Merge branch 'main' into pad

cgmanifests/generated/cgmanifest.json

@@ -196,7 +196,7 @@
       "component": {
         "type": "git",
         "git": {
-          "commitHash": "bc0d2e35909b8456abe32f3b30a49bb0c125e8b7",
+          "commitHash": "9c69a24bc2e20c8a511a4e6b06fd49639ec5300a",
           "repositoryUrl": "https://github.com/onnx/onnx-tensorrt.git"
         },
         "comments": "onnx_tensorrt"

cmake/deps.txt

@@ -36,8 +36,8 @@ microsoft_wil;https://github.com/microsoft/wil/archive/refs/tags/v1.0.230629.1.z
 mimalloc;https://github.com/microsoft/mimalloc/archive/refs/tags/v2.1.1.zip;d5ee7d34223d0567892db5179849939c8769dc41
 mp11;https://github.com/boostorg/mp11/archive/refs/tags/boost-1.82.0.zip;9bc9e01dffb64d9e0773b2e44d2f22c51aace063
 onnx;https://github.com/onnx/onnx/archive/refs/tags/v1.16.1.zip;2eb9198bb352757d5ff13977cbe0634898e0837c
-# Use the latest commit of 10.6-GA-ORT-DDS
-onnx_tensorrt;https://github.com/onnx/onnx-tensorrt/archive/bc0d2e35909b8456abe32f3b30a49bb0c125e8b7.zip;f233ae871ad82c023da62e5dd620639f00bc2d15
+# Use the latest commit of 10.7-GA
+onnx_tensorrt;https://github.com/onnx/onnx-tensorrt/archive/9c69a24bc2e20c8a511a4e6b06fd49639ec5300a.zip;ff1fe9af78eb129b4a4cdcb7450b7390b4436dd3
 protobuf;https://github.com/protocolbuffers/protobuf/archive/refs/tags/v21.12.zip;7cf2733949036c7d52fda017badcab093fe73bfa
 protoc_win64;https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-win64.zip;b4521f7ada5b260380f94c4bd7f1b7684c76969a
 protoc_win32;https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-win32.zip;3688010318192c46ce73213cdfb6b3e5656da874

cmake/onnxruntime.cmake

@@ -77,6 +77,7 @@ if(WIN32)
   onnxruntime_add_shared_library(onnxruntime
     ${SYMBOL_FILE}
     "${ONNXRUNTIME_ROOT}/core/dll/dllmain.cc"
+    "${ONNXRUNTIME_ROOT}/core/dll/delay_load_hook.cc"
     "${ONNXRUNTIME_ROOT}/core/dll/onnxruntime.rc"
   )
 elseif(onnxruntime_BUILD_APPLE_FRAMEWORK)

cmake/onnxruntime_nodejs.cmake

@@ -60,15 +60,26 @@ else()
     endif()
 endif()
 
+# a list of DLLs that the Node.js binding depends on
+set(NODEJS_DLL_DEPS)
+
 # setup providers
 if (onnxruntime_USE_CUDA)
     set(NODEJS_BINDING_USE_CUDA "--use_cuda")
 endif()
 if (onnxruntime_USE_DML)
     set(NODEJS_BINDING_USE_DML "--use_dml")
+    list(APPEND NODEJS_DLL_DEPS "$<TARGET_FILE_DIR:onnxruntime>/DirectML.dll")
 endif()
 if (onnxruntime_USE_WEBGPU)
     set(NODEJS_BINDING_USE_WEBGPU "--use_webgpu")
+    if (WIN32 AND onnxruntime_ENABLE_DAWN_BACKEND_D3D12)
+        list(APPEND NODEJS_DLL_DEPS "$<TARGET_FILE_DIR:dxcompiler>/dxil.dll")
+        list(APPEND NODEJS_DLL_DEPS "$<TARGET_FILE_DIR:dxcompiler>/dxcompiler.dll")
+    endif()
+    if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
+        list(APPEND NODEJS_DLL_DEPS "$<TARGET_FILE:dawn::webgpu_dawn>")
+    endif()
 endif()
 if (onnxruntime_USE_TENSORRT)
     set(NODEJS_BINDING_USE_TENSORRT "--use_tensorrt")
@@ -94,9 +105,12 @@ add_custom_target(js_common_npm_ci ALL
 
