Merge branch 'sycl' into llvmspirv_pulldown

Author: jsji

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

@@ -44,6 +44,7 @@ jobs:
     uses: ./.github/workflows/sycl-detect-changes.yml
 
   build:
+    name: Self build
     needs: [detect_changes]
     if: always() && success()
     uses: ./.github/workflows/sycl-linux-build.yml

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

@@ -179,6 +179,10 @@ on:
           - e2e
           - cts
           - benchmarks
+      toolchain_release_tag:
+        description: |
+          Tag of a "Nightly" release at https://github.com/intel/llvm/releases.
+        default: ''
 
       env:
         description: |
@@ -256,11 +260,6 @@ jobs:
       uses: actions/download-artifact@v4
       with:
         name: ${{ inputs.toolchain_artifact }}
-    - name: Debug prints [workflow_run]
-      if: inputs.toolchain_artifact != '' && github.event_name == 'workflow_run'
-      run: |
-        pwd
-        ls
     - name: Download SYCL toolchain [workflow_run]
       # NOTE: This is for `sycl-linux-precommit-aws.yml`.
       if: inputs.toolchain_artifact != '' && github.event_name == 'workflow_run'
@@ -299,6 +298,17 @@ jobs:
         rm -f ${{ inputs.toolchain_artifact_filename }}
         echo PATH=$PWD/toolchain/bin/:$PATH >> $GITHUB_ENV
         echo LD_LIBRARY_PATH=$PWD/toolchain/lib/:$LD_LIBRARY_PATH >> $GITHUB_ENV
+    - name: Download SYCL toolchain using release tag
+      if: inputs.toolchain_release_tag != ''
+      env:
+        TAG: ${{ inputs.toolchain_release_tag }}
+      shell: bash
+      run: |
+        mkdir toolchain
+        wget "https://github.com/intel/llvm/releases/download/$TAG/sycl_linux.tar.gz"
+        tar xf sycl_linux.tar.gz -C toolchain
+        echo PATH=$PWD/toolchain/bin/:$PATH >> $GITHUB_ENV
+        echo LD_LIBRARY_PATH=$PWD/toolchain/lib/:$LD_LIBRARY_PATH >> $GITHUB_ENV
     - run: which clang++ sycl-ls
     - run: sycl-ls --verbose
     - run: SYCL_UR_TRACE=1 sycl-ls

.github/workflows/sycl-nightly-benchmarking.yml

@@ -3,7 +3,8 @@ name: SYCL Nightly Benchmarking
 on:
   workflow_dispatch:
   schedule:
-    - cron: '0 2 * * *'
+    - cron: '0 0 * * *'
+    # 3 hours ahead of SYCL nightly
 
 permissions: read-all
 
@@ -31,7 +32,7 @@ jobs:
         include:
           - ref: ${{ github.sha }}
             save_name: 'Baseline'
-            preset: 'Minimal'
+            preset: 'Full'
     uses: ./.github/workflows/sycl-linux-run-tests.yml
     secrets: inherit
     with:

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

@@ -34,7 +34,7 @@ jobs:
     uses: ./.github/workflows/sycl-detect-changes.yml
 
   build-lin:
-    name: Linux (Self build + no-assertions)
+    name: Linux (GCC + no-assertions)
     if: github.repository == 'intel/llvm'
     uses: ./.github/workflows/sycl-linux-build.yml
     with:

clang/lib/Sema/SemaSYCL.cpp

@@ -6980,6 +6980,10 @@ void SYCLIntegrationHeader::emit(raw_ostream &O) {
 
   for (const KernelDesc &K : KernelDescs) {
     const size_t N = K.Params.size();
+    if (S.isFreeFunction(K.SyclKernel)) {
+      CurStart += N;
+      continue;
+    }
     PresumedLoc PLoc = S.getASTContext().getSourceManager().getPresumedLoc(
         S.getASTContext()
             .getSourceManager()

clang/test/CodeGenSYCL/free_function_int_header_rtc_mode.cpp

@@ -44,10 +44,9 @@ int main(){
 // CHECK-NEXT:    "{{.*}}__sycl_kernel_free_function_nd_rangePiii",
 // CHECK-NEXT:    "{{.*}}Kernel_Function",
 
-
-// CHECK: static constexpr const char* getName() { return "{{.*}}__sycl_kernel_free_function_singlePiii"; }
-// CHECK: static constexpr const char* getName() { return "{{.*}}__sycl_kernel_free_function_nd_rangePiii"; }
-// CHECK: static constexpr const char* getName() { return "{{.*}}Kernel_Function"; }
+// CHECK: _Z34__sycl_kernel_free_function_singlePiii
+// CHECK: _Z36__sycl_kernel_free_function_nd_rangePiii
+// CHECK: _ZTSZ4mainE15Kernel_Function
 
