Enable MacOS CI (#196)

.github/workflows/mac-cpu-arm64-build.yml

@@ -37,9 +37,9 @@ jobs:
         run: |
       - name: Build with CMake
         run: |
-          cmake -G "Ninja" -B build -S . -DCMAKE_BUILD_TYPE=Release -DUSE_CUDA=OFF
-          cmake --build build --config Release --parallel
-        continue-on-error: true
+          cmake --preset macos_cpu_release
+          cmake --build --preset macos_cpu_release --parallel
+        continue-on-error: false
 
       - name: Verify Build Artifacts
         if: always()

CMakeLists.txt

@@ -31,18 +31,7 @@ if("${CMAKE_C_COMPILER_ID}" STREQUAL "GNU" AND CMAKE_C_COMPILER_VERSION VERSION_
   message(FATAL_ERROR "GCC version must be greater than or equal to 8")
 endif()
 
-if("${CMAKE_C_COMPILER_ID}" STREQUAL "GNU" AND CMAKE_C_COMPILER_VERSION VERSION_LESS 10)
-  add_compile_definitions(USE_CXX17=1)
-  message("USE C++17")
-  set(CMAKE_CXX_STANDARD 17)
-elseif(USE_CUDA AND CMAKE_CUDA_COMPILER AND CMAKE_CUDA_COMPILER_VERSION VERSION_LESS 12)
-  add_compile_definitions(USE_CXX17=1)
-  message("USE C++17 Because of CUDA Version is less than 12")
-  set(CMAKE_CXX_STANDARD 17)
-else()
-  message("USE C++20")
-  set(CMAKE_CXX_STANDARD 20)
-endif()
+include(cmake/cxx_standard.cmake)
 
 set(GENERATORS_ROOT ${PROJECT_SOURCE_DIR}/src)
 set(MODELS_ROOT ${PROJECT_SOURCE_DIR}/src/models)
@@ -139,11 +128,6 @@ else()
   target_include_directories(onnxruntime-genai-static PUBLIC ${TOKENIZER_ROOT})
   target_link_libraries(onnxruntime-genai PRIVATE tokenizer)
   target_link_libraries(onnxruntime-genai-static PUBLIC tokenizer)
-  # if(NOT WIN32 AND USE_CXX17)
-  #   target_link_libraries(onnxruntime-genai PRIVATE stdc++fs)
-  #   target_link_libraries(onnxruntime-genai-static PRIVATE stdc++fs)
-  #   message("Linking filesystem")
-  # endif()
 endif()
 
 if(ENABLE_TESTS)
@@ -174,7 +158,7 @@ if(USE_CUDA AND CMAKE_CUDA_COMPILER)
   set_target_properties(onnxruntime-genai PROPERTIES LINKER_LANGUAGE CUDA)
   target_link_libraries(onnxruntime-genai PRIVATE cublasLt cublas curand cufft cudart)
   #  onnxruntime-genai-static is statically linked under Windows
-  if(NOT WIN32)
+  if(CMAKE_SYSTEM_NAME STREQUAL "Linux")
     set_target_properties(onnxruntime-genai-static PROPERTIES LINKER_LANGUAGE CUDA)
     target_link_libraries(onnxruntime-genai-static PRIVATE cublasLt cublas curand cufft cudart)
   endif()

CMakePresets.json

@@ -7,7 +7,9 @@
   },
   "include": [
     "cmake/presets/CMakeWinBuildPresets.json",
-    "cmake/presets/CMakeLinuxBuildPresets.json"
+    "cmake/presets/CMakeLinuxBuildPresets.json",
+    "cmake/presets/CMakeMacOSBuildPresets.json"
+
   ]
 
 }

cmake/cxx_standard.cmake

@@ -0,0 +1,12 @@
+if ("${CMAKE_C_COMPILER_ID}" STREQUAL "GNU" AND CMAKE_C_COMPILER_VERSION VERSION_LESS 10)
+    add_compile_definitions(USE_CXX17=1)
+    message("Test is using C++17 because GCC Version is less than 10")
+    set(CMAKE_CXX_STANDARD 17)
+elseif (USE_CUDA AND CMAKE_CUDA_COMPILER AND CMAKE_CUDA_COMPILER_VERSION VERSION_LESS 12)
+    add_compile_definitions(USE_CXX17=1)
+    message("Test is using C++17 Because CUDA Version is less than 12")
+    set(CMAKE_CXX_STANDARD 17)
+else ()
+    message("Test is using C++20")
+    set(CMAKE_CXX_STANDARD 20)
+endif ()