 add_custom_target(nodejs_binding_wrapper ALL
     COMMAND ${NPM_CLI} ci
-    COMMAND ${NPM_CLI} run build -- --onnxruntime-build-dir=${CMAKE_CURRENT_BINARY_DIR} --config=${CMAKE_BUILD_TYPE} --onnxruntime-generator=${CMAKE_GENERATOR}
-        --arch=${NODEJS_BINDING_ARCH} ${NODEJS_BINDING_USE_CUDA} ${NODEJS_BINDING_USE_DML} ${NODEJS_BINDING_USE_WEBGPU} ${NODEJS_BINDING_USE_TENSORRT}
-        ${NODEJS_BINDING_USE_COREML} ${NODEJS_BINDING_USE_QNN}
+    COMMAND ${NPM_CLI} run build -- "--onnxruntime-build-dir=${CMAKE_CURRENT_BINARY_DIR}"
+        --config=${CMAKE_BUILD_TYPE}
+        "--onnxruntime-generator=${CMAKE_GENERATOR}"
+        "--dll_deps=${NODEJS_DLL_DEPS}"
+        --arch=${NODEJS_BINDING_ARCH} ${NODEJS_BINDING_USE_CUDA} ${NODEJS_BINDING_USE_DML} ${NODEJS_BINDING_USE_WEBGPU}
+        ${NODEJS_BINDING_USE_TENSORRT} ${NODEJS_BINDING_USE_COREML} ${NODEJS_BINDING_USE_QNN}
     WORKING_DIRECTORY ${JS_NODE_ROOT}
     COMMENT "Using cmake-js to build OnnxRuntime Node.js binding")
 

cmake/onnxruntime_providers_webgpu.cmake

@@ -23,24 +23,42 @@
   onnxruntime_add_include_to_target(onnxruntime_providers_webgpu
     onnxruntime_common dawn::dawncpp_headers dawn::dawn_headers onnx onnx_proto flatbuffers::flatbuffers Boost::mp11 safeint_interface)
 
+  set(onnxruntime_providers_webgpu_dll_deps)
+
   if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
     target_link_libraries(onnxruntime_providers_webgpu dawn::webgpu_dawn)
 
-    if (onnxruntime_ENABLE_DELAY_LOADING_WIN_DLLS)
-      list(APPEND onnxruntime_DELAYLOAD_FLAGS "/DELAYLOAD:webgpu_dawn.dll")
-    endif()
+    if (WIN32)
+      if (onnxruntime_ENABLE_DELAY_LOADING_WIN_DLLS)
+        list(APPEND onnxruntime_DELAYLOAD_FLAGS "/DELAYLOAD:webgpu_dawn.dll")
+      endif()
 
-    # Copy webgpu_dawn.dll to the output directory
-    add_custom_command(
-      TARGET onnxruntime_providers_webgpu
-      POST_BUILD
-      COMMAND ${CMAKE_COMMAND} -E copy_if_different "$<TARGET_FILE:dawn::webgpu_dawn>" "$<TARGET_FILE_DIR:onnxruntime_providers_webgpu>"
-      VERBATIM )
+      list(APPEND onnxruntime_providers_webgpu_dll_deps "$<TARGET_FILE:dawn::webgpu_dawn>")
+    endif()
   else()
     if (NOT onnxruntime_USE_EXTERNAL_DAWN)
       target_link_libraries(onnxruntime_providers_webgpu dawn::dawn_native)
     endif()
     target_link_libraries(onnxruntime_providers_webgpu dawn::dawn_proc)
   endif()
 
