightly.yml

@@ -163,6 +163,7 @@ jobs:
     with:
       name: Intel GEN12 Graphics with Level Zero
       runner: '["Windows","gen12"]'
+      target_devices: level_zero:gpu
       sycl_toolchain_archive: ${{ needs.build-win.outputs.artifact_archive_name }}
 
   cuda-aws-start:

.github/workflows/sycl-post-commit.yml

@@ -102,7 +102,7 @@ jobs:
       && github.repository == 'intel/llvm'
     uses: ./.github/workflows/sycl-windows-build.yml
     with:
-      compiler: icx
+      cxx: icx
       build_configure_extra_args: -DCMAKE_C_FLAGS="/fp:precise /clang:-Wno-nonportable-include-path /clang:-Wno-cast-function-type-mismatch" -DCMAKE_CXX_FLAGS="/fp:precise /clang:-Wno-nonportable-include-path /clang:-Wno-cast-function-type-mismatch" -DCMAKE_EXE_LINKER_FLAGS=/manifest:no -DCMAKE_MODULE_LINKER_FLAGS=/manifest:no -DCMAKE_SHARED_LINKER_FLAGS=/manifest:no
       build_cache_suffix: icx
 
@@ -119,7 +119,7 @@ jobs:
       runner: '["Windows","gen12"]'
       target_devices: "level_zero:gpu"
       sycl_toolchain_archive: ${{ needs.build-win.outputs.artifact_archive_name }}
-      compiler: icx
+      cxx: icx
 
   macos_default:
     name: macOS

.github/workflows/sycl-windows-build.yml

@@ -28,7 +28,7 @@ on:
       e2e_binaries_artifact:
         type: string
         required: false
-      compiler:
+      cxx:
         type: string
         required: false
         default: "cl"
@@ -67,7 +67,7 @@ on:
         type: choice
         options:
           - 3
-      compiler:
+      cxx:
         type: choice
         options:
           - cl
@@ -104,7 +104,7 @@ jobs:
         arch: amd64
     - name: Setup oneAPI env
       uses: ./devops/actions/setup_windows_oneapi_env
-      if: ${{ always() && !cancelled() && inputs.compiler == 'icx' }}
+      if: ${{ always() && !cancelled() && inputs.cxx == 'icx' }}
     - name: Set env
       run: |
         git config --system core.longpaths true
@@ -132,8 +132,8 @@ jobs:
         IF NOT EXIST D:\github\_work\cache\${{inputs.build_cache_suffix}} MKDIR D:\github\_work\cache\${{inputs.build_cache_suffix}}
         python.exe src/buildbot/configure.py -o build ^
           --ci-defaults %ARGS% ^
-          "-DCMAKE_C_COMPILER=${{inputs.compiler}}" ^
-          "-DCMAKE_CXX_COMPILER=${{inputs.compiler}}" ^
+          "-DCMAKE_C_COMPILER=${{inputs.cxx}}" ^
+          "-DCMAKE_CXX_COMPILER=${{inputs.cxx}}" ^
           "-DCMAKE_INSTALL_PREFIX=%GITHUB_WORKSPACE%\install" ^
           -DCMAKE_CXX_COMPILER_LAUNCHER=ccache ^
           -DCMAKE_C_COMPILER_LAUNCHER=ccache ^
@@ -244,6 +244,7 @@ jobs:
         target_devices: all
         binaries_artifact: ${{ inputs.e2e_binaries_artifact }}
         extra_lit_opts: --param sycl_build_targets="spir"
+        cxx: ${{ inputs.cxx }}
 
     - name:  Detect hung tests
       if: always()

.github/workflows/sycl-windows-run-tests.yml

@@ -58,7 +58,7 @@ on:
         default: '{}'
         required: False
 
-      compiler:
+      cxx:
         type: string
         required: false
         default: "cl"
@@ -117,7 +117,7 @@ jobs:
         arch: amd64
     - name: Setup oneAPI env
       uses: ./devops/actions/setup_windows_oneapi_env
-      if: ${{ always() && !cancelled() && inputs.compiler == 'icx' }}
+      if: ${{ always() && !cancelled() && inputs.cxx == 'icx' }}
     - name: Set env
       run: |
         git config --system core.longpaths true
@@ -164,7 +164,7 @@ jobs:
         target_devices: ${{ inputs.target_devices }}
         extra_lit_opts: ${{ inputs.extra_lit_opts }}
         retention-days: ${{ inputs.artifact_retention_days }}
-        compiler: ${{ inputs.compiler }}
+        cxx: ${{ inputs.cxx }}
 