 // CHECK-RTC-NOT: free_function_single_kernel
 // CHECK-RTC-NOT: free_function_nd_range

devops/actions/run-tests/e2e/action.yml

@@ -60,18 +60,31 @@ runs:
     run: |
       cmake -GNinja -B./build-e2e -S./llvm/sycl/test-e2e -DCMAKE_CXX_COMPILER="${{ inputs.sycl_compiler || '$(which clang++)'}}" -DLLVM_LIT="$PWD/llvm/llvm/utils/lit/lit.py" ${{ steps.cmake_opts.outputs.opts }}
   - name: SYCL End-to-end tests
-    shell: bash {0}
+    id: run_e2e
+    continue-on-error: true
+    shell: bash
     env:
       LIT_OPTS: -v --no-progress-bar --show-unsupported --show-pass --show-xfail --max-time 3600 --time-tests --param print_features=True --param test-mode=${{ inputs.testing_mode }} --param sycl_devices=${{ inputs.target_devices }} ${{ inputs.extra_lit_opts }}
     run: |
       ninja -C build-e2e check-sycl-e2e > e2e.log 2>&1
-      exit_code=$?
+  # Two steps below are duplicated between Lin/Win actions, updates must change both
+  - name: E2E logs
+    if: ${{ always() }}
+    shell: bash
+    run: |
+      echo "::group::Show Full E2E Log"
       cat e2e.log
-      if [ $exit_code -ne 0 ]; then
-        # This is duplicated between lin/win, updates must change both.
+      echo "::endgroup::"
+  - name: Report E2E Failures
+    if: steps.run_e2e.outcome != 'success'
+    shell: bash
+    # For some reason Github uses the first line from the `run: |` section for
+    # the folded entry when displaying instead of this step's name.
+    run: |
+        # Report E2E Failures
         awk '/^Failed Tests|Unexpectedly Passed Tests|Unresolved tests|Timed Out Tests|Testing Time/{flag=1}/FAILED: CMakeFiles/{flag=0}flag' e2e.log >> $GITHUB_STEP_SUMMARY
-      fi
-      exit $exit_code
+        awk '/^Failed Tests|Unexpectedly Passed Tests|Unresolved tests|Timed Out Tests|Testing Time/{flag=1}/FAILED: CMakeFiles/{flag=0}flag' e2e.log
+        exit 1
 
   - name: Pack E2E binaries
     if: ${{ always() && !cancelled() && inputs.binaries_artifact != '' && inputs.testing_mode != 'run-only'}}

devops/actions/run-tests/windows/e2e/action.yml

@@ -72,20 +72,31 @@ runs:
     run: ls build-e2e > e2econf_files.txt
 
   - name: Run End-to-End tests
-    shell: bash {0}
+    id: run_e2e
+    continue-on-error: true
+    shell: bash
     env:
       LIT_OPTS: -v --no-progress-bar --show-unsupported --show-pass --show-xfail --max-time ${{ inputs.e2e_testing_mode == 'run-only' && 1200 || 3600 }} --time-tests --param print_features=True --param test-mode=${{ inputs.testing_mode }} --param sycl_devices=${{ inputs.target_devices }} ${{ inputs.extra_lit_opts }}
     run: |
-      # Run E2E tests.
       cmake --build build-e2e --target check-sycl-e2e > e2e.log 2>&1
-
-      exit_code=$?
+  # Two steps below are duplicated between Lin/Win actions, updates must change both
+  - name: E2E logs
+    if: ${{ always() }}
+    shell: bash
+    run: |
+      echo "::group::Show Full E2E Log"
       cat e2e.log
-      if [ $exit_code -ne 0 ]; then
-        # This is duplicated between lin/win, updates must change both.
+      echo "::endgroup::"
+  - name: Report E2E Failures
+    if: steps.run_e2e.outcome != 'success'
+    shell: bash
+    # For some reason Github uses the first line from the `run: |` section for
+    # the folded entry when displaying instead of this step's name.
+    run: |
+        # Report E2E Failures
         awk '/^Failed Tests|Unexpectedly Passed Tests|Unresolved tests|Timed Out Tests|Testing Time/{flag=1}/FAILED: CMakeFiles/{flag=0}flag' e2e.log >> $GITHUB_STEP_SUMMARY
-      fi
-      exit $exit_code
+        awk '/^Failed Tests|Unexpectedly Passed Tests|Unresolved tests|Timed Out Tests|Testing Time/{flag=1}/FAILED: CMakeFiles/{flag=0}flag' e2e.log
+        exit 1
 