cmake/presets/CMakeMacOSBuildPresets.json

@@ -0,0 +1,24 @@
+{
+  "version": 6,
+  "include": [
+    "CMakeMacOSConfigPresets.json"
+  ],
+  "buildPresets": [
+    {
+      "name": "macos_cpu_release",
+      "configurePreset": "macos_cpu_release"
+    },
+    {
+      "name": "macos_cpu_debug",
+      "configurePreset": "macos_cpu_debug"
+    },
+    {
+      "name": "macos_cpu_relwithdebinfo",
+      "configurePreset": "macos_cpu_relwithdebinfo"
+    },
+    {
+      "name": "macos_cpu_minsizerel",
+      "configurePreset": "macos_cpu_minsizerel"
+    }
+  ]
+}

cmake/presets/CMakeMacOSConfigPresets.json

@@ -0,0 +1,91 @@
+{
+  "version": 6,
+  "include": [
+    "CMakeLinuxDefaultConfigPresets.json"
+  ],
+  "configurePresets": [
+        {
+      "name": "macos_default",
+      "generator": "Ninja",
+      "binaryDir": "${sourceDir}/build/cpu",
+      "cacheVariables": {
+        "CMAKE_POSITION_INDEPENDENT_CODE": "ON",
+        "CMAKE_OSX_ARCHITECTURES": "arm64",
+        "USE_CUDA": "OFF"
+      },
+      "environment": {
+        "CC": "clang",
+        "CXX": "clang++"
+      },
+      "condition": {
+        "type": "equals",
+        "lhs": "${hostSystemName}",
+        "rhs": "Darwin"
+      }
+    },
+    {
+      "name": "macos_release_default",
+      "cacheVariables": {
+        "CMAKE_BUILD_TYPE": "Release",
+        "CMAKE_C_FLAGS": " -O3 -pipe",
+        "CMAKE_CXX_FLAGS": " -O3 -pipe"
+      }
+    },
+    {
+      "name": "macos_relwithdebinfo_default",
+      "cacheVariables": {
+        "CMAKE_BUILD_TYPE": "RelWithDebInfo",
+        "CMAKE_C_FLAGS": "-O3 -pipe -ggdb3",
+        "CMAKE_CXX_FLAGS": "-O3 -pipe -ggdb3"
+      }
+    },
+    {
+      "name": "macos_debug_default",
+      "cacheVariables": {
+        "CMAKE_BUILD_TYPE": "Debug",
+        "CMAKE_C_FLAGS": "-ggdb3 -O0",
+        "CMAKE_CXX_FLAGS": "-ggdb3 -O0 -D_GLIBCXX_DEBUG"
+      }
+    },
+    {
+      "name": "macos_minsizerel_default",
+      "cacheVariables": {
+        "CMAKE_BUILD_TYPE": "MinSizeRel",
+        "CMAKE_C_FLAGS": "-Os -pipe -ggdb3",
+        "CMAKE_CXX_FLAGS": "-Os -pipe -ggdb3"
+      }
+    },
+    {
+      "name": "macos_cpu_release",
+      "displayName": "macos cpu release",
+      "inherits": [
+        "macos_default",
+        "macos_release_default"
+      ]
+    },
+    {
+      "name": "macos_cpu_debug",
+      "displayName": "macos cpu debug",
+      "inherits": [
+        "macos_default",
+        "macos_debug_default"
+      ]
+    },
+    {
+      "name": "macos_cpu_relwithdebinfo",
+      "displayName": "macos cpu relwithdebinfo",
+      "inherits": [
+        "macos_default",
+        "macos_relwithdebinfo_default"
+      ]
+    },
+    {
+      "name": "macos_cpu_minsizerel",
+      "displayName": "macos cpu minsizerel",
+      "inherits": [
+        "macos_default",
+        "macos_minsizerel_default"
+      ]
+    }
+  ]
+}

src/json.cpp

@@ -202,7 +202,7 @@ void JSON::Parse_Array(Element& element) {
 
 double JSON::Parse_Number() {
   double value = NAN;
-#ifndef USE_CXX17
+#if !defined(USE_CXX17) && !defined(__APPLE__)
   auto result = std::from_chars(current_, end_, value);
   if (result.ec != std::errc{}) {
     throw std::runtime_error("Expecting number");

