Merge branch 'main' into carzh/export_for_inference_test

microsoft · Jul 30, 2024 · 331f9aa · 331f9aa
2 parents 32e321d + 0d7cf30
commit 331f9aa
Show file tree

Hide file tree

Showing 76 changed files with 1,903 additions and 597 deletions.
diff --git a/cmake/external/onnxruntime_external_deps.cmake b/cmake/external/onnxruntime_external_deps.cmake
@@ -46,9 +46,6 @@ if (onnxruntime_BUILD_UNIT_TESTS)
   if (CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
     set(gtest_disable_pthreads ON)
   endif()
-  if (${CMAKE_SYSTEM_NAME} MATCHES "AIX")
-    set(gtest_disable_pthreads ON CACHE BOOL "gtest_disable_pthreads" FORCE)
-  endif()
   set(INSTALL_GTEST OFF CACHE BOOL "" FORCE)
   if (IOS OR ANDROID)
     # on mobile platforms the absl flags class dumps the flag names (assumably for binary size), which breaks passing

diff --git a/cmake/patches/onnx/onnx.patch b/cmake/patches/onnx/onnx.patch
diff --git a/...icrosoft.ML.OnnxRuntime.ResNet50v2Sample/Microsoft.ML.OnnxRuntime.ResNet50v2Sample.csproj b/...icrosoft.ML.OnnxRuntime.ResNet50v2Sample/Microsoft.ML.OnnxRuntime.ResNet50v2Sample.csproj
@@ -8,7 +8,7 @@
 
   <ItemGroup>
     <PackageReference Include="Microsoft.ML.OnnxRuntime" Version="1.16.3" />
-    <PackageReference Include="Sixlabors.ImageSharp" Version="2.1.8" />
+    <PackageReference Include="Sixlabors.ImageSharp" Version="2.1.9" />
   </ItemGroup>
 
 </Project>
diff --git a/csharp/src/Microsoft.ML.OnnxRuntime/Microsoft.ML.OnnxRuntime.csproj b/csharp/src/Microsoft.ML.OnnxRuntime/Microsoft.ML.OnnxRuntime.csproj
@@ -23,14 +23,51 @@
   </PropertyGroup>
 
   <PropertyGroup Condition="'$(OrtPackageId)' == 'Microsoft.ML.OnnxRuntime.Training' AND
-                            '$(IncludeMobileTargets)' == 'true'">
+                              '$(IncludeMobileTargets)' == 'true'">
     <MobileTargets>net8.0-android</MobileTargets>
   </PropertyGroup>
 
   <PropertyGroup>
     <TargetFrameworks>$(BaseTargets);$(MobileTargets)</TargetFrameworks>
   </PropertyGroup>
 
+  <PropertyGroup>
+    <RootNamespace>Microsoft.ML.OnnxRuntime</RootNamespace>
+    <AssemblyName>Microsoft.ML.OnnxRuntime</AssemblyName>
+
+    <!-- Ignore any passed in value unless this is a release build. -->
+    <PackageVersion Condition="'$(IsReleaseBuild)' != 'true'"></PackageVersion>
+    <PackageVersion Condition="'$(PackageVersion)' == '' And '$(Configuration)' == 'Debug'">1.0.0</PackageVersion>
+    <PackageVersion Condition="'$(PackageVersion)' == '' ">0.0.0</PackageVersion>
+
+    <!-- Set the attributes for the managed dll -->
+    <!-- https://learn.microsoft.com/en-us/dotnet/standard/assembly/set-attributes-project-file -->
+    <GenerateAssemblyInfo>true</GenerateAssemblyInfo>
+    <AssemblyTitle>Microsoft.ML.OnnxRuntime C# Bindings</AssemblyTitle>
+    <Company>Microsoft</Company>
+    <Copyright>© Microsoft Corporation. All rights reserved.</Copyright>
+    <Description>This package contains ONNX Runtime for .Net platforms</Description>
+
+    <!-- NOTE: this is also used as the default for AssemblyVersion and FileVersion -->
+    <Version>$(PackageVersion)</Version>
+
+    <!-- Set the attributes for a nuget package -->
+    <!--- The package name is always hardcoded as the package created by this project only contains managed assemblies -->
+    <!--- The parameter OrtPackageId is only used for some conditional logic below -->
+    <Authors>Microsoft</Authors>
+    <PackageId>Microsoft.ML.OnnxRuntime.Managed</PackageId>
+    <PackageTags>ONNX;ONNX Runtime;Machine Learning</PackageTags>
+    <PackageProjectUrl>https://github.com/Microsoft/onnxruntime</PackageProjectUrl>
+    <PackageLicenseFile>LICENSE.txt</PackageLicenseFile>
+    <PackageIcon>ORT_icon_for_light_bg.png</PackageIcon>
+    <PackageReleaseNotes>
+      Release Def:
+        Branch: $(BUILD_SOURCEBRANCH)
+        Commit: $(BUILD_SOURCEVERSION)
+        Build: https://aiinfra.visualstudio.com/Lotus/_build/results?buildId=$(BUILD_BUILDID)
+    </PackageReleaseNotes>
+  </PropertyGroup>
+
   <PropertyGroup>
     <Platforms>AnyCPU;x86</Platforms>
     <LangVersion>default</LangVersion>
@@ -43,8 +80,6 @@
     <OnnxRuntimeCsharpRoot>$(OnnxRuntimeRoot)\csharp</OnnxRuntimeCsharpRoot>
     <TargetArchitecture Condition=" '$(TargetArchitecture)' == '' ">x64</TargetArchitecture>
 