+  if (WIN32 AND onnxruntime_ENABLE_DAWN_BACKEND_D3D12)
+    # Ensure dxil.dll and dxcompiler.dll exist in the output directory $<TARGET_FILE_DIR:dxcompiler>
+    add_dependencies(onnxruntime_providers_webgpu copy_dxil_dll)
+    add_dependencies(onnxruntime_providers_webgpu dxcompiler)
+
+    list(APPEND onnxruntime_providers_webgpu_dll_deps "$<TARGET_FILE_DIR:dxcompiler>/dxil.dll")
+    list(APPEND onnxruntime_providers_webgpu_dll_deps "$<TARGET_FILE_DIR:dxcompiler>/dxcompiler.dll")
+  endif()
+
+  if (onnxruntime_providers_webgpu_dll_deps)
+    # Copy dependency DLLs to the output directory
+    add_custom_command(
+      TARGET onnxruntime_providers_webgpu
+      POST_BUILD
+      COMMAND ${CMAKE_COMMAND} -E copy_if_different "${onnxruntime_providers_webgpu_dll_deps}" "$<TARGET_FILE_DIR:onnxruntime_providers_webgpu>"
+      COMMAND_EXPAND_LISTS
+      VERBATIM )
+  endif()
+
   set_target_properties(onnxruntime_providers_webgpu PROPERTIES FOLDER "ONNXRuntime")

cmake/onnxruntime_unittests.cmake

@@ -525,6 +525,9 @@ set (onnxruntime_global_thread_pools_test_SRC
 set (onnxruntime_webgpu_external_dawn_test_SRC
           ${TEST_SRC_DIR}/webgpu/external_dawn/main.cc)
 
+set (onnxruntime_webgpu_delay_load_test_SRC
+          ${TEST_SRC_DIR}/webgpu/delay_load/main.cc)
+
 # tests from lowest level library up.
 # the order of libraries should be maintained, with higher libraries being added first in the list
 
@@ -1864,4 +1867,13 @@ if (onnxruntime_USE_WEBGPU AND onnxruntime_USE_EXTERNAL_DAWN)
   onnxruntime_add_include_to_target(onnxruntime_webgpu_external_dawn_test dawn::dawncpp_headers dawn::dawn_headers)
 endif()
 
+if (onnxruntime_USE_WEBGPU AND WIN32 AND onnxruntime_BUILD_SHARED_LIB AND NOT CMAKE_SYSTEM_NAME STREQUAL "Emscripten" AND NOT onnxruntime_MINIMAL_BUILD)
+  AddTest(DYN
+          TARGET onnxruntime_webgpu_delay_load_test
+          SOURCES ${onnxruntime_webgpu_delay_load_test_SRC}
+          LIBS ${SYS_PATH_LIB}
+          DEPENDS ${all_dependencies}
+  )
+endif()
+
 include(onnxruntime_fuzz_test.cmake)

docs/ContribOperators.md

@@ -1625,7 +1625,7 @@ This version of the operator has been available since version 1 of the 'com.micr
 #### Type Constraints
 
 <dl>
-<dt><tt>T</tt> : tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(float16), tensor(float), tensor(double)</dt>
+<dt><tt>T</tt> : tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16)</dt>
 <dd>Constrain input and output types.</dd>
 </dl>
 

java/src/test/java/ai/onnxruntime/InferenceTest.java

@@ -737,7 +737,6 @@ public void testCoreML() throws OrtException {
     runProvider(OrtProvider.CORE_ML);
   }
 
-  @Disabled("DirectML Java API hasn't been supported yet")
   @Test
   @EnabledIfSystemProperty(named = "USE_DML", matches = "1")
   public void testDirectML() throws OrtException {

java/src/test/java/ai/onnxruntime/providers/ProviderOptionsTest.java

@@ -27,15 +27,13 @@
 import java.util.HashMap;
 import java.util.Map;
 import org.junit.jupiter.api.Test;
-import org.junit.jupiter.api.condition.DisabledIfSystemProperty;
 import org.junit.jupiter.api.condition.EnabledIfSystemProperty;
 
 public class ProviderOptionsTest {
   private static final OrtEnvironment env = TestHelpers.getOrtEnvironment();
 
