Merge branch 'sycl' into Alexandr-Konovalov/UR-counters-shared_ptr

Author: Alexandr-Konovalov

.github/CODEOWNERS

@@ -177,6 +177,7 @@ sycl/doc/syclgraph/ @intel/sycl-graphs-reviewers
 sycl/**/syclcompat/ @intel/syclcompat-lib-reviewers
 sycl/cmake/modules/AddSYCLLibraryUnitTest.cmake @intel/syclcompat-lib-reviewers
 sycl/include/syclcompat.hpp @intel/syclcompat-lib-reviewers
+sycl/test-e2e/syclcompat/ @intel/syclcompat-lib-reviewers
 
 # bindless images
 sycl/doc/extensions/experimental/sycl_ext_oneapi_bindless_images.asciidoc @intel/bindless-images-reviewers

.github/workflows/sycl-containers-igc-dev.yaml

@@ -7,12 +7,14 @@ on:
     paths:
       - 'devops/actions/build_container/**'
       - 'devops/scripts/**'
+      - '!devops/scripts/benchmarks/**'
       - 'devops/dependencies-igc-dev.json'
       - '.github/workflows/sycl-containers-igc-dev.yaml'
   pull_request:
     paths:
       - 'devops/actions/build_container/**'
       - 'devops/scripts/**'
+      - '!devops/scripts/benchmarks/**'
       - 'devops/dependencies-igc-dev.json'
       - '.github/workflows/sycl-containers-igc-dev.yaml'
 

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

@@ -32,6 +32,10 @@ on:
       build_artifact_suffix:
         type: string
         required: true
+      build_target:
+        type: string
+        required: false
+        default: sycl-toolchain
       artifact_archive_name:
         type: string
         default: llvm_sycl.tar.zst
@@ -177,7 +181,8 @@ jobs:
           -DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=SPIRV
     - name: Compile
       id: build
-      run: cmake --build $GITHUB_WORKSPACE/build --target sycl-toolchain
+      # Emulate default value for manual dispatch as we've run out of available arguments.
+      run: cmake --build $GITHUB_WORKSPACE/build --target ${{ inputs.build_target || 'sycl-toolchain' }}
     - name: check-llvm
       if: always() && !cancelled() && contains(inputs.changes, 'llvm')
       run: |

.github/workflows/sycl-nightly.yml

@@ -33,6 +33,7 @@ jobs:
       build_cache_suffix: sprod_shared
       build_artifact_suffix: sprod_shared
       build_configure_extra_args: '--shared-libs --hip --cuda --native_cpu --no-assertions'
+      build_target: all
 
       artifact_archive_name: sycl_linux_shared.tar.zst
 
@@ -151,6 +152,7 @@ jobs:
       artifact_archive_name: sycl_windows.tar.gz
       # Disable the spirv-dis requirement as to not require SPIR-V Tools.
       build_configure_extra_args: -DLLVM_SPIRV_ENABLE_LIBSPIRV_DIS=off
+      build_target: all
 
   e2e-win:
     needs: build-win

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

@@ -10,6 +10,10 @@ on:
       build_configure_extra_args:
         type: string
         required: false
+      build_target:
+        type: string
+        required: false
+        default: sycl-toolchain
       changes:
         type: string
         description: 'Filter matches for the changed files in the PR'
@@ -58,6 +62,11 @@ on:
       build_configure_extra_args:
         type: string
         required: false
+      build_target:
+        type: choice
+        options:
+          - "sycl-toolchain"
+          - "all"
       artifact_archive_name:
         type: choice
         options:
@@ -143,7 +152,7 @@ jobs:
       id: build
       shell: bash
       run: |
-        cmake --build build --target sycl-toolchain
+        cmake --build build --target ${{ inputs.build_target }}
     - name: check-llvm
       if: always() && !cancelled() && contains(inputs.changes, 'llvm')
       shell: bash

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

@@ -164,7 +164,8 @@ jobs:
     - name: Test adapter specific
       env:
         ZE_ENABLE_LOADER_DEBUG_TRACE: 1
-      run: ctest -C ${{matrix.build_type}} --test-dir build --output-on-failure -L "adapter-specific" -E "memcheck" --timeout 600 -VV
+        LIT_OPTS: "--timeout 120 --filter-out 'adapters/level_zero/memcheck.test'"
+      run: cmake --build build -j $(nproc) -- check-unified-runtime-adapter
       # Don't run adapter specific tests when building multiple adapters
       if: ${{ matrix.adapter.other_name == '' }}
 

devops/scripts/benchmarks/benches/compute.py

@@ -149,15 +149,17 @@ def benchmarks(self) -> list[Benchmark]:
             for in_order_queue in [0, 1]:
                 for measure_completion in [0, 1]:
                     for use_events in [0, 1]:
-                        benches.append(
-                            SubmitKernel(
-                                self,
-                                runtime,
-                                in_order_queue,
-                                measure_completion,
-                                use_events,
+                        for kernel_exec_time in [1, 20]:
+                            benches.append(
+                                SubmitKernel(
+                                    self,
+                                    runtime,
+                                    in_order_queue,
+                                    measure_completion,
+                                    use_events,
+                                    kernel_exec_time,
+                                )
                             )
-                        )
 
