[SYCL] Add support for structs containing special types as free function kernel parameters

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

@@ -339,10 +339,6 @@ jobs:
         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
-      if: ${{ inputs.e2e_binaries_spirv_backend_artifact && !cancelled() && steps.build.conclusion == 'success' }}
-      run: rm -rf build-e2e
-
     - name: Build E2E tests with SPIR-V Backend
       if: ${{ inputs.e2e_binaries_spirv_backend_artifact && !cancelled() && steps.build.conclusion == 'success' }}
       uses: ./devops/actions/run-tests/e2e
@@ -354,10 +350,6 @@ jobs:
         sycl_compiler: $GITHUB_WORKSPACE/toolchain/bin/clang++
         extra_lit_opts: --param spirv-backend=True
 
-    - name: Remove E2E tests before preview-mode run
-      if: ${{ inputs.e2e_binaries_preview_artifact && !cancelled() && steps.build.conclusion == 'success' }}
-      run: rm -rf build-e2e
-
     - name: Build E2E tests in Preview Mode
       if: ${{ inputs.e2e_binaries_preview_artifact && !cancelled() && steps.build.conclusion == 'success' }}
       uses: ./devops/actions/run-tests/e2e

.github/workflows/sycl-nightly.yml

@@ -222,23 +222,20 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - name: Intel L0 Gen12 GPU
+          - name: Intel Gen12 GPU
             runner: '["Windows", "gen12"]'
-            target_devices: level_zero:gpu
 
-          - name: Intel L0 Arc GPU
+          - name: Intel Arc GPU
             runner: '["Windows", "arc"]'
-            target_devices: level_zero:gpu
 
-          - name: Intel L0 Battlemage GPU
+          - name: Intel Battlemage GPU
             runner: '["Windows", "bmg"]'
-            target_devices: level_zero:gpu
 
     uses: ./.github/workflows/sycl-windows-run-tests.yml
     with:
       name: ${{ matrix.name }}
       runner: ${{ matrix.runner }}
-      target_devices: ${{ matrix.target_devices }}
+      target_devices: level_zero:gpu
       toolchain_artifact_filename: ${{ needs.build-win.outputs.toolchain_artifact_filename }}
 
   cuda-aws-start:

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

@@ -126,7 +126,7 @@ jobs:
 
     - name: Download DPC++
       run: |
-        wget -O dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-12-12/sycl_linux.tar.gz
+        wget -O dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2025-11-28/sycl_linux.tar.gz
         mkdir -p dpcpp_compiler
         tar -xvf dpcpp_compiler.tar.gz -C dpcpp_compiler
 

buildbot/compile.py

@@ -31,15 +31,18 @@ def do_compile(args):
         "cmake",
         "--build",
         abs_obj_dir,
+    ]
+
+    if args.verbose:
+        cmake_cmd.append("--verbose")
+
+    cmake_cmd += [
         "--",
         args.build_target,
         "-j",
         str(cpu_count),
     ]
 
-    if args.verbose:
-        cmake_cmd.append("--verbose")
-
     print("[Cmake Command]: {}".format(" ".join(cmake_cmd)))
 
     subprocess.check_call(cmake_cmd, cwd=abs_obj_dir)

buildbot/configure.py

@@ -21,7 +21,7 @@ def do_configure(args, passthrough_args):
     if not os.path.isdir(abs_obj_dir):
         os.makedirs(abs_obj_dir)
 
-    llvm_external_projects = "sycl;llvm-spirv;opencl;xpti;xptifw"
+    llvm_external_projects = "sycl;llvm-spirv;opencl;xpti;xptifw;compiler-rt"
 
     # libdevice build requires a working SYCL toolchain, which is not the case
     # with macOS target right now.

clang/include/clang/Sema/SemaSYCL.h

@@ -65,7 +65,8 @@ class SYCLIntegrationHeader {
     kind_work_group_memory,
     kind_dynamic_work_group_memory,
     kind_dynamic_accessor,
-    kind_last = kind_dynamic_accessor
+    kind_struct_with_special_type, // structs that contain special types
+    kind_last = kind_struct_with_special_type
   };
 
 public:
@@ -118,6 +119,9 @@ class SYCLIntegrationHeader {
   /// integration header is required.
   void addHostPipeRegistration() { NeedToEmitHostPipeRegistration = true; }
 
