Merge branch 'main' into yc/1218-libcxx-mess

Author: yingcong-wu

.github/docker/install_dpcpp.sh

@@ -16,5 +16,5 @@ if [ "${SKIP_DPCPP_BUILD}" ]; then
 fi
 
 mkdir -p ${DPCPP_PATH}/dpcpp_compiler
-wget -O ${DPCPP_PATH}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-09-27/sycl_linux.tar.gz
+wget -O ${DPCPP_PATH}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-12-12/sycl_linux.tar.gz
 tar -xvf ${DPCPP_PATH}/dpcpp_compiler.tar.gz -C ${DPCPP_PATH}/dpcpp_compiler

.github/workflows/benchmarks-reusable.yml

@@ -156,14 +156,49 @@ jobs:
     - name: Install UR
       run: cmake --install ${{github.workspace}}/ur_build
 
+    - name: Checkout UMF
+      uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11 # v4.1.1
+      with:
+        repository: oneapi-src/unified-memory-framework
+        ref: main
+        path: umf-repo
+        fetch-depth: 1
+        fetch-tags: false
+
+    - name: Configure UMF
+      run: >
+        cmake -DCMAKE_BUILD_TYPE=Release
+        -S${{github.workspace}}/umf-repo
+        -B${{github.workspace}}/umf_build
+        -DUMF_BUILD_BENCHMARKS=ON
+        -DUMF_TESTS_FAIL_ON_SKIP=ON
+
+    - name: Build UMF
+      run: cmake --build ${{github.workspace}}/umf_build -j $(nproc)
+
+    - name: Compute core range
+      run: |
+        # Compute the core range for the first NUMA node, skipping the first 4 cores.
+        # This is to avoid the first cores that the kernel is likely to schedule more work on.
+        CORES=$(lscpu | awk '
+          /NUMA node0 CPU|On-line CPU/ {line=$0}
+          END {
+            split(line, a, " ")
+            split(a[4], b, ",")
+            sub(/^0/, "4", b[1])
+            print b[1]
+          }')
+        echo "CORES=$CORES" >> $GITHUB_ENV
+
     - name: Run benchmarks
       working-directory: ${{ github.workspace }}/ur-repo/
       id: benchmarks
       run: >
-        numactl -N 0 ${{ github.workspace }}/ur-repo/scripts/benchmarks/main.py
+        taskset -c ${{ env.CORES }} ${{ github.workspace }}/ur-repo/scripts/benchmarks/main.py
         ~/bench_workdir
         --sycl ${{ github.workspace }}/sycl_build
         --ur ${{ github.workspace }}/ur_install
+        --umf ${{ github.workspace }}/umf_build
         --adapter ${{ matrix.adapter.str_name }}
         ${{ inputs.upload_report && '--output-html' || '' }}
         ${{ inputs.bench_script_params }}

.github/workflows/build-fuzz-reusable.yml

@@ -35,7 +35,7 @@ jobs:
 
     - name: Download DPC++
       run: |
-        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-01-29/sycl_linux.tar.gz
+        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-12-12/sycl_linux.tar.gz
         mkdir dpcpp_compiler
         tar -xvf ${{github.workspace}}/dpcpp_compiler.tar.gz -C dpcpp_compiler
 

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

@@ -77,7 +77,7 @@ jobs:
 
     - name: Download DPC++
       run: |
-        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-01-29/sycl_linux.tar.gz
+        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-12-12/sycl_linux.tar.gz
         mkdir dpcpp_compiler
         tar -xvf ${{github.workspace}}/dpcpp_compiler.tar.gz -C dpcpp_compiler
 

.github/workflows/cmake.yml

@@ -76,7 +76,7 @@ jobs:
       if: matrix.os == 'ubuntu-22.04'
       run: |
         sudo apt install libncurses5
-        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-09-27/sycl_linux.tar.gz
+        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-12-12/sycl_linux.tar.gz
         mkdir -p ${{github.workspace}}/dpcpp_compiler
         tar -xvf ${{github.workspace}}/dpcpp_compiler.tar.gz -C ${{github.workspace}}/dpcpp_compiler
 

.github/workflows/multi_device.yml

@@ -33,7 +33,7 @@ jobs:
 
     - name: Download DPC++
       run: |
-        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-01-29/sycl_linux.tar.gz
+        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-12-12/sycl_linux.tar.gz
         mkdir dpcpp_compiler
         tar -xvf ${{github.workspace}}/dpcpp_compiler.tar.gz -C dpcpp_compiler
 
@@ -58,8 +58,7 @@ jobs:
 
     - name: Test adapter specific
       working-directory: ${{github.workspace}}/build
-      run: ctest -C ${{matrix.build_type}} --output-on-failure -L "adapter-specific" -E "test-adapter-level_zero_multi_queue" --timeout 180
-      # TODO: investigate why test-adapter-level_zero_multi_queue fails on newer driver
+      run: ctest -C ${{matrix.build_type}} --output-on-failure -L "adapter-specific" --timeout 180
 