-    <RootNamespace>Microsoft.ML.OnnxRuntime</RootNamespace>
-    <AssemblyName>Microsoft.ML.OnnxRuntime</AssemblyName>
     <EnableDefaultItems>false</EnableDefaultItems>
     <EnableDefaultCompileItems>false</EnableDefaultCompileItems>
     <DebugType>portable</DebugType>
@@ -54,35 +89,15 @@
     on their device is not built for training, an exception will be thrown with the following message -
     "Training is disabled in the current build. Please build onnxruntime from source with the build flags
     enable_training_apis. "-->
-   <EnableTrainingApis Condition="'$(EnableTrainingApis)' == ''">true</EnableTrainingApis>
+    <EnableTrainingApis Condition="'$(EnableTrainingApis)' == ''">true</EnableTrainingApis>
 
-    <!--- The package name is always hardcoded as the package created by this project only contains managed assemblies -->
-    <!--- The parameter OrtPackageId is only used for some conditional logic below -->
-    <PackageId>Microsoft.ML.OnnxRuntime.Managed</PackageId>
-    <Authors>Microsoft</Authors>
-    <PackageVersion Condition=" '$(PackageVersion)' == '' And '$(Configuration)' == 'Debug' ">1.0.0</PackageVersion>
-    <PackageVersion Condition=" '$(PackageVersion)' == '' ">0.0.0</PackageVersion>
-    <Version>$(PackageVersion)</Version>
-    <Description>This package contains ONNX Runtime for .Net platforms</Description>
-    <PackageTags>ONNX;ONNX Runtime;Machine Learning</PackageTags>
-    <PackageProjectUrl>https://github.com/Microsoft/onnxruntime</PackageProjectUrl>
-    <Copyright>© Microsoft Corporation. All rights reserved.</Copyright>
-    <PackageLicenseFile>LICENSE.txt</PackageLicenseFile>
-    <PackageIcon>ORT_icon_for_light_bg.png</PackageIcon>
-    <PackageReleaseNotes>
-      Release Def:
-        Branch: $(BUILD_SOURCEBRANCH)
-        Commit: $(BUILD_SOURCEVERSION)
-        Build: https://aiinfra.visualstudio.com/Lotus/_build/results?buildId=$(BUILD_BUILDID)
-    </PackageReleaseNotes>
     <!-- sourcelink flags -->
     <PublishRepositoryUrl>true</PublishRepositoryUrl>
 
     <!-- Optional: Embed source files that are not tracked by the source control manager in the PDB -->
     <!--EmbedUntrackedSources>true</EmbedUntrackedSources-->
 
     <GenerateTargetFrameworkAttribute>false</GenerateTargetFrameworkAttribute>
-    <GenerateAssemblyInfo>false</GenerateAssemblyInfo>
     <AllowedOutputExtensionsInPackageBuildOutputFolder>$(AllowedOutputExtensionsInPackageBuildOutputFolder);.pdb</AllowedOutputExtensionsInPackageBuildOutputFolder>
     <Configurations>Debug;Release;RelWithDebInfo</Configurations>
 
@@ -158,10 +173,6 @@
     <OrtConstants>$(OrtConstants);__ENABLE_COREML__</OrtConstants>
   </PropertyGroup>
 
-  <PropertyGroup Condition="'$(IsXamarinTarget)'=='true'">
-    <OrtConstants>$(OrtConstants);__XAMARIN__</OrtConstants>
-  </PropertyGroup>
-
   <PropertyGroup>
     <DefineConstants>$(DefineConstants);$(OrtConstants)</DefineConstants>
   </PropertyGroup>

diff --git a/include/onnxruntime/core/session/onnxruntime_cxx_api.h b/include/onnxruntime/core/session/onnxruntime_cxx_api.h
@@ -2216,7 +2216,7 @@ struct ShapeInferContext {
 
   size_t GetInputCount() const { return input_shapes_.size(); }
 
-  Status SetOutputShape(size_t indice, const Shape& shape);
+  Status SetOutputShape(size_t indice, const Shape& shape, ONNXTensorElementDataType type = ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT);
 
   int64_t GetAttrInt(const char* attr_name);
 

diff --git a/include/onnxruntime/core/session/onnxruntime_cxx_inline.h b/include/onnxruntime/core/session/onnxruntime_cxx_inline.h
@@ -1998,9 +1998,10 @@ inline ShapeInferContext::ShapeInferContext(const OrtApi* ort_api,
   }
 }
 