     - name: Run SYCL CTS Tests
       if: inputs.tests_selector == 'cts'

.github/workflows/ur-build-hw.yml

@@ -13,11 +13,6 @@ on:
       runner_name:
         required: true
         type: string
-      platform:
-        description: "Platform string, `UR_CTS_ADAPTER_PLATFORM` will be set to this."
-        required: false
-        type: string
-        default: ""
       static_loader:
         required: false
         type: string
@@ -46,11 +41,6 @@ on:
       runner_name:
         required: true
         type: string
-      platform:
-        description: "Platform string, `UR_CTS_ADAPTER_PLATFORM` will be set to this."
-        required: false
-        type: string
-        default: ""
       static_loader:
         required: false
         type: string
@@ -89,7 +79,6 @@ jobs:
           {
             name: "${{inputs.adapter_name}}",
             other_name: "${{inputs.other_adapter_name}}",
-            platform: "${{inputs.platform}}",
             static_Loader: "${{inputs.static_loader}}",
             static_adapter: "${{inputs.static_loader}}"
           }
@@ -105,7 +94,6 @@ jobs:
     steps:
     # TODO:
     # - investigate if DUR_CONFORMANCE_AMD_ARCH could be removed
-    # - find better way to handle platform param (e.g. "Intel(R) OpenCL" -> "opencl")
     # - switch to Ninja generator in CMake
     # - downloading DPC++ should be integrated somehow; most likely use nightly release.
     #
@@ -184,7 +172,7 @@ jobs:
     - name: Test adapters
       env:
         ZE_ENABLE_LOADER_DEBUG_TRACE: 1
-      run: env UR_CTS_ADAPTER_PLATFORM="${{matrix.adapter.platform}}" ctest -C ${{matrix.build_type}} --test-dir build --output-on-failure -L "conformance" --timeout 600 -VV
+      run: ctest -C ${{matrix.build_type}} --test-dir build --output-on-failure -L "conformance" --timeout 600 -VV
 
     - name: Get information about platform
       if: ${{ always() }}

.github/workflows/ur-precommit.yml

@@ -77,12 +77,10 @@ jobs:
             image_options: -u 1001 --privileged --cap-add SYS_ADMIN --gpus all
           - name: OPENCL
             runner: UR_OPENCL
-            platform: "Intel(R) OpenCL"
             docker_image: "ghcr.io/intel/llvm/ubuntu2204_build:latest"
             image_options: -u 1001 --device=/dev/dri --device=/dev/kfd --privileged --cap-add SYS_ADMIN
           - name: OPENCL
             runner: UR_OPENCL
-            platform: "OPENCL:Intel(R) OpenCL"
             other_adapter: NATIVE_CPU
             docker_image: "ghcr.io/intel/llvm/ubuntu2204_build:latest"
             image_options: -u 1001 --device=/dev/dri --device=/dev/kfd --privileged --cap-add SYS_ADMIN
@@ -96,7 +94,6 @@ jobs:
       runner_name: ${{ matrix.runner }}
       static_loader: ${{ matrix.static || 'OFF' }}
       static_adapter: ${{ matrix.static || 'OFF' }}
-      platform: ${{ matrix.platform || '' }}
       other_adapter_name: ${{ matrix.other_adapter || '' }}
       docker_image: ${{ matrix.docker_image || 'ghcr.io/intel/llvm/ubuntu2404_intel_drivers:alldeps'}}
       image_options: ${{ matrix.image_options || '' }}

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -12865,8 +12865,13 @@ def err_registered_kernels_name_already_registered : Error<
   "free function kernel has already been registered with '%0'; cannot register with '%1'">;
 def err_not_sycl_free_function : Error<
   "attempting to register a function that is not a SYCL free function as '%0'">;
-def err_free_function_variadic_args: Error<
+def err_free_function_with_default_arg : Error<
+  "a function with a default argument value cannot be used to define SYCL free function kernel">;
+def err_free_function_variadic_args : Error<
   "free function kernel cannot be a variadic function">;
+def err_free_function_return_type : Error<
+  "SYCL free function kernel should have return type 'void'">;
+
 
 // SYCL kernel entry point diagnostics
 def err_sycl_entry_point_invalid : Error<

clang/include/clang/Driver/Options.td

@@ -7300,6 +7300,8 @@ defm sycl_allow_device_image_dependencies: BoolOptionWithoutMarshalling<"f", "sy
 def fsycl_dump_device_code_EQ : Joined<["-"], "fsycl-dump-device-code=">,
   Flags<[NoXarchOption]>,
   HelpText<"Dump device code into the user provided directory.">;
+def fsyclbin : Flag<["-"], "fsyclbin">,
+  HelpText<"Create a SYCLBIN file">;
 } // let Group = sycl_Group
 