   @Test
   @EnabledIfSystemProperty(named = "USE_CUDA", matches = "1")
-  @DisabledIfSystemProperty(named = "NO_CUDA_TEST", matches = "1")
   public void testCUDAOptions() throws OrtException {
     // Test standard options
     OrtCUDAProviderOptions cudaOpts = new OrtCUDAProviderOptions(0);
@@ -63,7 +61,6 @@ public void testCUDAOptions() throws OrtException {
 
   @Test
   @EnabledIfSystemProperty(named = "USE_TENSORRT", matches = "1")
-  @DisabledIfSystemProperty(named = "NO_CUDA_TEST", matches = "1")
   public void testTensorRT() throws OrtException {
     // Test standard options
     OrtTensorRTProviderOptions rtOpts = new OrtTensorRTProviderOptions(0);

js/node/CMakeLists.txt

@@ -113,10 +113,12 @@ endif()
 if (WIN32)
   file(COPY ${ONNXRUNTIME_WIN_BIN_DIR}/onnxruntime.dll
       DESTINATION ${dist_folder})
-  if (USE_DML)
-    file(COPY ${ONNXRUNTIME_WIN_BIN_DIR}/DirectML.dll
-      DESTINATION ${dist_folder})
-  endif ()
+  if (ORT_NODEJS_DLL_DEPS)
+    foreach(dll ${ORT_NODEJS_DLL_DEPS})
+      file(COPY ${dll} DESTINATION ${dist_folder})
+    endforeach()
+  endif()
+
 elseif (APPLE)
   file(COPY ${ONNXRUNTIME_BUILD_DIR}/libonnxruntime.dylib
       DESTINATION ${dist_folder} FOLLOW_SYMLINK_CHAIN)

js/node/script/build.ts

@@ -39,6 +39,8 @@ const USE_TENSORRT = !!buildArgs.use_tensorrt;
 const USE_COREML = !!buildArgs.use_coreml;
 // --use_qnn
 const USE_QNN = !!buildArgs.use_qnn;
+// --dll_deps=
+const DLL_DEPS = buildArgs.dll_deps;
 
 // build path
 const ROOT_FOLDER = path.join(__dirname, '..');
@@ -82,6 +84,9 @@ if (USE_COREML) {
 if (USE_QNN) {
   args.push('--CDUSE_QNN=ON');
 }
+if (DLL_DEPS) {
+  args.push(`--CDORT_NODEJS_DLL_DEPS=${DLL_DEPS}`);
+}
 
 // set CMAKE_OSX_ARCHITECTURES for macOS build
 if (os.platform() === 'darwin') {

js/node/src/inference_session_wrap.cc

@@ -4,7 +4,6 @@
 #include "onnxruntime_cxx_api.h"
 
 #include "common.h"
-#include "directml_load_helper.h"
 #include "inference_session_wrap.h"
 #include "run_options_helper.h"
 #include "session_options_helper.h"
@@ -19,9 +18,6 @@ Napi::FunctionReference& InferenceSessionWrap::GetTensorConstructor() {
 }
 
 Napi::Object InferenceSessionWrap::Init(Napi::Env env, Napi::Object exports) {
-#if defined(USE_DML) && defined(_WIN32)
-  LoadDirectMLDll(env);
-#endif
   // create ONNX runtime env
   Ort::InitApi();
   ORT_NAPI_THROW_ERROR_IF(

js/node/src/tensor_helper.cc

@@ -53,24 +53,24 @@ constexpr size_t DATA_TYPE_ELEMENT_SIZE_MAP[] = {
 static_assert(sizeof(DATA_TYPE_ELEMENT_SIZE_MAP) == sizeof(size_t) * ONNX_TENSOR_ELEMENT_DATA_TYPE_COUNT,
               "definition not matching");
 
