feat: add directml support (#1153)

CMakeLists.txt

@@ -30,6 +30,7 @@ option(SHERPA_ONNX_ENABLE_JNI "Whether to build JNI internface" OFF)
 option(SHERPA_ONNX_ENABLE_C_API "Whether to build C API" ON)
 option(SHERPA_ONNX_ENABLE_WEBSOCKET "Whether to build webscoket server/client" ON)
 option(SHERPA_ONNX_ENABLE_GPU "Enable ONNX Runtime GPU support" OFF)
+option(SHERPA_ONNX_ENABLE_DIRECTML "Enable ONNX Runtime DirectML support" OFF)
 option(SHERPA_ONNX_ENABLE_WASM "Whether to enable WASM" OFF)
 option(SHERPA_ONNX_ENABLE_WASM_TTS "Whether to enable WASM for TTS" OFF)
 option(SHERPA_ONNX_ENABLE_WASM_ASR "Whether to enable WASM for ASR" OFF)
@@ -94,6 +95,19 @@ to install CUDA toolkit if you have not installed it.")
   endif()
 endif()
 
+if(SHERPA_ONNX_ENABLE_DIRECTML)
+  message(WARNING "\
+Compiling with DirectML enabled. Please make sure Windows 10 SDK
+is installed on your system. Otherwise, you will get errors at runtime.
+Please refer to
+  https://onnxruntime.ai/docs/execution-providers/DirectML-ExecutionProvider.html#requirements
+to install Windows 10 SDK if you have not installed it.")
+  if(NOT BUILD_SHARED_LIBS)
+    message(STATUS "Set BUILD_SHARED_LIBS to ON since SHERPA_ONNX_ENABLE_DIRECTML is ON")
+    set(BUILD_SHARED_LIBS ON CACHE BOOL "" FORCE)
+  endif()
+endif()
+
 # see https://cmake.org/cmake/help/latest/prop_tgt/MSVC_RUNTIME_LIBRARY.html
 # https://stackoverflow.com/questions/14172856/compile-with-mt-instead-of-md-using-cmake
 if(MSVC)
@@ -160,6 +174,14 @@ else()
   add_definitions(-DSHERPA_ONNX_ENABLE_TTS=0)
 endif()
 
+if(SHERPA_ONNX_ENABLE_DIRECTML)
+  message(STATUS "DirectML is enabled")
+  add_definitions(-DSHERPA_ONNX_ENABLE_DIRECTML=1)
+else()
+  message(WARNING "DirectML is disabled")
+  add_definitions(-DSHERPA_ONNX_ENABLE_DIRECTML=0)
+endif()
+
 if(SHERPA_ONNX_ENABLE_WASM_TTS)
   if(NOT SHERPA_ONNX_ENABLE_TTS)
     message(FATAL_ERROR "Please set SHERPA_ONNX_ENABLE_TTS to ON if you want to build wasm TTS")

