[SYCL] Enable builds on macOS host (#6706)

This patch contains minimal changes to make compiler toolchain builds
succeed on macOS:
- Add stubs for platform-dependent functions
- Disable level zero plugin for macOS exclusively
- Fix platform-dependent library names
- Remove get_device_count_by_type tool (which is unused, and rather than
fixing it, removal is much simpler)
- Add minimal CI in post-commit to prevent build breakages in future

The above changes are far from fully working SYCL on Darwin targets,
this patch only fixes toolchain builds, and does not attempt to make the
toolchain itself functional.

Fixes #258
Fixes #982

.github/workflows/sycl_macos_build_and_test.yml

@@ -0,0 +1,73 @@
+name: Reusable SYCL macOS build and test workflow
+
+on:
+  workflow_call:
+    inputs:
+      build_ref:
+        type: string
+        required: false
+      build_cache_suffix:
+        type: string
+        required: false
+        default: "default"
+      build_cache_size:
+        type: string
+        required: false
+        default: 2G
+      build_configure_extra_args:
+        type: string
+        required: false
+        default: ""
+      build_artifact_suffix:
+        type: string
+        required: false
+        default: "default"
+
+jobs:
+  build:
+    name: Build
+    runs-on: macos-12
+    steps:
+    - name: Install dependencies
+      run: brew install ccache ninja
+    - uses: actions/checkout@v3
+      with:
+        ref: ${{ inputs.build_ref }}
+        path: src
+    - uses: actions/cache@v3
+      with:
+        path: build_cache_${{ inputs.build_cache_suffix }}
+        key: sycl-${{ runner.os }}-${{ inputs.build_cache_suffix }}-${{ github.sha }}
+        restore-keys: sycl-${{ runner.os }}-${{ inputs.build_cache_suffix }}-
+    - name: Configure
+      env:
+        CACHE_SUFFIX: ${{ inputs.build_cache_suffix }}
+        CACHE_SIZE: ${{ inputs.build_cache_size }}
+        ARGS: ${{ inputs.build_configure_extra_args }}
+      run: |
+        mkdir -p $GITHUB_WORKSPACE/build_cache_$CACHE_SUFFIX
+        mkdir -p $GITHUB_WORKSPACE/build
+        cd $GITHUB_WORKSPACE/build
+        python3 $GITHUB_WORKSPACE/src/buildbot/configure.py -w $GITHUB_WORKSPACE \
+          -s $GITHUB_WORKSPACE/src -o $GITHUB_WORKSPACE/build -t Release \
+          --ci-defaults $ARGS \
+          --cmake-opt="-DLLVM_CCACHE_BUILD=ON" \
+          --cmake-opt="-DLLVM_CCACHE_DIR=$GITHUB_WORKSPACE/build_cache_$CACHE_SUFFIX" \
+          --cmake-opt="-DLLVM_CCACHE_MAXSIZE=$CACHE_SIZE" \
+          --cmake-opt="-DLLVM_INSTALL_UTILS=ON" \
+          --cmake-opt="-DSYCL_PI_TESTS=OFF"
+    - name: Compile
+      id: build
+      run: cmake --build $GITHUB_WORKSPACE/build --target sycl-toolchain
+    - name: Install
+      run: |
+        cmake --build $GITHUB_WORKSPACE/build --target deploy-sycl-toolchain
+
+    - name: Pack toolchain
+      run: tar -cJf llvm_sycl.tar.xz -C $GITHUB_WORKSPACE/build/install .
+    - name: Upload toolchain
+      uses: actions/upload-artifact@v2
+      with:
+        name: sycl_macos_${{ inputs.build_artifact_suffix }}
+        path: llvm_sycl.tar.xz
+

.github/workflows/sycl_post_commit.yml

@@ -113,3 +113,8 @@ jobs:
     name: Windows
     if: github.repository == 'intel/llvm'
     uses: ./.github/workflows/sycl_windows_build_and_test.yml
+
+  macos_default:
+    name: macOS
+    if: github.repository == 'intel/llvm'
+    uses: ./.github/workflows/sycl_macos_build_and_test.yml

buildbot/configure.py

@@ -13,7 +13,12 @@ def do_configure(args):
     if not os.path.isdir(abs_obj_dir):
       os.makedirs(abs_obj_dir)
 
