Use generic ov::Tensor for mapping buffers

The set_input_tensor uses the generic type ov::Tensor for it's function
argument. Thus, the remote_tensor type gets passed as ov::Tensor.
Hence, the tensor_handle_map_ can use ov::Tensor for the tensor mappings.

Signed-off-by: Anoob Anto Kodankandath <[email protected]>

Author: akodanka

litert/vendors/intel_openvino/dispatch/device_context.h

@@ -42,9 +42,9 @@ class LiteRtDispatchDeviceContextT {
       LiteRtTensorBufferHandle tensor_buffer_handle);
 
 #if defined(LITERT_WINDOWS_OS)
-  litert::Expected<ov::intel_npu::level_zero::ZeroBufferTensor> getRemoteTensor(
+  litert::Expected<ov::intel_npu::level_zero::ZeroBufferTensor> getOvTensor(
 #else
-  litert::Expected<ov::RemoteTensor> getRemoteTensor(
+  litert::Expected<ov::Tensor> getOvTensor(
 #endif
       const LiteRtTensorBufferHandle& handle) const {
     auto it = tensor_handle_map_.find(handle);
@@ -68,7 +68,7 @@ class LiteRtDispatchDeviceContextT {
                      ov::intel_npu::level_zero::ZeroBufferTensor>
       tensor_handle_map_;
 #else
-  std::unordered_map<LiteRtTensorBufferHandle, ov::RemoteTensor>
+  std::unordered_map<LiteRtTensorBufferHandle, ov::Tensor>
       tensor_handle_map_;
 #endif
   uint64_t next_handle_;

litert/vendors/intel_openvino/dispatch/invocation_context.cc

@@ -103,33 +103,33 @@ litert::Expected<void> LiteRtDispatchInvocationContextT::AttachInput(
     int graph_input_index, LiteRtTensorBufferHandle tensor_buffer_handle) {
 #if defined(LITERT_WINDOWS_OS)
   LITERT_ASSIGN_OR_RETURN(
-      ov::intel_npu::level_zero::ZeroBufferTensor remote_tensor,
-      device_context_.getRemoteTensor(tensor_buffer_handle));
+      ov::intel_npu::level_zero::ZeroBufferTensor ov_tensor,
+      device_context_.getOvTensor(tensor_buffer_handle));
 #else
   LITERT_ASSIGN_OR_RETURN(
-      ov::RemoteTensor remote_tensor,
-      device_context_.getRemoteTensor(tensor_buffer_handle));
+      ov::Tensor ov_tensor,
+      device_context_.getOvTensor(tensor_buffer_handle));
 #endif
   // TODO: visit this if need to maintain graph indices for inputs and outputs
   // in dispatch_api
-  infer_request_.set_input_tensor(graph_input_index, remote_tensor);
+  infer_request_.set_input_tensor(graph_input_index, ov_tensor);
   return {};
 }
 
 litert::Expected<void> LiteRtDispatchInvocationContextT::AttachOutput(
     int graph_output_index, LiteRtTensorBufferHandle tensor_buffer_handle) {
 #if defined(LITERT_WINDOWS_OS)
   LITERT_ASSIGN_OR_RETURN(
-      ov::intel_npu::level_zero::ZeroBufferTensor remote_tensor,
-      device_context_.getRemoteTensor(tensor_buffer_handle));
+      ov::intel_npu::level_zero::ZeroBufferTensor ov_tensor,
+      device_context_.getOvTensor(tensor_buffer_handle));
 #else
   LITERT_ASSIGN_OR_RETURN(
-      ov::RemoteTensor remote_tensor,
-      device_context_.getRemoteTensor(tensor_buffer_handle));
+      ov::Tensor ov_tensor,
+      device_context_.getOvTensor(tensor_buffer_handle));
 #endif
   // TODO: visit this if need to maintain graph indices for inputs and outputs
   // in dispatch_api
-  infer_request_.set_output_tensor(graph_output_index, remote_tensor);
+  infer_request_.set_output_tensor(graph_output_index, ov_tensor);
   return {};
 }