RyzenAI fallback to GPU not CPU

Author: saddam213

TensorStack.Providers.RyzenAI/Provider.cs

@@ -118,14 +118,14 @@ public static ExecutionProvider GetProvider(DeviceType deviceType, GraphOptimiza
 
 
         /// <summary>
-        /// Gets the RyzenAI provider this DeviceType, DeviceId.
+        /// Gets the RyzenAI provider for NPU if supported, else DirectML GPU fallback.
         /// </summary>
         /// <param name="deviceType">Type of the device.</param>
         /// <param name="deviceId">The device identifier.</param>
         /// <param name="optimizationLevel">The optimization level.</param>
-        public static ExecutionProvider GetProvider(DeviceType deviceType, int deviceId, GraphOptimizationLevel optimizationLevel = GraphOptimizationLevel.ORT_ENABLE_ALL)
+        public static ExecutionProvider GetProvider(DeviceType deviceType, int fallbackDeviceId, GraphOptimizationLevel optimizationLevel = GraphOptimizationLevel.ORT_ENABLE_ALL)
         {
-            return GetDevice(deviceType, deviceId).GetProvider(optimizationLevel);
+            return GetDevice(deviceType, fallbackDeviceId).GetProvider(optimizationLevel);
         }
 
 
@@ -146,13 +146,13 @@ public static ExecutionProvider GetProvider(this Device device, GraphOptimizatio
 
 
         /// <summary>
-        /// Gets the RyzenAI provider for this DeviceId.
+        /// Gets the RyzenAI provider for NPU if supported, else DirectML fallback.
         /// </summary>
-        /// <param name="deviceId">The device identifier.</param>
+        /// <param name="fallbackDeviceId">The fallback device identifier.</param>
         /// <param name="optimizationLevel">The optimization level.</param>
-        private static ExecutionProvider CreateRyzenProvider(int deviceId, GraphOptimizationLevel optimizationLevel = GraphOptimizationLevel.ORT_ENABLE_ALL)
+        private static ExecutionProvider CreateRyzenProvider(int fallbackDeviceId, GraphOptimizationLevel optimizationLevel = GraphOptimizationLevel.ORT_ENABLE_ALL)
         {
-            var memoryInfo = new OrtMemoryInfo(OrtMemoryInfo.allocatorCPU, OrtAllocatorType.DeviceAllocator, deviceId, OrtMemType.Default);
+            var memoryInfo = new OrtMemoryInfo(OrtMemoryInfo.allocatorCPU, OrtAllocatorType.DeviceAllocator, fallbackDeviceId, OrtMemType.Default);
             return new ExecutionProvider(_providerName, memoryInfo, configuration =>
             {
                 var sessionOptions = new SessionOptions
@@ -166,6 +166,7 @@ private static ExecutionProvider CreateRyzenProvider(int deviceId, GraphOptimiza
 
                 sessionOptions.AddSessionConfigEntries(configuration.SessionOptions);
                 sessionOptions.RegisterCustomOpLibrary("onnx_custom_ops.dll");
+                sessionOptions.AppendExecutionProvider_DML();
                 sessionOptions.AppendExecutionProvider_CPU();
                 return sessionOptions;
             });

TensorStack.sln

@@ -29,6 +29,8 @@ Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "TensorStack.TextGeneration"
 EndProject
 Project("{2150E333-8FDC-42A3-9474-1A3956D46DE8}") = "Solution Items", "Solution Items", "{02EA681E-C7D8-13C7-8484-4AC65E1B71E8}"
 	ProjectSection(SolutionItems) = preProject
+		BuildDemos.bat = BuildDemos.bat
+		BuildRelease.bat = BuildRelease.bat
 		Directory.Build.props = Directory.Build.props
 	EndProjectSection
 EndProject
@@ -50,6 +52,8 @@ Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "TensorStack.Example.Extract
 EndProject
 Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "TensorStack.Example.TextGeneration", "Examples\TensorStack.Example.TextGeneration\TensorStack.Example.TextGeneration.csproj", "{0B9E97D9-76FA-43BF-8217-7DBE5536EB88}"
 EndProject
+Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "TensorStack.Providers.RyzenAI", "TensorStack.Providers.RyzenAI\TensorStack.Providers.RyzenAI.csproj", "{93621B62-037D-44FA-9FA5-7A9A5566F7FE}"
+EndProject
 Global
 	GlobalSection(SolutionConfigurationPlatforms) = preSolution
 		Debug_CUDA|Any CPU = Debug_CUDA|Any CPU
@@ -218,6 +222,14 @@ Global
 		{0B9E97D9-76FA-43BF-8217-7DBE5536EB88}.Release_CUDA|Any CPU.Build.0 = Release|Any CPU
 		{0B9E97D9-76FA-43BF-8217-7DBE5536EB88}.Release|Any CPU.ActiveCfg = Release|Any CPU
 		{0B9E97D9-76FA-43BF-8217-7DBE5536EB88}.Release|Any CPU.Build.0 = Release|Any CPU
+		{93621B62-037D-44FA-9FA5-7A9A5566F7FE}.Debug_CUDA|Any CPU.ActiveCfg = Debug|Any CPU
+		{93621B62-037D-44FA-9FA5-7A9A5566F7FE}.Debug_CUDA|Any CPU.Build.0 = Debug|Any CPU
+		{93621B62-037D-44FA-9FA5-7A9A5566F7FE}.Debug|Any CPU.ActiveCfg = Debug|Any CPU
+		{93621B62-037D-44FA-9FA5-7A9A5566F7FE}.Debug|Any CPU.Build.0 = Debug|Any CPU
+		{93621B62-037D-44FA-9FA5-7A9A5566F7FE}.Release_CUDA|Any CPU.ActiveCfg = Release|Any CPU
+		{93621B62-037D-44FA-9FA5-7A9A5566F7FE}.Release_CUDA|Any CPU.Build.0 = Release|Any CPU
+		{93621B62-037D-44FA-9FA5-7A9A5566F7FE}.Release|Any CPU.ActiveCfg = Release|Any CPU
+		{93621B62-037D-44FA-9FA5-7A9A5566F7FE}.Release|Any CPU.Build.0 = Release|Any CPU
 	EndGlobalSection
 	GlobalSection(SolutionProperties) = preSolution
 		HideSolutionNode = FALSE