-    llvm_external_projects = 'sycl;llvm-spirv;opencl;libdevice;xpti;xptifw'
+    llvm_external_projects = 'sycl;llvm-spirv;opencl;xpti;xptifw'
+
+    # libdevice build requires a working SYCL toolchain, which is not the case
+    # with macOS target right now.
+    if sys.platform != "darwin":
+        llvm_external_projects += ';libdevice'
 
     libclc_amd_target_names = ';amdgcn--;amdgcn--amdhsa'
     libclc_nvidia_target_names = ';nvptx64--;nvptx64--nvidiacl'
@@ -39,11 +44,14 @@ def do_configure(args):
     llvm_enable_sphinx = 'OFF'
     llvm_build_shared_libs = 'OFF'
     llvm_enable_lld = 'OFF'
-    sycl_enabled_plugins = ["opencl", "level_zero"]
+    sycl_enabled_plugins = ["opencl"]
 
     sycl_enable_xpti_tracing = 'ON'
     xpti_enable_werror = 'OFF'
 
+    if sys.platform != "darwin":
+        sycl_enabled_plugins.append("level_zero")
+
     # lld is needed on Windows or for the HIP plugin on AMD
     if platform.system() == 'Windows' or (args.hip and args.hip_platform == 'AMD'):
         llvm_enable_projects += ';lld'
@@ -104,18 +112,19 @@ def do_configure(args):
 
         # For clang-format, clang-tidy and code coverage
         llvm_enable_projects += ";clang-tools-extra;compiler-rt"
-        # libclc is required for CI validation
-        if 'libclc' not in llvm_enable_projects:
-            llvm_enable_projects += ';libclc'
-        # libclc passes `--nvvm-reflect-enable=false`, build NVPTX to enable it
-        if 'NVPTX' not in llvm_targets_to_build:
-            llvm_targets_to_build += ';NVPTX'
-        # Add both NVIDIA and AMD libclc targets
-        if libclc_amd_target_names not in libclc_targets_to_build:
-            libclc_targets_to_build += libclc_amd_target_names
-        if libclc_nvidia_target_names not in libclc_targets_to_build:
-            libclc_targets_to_build += libclc_nvidia_target_names
-        libclc_gen_remangled_variants = 'ON'
+        if sys.platform != "darwin":
+            # libclc is required for CI validation
+            if 'libclc' not in llvm_enable_projects:
+                llvm_enable_projects += ';libclc'
+            # libclc passes `--nvvm-reflect-enable=false`, build NVPTX to enable it
+            if 'NVPTX' not in llvm_targets_to_build:
+                llvm_targets_to_build += ';NVPTX'
+            # Add both NVIDIA and AMD libclc targets
+            if libclc_amd_target_names not in libclc_targets_to_build:
+                libclc_targets_to_build += libclc_amd_target_names
+            if libclc_nvidia_target_names not in libclc_targets_to_build:
+                libclc_targets_to_build += libclc_nvidia_target_names
+            libclc_gen_remangled_variants = 'ON'
 
     if args.enable_plugin:
         sycl_enabled_plugins += args.enable_plugin

sycl/cmake/modules/AddSYCL.cmake

@@ -13,7 +13,7 @@ function(add_sycl_library LIB_NAME TYPE)
     add_dependencies(sycl-toolchain ${LIB_NAME})
   endif()
 
-  if (ARG_LINKER_SCRIPT AND UNIX)
+  if (ARG_LINKER_SCRIPT AND UNIX AND NOT APPLE)
     target_link_libraries(${LIB_NAME} PRIVATE
       "-Wl,--version-script=${ARG_LINKER_SCRIPT}")
   endif()