   # Github CI doesn't support containers on Windows, so we cannot guarantee
   # that paths are the same between building and running systems. To avoid

devops/dependencies-igc-dev.json

@@ -1,10 +1,10 @@
 {
   "linux": {
     "igc_dev": {
-      "github_tag": "igc-dev-e22e2a8",
-      "version": "e22e2a8",
-      "updated_at": "2025-07-31T01:11:27Z",
-      "url": "https://api.github.com/repos/intel/intel-graphics-compiler/actions/artifacts/3654137925/zip",
+      "github_tag": "igc-dev-04d2e53",
+      "version": "04d2e53",
+      "updated_at": "2025-08-04T02:10:56Z",
+      "url": "https://api.github.com/repos/intel/intel-graphics-compiler/actions/artifacts/3677914099/zip",
       "root": "{DEPS_ROOT}/opencl/runtime/linux/oclgpu"
     }
   }

devops/scripts/benchmarks/benches/compute.py

@@ -51,7 +51,7 @@ def git_url(self) -> str:
         return "https://github.com/intel/compute-benchmarks.git"
 
     def git_hash(self) -> str:
-        return "83b9ae3ebb3563552409f3a317cdc1cf3d3ca6bd"
+        return "c9e135d4f26dd6badd83009f92f25d6285fc1e21"
 
     def setup(self) -> None:
         if options.sycl is None:
@@ -73,6 +73,8 @@ def setup(self) -> None:
             f"-DBUILD_SYCL=ON",
             f"-DSYCL_COMPILER_ROOT={options.sycl}",
             f"-DALLOW_WARNINGS=ON",
+            f"-DCMAKE_CXX_COMPILER=clang++",
+            f"-DCMAKE_C_COMPILER=clang",
         ]
 
         if options.ur_adapter == "cuda":
@@ -204,6 +206,17 @@ def benchmarks(self) -> list[Benchmark]:
 
             # Add GraphApiSubmitGraph benchmarks
             for in_order_queue in [0, 1]:
+                benches.append(
+                    GraphApiSubmitGraph(
+                        self,
+                        runtime,
+                        in_order_queue,
+                        self.submit_graph_num_kernels[-1],
+                        0,
+                        useEvents=0,
+                        useHostTasks=1,
+                    )
+                )
                 for num_kernels in self.submit_graph_num_kernels:
                     for measure_completion_time in [0, 1]:
                         for use_events in [0, 1]:
@@ -215,6 +228,7 @@ def benchmarks(self) -> list[Benchmark]:
                                     num_kernels,
                                     measure_completion_time,
                                     use_events,
+                                    useHostTasks=0,
                                 )
                             )
 
@@ -840,22 +854,25 @@ def __init__(
         numKernels,
         measureCompletionTime,
         useEvents,
+        useHostTasks,
     ):
         self.inOrderQueue = inOrderQueue
         self.numKernels = numKernels
         self.measureCompletionTime = measureCompletionTime
         self.useEvents = useEvents
+        self.useHostTasks = useHostTasks
         self.ioq_str = "in order" if self.inOrderQueue else "out of order"
         self.measure_str = (
             " with measure completion" if self.measureCompletionTime else ""
         )
         self.use_events_str = f" with events" if self.useEvents else ""
+        self.host_tasks_str = f" use host tasks" if self.useHostTasks else ""
         super().__init__(
             bench, f"graph_api_benchmark_{runtime.value}", "SubmitGraph", runtime
         )
 
     def explicit_group(self):
-        return f"SubmitGraph {self.ioq_str}{self.measure_str}{self.use_events_str}, {self.numKernels} kernels"
+        return f"SubmitGraph {self.ioq_str}{self.measure_str}{self.use_events_str}{self.host_tasks_str}, {self.numKernels} kernels"
 
     def description(self) -> str:
         return (
@@ -864,10 +881,10 @@ def description(self) -> str:
         )
 
     def name(self):
-        return f"graph_api_benchmark_{self.runtime.value} SubmitGraph{self.use_events_str} numKernels:{self.numKernels} ioq {self.inOrderQueue} measureCompletion {self.measureCompletionTime}"
+        return f"graph_api_benchmark_{self.runtime.value} SubmitGraph{self.use_events_str}{self.host_tasks_str} numKernels:{self.numKernels} ioq {self.inOrderQueue} measureCompletion {self.measureCompletionTime}"
 
     def display_name(self) -> str:
-        return f"{self.runtime.value.upper()} SubmitGraph {self.ioq_str}{self.measure_str}{self.use_events_str}, {self.numKernels} kernels"
+        return f"{self.runtime.value.upper()} SubmitGraph {self.ioq_str}{self.measure_str}{self.use_events_str}{self.host_tasks_str}, {self.numKernels} kernels"
 
     def get_tags(self):
         return [
@@ -888,6 +905,7 @@ def bin_args(self) -> list[str]:
             "--KernelExecutionTime=1",
             f"--UseEvents={self.useEvents}",
             "--UseExplicit=0",
+            f"--UseHostTasks={self.useHostTasks}",
         ]