-inline Status ShapeInferContext::SetOutputShape(size_t indice, const Shape& shape) {
+inline Status ShapeInferContext::SetOutputShape(size_t indice, const Shape& shape, ONNXTensorElementDataType type) {
   OrtTensorTypeAndShapeInfo* info = {};
   ORT_CXX_RETURN_ON_API_FAIL(ort_api_->CreateTensorTypeAndShapeInfo(&info));
+  ORT_CXX_RETURN_ON_API_FAIL(ort_api_->SetTensorElementType(info, type));
 
   using InfoPtr = std::unique_ptr<OrtTensorTypeAndShapeInfo, std::function<void(OrtTensorTypeAndShapeInfo*)>>;
 

diff --git a/include/onnxruntime/core/session/onnxruntime_session_options_config_keys.h b/include/onnxruntime/core/session/onnxruntime_session_options_config_keys.h
@@ -265,6 +265,10 @@ static const char* const kOrtSessionOptionEpContextFilePath = "ep.context_file_p
 // "1": dump the EP context into the Onnx model. (default).
 static const char* const kOrtSessionOptionEpContextEmbedMode = "ep.context_embed_mode";
 
+// Specify the EPContext node name prefix to make it unique
+// in case user need to merge/connect multiple EPContext nodes in one model
+static const char* const kOrtSessionOptionEpContextNodeNamePrefix = "ep.context_node_name_prefix";
+
 // Gemm fastmath mode provides fp32 gemm acceleration with bfloat16 based matmul.
 // Option values:
 // - "0": Gemm FastMath mode is not enabled. [DEFAULT]

diff --git a/js/node/src/tensor_helper.cc b/js/node/src/tensor_helper.cc
@@ -38,13 +38,13 @@ constexpr size_t DATA_TYPE_ELEMENT_SIZE_MAP[] = {
     2, // ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT16
     2, // ONNX_TENSOR_ELEMENT_DATA_TYPE_INT16
     4, // ONNX_TENSOR_ELEMENT_DATA_TYPE_INT32
-    8, // ONNX_TENSOR_ELEMENT_DATA_TYPE_INT64         INT64 not working in Javascript
+    8, // ONNX_TENSOR_ELEMENT_DATA_TYPE_INT64
     0, // ONNX_TENSOR_ELEMENT_DATA_TYPE_STRING        N/A
     1, // ONNX_TENSOR_ELEMENT_DATA_TYPE_BOOL
-    0, // ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT16       FLOAT16 not working in Javascript
+    2, // ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT16
     8, // ONNX_TENSOR_ELEMENT_DATA_TYPE_DOUBLE
     4, // ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT32
-    8, // ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT64        UINT64 not working in Javascript
+    8, // ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT64
     0, // ONNX_TENSOR_ELEMENT_DATA_TYPE_COMPLEX64     not supported
     0, // ONNX_TENSOR_ELEMENT_DATA_TYPE_COMPLEX128    not supported
     0  // ONNX_TENSOR_ELEMENT_DATA_TYPE_BFLOAT16      not supported
@@ -60,13 +60,13 @@ constexpr napi_typedarray_type DATA_TYPE_TYPEDARRAY_MAP[] = {
     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         INT64 not working i
+    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_typedarray_type)(-1), // ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT16       FLOAT16 not working
+    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        UINT64 not working
+    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
@@ -182,9 +182,7 @@ Ort::Value NapiValueToOrtValue(Napi::Env env, Napi::Value value, OrtMemoryInfo *
 
     char *buffer = reinterpret_cast<char *>(tensorDataTypedArray.ArrayBuffer().Data());
     size_t bufferByteOffset = tensorDataTypedArray.ByteOffset();
-    // there is a bug in TypedArray::ElementSize(): https://github.com/nodejs/node-addon-api/pull/705
-    // TODO: change to TypedArray::ByteLength() in next node-addon-api release.
-    size_t bufferByteLength = tensorDataTypedArray.ElementLength() * DATA_TYPE_ELEMENT_SIZE_MAP[elemType];
+    size_t bufferByteLength = tensorDataTypedArray.ByteLength();
     return Ort::Value::CreateTensor(memory_info, buffer + bufferByteOffset, bufferByteLength,
                                     dims.empty() ? nullptr : &dims[0], dims.size(), elemType);
   }