sycl/include/sycl/detail/pi.hpp

@@ -67,12 +67,20 @@ bool trace(TraceLevel level);
 #define __SYCL_CUDA_PLUGIN_NAME "pi_cuda.dll"
 #define __SYCL_ESIMD_EMULATOR_PLUGIN_NAME "pi_esimd_emulator.dll"
 #define __SYCL_HIP_PLUGIN_NAME "libpi_hip.dll"
-#else
+#elif defined(__SYCL_RT_OS_LINUX)
 #define __SYCL_OPENCL_PLUGIN_NAME "libpi_opencl.so"
 #define __SYCL_LEVEL_ZERO_PLUGIN_NAME "libpi_level_zero.so"
 #define __SYCL_CUDA_PLUGIN_NAME "libpi_cuda.so"
 #define __SYCL_ESIMD_EMULATOR_PLUGIN_NAME "libpi_esimd_emulator.so"
 #define __SYCL_HIP_PLUGIN_NAME "libpi_hip.so"
+#elif defined(__SYCL_RT_OS_DARWIN)
+#define __SYCL_OPENCL_PLUGIN_NAME "libpi_opencl.dylib"
+#define __SYCL_LEVEL_ZERO_PLUGIN_NAME "libpi_level_zero.dylib"
+#define __SYCL_CUDA_PLUGIN_NAME "libpi_cuda.dylib"
+#define __SYCL_ESIMD_EMULATOR_PLUGIN_NAME "libpi_esimd_emulator.dylib"
+#define __SYCL_HIP_PLUGIN_NAME "libpi_hip.dylib"
+#else
+#error "Unsupported OS"
 #endif
 
 // Report error and no return (keeps compiler happy about no return statements).

sycl/source/CMakeLists.txt

@@ -63,10 +63,12 @@ function(add_sycl_rt_library LIB_NAME LIB_OBJ_NAME)
   else()
     target_compile_options(${LIB_OBJ_NAME} PUBLIC
                            -fvisibility=hidden -fvisibility-inlines-hidden)
-    set(linker_script "${CMAKE_CURRENT_SOURCE_DIR}/ld-version-script.txt")
-    target_link_libraries(
-        ${LIB_NAME} PRIVATE "-Wl,--version-script=${linker_script}")
-    set_target_properties(${LIB_NAME} PROPERTIES LINK_DEPENDS ${linker_script})
+    if (NOT APPLE)
+      set(linker_script "${CMAKE_CURRENT_SOURCE_DIR}/ld-version-script.txt")
+      target_link_libraries(
+          ${LIB_NAME} PRIVATE "-Wl,--version-script=${linker_script}")
+      set_target_properties(${LIB_NAME} PROPERTIES LINK_DEPENDS ${linker_script})
+    endif()
     if (SYCL_ENABLE_XPTI_TRACING)
       target_link_libraries(${LIB_NAME} PRIVATE dl)
     endif()

sycl/source/detail/online_compiler/online_compiler.cpp

@@ -143,7 +143,7 @@ compileToSPIRV(const std::string &Source, sycl::info::device_type DeviceType,
 
   uint32_t NumOutputs = 0;
   byte **Outputs = nullptr;
-  size_t *OutputLengths = nullptr;
+  uint64_t *OutputLengths = nullptr;
   char **OutputNames = nullptr;
 
   const byte *Sources[] = {reinterpret_cast<const byte *>(Source.c_str())};

sycl/source/detail/os_util.cpp

@@ -233,6 +233,8 @@ OSModuleHandle OSUtil::getOSModuleHandle(const void *VirtAddr) {
   return reinterpret_cast<OSModuleHandle>(Res.dli_fbase);
 }
 
+std::string OSUtil::getCurrentDSODir() { return ""; }
+
 #endif // __SYCL_RT_OS
 
 size_t OSUtil::getOSMemSize() {
@@ -288,7 +290,7 @@ int OSUtil::makeDir(const char *Dir) {
   do {
     pos = Path.find_first_of("/\\", ++pos);
     CurPath = Path.substr(0, pos);
-#if defined(__SYCL_RT_OS_LINUX)
+#if defined(__SYCL_RT_OS_POSIX_SUPPORT)
     auto Res = mkdir(CurPath.c_str(), 0777);
 #else
     auto Res = _mkdir(CurPath.c_str());

sycl/source/detail/persistent_device_code_cache.cpp

@@ -14,7 +14,7 @@
 #include <cstdio>
 #include <optional>
 
-#if defined(__SYCL_RT_OS_LINUX)
+#if defined(__SYCL_RT_OS_POSIX_SUPPORT)
 #include <unistd.h>
 #else
 #include <direct.h>

sycl/source/detail/platform_util.cpp

@@ -18,6 +18,8 @@
 #endif
 #elif defined(__SYCL_RT_OS_WINDOWS)
 #include <intrin.h>
+#elif defined(__SYCL_RT_OS_DARWIN)
+#include <cpuid.h>
 #endif
 
