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

Author: mikolaj-pirog

.github/workflows/sycl-detect-changes.yml

@@ -84,6 +84,10 @@ jobs:
               - .github/workflows/ur-*
             ur_cuda_adapter:
               - 'unified-runtime/source/adapters/cuda/**'
+            ur_offload_adapter:
+              - 'unified-runtime/include/**'
+              - 'unified-runtime/source/adapters/offload/**'
+              - '.github/workflows/ur-build-offload.yml'
 
       - name: Set output
         id: result
@@ -94,7 +98,7 @@ jobs:
               return '${{ steps.changes.outputs.changes }}';
             }
             // Treat everything as changed for huge PRs.
-            return ["llvm", "llvm_spirv", "clang", "sycl_jit", "xptifw", "libclc", "sycl", "ci", "esimd", "ur", "ur_cuda_adapter"];
+            return ["llvm", "llvm_spirv", "clang", "sycl_jit", "xptifw", "libclc", "sycl", "ci", "esimd", "ur", "ur_cuda_adapter", "ur_offload_adapter"];
 
       - run: echo '${{ steps.result.outputs.result }}'
 

.github/workflows/sycl-docs.yml

@@ -53,7 +53,7 @@ jobs:
         touch .nojekyll
         # Update benchmarking dashboard configuration
         cat << EOF > benchmarks/config.js
-        remoteDataUrl = 'https://raw.githubusercontent.com/intel/llvm-ci-perf-results/refs/heads/unify-ci/data.json';
+        remoteDataUrl = 'https://raw.githubusercontent.com/intel/llvm-ci-perf-results/refs/heads/unify-ci/';
         defaultCompareNames = ["Baseline_PVC_L0"];
         EOF
     # Upload the generated docs as an artifact and deploy to GitHub Pages.

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

@@ -105,9 +105,10 @@ jobs:
             runner: '["Linux", "arc"]'
             image: ghcr.io/intel/llvm/ubuntu2404_intel_drivers:devigc
             image_options: -u 1001 --device=/dev/dri -v /dev/dri/by-path:/dev/dri/by-path --privileged --cap-add SYS_ADMIN
-            target_devices: level_zero:gpu;opencl:gpu
+            target_devices: level_zero:gpu
             extra_lit_opts: --param matrix-xmx8=True
             use_igc_dev: true
+            env: '{"LIT_FILTER":"Matrix/"}'
           - name: E2E tests on Intel Ponte Vecchio GPU
             runner: '["Linux", "pvc"]'
             image_options: -u 1001 --device=/dev/dri -v /dev/dri/by-path:/dev/dri/by-path --privileged --cap-add SYS_ADMIN
@@ -116,8 +117,9 @@ jobs:
             runner: '["Linux", "pvc"]'
             image: ghcr.io/intel/llvm/ubuntu2404_intel_drivers:devigc
             image_options: -u 1001 --device=/dev/dri -v /dev/dri/by-path:/dev/dri/by-path --privileged --cap-add SYS_ADMIN
-            target_devices: level_zero:gpu;opencl:gpu
+            target_devices: level_zero:gpu
             use_igc_dev: true
+            env: '{"LIT_FILTER":"Matrix/"}'
           - name: Intel Battlemage Graphics
             runner: '["Linux", "bmg"]'
             image_options: -u 1001 --device=/dev/dri -v /dev/dri/by-path:/dev/dri/by-path --privileged --cap-add SYS_ADMIN
@@ -163,7 +165,7 @@ jobs:
         'false' }}
       # Run only if the PR does not have the 'ci-no-devigc' label.
       skip_run: ${{matrix.use_igc_dev && contains(github.event.pull_request.labels.*.name, 'ci-no-devigc') || 'false'}}
-      env: ${{ contains(needs.detect_changes.outputs.filters, 'esimd') && '{}' || '{"LIT_FILTER_OUT":"ESIMD/"}' }}
+      env: ${{ matrix.env || (contains(needs.detect_changes.outputs.filters, 'esimd') && '{}' || '{"LIT_FILTER_OUT":"ESIMD/"}') }}
 
   test-perf:
     needs: [build, detect_changes]

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

@@ -0,0 +1,54 @@
+name: UR - Build offload adapter
+
+permissions: read-all
+
+on: [ workflow_call, workflow_dispatch ]
+
+jobs:
+  offload_build:
+    name: Build
+    strategy:
+      fail-fast: false
+      matrix:
+        build_type: [Release]
+        compiler: [{c: gcc, cxx: g++}]
+
+    runs-on: [ "Linux", "build" ]
+    container:
+      image: 'ghcr.io/intel/llvm/ubuntu2404_base'
+
+    steps:
+    - name: Checkout LLVM
+      uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11 # v4.1.1
+
+    - name: Install liboffload
+      run: sudo apt-get update -qq && sudo apt-get install --no-install-recommends -yqq liboffload-21 liboffload-21-dev
+
+    - name: Configure Unified Runtime project
+      # ">" is used to avoid adding "\" at the end of each line; this command is quite long
+      run: >
+        cmake
+        -S unified-runtime
+        -B $GITHUB_WORKSPACE/build
+        -DCMAKE_C_COMPILER=${{matrix.compiler.c}}
+        -DCMAKE_CXX_COMPILER=${{matrix.compiler.cxx}}
+        -DCMAKE_BUILD_TYPE=${{matrix.build_type}}
+        -DUR_ENABLE_TRACING=ON
+        -DUR_DEVELOPER_MODE=ON
+        -DUR_BUILD_TESTS=OFF
+        -DUR_BUILD_ADAPTER_OFFLOAD=ON
+        -DUR_OFFLOAD_INSTALL_DIR="/usr/lib/llvm-21"
+        -DUR_OFFLOAD_INCLUDE_DIR="/usr/lib/llvm-21/include"
+        -DCMAKE_INSTALL_PREFIX=$GITHUB_WORKSPACE/install
+
+    - name: Build
+      # This is so that device binaries can find the sycl runtime library
+      run: cmake --build $GITHUB_WORKSPACE/build -j $(nproc)
+
+    - name: Install
+      # This is to check that install command does not fail
+      run: cmake --install $GITHUB_WORKSPACE/build
+
+    - name: Get information about platform
+      if: ${{ always() }}
+      run: ./unified-runtime/.github/scripts/get_system_info.sh

