don't provide inputs that are not CoreML model inputs to CoreML

microsoft · Aug 17, 2023 · 2efa90b · 2efa90b
1 parent 5af548a
commit 2efa90b
Show file tree

Hide file tree

Showing 2 changed files with 20 additions and 12 deletions.
diff --git a/onnxruntime/core/providers/coreml/coreml_execution_provider.cc b/onnxruntime/core/providers/coreml/coreml_execution_provider.cc
@@ -104,7 +104,7 @@ common::Status CoreMLExecutionProvider::Compile(const std::vector<FusedNodeAndGr
       for (size_t i = 0, end = input_defs.size(); i < end; ++i) {
         onnx_input_names[i] = input_defs[i]->Name();
       }
-      coreml_model->SetInputs(std::move(onnx_input_names));
+      coreml_model->SetOnnxInputs(std::move(onnx_input_names));
     }
 
     {
@@ -113,7 +113,7 @@ common::Status CoreMLExecutionProvider::Compile(const std::vector<FusedNodeAndGr
       for (size_t i = 0, end = output_defs.size(); i < end; ++i) {
         onnx_output_names[i] = output_defs[i]->Name();
       }
-      coreml_model->SetOutputs(std::move(onnx_output_names));
+      coreml_model->SetOnnxOutputs(std::move(onnx_output_names));
     }
 
     coreml_models_.emplace(fused_node.Name(), std::move(coreml_model));
@@ -136,8 +136,8 @@ common::Status CoreMLExecutionProvider::Compile(const std::vector<FusedNodeAndGr
       const size_t num_outputs = ctx.GetOutputCount();
 
       coreml::Model* model = reinterpret_cast<coreml::Model*>(state);
-      const auto& model_inputs = model->GetInputs();
-      const auto& model_outputs = model->GetOutputs();
+      const auto& model_inputs = model->GetOnnxInputs();
+      const auto& model_outputs = model->GetOnnxOutputs();
 
       ORT_RETURN_IF_NOT(model_inputs.size() <= num_inputs, "Inconsistent input sizes");
       ORT_RETURN_IF_NOT(model_outputs.size() == num_outputs, "Inconsistent output sizes");
@@ -146,14 +146,22 @@ common::Status CoreMLExecutionProvider::Compile(const std::vector<FusedNodeAndGr
       inputs.reserve(model_inputs.size());
       for (size_t i = 0; i < model_inputs.size(); i++) {
         const auto& input_name = model_inputs[i];
+        const auto* input_info = model->TryGetInputOutputInfo(input_name);
+        if (input_info == nullptr) {
+          // The CoreML model may not have an actual input that corresponds to this one.
+          // E.g., when the input is an initializer that already got copied to the CoreML model.
+          // If there's no CoreML model input, we don't need to provide this input to CoreML.
+          continue;
+        }
+
         auto input_tensor = ctx.GetInput(i);
         auto tensor_info = input_tensor.GetTensorTypeAndShapeInfo();
         auto shape = tensor_info.GetShape();
 
         // Disallow inputs with dynamic shape which actually have zero elements.
         // CoreML doesn't consistently handle this well (e.g., there may be runtime errors).
-        if (const auto* model_input_info = model->TryGetInputOutputInfo(input_name); model_input_info != nullptr) {
-          const auto& inferred_shape = model_input_info->shape;
+        {
+          const auto& inferred_shape = input_info->shape;
           ORT_RETURN_IF(!coreml::IsStaticShape(inferred_shape) && coreml::DoesShapeSpecifyZeroElements(shape),
                         "Input (", input_name, ") has a dynamic shape (", coreml::Shape2String(inferred_shape),
                         ") but the runtime shape (", coreml::Shape2String(shape),

diff --git a/onnxruntime/core/providers/coreml/model/model.h b/onnxruntime/core/providers/coreml/model/model.h
@@ -51,11 +51,11 @@ class Model {
   OrtMutex& GetMutex() { return mutex_; }
 
   // Input and output names in the onnx model's order
-  const std::vector<std::string>& GetInputs() const { return inputs_; }
-  void SetInputs(std::vector<std::string>&& inputs) { inputs_ = std::move(inputs); }
+  const std::vector<std::string>& GetOnnxInputs() const { return onnx_inputs_; }
+  void SetOnnxInputs(std::vector<std::string>&& inputs) { onnx_inputs_ = std::move(inputs); }
 
-  const std::vector<std::string>& GetOutputs() const { return outputs_; }
-  void SetOutputs(std::vector<std::string>&& outputs) { outputs_ = std::move(outputs); }
+  const std::vector<std::string>& GetOnnxOutputs() const { return onnx_outputs_; }
+  void SetOnnxOutputs(std::vector<std::string>&& outputs) { onnx_outputs_ = std::move(outputs); }
 
   const OnnxTensorInfo* TryGetInputOutputInfo(const std::string& name) const;
   const OnnxTensorInfo& GetInputOutputInfo(const std::string& name) const;
@@ -65,8 +65,8 @@ class Model {
   std::unordered_set<std::string> scalar_outputs_;
   std::unordered_set<std::string> int64_outputs_;
 
-  std::vector<std::string> inputs_;
-  std::vector<std::string> outputs_;
+  std::vector<std::string> onnx_inputs_;
+  std::vector<std::string> onnx_outputs_;
 
   std::unordered_map<std::string, OnnxTensorInfo> input_output_info_;