     - name: Test adapters
       working-directory: ${{github.workspace}}/build

CMakeLists.txt

@@ -4,7 +4,7 @@
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
 cmake_minimum_required(VERSION 3.20.0 FATAL_ERROR)
-project(unified-runtime VERSION 0.11.0)
+project(unified-runtime VERSION 0.12.0)
 
 # Check if unified runtime is built as a standalone project.
 if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR OR UR_STANDALONE_BUILD)
@@ -27,7 +27,7 @@ option(UR_USE_ASAN "enable AddressSanitizer" OFF)
 option(UR_USE_UBSAN "enable UndefinedBehaviorSanitizer" OFF)
 option(UR_USE_MSAN "enable MemorySanitizer" OFF)
 option(UR_USE_TSAN "enable ThreadSanitizer" OFF)
-option(UR_USE_CFI "enable Control Flow Integrity checks (requires clang and implies -flto)" ON)
+option(UR_USE_CFI "enable Control Flow Integrity checks (requires clang and implies -flto)" OFF)
 option(UR_ENABLE_TRACING "enable api tracing through xpti" OFF)
 option(UR_ENABLE_SANITIZER "enable device sanitizer" ON)
 option(UR_ENABLE_SYMBOLIZER "enable symoblizer for sanitizer" OFF)

README.md

@@ -130,7 +130,7 @@ List of options provided by CMake:
 | UR_USE_TSAN | Enable ThreadSanitizer | ON/OFF | OFF |
 | UR_USE_UBSAN | Enable UndefinedBehavior Sanitizer | ON/OFF | OFF |
 | UR_USE_MSAN | Enable MemorySanitizer (clang only) | ON/OFF | OFF |
-| UR_USE_CFI | Enable Control Flow Integrity checks (clang only, also enables lto) | ON/OFF | ON |
+| UR_USE_CFI | Enable Control Flow Integrity checks (clang only, also enables lto) | ON/OFF | OFF |
 | UR_ENABLE_TRACING | Enable XPTI-based tracing layer | ON/OFF | OFF |
 | UR_ENABLE_SANITIZER | Enable device sanitizer layer | ON/OFF | ON |
 | UR_CONFORMANCE_TARGET_TRIPLES | SYCL triples to build CTS device binaries for | Comma-separated list | spir64 |

cmake/FetchLevelZero.cmake

@@ -104,14 +104,14 @@ if (UR_COMPUTE_RUNTIME_TAG STREQUAL "")
 set(UR_COMPUTE_RUNTIME_TAG 24.39.31294.12)
 endif()
 include(FetchContent)
-# Sparse fetch only the dir with level zero headers to avoid pulling in the entire compute-runtime.
-FetchContentSparse_Declare(compute-runtime-level-zero-headers ${UR_COMPUTE_RUNTIME_REPO} "${UR_COMPUTE_RUNTIME_TAG}" "level_zero/include")
-FetchContent_GetProperties(compute-runtime-level-zero-headers)
-if(NOT compute-runtime-level-zero-headers_POPULATED)
-  FetchContent_Populate(compute-runtime-level-zero-headers)
+# Sparse fetch only the dir with level zero headers for experimental features to avoid pulling in the entire compute-runtime.
+FetchContentSparse_Declare(exp-headers ${UR_COMPUTE_RUNTIME_REPO} "${UR_COMPUTE_RUNTIME_TAG}" "level_zero/include")
+FetchContent_GetProperties(exp-headers)
+if(NOT exp-headers_POPULATED)
+  FetchContent_Populate(exp-headers)
 endif()
 add_library(ComputeRuntimeLevelZero-Headers INTERFACE)
-set(COMPUTE_RUNTIME_LEVEL_ZERO_INCLUDE "${compute-runtime-level-zero-headers_SOURCE_DIR}/../..")
+set(COMPUTE_RUNTIME_LEVEL_ZERO_INCLUDE "${exp-headers_SOURCE_DIR}/../..")
 message(STATUS "Level Zero Adapter: Using Level Zero headers from ${COMPUTE_RUNTIME_LEVEL_ZERO_INCLUDE}")
 target_include_directories(ComputeRuntimeLevelZero-Headers
     INTERFACE "$<BUILD_INTERFACE:${COMPUTE_RUNTIME_LEVEL_ZERO_INCLUDE}>"

cmake/FindRocmAgentEnumerator.cmake

@@ -9,7 +9,7 @@
 #                                  rocm_agent_enumerator is found.
 #
 
-find_program(ROCM_AGENT_ENUMERATOR NAMES rocm_agent_enumerator)
+find_program(ROCM_AGENT_ENUMERATOR NAMES rocm_agent_enumerator PATHS /opt/rocm/bin)
 
 if(ROCM_AGENT_ENUMERATOR)
     set(ROCM_AGENT_ENUMERATOR_FOUND TRUE)