+  /// Set the ParentStruct field
+  void setParentStruct(ParmVarDecl *parent);
+
 private:
   // Kernel actual parameter descriptor.
   struct KernelParamDesc {
@@ -205,6 +209,20 @@ class SYCLIntegrationHeader {
   /// Keeps track of whether declaration of __sycl_host_pipe_registration
   /// type and __sycl_host_pipe_registrar variable are required to emit.
   bool NeedToEmitHostPipeRegistration = false;
+
+  // For free function kernels, keeps track of the parameter that is currently
+  // being analyzed if it is a struct that contains special types.
+  ParmVarDecl *ParentStruct = nullptr;
+
+  // For every struct that contains a special type which is given by
+  // the ParentStruct field above, record the offset and size of its fields
+  // at any nesting level. Store the information in the variable below.
+  llvm::DenseMap<ParmVarDecl *, llvm::SmallVector<std::pair<size_t, size_t>>>
+      OffsetSizeInfo;
+  // Likewise for the kind of a field i.e accessor, std_layout etc...
+  llvm::DenseMap<ParmVarDecl *,
+                 llvm::SmallVector<SYCLIntegrationHeader::kernel_param_kind_t>>
+      KindInfo;
 };
 
 class SYCLIntegrationFooter {
@@ -267,6 +285,10 @@ class SemaSYCL : public SemaBase {
 
   llvm::DenseSet<const FunctionDecl *> FreeFunctionDeclarations;
 
+  // A map that keeps track of all structs encountered with
+  // special types inside. Relevant for free function kernels only.
+  llvm::DenseSet<const RecordDecl *> StructsWithSpecialTypes;
+
 public:
   SemaSYCL(Sema &S);
 
@@ -317,6 +339,13 @@ class SemaSYCL : public SemaBase {
     SYCLKernelFunctions.insert(FD);
   }
 
+  /// Add ParentStruct to StructsWithSpecialTypes.
+  void addStructWithSpecialType(const RecordDecl *ParentStruct) {
+    StructsWithSpecialTypes.insert(ParentStruct);
+  }
+
+  auto &getStructsWithSpecialType() const { return StructsWithSpecialTypes; }
+
   /// Lazily creates and returns SYCL integration header instance.
   SYCLIntegrationHeader &getSyclIntegrationHeader() {
     if (SyclIntHeader == nullptr)

clang/lib/CodeGen/BackendUtil.cpp

@@ -1213,6 +1213,9 @@ void EmitAssemblyHelper::RunOptimizationPipeline(
       PB.registerPipelineStartEPCallback(
           [Options](ModulePassManager &MPM, OptimizationLevel Level) {
             MPM.addPass(InstrProfilingLoweringPass(*Options, false));
+            // The profiling pass adds SYCL device globals so we need to run
+            // the compile-time properties pass to update the metadata.
+            MPM.addPass(CompileTimePropertiesPass());
           });
 
     // TODO: Consider passing the MemoryProfileOutput to the pass builder via

clang/lib/Driver/ToolChains/Clang.cpp

@@ -5423,6 +5423,9 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
       CmdArgs.push_back("-fsycl-is-device");
       CmdArgs.push_back("-fdeclare-spirv-builtins");
 
+      // Set the atomic profile update flag to increment counters atomically.
+      CmdArgs.push_back("-fprofile-update=atomic");
+
       // Set O2 optimization level by default
       if (!Args.getLastArg(options::OPT_O_Group))
         CmdArgs.push_back("-O2");

clang/lib/Driver/ToolChains/SYCL.cpp

@@ -1371,11 +1371,7 @@ static ArrayRef<options::ID> getUnsupportedOpts() {
       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_coverage,             // --coverage
-      options::OPT_fno_coverage_mapping, // -f[no-]coverage-mapping
-      options::OPT_fprofile_instr_generate,
-      options::OPT_fprofile_instr_generate_EQ,
+      options::OPT_coverage,          // --coverage
       options::OPT_fprofile_arcs,
       options::OPT_fno_profile_arcs,           // -f[no-]profile-arcs
       options::OPT_fno_profile_instr_generate, // -f[no-]profile-instr-generate