cmake/onnxruntime-win-x64-directml.cmake

@@ -0,0 +1,161 @@
+# Copyright (c)  2022-2023  Xiaomi Corporation
+message(STATUS "CMAKE_SYSTEM_NAME: ${CMAKE_SYSTEM_NAME}")
+message(STATUS "CMAKE_SYSTEM_PROCESSOR: ${CMAKE_SYSTEM_PROCESSOR}")
+message(STATUS "CMAKE_VS_PLATFORM_NAME: ${CMAKE_VS_PLATFORM_NAME}")
+
+if(NOT CMAKE_SYSTEM_NAME STREQUAL Windows)
+  message(FATAL_ERROR "This file is for Windows only. Given: ${CMAKE_SYSTEM_NAME}")
+endif()
+
+if(NOT (CMAKE_VS_PLATFORM_NAME STREQUAL X64 OR CMAKE_VS_PLATFORM_NAME STREQUAL x64))
+  message(FATAL_ERROR "This file is for Windows x64 only. Given: ${CMAKE_VS_PLATFORM_NAME}")
+endif()
+
+if(NOT BUILD_SHARED_LIBS)
+  message(FATAL_ERROR "This file is for building shared libraries. BUILD_SHARED_LIBS: ${BUILD_SHARED_LIBS}")
+endif()
+
+if(NOT SHERPA_ONNX_ENABLE_DIRECTML)
+  message(FATAL_ERROR "This file is for DirectML. Given SHERPA_ONNX_ENABLE_DIRECTML: ${SHERPA_ONNX_ENABLE_DIRECTML}")
+endif()
+
+set(onnxruntime_URL  "https://globalcdn.nuget.org/packages/microsoft.ml.onnxruntime.directml.1.14.1.nupkg")
+set(onnxruntime_URL2 "https://huggingface.co/csukuangfj/sherpa-onnx-cmake-deps/resolve/main/microsoft.ml.onnxruntime.directml.1.14.1.nupkg")
+set(onnxruntime_HASH "SHA256=c8ae7623385b19cd5de968d0df5383e13b97d1b3a6771c9177eac15b56013a5a")
+
+# If you don't have access to the Internet,
+# please download onnxruntime to one of the following locations.
+# You can add more if you want.
+set(possible_file_locations
+    $ENV{HOME}/Downloads/microsoft.ml.onnxruntime.directml.1.14.1.nupkg
+    ${PROJECT_SOURCE_DIR}/microsoft.ml.onnxruntime.directml.1.14.1.nupkg
+    ${PROJECT_BINARY_DIR}/microsoft.ml.onnxruntime.directml.1.14.1.nupkg
+    /tmp/microsoft.ml.onnxruntime.directml.1.14.1.nupkg
+)
+
+foreach(f IN LISTS possible_file_locations)
+  if(EXISTS ${f})
+    set(onnxruntime_URL  "${f}")
+    file(TO_CMAKE_PATH "${onnxruntime_URL}" onnxruntime_URL)
+    message(STATUS "Found local downloaded onnxruntime: ${onnxruntime_URL}")
+    set(onnxruntime_URL2)
+    break()
+  endif()
+endforeach()
+
+FetchContent_Declare(onnxruntime
+  URL
+    ${onnxruntime_URL}
+    ${onnxruntime_URL2}
+  URL_HASH          ${onnxruntime_HASH}
+)
+
+FetchContent_GetProperties(onnxruntime)
+if(NOT onnxruntime_POPULATED)
+  message(STATUS "Downloading onnxruntime from ${onnxruntime_URL}")
+  FetchContent_Populate(onnxruntime)
+endif()
+message(STATUS "onnxruntime is downloaded to ${onnxruntime_SOURCE_DIR}")
+
+find_library(location_onnxruntime onnxruntime
+  PATHS
+  "${onnxruntime_SOURCE_DIR}/runtimes/win-x64/native"
+  NO_CMAKE_SYSTEM_PATH
+)
+
+message(STATUS "location_onnxruntime: ${location_onnxruntime}")
+
+add_library(onnxruntime SHARED IMPORTED)
+
+set_target_properties(onnxruntime PROPERTIES
+  IMPORTED_LOCATION ${location_onnxruntime}
+  INTERFACE_INCLUDE_DIRECTORIES "${onnxruntime_SOURCE_DIR}/build/native/include"
+)
+
+set_property(TARGET onnxruntime
+  PROPERTY
+    IMPORTED_IMPLIB "${onnxruntime_SOURCE_DIR}/runtimes/win-x64/native/onnxruntime.lib"
+)
+
+file(COPY ${onnxruntime_SOURCE_DIR}/runtimes/win-x64/native/onnxruntime.dll
+  DESTINATION
+    ${CMAKE_BINARY_DIR}/bin/${CMAKE_BUILD_TYPE}
+)
+
+file(GLOB onnxruntime_lib_files "${onnxruntime_SOURCE_DIR}/runtimes/win-x64/native/onnxruntime.*")
+
+message(STATUS "onnxruntime lib files: ${onnxruntime_lib_files}")
+
+if(SHERPA_ONNX_ENABLE_PYTHON)
+  install(FILES ${onnxruntime_lib_files} DESTINATION ..)
+else()
+  install(FILES ${onnxruntime_lib_files} DESTINATION lib)
+endif()
+
+install(FILES ${onnxruntime_lib_files} DESTINATION bin)
+
+# Setup DirectML
+
+set(directml_URL "https://www.nuget.org/api/v2/package/Microsoft.AI.DirectML/1.15.0")
+set(directml_HASH "SHA256=10d175f8e97447712b3680e3ac020bbb8eafdf651332b48f09ffee2eec801c23")
+
+set(possible_directml_file_locations
+    $ENV{HOME}/Downloads/Microsoft.AI.DirectML.1.15.0.nupkg
+    ${PROJECT_SOURCE_DIR}/Microsoft.AI.DirectML.1.15.0.nupkg
+    ${PROJECT_BINARY_DIR}/Microsoft.AI.DirectML.1.15.0.nupkg
+    /tmp/Microsoft.AI.DirectML.1.15.0.nupkg
+)
+
+foreach(f IN LISTS possible_directml_file_locations)
+  if(EXISTS ${f})
+    set(directml_URL  "${f}")
+    file(TO_CMAKE_PATH "${directml_URL}" directml_URL)
+    message(STATUS "Found local downloaded DirectML: ${directml_URL}")
+    break()
+  endif()
+endforeach()
+
+FetchContent_Declare(directml
+  URL
+    ${directml_URL}
+  URL_HASH ${directml_HASH}
+)
+
+FetchContent_GetProperties(directml)
+if(NOT directml_POPULATED)
+  message(STATUS "Downloading DirectML from ${directml_URL}")
+  FetchContent_Populate(directml)
+endif()
+message(STATUS "DirectML is downloaded to ${directml_SOURCE_DIR}")
+
+find_library(location_directml DirectML
+  PATHS
+  "${directml_SOURCE_DIR}/bin/x64-win"
+  NO_CMAKE_SYSTEM_PATH
+)
+
+message(STATUS "location_directml: ${location_directml}")
+
+add_library(directml SHARED IMPORTED)
+
+set_target_properties(directml PROPERTIES
+  IMPORTED_LOCATION ${location_directml}
+  INTERFACE_INCLUDE_DIRECTORIES "${directml_SOURCE_DIR}/bin/x64-win"
+)
+
+set_property(TARGET directml
+  PROPERTY
+    IMPORTED_IMPLIB "${directml_SOURCE_DIR}/bin/x64-win/DirectML.lib"
+)
+
+file(COPY ${directml_SOURCE_DIR}/bin/x64-win/DirectML.dll
+  DESTINATION
+    ${CMAKE_BINARY_DIR}/bin/${CMAKE_BUILD_TYPE}
+)
+
+file(GLOB directml_lib_files "${directml_SOURCE_DIR}/bin/x64-win/DirectML.*")
+
+message(STATUS "DirectML lib files: ${directml_lib_files}")
+
+install(FILES ${directml_lib_files} DESTINATION lib)
+install(FILES ${directml_lib_files} DESTINATION bin)