 // FIXME: -fsycl-explicit-simd is deprecated. remove it when support is dropped.

clang/lib/Driver/Driver.cpp

@@ -1226,9 +1226,11 @@ 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.
+  // Use of -fsyclbin enables SYCL device compilation.
   bool IsSYCL = C.getInputArgs().hasFlag(options::OPT_fsycl,
                                          options::OPT_fno_sycl, false) ||
-                C.getInputArgs().hasArg(options::OPT_fsycl_device_only);
+                C.getInputArgs().hasArgNoClaim(options::OPT_fsycl_device_only,
+                                               options::OPT_fsyclbin);
 
   auto argSYCLIncompatible = [&](OptSpecifier OptId) {
     if (!IsSYCL)
@@ -3468,7 +3470,7 @@ void Driver::BuildInputs(const ToolChain &TC, DerivedArgList &Args,
   Arg *InputTypeArg = nullptr;
   bool IsSYCL =
       Args.hasFlag(options::OPT_fsycl, options::OPT_fno_sycl, false) ||
-      Args.hasArg(options::OPT_fsycl_device_only);
+      Args.hasArgNoClaim(options::OPT_fsycl_device_only, options::OPT_fsyclbin);
 
   // The last /TC or /TP option sets the input type to C or C++ globally.
   if (Arg *TCTP = Args.getLastArgNoClaim(options::OPT__SLASH_TC,
@@ -7955,6 +7957,19 @@ Action *Driver::BuildOffloadingActions(Compilation &C,
         C.MakeAction<LinkJobAction>(OffloadActions, types::TY_HIP_FATBIN);
     DDep.add(*FatbinAction, *C.getSingleOffloadToolChain<Action::OFK_HIP>(),
              nullptr, Action::OFK_HIP);
+  } else if (C.isOffloadingHostKind(Action::OFK_SYCL) &&
+             Args.hasArg(options::OPT_fsyclbin)) {
+    // With '-fsyclbin', package all the offloading actions into a single output
+    // that is sent to the clang-linker-wrapper.
+    Action *PackagerAction =
+        C.MakeAction<OffloadPackagerJobAction>(OffloadActions, types::TY_Image);
+    ActionList PackagerActions;
+    PackagerActions.push_back(PackagerAction);
+    Action *LinkAction =
+        C.MakeAction<LinkerWrapperJobAction>(PackagerActions, types::TY_Image);
+    DDep.add(*LinkAction, *C.getSingleOffloadToolChain<Action::OFK_Host>(),
+             nullptr, C.getActiveOffloadKinds());
+    return C.MakeAction<OffloadAction>(DDep, types::TY_Nothing);
   } else {
     // Package all the offloading actions into a single output that can be
     // embedded in the host and linked.
@@ -9365,8 +9380,10 @@ const char *Driver::GetNamedOutputPath(Compilation &C, const JobAction &JA,
   if (AtTopLevel && !isa<DsymutilJobAction>(JA) && !isa<VerifyJobAction>(JA)) {
     if (Arg *FinalOutput = C.getArgs().getLastArg(options::OPT_o))
       return C.addResultFile(FinalOutput->getValue(), &JA);
-    // Output to destination for -fsycl-device-only and Windows -o
-    if (offloadDeviceOnly() && JA.getOffloadingDeviceKind() == Action::OFK_SYCL)
+    // Output to destination for -fsycl-device-only/-fsyclbin and Windows -o
+    if ((offloadDeviceOnly() ||
+         C.getArgs().hasArgNoClaim(options::OPT_fsyclbin)) &&
+        JA.getOffloadingDeviceKind() == Action::OFK_SYCL)
       if (Arg *FinalOutput = C.getArgs().getLastArg(options::OPT__SLASH_o))
         return C.addResultFile(FinalOutput->getValue(), &JA);
   }
@@ -9535,6 +9552,20 @@ const char *Driver::GetNamedOutputPath(Compilation &C, const JobAction &JA,
   else
     BaseName = llvm::sys::path::filename(BasePath);
 