 namespace sycl {
@@ -27,7 +29,7 @@ namespace detail {
 #if defined(__x86_64__) || defined(__i386__)
 // Used by methods that duplicate OpenCL behaviour in order to get CPU info
 static void cpuid(uint32_t *CPUInfo, uint32_t Type, uint32_t SubType = 0) {
-#if defined(__SYCL_RT_OS_LINUX)
+#if defined(__SYCL_RT_OS_LINUX) || defined(__SYCL_RT_OS_DARWIN)
   __cpuid_count(Type, SubType, CPUInfo[0], CPUInfo[1], CPUInfo[2], CPUInfo[3]);
 #elif defined(__SYCL_RT_OS_WINDOWS)
   __cpuidex(reinterpret_cast<int *>(CPUInfo), Type, SubType);
@@ -115,7 +117,7 @@ uint32_t PlatformUtil::getNativeVectorWidth(PlatformUtil::TypeIndex TIndex) {
   // AVX512 has 64 byte (ZMM) registers
   static constexpr uint32_t VECTOR_WIDTH_AVX512[] = {64, 32, 16, 8, 16, 8, 0};
 
-#if defined(__SYCL_RT_OS_LINUX)
+#if defined(__SYCL_RT_OS_LINUX) || defined(__SYCL_RT_OS_DARWIN)
   if (__builtin_cpu_supports("avx512f"))
     return VECTOR_WIDTH_AVX512[Index];
   if (__builtin_cpu_supports("avx2"))

sycl/test/CMakeLists.txt

@@ -31,7 +31,6 @@ list(APPEND SYCL_TEST_DEPS
   sycl-toolchain
   FileCheck
   not
-  get_device_count_by_type
   llvm-config
   llvm-cxxdump
   llvm-dis

sycl/tools/CMakeLists.txt

@@ -15,51 +15,3 @@ if (SYCL_ENABLE_XPTI_TRACING)
   endif()
 endif()
 
-# TODO: move each tool in its own sub-directory
-add_executable(get_device_count_by_type get_device_count_by_type.cpp)
-add_dependencies(get_device_count_by_type
-  level-zero-loader
-)
-
-if(MSVC)
-  set(LEVEL_ZERO_LIBRARY
-    "${LLVM_LIBRARY_OUTPUT_INTDIR}/${CMAKE_STATIC_LIBRARY_PREFIX}ze_loader${CMAKE_STATIC_LIBRARY_SUFFIX}")
-else()
-  set(LEVEL_ZERO_LIBRARY
-    "${LLVM_LIBRARY_OUTPUT_INTDIR}/${CMAKE_SHARED_LIBRARY_PREFIX}ze_loader${CMAKE_SHARED_LIBRARY_SUFFIX}")
-endif()
-
-if ("cuda" IN_LIST SYCL_ENABLE_PLUGINS)
-  set(SYCL_BUILD_PI_CUDA ON)
-endif()
-if ("hip" IN_LIST SYCL_ENABLE_PLUGINS)
-  set(SYCL_BUILD_PI_HIP ON)
-endif()
-
-target_link_libraries(get_device_count_by_type
-  PRIVATE
-    OpenCL-Headers
-    LevelZeroLoader::Headers
-    OpenCL-ICD
-    ${LEVEL_ZERO_LIBRARY}
-    # The CUDA and HIP for NVIDA plugins need cudadrv
-    $<$<OR:$<BOOL:${SYCL_BUILD_PI_CUDA}>,$<AND:$<BOOL:${SYCL_BUILD_PI_HIP}>,$<STREQUAL:${SYCL_BUILD_PI_HIP_PLATFORM},NVIDIA>>>:cudadrv>
-    # The HIP for AMD plugin needs rocmdrv
-    $<$<AND:$<BOOL:${SYCL_BUILD_PI_HIP}>,$<STREQUAL:${SYCL_BUILD_PI_HIP_PLATFORM},AMD>>:rocmdrv>
-    # The HIP for NVIDIA plugin also needs cudart
-    $<$<AND:$<BOOL:${SYCL_BUILD_PI_HIP}>,$<STREQUAL:${SYCL_BUILD_PI_HIP_PLATFORM},NVIDIA>>:cudart>
-)
-target_compile_definitions(get_device_count_by_type
-  PRIVATE
-    $<$<BOOL:${SYCL_BUILD_PI_CUDA}>:USE_PI_CUDA>
-    $<$<BOOL:${SYCL_BUILD_PI_HIP}>:USE_PI_HIP>
-    # For HIP set defines depending on the platform
-    $<$<AND:$<BOOL:${SYCL_BUILD_PI_HIP}>,$<STREQUAL:${SYCL_BUILD_PI_HIP_PLATFORM},AMD>>:__HIP_PLATFORM_AMD__>
-    $<$<AND:$<BOOL:${SYCL_BUILD_PI_HIP}>,$<STREQUAL:${SYCL_BUILD_PI_HIP_PLATFORM},NVIDIA>>:__HIP_PLATFORM_NVIDIA__>
-)
-
-if(SYCL_BUILD_PI_HIP)
-  target_include_directories(get_device_count_by_type
-    PRIVATE
-    ${SYCL_BUILD_PI_HIP_INCLUDE_DIR})
-endif()

