Create factory API

intel · Jul 16, 2024 · f2ff0cd · f2ff0cd
1 parent 1931246
commit f2ff0cd
Show file tree

Hide file tree

Showing 2 changed files with 40 additions and 8 deletions.
diff --git a/examples/cpp/main.cpp b/examples/cpp/main.cpp
@@ -14,8 +14,6 @@ int main() {
     std::cout << "Create a ModelFactory" << std::endl;
     auto factory = std::make_shared<ModelFactory>("NPU");
 
-    auto context = factory->get_context();
-
     // create parameter
     auto input = factory->parameter({batch, inC}, ov::element::f16);
     auto weights = factory->parameter({outC, inC}, ov::element::f16);
@@ -34,12 +32,10 @@ int main() {
     // factory->saveModel("matmul.xml");
 
     std::cout << "Creating a remote tensor" << std::endl;
-    auto input_buffer = context.create_l0_host_tensor(ov::element::f16, {batch, inC}, ov::intel_npu::TensorType::INPUT);
-    auto weights_buffer =
-            context.create_l0_host_tensor(ov::element::f16, {outC, inC}, ov::intel_npu::TensorType::INPUT);
-    auto bias_buffer = context.create_l0_host_tensor(ov::element::f16, {1, outC}, ov::intel_npu::TensorType::INPUT);
-    auto output_buffer =
-            context.create_l0_host_tensor(ov::element::f16, {batch, outC}, ov::intel_npu::TensorType::OUTPUT);
+    auto input_buffer = factory->createRemoteInputTensor(0);
+    auto weights_buffer = factory->createRemoteInputTensor(1);
+    auto bias_buffer = factory->createRemoteInputTensor(2);
+    auto output_buffer = factory->createRemoteOutputTensor(0);
 
     std::memset(input_buffer.get(), 0, input_buffer.get_byte_size());
     std::memset(weights_buffer.get(), 0, weights_buffer.get_byte_size());

diff --git a/include/intel_npu_acceleration_library/inference.h b/include/intel_npu_acceleration_library/inference.h
@@ -173,6 +173,42 @@ class OVInferenceModel {
         }
     }
 
+    /**
+     * @brief Create a Remote Tensor object
+     *
+     * @param type element type
+     * @param shape element shape
+     * @param tensor_type element tensor type: INPUT, OUTPUT, BIND
+     * @return auto
+     */
+    auto createRemoteTensor(const ov::element::Type type, const ov::Shape& shape,
+                            const ov::intel_npu::TensorType tensor_type) {
+        ov::intel_npu::level_zero::ZeroContext context = get_context();
+        return context.create_l0_host_tensor(type, shape, tensor_type);
+    }
+
+    /**
+     * @brief Create a Remote Tensor object
+     *
+     * @param idx index of the input tensor
+     * @return auto
+     */
+    auto createRemoteInputTensor(size_t idx) {
+        auto tensor = infer_request.get_input_tensor(idx);
+        return createRemoteTensor(tensor.get_element_type(), tensor.get_shape(), ov::intel_npu::TensorType::INPUT);
+    }
+
+    /**
+     * @brief Create a Remote Tensor object
+     *
+     * @param idx index of the output tensor
+     * @return auto
+     */
+    auto createRemoteOutputTensor(size_t idx) {
+        auto tensor = infer_request.get_output_tensor(idx);
+        return createRemoteTensor(tensor.get_element_type(), tensor.get_shape(), ov::intel_npu::TensorType::OUTPUT);
+    }
+
     /**
      * @brief Get model input tensor
      *