.github/workflows/ur-precommit.yml

@@ -99,6 +99,13 @@ jobs:
       docker_image: ${{ matrix.docker_image || 'ghcr.io/intel/llvm/ubuntu2404_intel_drivers:alldeps'}}
       image_options: ${{ matrix.image_options || '' }}
 
+# TODO: Enable once the apt package at https://apt.llvm.org/noble/pool/main/l/llvm-toolchain-snapshot/ is updated
+#  offload_build:
+#    name: Adapters (Offload)
+#    needs: [detect_changes, source_checks]
+#    if: ${{ always() && !cancelled() && contains(needs.detect_changes.outputs.filters, 'ur_offload_adapter') }}
+#    uses: ./.github/workflows/ur-build-offload.yml
+
   macos:
     name: MacOS build only
     needs: [detect_changes, source_checks]

clang/include/clang/Basic/Attr.td

@@ -1625,6 +1625,14 @@ def SYCLDevice : InheritableAttr {
   let Documentation = [SYCLDeviceDocs];
 }
 
+def SYCLDeviceOnly : InheritableAttr {
+  let Spellings = [Clang<"sycl_device_only">];
+  let Subjects = SubjectList<[Function]>;
+  let LangOpts = [SYCLIsDevice, SilentlyIgnoreSYCLIsHost];
+  let Documentation = [SYCLDeviceOnlyDocs];
+}
+def : MutualExclusions<[SYCLDevice, SYCLDeviceOnly]>;
+
 def SYCLGlobalVar : InheritableAttr {
   let Spellings = [GNU<"sycl_global_var">];
   let Subjects = SubjectList<[GlobalStorageNonLocalVar], ErrorDiag>;

clang/include/clang/Basic/AttrDocs.td

@@ -4522,6 +4522,20 @@ implicitly inherit this attribute.
   }];
 }
 
+def SYCLDeviceOnlyDocs : Documentation {
+  let Category = DocCatFunction;
+  let Heading = "sycl_device_only";
+  let Content = [{
+This attribute can only be applied to functions and indicates that the function
+is only available for the device. It allows functions marked with it to
+overload existing functions without the attribute, in which case the overload
+with the attribute will be used on the device side and the overload without
+will be used on the host side. Note: as opposed to ``sycl_device`` this does
+not mark the function as being exported, both attributes are incompatible and
+can't be used together.
+  }];
+}
+
 def RISCVInterruptDocs : Documentation {
   let Category = DocCatFunction;
   let Heading = "interrupt (RISC-V)";

clang/include/clang/Driver/Options.td

@@ -9051,10 +9051,6 @@ def fenable_sycl_dae : Flag<["-"], "fenable-sycl-dae">,
 def fsycl_enable_int_header_diags: Flag<["-"], "fsycl-enable-int-header-diags">,
   HelpText<"Enable diagnostics that require the SYCL integration header.">,
   MarshallingInfoFlag<LangOpts<"SYCLEnableIntHeaderDiags">>;
-def fsycl_is_native_cpu : Flag<["-"], "fsycl-is-native-cpu">,
-  HelpText<"Perform device compilation for Native CPU.">,
-  Visibility<[CC1Option]>,
-  MarshallingInfoFlag<LangOpts<"SYCLIsNativeCPU">>;
 // TODO: This option can be removed once a fix goes in that can
 // work with the community changes for using the alloca address space.
 defm offload_use_alloca_addrspace_for_srets : BoolFOption<"offload-use-alloca-addrspace-for-srets",

clang/lib/AST/Decl.cpp

@@ -3729,6 +3729,13 @@ unsigned FunctionDecl::getBuiltinID(bool ConsiderWrapperFunctions) const {
       !(BuiltinID == Builtin::BIprintf || BuiltinID == Builtin::BImalloc))
     return 0;
 
+  // SYCL doesn't have a device-side standard library. SYCLDeviceOnlyAttr may
+  // be used to provide device-side definitions of standard functions, so
+  // anything with that attribute shouldn't be treated as a builtin.
+  if (Context.getLangOpts().isSYCL() && hasAttr<SYCLDeviceOnlyAttr>()) {
+    return 0;
+  }
+
   // As AMDGCN implementation of OpenMP does not have a device-side standard
   // library, none of the predefined library functions except printf and malloc
   // should be treated as a builtin i.e. 0 should be returned for them.

clang/lib/AST/MicrosoftMangle.cpp

@@ -3226,7 +3226,7 @@ void MicrosoftCXXNameMangler::mangleCallingConvention(CallingConv CC,
       // target triple (eg for Windows) as host code.
       // FIXME: 1.) provide mangling if needed
       //        2.) check if other conventions need to be handled.
-      if (!getASTContext().getLangOpts().SYCLIsNativeCPU)
+      if (!getASTContext().getTargetInfo().getTriple().isNativeCPU())
         // Currently we only allow this convention in
         // SYCLNativeCPU and raise the usual error otherwise.
         llvm_unreachable("Unsupported CC for mangling");