-constexpr napi_typedarray_type DATA_TYPE_TYPEDARRAY_MAP[] = {
-    (napi_typedarray_type)(-1),  // ONNX_TENSOR_ELEMENT_DATA_TYPE_UNDEFINED     not supported
-    napi_float32_array,          // ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT
-    napi_uint8_array,            // ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT8
-    napi_int8_array,             // ONNX_TENSOR_ELEMENT_DATA_TYPE_INT8
-    napi_uint16_array,           // ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT16
-    napi_int16_array,            // ONNX_TENSOR_ELEMENT_DATA_TYPE_INT16
-    napi_int32_array,            // ONNX_TENSOR_ELEMENT_DATA_TYPE_INT32
-    napi_bigint64_array,         // ONNX_TENSOR_ELEMENT_DATA_TYPE_INT64
-    (napi_typedarray_type)(-1),  // ONNX_TENSOR_ELEMENT_DATA_TYPE_STRING        not supported
-    napi_uint8_array,            // ONNX_TENSOR_ELEMENT_DATA_TYPE_BOOL
-    napi_uint16_array,           // ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT16       FLOAT16 uses Uint16Array
-    napi_float64_array,          // ONNX_TENSOR_ELEMENT_DATA_TYPE_DOUBLE
-    napi_uint32_array,           // ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT32
-    napi_biguint64_array,        // ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT64
-    (napi_typedarray_type)(-1),  // ONNX_TENSOR_ELEMENT_DATA_TYPE_COMPLEX64     not supported
-    (napi_typedarray_type)(-1),  // ONNX_TENSOR_ELEMENT_DATA_TYPE_COMPLEX128    not supported
-    (napi_typedarray_type)(-1)   // ONNX_TENSOR_ELEMENT_DATA_TYPE_BFLOAT16      not supported
+constexpr std::underlying_type_t<napi_typedarray_type> DATA_TYPE_TYPEDARRAY_MAP[] = {
+    std::underlying_type_t<napi_typedarray_type>(-1),  // ONNX_TENSOR_ELEMENT_DATA_TYPE_UNDEFINED     not supported
+    napi_float32_array,                                // ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT
+    napi_uint8_array,                                  // ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT8
+    napi_int8_array,                                   // ONNX_TENSOR_ELEMENT_DATA_TYPE_INT8
+    napi_uint16_array,                                 // ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT16
+    napi_int16_array,                                  // ONNX_TENSOR_ELEMENT_DATA_TYPE_INT16
+    napi_int32_array,                                  // ONNX_TENSOR_ELEMENT_DATA_TYPE_INT32
+    napi_bigint64_array,                               // ONNX_TENSOR_ELEMENT_DATA_TYPE_INT64
+    std::underlying_type_t<napi_typedarray_type>(-1),  // ONNX_TENSOR_ELEMENT_DATA_TYPE_STRING        not supported
+    napi_uint8_array,                                  // ONNX_TENSOR_ELEMENT_DATA_TYPE_BOOL
+    napi_uint16_array,                                 // ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT16       FLOAT16 uses Uint16Array
+    napi_float64_array,                                // ONNX_TENSOR_ELEMENT_DATA_TYPE_DOUBLE
+    napi_uint32_array,                                 // ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT32
+    napi_biguint64_array,                              // ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT64
+    std::underlying_type_t<napi_typedarray_type>(-1),  // ONNX_TENSOR_ELEMENT_DATA_TYPE_COMPLEX64     not supported
+    std::underlying_type_t<napi_typedarray_type>(-1),  // ONNX_TENSOR_ELEMENT_DATA_TYPE_COMPLEX128    not supported
+    std::underlying_type_t<napi_typedarray_type>(-1)   // ONNX_TENSOR_ELEMENT_DATA_TYPE_BFLOAT16      not supported
 };
 static_assert(sizeof(DATA_TYPE_TYPEDARRAY_MAP) == sizeof(napi_typedarray_type) * ONNX_TENSOR_ELEMENT_DATA_TYPE_COUNT,
               "definition not matching");
@@ -98,7 +98,20 @@ static_assert(sizeof(DATA_TYPE_ID_TO_NAME_MAP) == sizeof(const char*) * ONNX_TEN
               "definition not matching");
 
