Add D3D12 agility SDK dependency to fix crashes on long prompts

.github/workflows/win-directml-x64-build.yml

@@ -21,6 +21,9 @@ env:
   dml_dir: "Microsoft.AI.DirectML.1.14.2"
   dml_zip: "Microsoft.AI.DirectML.1.14.2.zip"
   dml_url: "https://www.nuget.org/api/v2/package/Microsoft.AI.DirectML/1.14.2"
+  d3d12_dir: "Microsoft.Direct3D.D3D12.1.614.0"
+  d3d12_zip: "Microsoft.Direct3D.D3D12.1.614.0.zip"
+  d3d12_url: "https://www.nuget.org/api/v2/package/Microsoft.Direct3D.D3D12/1.614.0"
   binaryDir: 'build/directml'
 
 
@@ -46,6 +49,10 @@ jobs:
       run: |
         Invoke-WebRequest -Uri $env:dml_url -OutFile $env:dml_zip
 
+    - name: Download the D3D12 Agility SDK
+      run: |
+        Invoke-WebRequest -Uri $env:d3d12_url -OutFile $env:d3d12_zip
+
     - name: Unzip OnnxRuntime
       run: |
         Expand-Archive $env:ort_zip -DestinationPath $env:ort_dir
@@ -56,13 +63,19 @@ jobs:
         Expand-Archive $env:dml_zip -DestinationPath $env:dml_dir
         Remove-Item -Path $env:dml_zip
 
+    - name: Unzip the D3D12 Agility SDK
+      run: |
+        Expand-Archive $env:d3d12_zip -DestinationPath $env:d3d12_dir
+        Remove-Item -Path $env:d3d12_zip
+
     - name: Move the files to the ort directory
       run: |
         mkdir ort/lib
         mkdir ort/include
         mv $env:ort_dir\runtimes\win-x64\native\onnxruntime.dll ort\lib
         mv $env:ort_dir\runtimes\win-x64\native\onnxruntime.lib ort\lib
         mv $env:dml_dir\bin\x64-win\DirectML.dll ort\lib
+        mv $env:d3d12_dir\build\native\bin\x64\D3D12Core.dll ort\lib
         mv $env:ort_dir\build\native\include\dml_provider_factory.h ort\include
         mv $env:ort_dir\build\native\include\onnxruntime_c_api.h ort\include
         mv $env:dml_dir\include\DirectML.h ort\include

.pipelines/stages/jobs/nuget-packaging-job.yml

@@ -106,6 +106,13 @@ jobs:
   - name: ort_arm64_url
     value: "https://github.com/microsoft/onnxruntime/releases/download/v1.17.3/onnxruntime-win-arm64-1.17.3.zip"
 
+  - name: d3d12_dir
+    value: 'Microsoft.Direct3D.D3D12.1.614.0'
+  - name: d3d12_zip
+    value: 'Microsoft.Direct3D.D3D12.1.614.0.zip'
+  - name: d3d12_url
+    value: "https://www.nuget.org/api/v2/package/Microsoft.Direct3D.D3D12/1.614.0"
+
   workspace:
     clean: all
   steps:

.pipelines/stages/jobs/py-packaging-job.yml

@@ -120,6 +120,13 @@ jobs:
   - name: ort_arm64_url
     value: "https://github.com/microsoft/onnxruntime/releases/download/v1.17.3/onnxruntime-win-arm64-1.17.3.zip"
 
+  - name: d3d12_dir
+    value: 'Microsoft.Direct3D.D3D12.1.614.0'
+  - name: d3d12_zip
+    value: 'Microsoft.Direct3D.D3D12.1.614.0.zip'
+  - name: d3d12_url
+    value: "https://www.nuget.org/api/v2/package/Microsoft.Direct3D.D3D12/1.614.0"
+
   steps:
   - script: |
       MKDIR $(Agent.ToolsDirectory)\Python\3.12.3\arm64 

.pipelines/stages/jobs/steps/capi-win-step.yml

@@ -54,6 +54,12 @@ steps:
       Get-ChildItem -Recurse $(dml_dir)
       mv $(dml_dir)\bin\x64-win\DirectML.dll ort\lib
       mv $(dml_dir)\include\DirectML.h ort\include
+
+      Invoke-WebRequest -Uri $(d3d12_url) -OutFile $(d3d12_zip)
+      Expand-Archive $(d3d12_zip) -DestinationPath $(d3d12_dir)
+      Remove-Item -Path $(d3d12_zip)
+      Get-ChildItem -Recurse $(d3d12_dir)
+      mv $(d3d12_dir)\build\native\bin\x64\D3D12Core.dll ort\lib
     workingDirectory: '$(Build.Repository.LocalPath)'
     continueOnError: true
 

CMakeLists.txt

@@ -87,6 +87,7 @@ endif()
 
 if(USE_DML)
   list(APPEND onnxruntime_libs "${ORT_LIB_DIR}/DirectML.dll")
+  list(APPEND onnxruntime_libs "${ORT_LIB_DIR}/D3D12Core.dll")
   target_include_directories(onnxruntime-genai PRIVATE $<TARGET_PROPERTY:${WIL_TARGET},INTERFACE_INCLUDE_DIRECTORIES>)
   target_include_directories(onnxruntime-genai PRIVATE $<TARGET_PROPERTY:${DIRECTX_HEADERS_TARGET},INTERFACE_INCLUDE_DIRECTORIES>/directx)
   target_include_directories(onnxruntime-genai PRIVATE $<TARGET_PROPERTY:${DIRECTX_HEADERS_TARGET},INTERFACE_INCLUDE_DIRECTORIES>)

src/dml/dml_helpers.cpp

@@ -80,7 +80,7 @@ static ComPtr<IDXGIAdapter1> CreatePerformantAdapter() {
   return filtered_adapters.front();
 }
 