cmake/onnxruntime.cmake

@@ -95,7 +95,10 @@ function(download_onnxruntime)
       include(onnxruntime-win-arm64)
     else()
       # for 64-bit windows (x64)
-      if(BUILD_SHARED_LIBS)
+      if(SHERPA_ONNX_ENABLE_DIRECTML)
+        message(STATUS "Use DirectML")
+        include(onnxruntime-win-x64-directml)
+      elseif(BUILD_SHARED_LIBS)
         message(STATUS "Use dynamic onnxruntime libraries")
         if(SHERPA_ONNX_ENABLE_GPU)
           include(onnxruntime-win-x64-gpu)

sherpa-onnx/csrc/provider.cc

@@ -26,6 +26,8 @@ Provider StringToProvider(std::string s) {
     return Provider::kNNAPI;
   } else if (s == "trt") {
     return Provider::kTRT;
+  } else if (s == "directml") {
+    return Provider::kDirectML;
   } else {
     SHERPA_ONNX_LOGE("Unsupported string: %s. Fallback to cpu", s.c_str());
     return Provider::kCPU;

sherpa-onnx/csrc/provider.h

@@ -14,12 +14,13 @@ namespace sherpa_onnx {
 // https://github.com/microsoft/onnxruntime/blob/main/java/src/main/java/ai/onnxruntime/OrtProvider.java
 // for a list of available providers
 enum class Provider {
-  kCPU = 0,      // CPUExecutionProvider
-  kCUDA = 1,     // CUDAExecutionProvider
-  kCoreML = 2,   // CoreMLExecutionProvider
-  kXnnpack = 3,  // XnnpackExecutionProvider
-  kNNAPI = 4,    // NnapiExecutionProvider
-  kTRT = 5,     // TensorRTExecutionProvider
+  kCPU = 0,       // CPUExecutionProvider
+  kCUDA = 1,      // CUDAExecutionProvider
+  kCoreML = 2,    // CoreMLExecutionProvider
+  kXnnpack = 3,   // XnnpackExecutionProvider
+  kNNAPI = 4,     // NnapiExecutionProvider
+  kTRT = 5,       // TensorRTExecutionProvider
+  kDirectML = 6,  // DmlExecutionProvider
 };
 
 /**

sherpa-onnx/csrc/session.cc

@@ -19,6 +19,10 @@
 #include "nnapi_provider_factory.h"  // NOLINT
 #endif
 
+#if defined(_WIN32) && SHERPA_ONNX_ENABLE_DIRECTML == 1
+#include "dml_provider_factory.h"  // NOLINT
+#endif
+
 namespace sherpa_onnx {
 
 static void OrtStatusFailure(OrtStatus *status, const char *s) {
@@ -167,6 +171,24 @@ static Ort::SessionOptions GetSessionOptionsImpl(
       }
       break;
     }
+    case Provider::kDirectML: {
+#if defined(_WIN32) && SHERPA_ONNX_ENABLE_DIRECTML == 1
+      sess_opts.DisableMemPattern();
+      sess_opts.SetExecutionMode(ORT_SEQUENTIAL);
+      int32_t device_id = 0;
+      OrtStatus *status =
+          OrtSessionOptionsAppendExecutionProvider_DML(sess_opts, device_id);
+      if (status) {
+        const auto &api = Ort::GetApi();
+        const char *msg = api.GetErrorMessage(status);
+        SHERPA_ONNX_LOGE("Failed to enable DirectML: %s. Fallback to cpu", msg);
+        api.ReleaseStatus(status);
+      }
+#else
+      SHERPA_ONNX_LOGE("DirectML is for Windows only. Fallback to cpu!");
+#endif
+      break;
+    }
     case Provider::kCoreML: {
 #if defined(__APPLE__)
       uint32_t coreml_flags = 0;