 const std::unordered_map<std::string, ONNXTensorElementDataType> DATA_TYPE_NAME_TO_ID_MAP = {
-    {"float32", ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT}, {"uint8", ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT8}, {"int8", ONNX_TENSOR_ELEMENT_DATA_TYPE_INT8}, {"uint16", ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT16}, {"int16", ONNX_TENSOR_ELEMENT_DATA_TYPE_INT16}, {"int32", ONNX_TENSOR_ELEMENT_DATA_TYPE_INT32}, {"int64", ONNX_TENSOR_ELEMENT_DATA_TYPE_INT64}, {"string", ONNX_TENSOR_ELEMENT_DATA_TYPE_STRING}, {"bool", ONNX_TENSOR_ELEMENT_DATA_TYPE_BOOL}, {"float16", ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT16}, {"float64", ONNX_TENSOR_ELEMENT_DATA_TYPE_DOUBLE}, {"uint32", ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT32}, {"uint64", ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT64}};
+    {"float32", ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT},
+    {"uint8", ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT8},
+    {"int8", ONNX_TENSOR_ELEMENT_DATA_TYPE_INT8},
+    {"uint16", ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT16},
+    {"int16", ONNX_TENSOR_ELEMENT_DATA_TYPE_INT16},
+    {"int32", ONNX_TENSOR_ELEMENT_DATA_TYPE_INT32},
+    {"int64", ONNX_TENSOR_ELEMENT_DATA_TYPE_INT64},
+    {"string", ONNX_TENSOR_ELEMENT_DATA_TYPE_STRING},
+    {"bool", ONNX_TENSOR_ELEMENT_DATA_TYPE_BOOL},
+    {"float16", ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT16},
+    {"float64", ONNX_TENSOR_ELEMENT_DATA_TYPE_DOUBLE},
+    {"uint32", ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT32},
+    {"uint64", ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT64},
+};
 
 // currently only support tensor
 Ort::Value NapiValueToOrtValue(Napi::Env env, Napi::Value value, OrtMemoryInfo* cpu_memory_info, OrtMemoryInfo* webgpu_memory_info) {
@@ -181,7 +194,7 @@ Ort::Value NapiValueToOrtValue(Napi::Env env, Napi::Value value, OrtMemoryInfo*
                                   "Tensor.data must be a typed array for numeric tensor.");
 
       auto tensorDataTypedArray = tensorDataValue.As<Napi::TypedArray>();
-      auto typedArrayType = tensorDataValue.As<Napi::TypedArray>().TypedArrayType();
+      std::underlying_type_t<napi_typedarray_type> typedArrayType = tensorDataValue.As<Napi::TypedArray>().TypedArrayType();
       ORT_NAPI_THROW_TYPEERROR_IF(DATA_TYPE_TYPEDARRAY_MAP[elemType] != typedArrayType, env,
                                   "Tensor.data must be a typed array (", DATA_TYPE_TYPEDARRAY_MAP[elemType], ") for ",
                                   tensorTypeString, " tensors, but got typed array (", typedArrayType, ").");
@@ -294,7 +307,7 @@ Napi::Value OrtValueToNapiValue(Napi::Env env, Ort::Value&& value) {
       }
       napi_value typedArrayData;
       napi_status status =
-          napi_create_typedarray(env, DATA_TYPE_TYPEDARRAY_MAP[elemType], size, arrayBuffer, 0, &typedArrayData);
+          napi_create_typedarray(env, (napi_typedarray_type)DATA_TYPE_TYPEDARRAY_MAP[elemType], size, arrayBuffer, 0, &typedArrayData);
       NAPI_THROW_IF_FAILED(env, status, Napi::Value);
 
       // new Tensor(type, typedArrayData, dims)