sycl/tools/sycl-trace/CMakeLists.txt

@@ -10,11 +10,23 @@ link_llvm_libs(sycl-trace
   LLVMSupport
 )
 
+if ("level_zero" IN_LIST SYCL_ENABLE_PLUGINS)
+  set(EXTRA_SRC
+    ze_trace_collector.cpp
+  )
+endif()
+
+if ("cuda" IN_LIST SYCL_ENABLE_PLUGINS)
+  set(EXTRA_SRC
+    ${EXTRA_SRC}
+    cuda_trace_collector.cpp
+  )
+endif()
+
 add_library(sycl_pi_trace_collector SHARED
   collector.cpp
   pi_trace_collector.cpp
-  ze_trace_collector.cpp
-  $<$<BOOL:${SYCL_BUILD_PI_CUDA}>:cuda_trace_collector.cpp>
+  ${EXTRA_SRC}
 )
 
 find_package(Python3 REQUIRED)
@@ -30,18 +42,23 @@ add_custom_target(pi-pretty-printers
   )
 
 # To get L0 loader
-add_dependencies(sycl_pi_trace_collector pi_level_zero)
+if ("level_zero" IN_LIST SYCL_ENABLE_PLUGINS)
+  add_dependencies(sycl_pi_trace_collector pi_level_zero)
 
-target_link_libraries(sycl_pi_trace_collector PRIVATE LevelZeroLoader::Headers)
+  target_link_libraries(sycl_pi_trace_collector PRIVATE LevelZeroLoader::Headers)
+  target_compile_definitions(sycl_pi_trace_collector PRIVATE SYCL_HAS_LEVEL_ZERO)
 
-add_custom_target(ze-pretty-printers
-  COMMAND ${Python3_EXECUTABLE}
-  ${CMAKE_CURRENT_SOURCE_DIR}/generate_ze_pretty_printers.py
-  ${SYCL_INCLUDE_BUILD_DIR}/sycl/level_zero/ze_api.h
-  DEPENDS pi_level_zero
-  BYPRODUCTS
-    ${CMAKE_CURRENT_BINARY_DIR}/ze_printers.def
-  )
+  add_custom_target(ze-pretty-printers
+    COMMAND ${Python3_EXECUTABLE}
+    ${CMAKE_CURRENT_SOURCE_DIR}/generate_ze_pretty_printers.py
+    ${SYCL_INCLUDE_BUILD_DIR}/sycl/level_zero/ze_api.h
+    DEPENDS pi_level_zero
+    BYPRODUCTS
+      ${CMAKE_CURRENT_BINARY_DIR}/ze_printers.def
+    )
+
+  add_dependencies(sycl_pi_trace_collector ze-pretty-printers)
+endif()
 
 target_compile_definitions(sycl_pi_trace_collector PRIVATE XPTI_CALLBACK_API_EXPORTS)
 target_link_libraries(sycl_pi_trace_collector PRIVATE xptifw)
@@ -56,7 +73,7 @@ target_include_directories(sycl_pi_trace_collector PRIVATE
     "${CMAKE_CURRENT_BINARY_DIR}"
 )
 
-add_dependencies(sycl_pi_trace_collector pi-pretty-printers ze-pretty-printers)
+add_dependencies(sycl_pi_trace_collector pi-pretty-printers)
 
 if(SYCL_BUILD_PI_CUDA)