src/models/onnxruntime_inline.h

@@ -151,7 +151,14 @@ inline std::unique_ptr<Allocator> Allocator::Create(const OrtSession& sess, cons
 }
 
 inline void SetCurrentGpuDeviceId(int device_id) {
+#ifdef __APPLE__
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wunused-result"
+#endif
   api->SetCurrentGpuDeviceId(device_id);
+#ifdef __APPLE__
+#pragma clang diagnostic pop
+#endif
 }
 
 inline int GetCurrentGpuDeviceId() {

src/python/CMakeLists.txt

@@ -19,8 +19,9 @@ pybind11_add_module(python ${python_srcs})
 target_include_directories(python PRIVATE ${ORT_HEADER_DIR})
 target_link_directories(python PRIVATE ${ORT_LIB_DIR})
 target_link_libraries(python PRIVATE onnxruntime-genai-static ${ONNXRUNTIME_LIB})
-if(NOT WIN32)
+if(CMAKE_SYSTEM_NAME STREQUAL "Linux")
   set_property(TARGET python APPEND_STRING PROPERTY LINK_FLAGS " -Xlinker -rpath=\\$ORIGIN")
+  target_link_libraries(python PRIVATE stdc++fs)
 endif()
 set_target_properties(python PROPERTIES OUTPUT_NAME "onnxruntime_genai")
 
@@ -29,10 +30,6 @@ if(CMAKE_GENERATOR_TOOLSET MATCHES "Visual Studio")
   target_compile_options(python PRIVATE "/sdl" PRIVATE "/Qspectre")
 endif()
 
-if(NOT WIN32)
-   target_link_libraries(python PRIVATE stdc++fs)
-endif()
-
 if(USE_CUDA AND CMAKE_CUDA_COMPILER)
   cmake_policy(SET CMP0104 OLD)
   enable_language(CUDA)

src/python/python.cpp

@@ -72,7 +72,12 @@ struct PyGeneratorParams : GeneratorParams {
 
     if (py_whisper_input_features_.size() != 0) {
       GeneratorParams::Whisper& whisper = inputs.emplace<GeneratorParams::Whisper>();
+#ifdef __APPLE__
+      std::span shape(reinterpret_cast<const int64_t*>(py_whisper_input_features_.shape()),
+                      py_whisper_input_features_.ndim());
+#else
       std::span<const int64_t> shape(py_whisper_input_features_.shape(), py_whisper_input_features_.ndim());
+#endif
       whisper.input_features = OrtValue::CreateTensor<float>(Ort::Allocator::GetWithDefaultOptions().GetInfo(), ToSpan(py_whisper_input_features_), shape);
       whisper.decoder_input_ids = ToSpan(py_whisper_decoder_input_ids_);
       batch_size = 1;

src/tokenizer/CMakeLists.txt

@@ -27,10 +27,6 @@ FetchContent_MakeAvailable(simdjson)
 
 add_library(tokenizer STATIC ${tokenizer_srcs})
 
-if(NOT WIN32)
-   target_link_libraries(tokenizer PRIVATE stdc++fs)
-endif()
-
 message(STATUS "GSL_SOURCE_DIR: ${GSL_SOURCE_DIR}")
 target_include_directories(tokenizer PUBLIC
    ${CMAKE_CURRENT_SOURCE_DIR}
@@ -40,6 +36,7 @@ target_include_directories(tokenizer PUBLIC
 target_compile_definitions(tokenizer PRIVATE _SILENCE_ALL_CXX20_DEPRECATION_WARNINGS)
 
 if(CMAKE_SYSTEM_NAME STREQUAL "Linux")
+  target_link_libraries(tokenizer PRIVATE stdc++fs)
   set_target_properties(tokenizer PROPERTIES POSITION_INDEPENDENT_CODE ON)
   set_target_properties(simdjson PROPERTIES POSITION_INDEPENDENT_CODE ON)
 endif()

test/CMakeLists.txt

@@ -40,7 +40,7 @@ target_link_libraries(unit_tests PRIVATE
   GTest::gtest_main
 )
 
-if(NOT WIN32)
+if(CMAKE_SYSTEM_NAME STREQUAL "Linux")
   target_link_libraries(unit_tests PRIVATE stdc++fs)
 endif()