+  // When compiling with -fsyclbin, maintain a simple output file name for the
+  // resulting image.  A '.syclbin' extension is used to represent the resulting
+  // output file.
+  if (JA.getOffloadingDeviceKind() == Action::OFK_SYCL &&
+      C.getArgs().hasArgNoClaim(options::OPT_fsyclbin) &&
+      JA.getType() == types::TY_Image) {
+    SmallString<128> SYCLBinOutput(getDefaultImageName());
+    if (IsCLMode())
+      // Use BaseName for the syclbin output name.
+      SYCLBinOutput = BaseName;
+    llvm::sys::path::replace_extension(SYCLBinOutput, ".syclbin");
+    return C.addResultFile(C.getArgs().MakeArgString(SYCLBinOutput), &JA);
+  }
+
   // Determine what the derived output name should be.
   const char *NamedOutput;
 

clang/lib/Driver/ToolChains/Clang.cpp

@@ -11499,6 +11499,9 @@ void LinkerWrapper::ConstructJob(Compilation &C, const JobAction &JA,
         CmdArgs.push_back(
             Args.MakeArgString("--sycl-target-link-options=" + LinkOptString));
     }
+    // Add option to enable creating of the .syclbin file.
+    if (Args.hasArg(options::OPT_fsyclbin))
+      CmdArgs.push_back(Args.MakeArgString("--syclbin"));
   }
 
   // Construct the link job so we can wrap around it.

clang/lib/Sema/SPIRVBuiltins.td

@@ -850,7 +850,7 @@ foreach AS = [GlobalAS, LocalAS, PrivateAS] in {
   def : SPVBuiltin<"GenericCastToPtrExplicit", [PointerType<Char, AS>, PointerType<Char, GenericAS>], Attr.Const>;
 }
 
-foreach Ty = [Void, ConstType<Void>, VolatileType<Void>, VolatileType<ConstType<Void>>] in {
+foreach Ty = [Void, ConstType<Void>] in {
   def : SPVBuiltin<"GenericCastToPtrExplicit_ToGlobal", [PointerType<Ty, GlobalAS>, PointerType<Ty, DefaultAS>, Int], Attr.Const>;
   def : SPVBuiltin<"GenericCastToPtrExplicit_ToLocal", [PointerType<Ty, LocalAS>, PointerType<Ty, DefaultAS>, Int], Attr.Const>;
   def : SPVBuiltin<"GenericCastToPtrExplicit_ToPrivate", [PointerType<Ty, PrivateAS>, PointerType<Ty, DefaultAS>, Int], Attr.Const>;

clang/lib/Sema/SemaSYCL.cpp

@@ -1159,21 +1159,13 @@ static target getAccessTarget(QualType FieldTy,
       AccTy->getTemplateArgs()[3].getAsIntegral().getExtValue());
 }
 
-// FIXME: Free functions must have void return type and be declared at file
-// scope, outside any namespaces.
 bool SemaSYCL::isFreeFunction(const FunctionDecl *FD) {
   for (auto *IRAttr : FD->specific_attrs<SYCLAddIRAttributesFunctionAttr>()) {
     SmallVector<std::pair<std::string, std::string>, 4> NameValuePairs =
         IRAttr->getAttributeNameValuePairs(getASTContext());
     for (const auto &NameValuePair : NameValuePairs) {
       if (NameValuePair.first == "sycl-nd-range-kernel" ||
           NameValuePair.first == "sycl-single-task-kernel") {
-        if (!FD->getReturnType()->isVoidType()) {
-          llvm::report_fatal_error(
-              "Only functions at file scope with void return "
-              "type are permitted as free functions");
-          return false;
-        }
         return true;
       }
     }
@@ -5791,12 +5783,29 @@ void SemaSYCL::MarkDevices() {
   }
 }
 
+static bool CheckFreeFunctionDiagnostics(Sema &S, FunctionDecl *FD) {
+  if (FD->isVariadic()) {
+    return S.Diag(FD->getLocation(), diag::err_free_function_variadic_args);
+  }
+
+  if (!FD->getReturnType()->isVoidType()) {
+    return S.Diag(FD->getLocation(), diag::err_free_function_return_type);
+  }
+
+  for (ParmVarDecl *Param : FD->parameters()) {
+    if (Param->hasDefaultArg()) {
+      return S.Diag(Param->getLocation(),
+                    diag::err_free_function_with_default_arg)
+             << Param->getSourceRange();
+    }
+  }
+  return false;
+}
+
 void SemaSYCL::ProcessFreeFunction(FunctionDecl *FD) {
   if (isFreeFunction(FD)) {
-    if (FD->isVariadic()) {
-      Diag(FD->getLocation(), diag::err_free_function_variadic_args);
+    if (CheckFreeFunctionDiagnostics(SemaRef, FD))
       return;
-    }
     SyclKernelDecompMarker DecompMarker(*this);
     SyclKernelFieldChecker FieldChecker(*this);
     SyclKernelUnionChecker UnionChecker(*this);