         # Add SinKernelGraph benchmarks
         for runtime in self.enabled_runtimes():
@@ -332,11 +334,20 @@ def teardown(self):
 
 
 class SubmitKernel(ComputeBenchmark):
-    def __init__(self, bench, runtime: RUNTIMES, ioq, MeasureCompletion=0, UseEvents=0):
+    def __init__(
+        self,
+        bench,
+        runtime: RUNTIMES,
+        ioq,
+        MeasureCompletion=0,
+        UseEvents=0,
+        KernelExecTime=1,
+    ):
         self.ioq = ioq
         self.runtime = runtime
         self.MeasureCompletion = MeasureCompletion
         self.UseEvents = UseEvents
+        self.KernelExecTime = KernelExecTime
         self.NumKernels = 10
         super().__init__(
             bench, f"api_overhead_benchmark_{runtime.value}", "SubmitKernel"
@@ -353,7 +364,11 @@ def name(self):
         # to match the existing already stored results
         events_str = " not using events" if not self.UseEvents else ""
 
-        return f"api_overhead_benchmark_{self.runtime.value} SubmitKernel {order}{completion_str}{events_str}"
+        kernel_exec_time_str = (
+            f" KernelExecTime={self.KernelExecTime}" if self.KernelExecTime != 1 else ""
+        )
+
+        return f"api_overhead_benchmark_{self.runtime.value} SubmitKernel {order}{completion_str}{events_str}{kernel_exec_time_str}"
 
     def display_name(self) -> str:
         order = "in order" if self.ioq else "out of order"
@@ -362,6 +377,8 @@ def display_name(self) -> str:
             info.append("with measure completion")
         if self.UseEvents:
             info.append("using events")
+        if self.KernelExecTime != 1:
+            info.append(f"KernelExecTime={self.KernelExecTime}")
         additional_info = f" {' '.join(info)}" if info else ""
         return f"{self.runtime.value.upper()} SubmitKernel {order}{additional_info}, NumKernels {self.NumKernels}"
 
@@ -373,7 +390,11 @@ def explicit_group(self):
         # to match the existing already stored results
         events_str = " not using events" if not self.UseEvents else ""
 
-        return f"SubmitKernel {order}{completion_str}{events_str}"
+        kernel_exec_time_str = (
+            f" KernelExecTime={self.KernelExecTime}" if self.KernelExecTime != 1 else ""
+        )
+
+        return f"SubmitKernel {order}{completion_str}{events_str}{kernel_exec_time_str}"
 
     def description(self) -> str:
         order = "in-order" if self.ioq else "out-of-order"
@@ -386,6 +407,7 @@ def description(self) -> str:
         return (
             f"Measures CPU time overhead of submitting {order} kernels through {runtime_name} API{completion_desc}. "
             f"Runs {self.NumKernels} simple kernels with minimal execution time to isolate API overhead from kernel execution time."
+            f"Each kernel executes for approximately {self.KernelExecTime} micro seconds."
         )
 
     def range(self) -> tuple[float, float]:
@@ -398,7 +420,7 @@ def bin_args(self) -> list[str]:
             "--iterations=100000",
             "--Profiling=0",
             f"--NumKernels={self.NumKernels}",
-            "--KernelExecTime=1",
+            f"--KernelExecTime={self.KernelExecTime}",
             f"--UseEvents={self.UseEvents}",
         ]
 

libclc/CMakeLists.txt

@@ -586,6 +586,7 @@ foreach( t ${LIBCLC_TARGETS_TO_BUILD} )
         ARCH_SUFFIX libspirv-${arch_suffix}
         TRIPLE ${clang_triple}
         TARGET_ENV libspirv-
+        REMANGLE ${LIBCLC_GENERATE_REMANGLED_VARIANTS}
         COMPILE_FLAGS ${spirv_build_flags}
         OPT_FLAGS ${opt_flags}
         LIB_FILES ${libspirv_lib_files}

libclc/cmake/modules/AddLibclc.cmake

@@ -309,6 +309,8 @@ endfunction()
 #      Optimization options (for opt)
 #  * TARGET_ENV <string>
 #      Prefix to give the final builtin library aliases
+#  * REMANGLE <string>
+#      Bool string indicating whether remangler will be run
 #  * ALIASES <string> ...
 #      List of aliases
 #  * INTERNAL_LINK_DEPENDENCIES <target> ...
@@ -318,7 +320,7 @@ endfunction()
 function(add_libclc_builtin_set)
   cmake_parse_arguments(ARG
     "CLC_INTERNAL"
-    "ARCH;TRIPLE;ARCH_SUFFIX;TARGET_ENV;PARENT_TARGET"
+    "ARCH;TRIPLE;ARCH_SUFFIX;TARGET_ENV;REMANGLE;PARENT_TARGET"
     "LIB_FILES;GEN_FILES;COMPILE_FLAGS;OPT_FLAGS;ALIASES;INTERNAL_LINK_DEPENDENCIES"
     ${ARGN}
   )
@@ -483,7 +485,7 @@ function(add_libclc_builtin_set)
   install( FILES ${builtins_lib} DESTINATION ${CMAKE_INSTALL_DATADIR}/clc )
 
   # Generate remangled variants if requested
-  if( LIBCLC_GENERATE_REMANGLED_VARIANTS )
+  if( ARG_REMANGLE )
     set( dummy_in ${LIBCLC_LIBRARY_OUTPUT_INTDIR}/libclc_dummy_in.cc )
     add_custom_command( OUTPUT ${dummy_in}
       COMMAND ${CMAKE_COMMAND} -E make_directory ${LIBCLC_LIBRARY_OUTPUT_INTDIR}