-DmlObjects CreateDmlObjects() {
+DmlObjects CreateDmlObjects(const std::string& current_module_path) {
   D3D12_COMMAND_QUEUE_DESC command_queue_description = {
       D3D12_COMMAND_LIST_TYPE_COMPUTE,
       0,
@@ -91,7 +91,21 @@ DmlObjects CreateDmlObjects() {
   DmlObjects dml_objects;
 
   auto adapter = CreatePerformantAdapter();
-  THROW_IF_FAILED(D3D12CreateDevice(adapter.Get(), D3D_FEATURE_LEVEL_11_0, IID_PPV_ARGS(&dml_objects.d3d12_device)));
+
+  ComPtr<ID3D12SDKConfiguration1> d3d12_sdk_config;
+  ComPtr<ID3D12DeviceFactory> d3d12_factory;
+
+  // Get the version from https://devblogs.microsoft.com/directx/directx12agility/. We are currently using 1.614.0.
+  constexpr uint32_t agility_sdk_version = 614;
+
+  if (SUCCEEDED(D3D12GetInterface(CLSID_D3D12SDKConfiguration, IID_PPV_ARGS(&d3d12_sdk_config))) &&
+      SUCCEEDED(d3d12_sdk_config->CreateDeviceFactory(agility_sdk_version, current_module_path.c_str(), IID_PPV_ARGS(&d3d12_factory)))) {
+    THROW_IF_FAILED(d3d12_factory->CreateDevice(adapter.Get(), D3D_FEATURE_LEVEL_11_0, IID_PPV_ARGS(&dml_objects.d3d12_device)));
+  } else {
+    printf("Warning: Unable to create a device from version 1.614.0 of the DirectX 12 Agility SDK. You can still use this library, but some scenarios may not work.\n");
+    THROW_IF_FAILED(D3D12CreateDevice(adapter.Get(), D3D_FEATURE_LEVEL_11_0, IID_PPV_ARGS(&dml_objects.d3d12_device)));
+  }
+
   THROW_IF_FAILED(dml_objects.d3d12_device->CreateCommandQueue(&command_queue_description, IID_PPV_ARGS(&dml_objects.command_queue)));
   THROW_IF_FAILED(dml_objects.d3d12_device->CreateCommandAllocator(D3D12_COMMAND_LIST_TYPE_DIRECT, IID_PPV_ARGS(&dml_objects.command_allocator)));
   THROW_IF_FAILED(dml_objects.d3d12_device->CreateCommandList(0, D3D12_COMMAND_LIST_TYPE_DIRECT, dml_objects.command_allocator.Get(), nullptr, IID_PPV_ARGS(&dml_objects.command_list)));

src/dml/dml_helpers.h

@@ -31,7 +31,7 @@ struct DmlObjects {
 };
 
 namespace DmlHelpers {
-DmlObjects CreateDmlObjects();
+DmlObjects CreateDmlObjects(const std::string& current_module_path);
 
 DmlReusedCommandListState BuildReusableCommandList(
     IDMLDevice* dml_device,

src/models/model.cpp

@@ -18,16 +18,16 @@
 
 EXTERN_C IMAGE_DOS_HEADER __ImageBase;
 
-static std::wstring CurrentModulePath() {
-  wchar_t path[MAX_PATH];
-  GetModuleFileNameW((HINSTANCE)&__ImageBase, path, _countof(path));
+static std::string CurrentModulePath() {
+  char path[MAX_PATH];
+  GetModuleFileNameA((HINSTANCE)&__ImageBase, path, _countof(path));
 
-  wchar_t absolute_path[MAX_PATH];
-  wchar_t* name;
-  GetFullPathNameW(path, _countof(path), absolute_path, &name);
+  char absolute_path[MAX_PATH];
+  char* name;
+  GetFullPathNameA(path, _countof(path), absolute_path, &name);
 
   auto idx = std::distance(absolute_path, name);
-  auto out_path = std::wstring(absolute_path);
+  auto out_path = std::string(absolute_path);
   out_path.resize(idx);
 
   return out_path;
@@ -345,13 +345,14 @@ void Model::CreateSessionOptions() {
       ort_options.AppendExecutionProvider_ROCM(ort_provider_options);
 #if USE_DML
     } else if (provider_options.name == "dml") {
-      dml_objects_ = DmlHelpers::CreateDmlObjects();
+      auto current_module_path = CurrentModulePath();
+      dml_objects_ = DmlHelpers::CreateDmlObjects(current_module_path);
 
-      auto directml_dll = CurrentModulePath() + L"DirectML.dll";
-      wil::unique_hmodule smart_directml_dll(LoadLibraryExW(directml_dll.c_str(), nullptr, 0));
+      auto directml_dll = current_module_path + "DirectML.dll";
+      wil::unique_hmodule smart_directml_dll(LoadLibraryExA(directml_dll.c_str(), nullptr, 0));
       THROW_LAST_ERROR_IF(!smart_directml_dll);
 
-      if (LoadLibraryExW(directml_dll.c_str(), nullptr, 0) == NULL) {
+      if (LoadLibraryExA(directml_dll.c_str(), nullptr, 0) == NULL) {
         throw std::runtime_error("DirectML.dll not found");
       }