[NNAdapter][IntelOpenVINO] Init support for OpenVINO by csy0225 · Pull Request #8552 · PaddlePaddle/Paddle-Lite

csy0225 · 2022-03-01T12:11:04Z

编译命令
./lite/tools/build_linux.sh --arch=x86 --with_nnadapter=ON --nnadapter_with_intel_openvino=ON --nnadapter_intel_openvino_sdk_root=${nnadapter_intel_openvino_sdk_root}

paddle-bot-old · 2022-03-01T12:11:09Z

Thanks for your contribution!

hong19860320 · 2022-03-01T13:01:21Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/converter/batch_normalization.cc

+  auto mean_node = converter->ConvertToOutputNode(mean_operand);
+  auto variance_node = converter->ConvertToOutputNode(variance_operand);
+  // Create <BatchNormInference> Node for Intel OpenVINO
+  std::shared_ptr<Node> node =


auto batch_norm_op

hong19860320 · 2022-03-01T13:01:37Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/converter/batch_normalization.cc

+  auto beta_node = converter->ConvertToOutputNode(bias_operand);
+  auto mean_node = converter->ConvertToOutputNode(mean_operand);
+  auto variance_node = converter->ConvertToOutputNode(variance_operand);
+  // Create <BatchNormInference> Node for Intel OpenVINO


这一类注释可以去掉

hong19860320 · 2022-03-01T13:03:00Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/converter/converter.cc

+  return output_node;
+}
+
+std::shared_ptr<OutputNode> Converter::ConvertToOutputNode(


直接用ConvertOperand就行了，不用强调output node

hong19860320 · 2022-03-01T13:05:21Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/converter/batch_normalization.cc

+                              core::Operation* operation) {
+  BATCH_NORMALIZATION_OPERATION_EXTRACT_INPUTS_OUTPUTS
+
+  // Convert operand to Intel OpenVINO's OutputNode


Convert to OpenVINO nodes

hong19860320 · 2022-03-01T13:05:34Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/converter/batch_normalization.cc

+  BATCH_NORMALIZATION_OPERATION_EXTRACT_INPUTS_OUTPUTS
+
+  // Convert operand to Intel OpenVINO's OutputNode
+  auto input_node = converter->GetMappedOutputNode(input_operand);


直接用GetMappedNode

未修改，OutputNode用来修饰方法，变量名用tensor代替

hong19860320 · 2022-03-01T13:48:09Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/engine.h

+  std::shared_ptr<ov::Core> ov_core_{nullptr};
+  std::map<core::Operand*, std::vector<std::shared_ptr<OutputNode>>>
+      output_nodes_;
+  std::vector<std::shared_ptr<default_opset::Parameter>> parameter_nodes_;


为啥parameter_nodes_，这里不加 ov_前缀？是不是得统一下？我建议都不加

hong19860320 · 2022-03-01T13:48:30Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/engine.h

+  Context* context_{nullptr};
+  std::vector<NNAdapterOperandType> input_types_;
+  std::vector<NNAdapterOperandType> output_types_;
+  std::shared_ptr<ov::Core> ov_core_{nullptr};


runtime_core_

hong19860320 · 2022-03-01T13:48:43Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/engine.h

+      output_nodes_;
+  std::vector<std::shared_ptr<default_opset::Parameter>> parameter_nodes_;
+  std::vector<std::shared_ptr<Node>> result_nodes_;
+  std::shared_ptr<ov::CompiledModel> compiled_ov_model_{nullptr};


compiled_model_

hong19860320 · 2022-03-01T13:49:53Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/utility.cc

+
+ElementType ConvertToOVElementType(
+    const NNAdapterOperandPrecisionCode& precision_code) {
+  std::map<NNAdapterOperandPrecisionCode, ElementType> precision_map{


为啥要用map，用switch 不是更简单更快吗？

hong19860320 · 2022-03-01T13:51:52Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/utility.cc

+
+#define FUNCTION_ADD_CONST_OUTPUT_NODE_DEFINE(type, element_type) \
+  template <>                                                     \
+  std::shared_ptr<OutputNode> AddConstOutputNode<type>(           \


用CreateOVConstantNode或AddOVConstantNode？

hong19860320 · 2022-03-01T13:54:38Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/converter/converter.cc

+    UpdateOutputNodeMap(operand, output_node);
+    return output_node;
+  }
+  NNADAPTER_LOG(FATAL) << "Only constant and model input operands can be "


直接用OpenVINO吧，不要加Intel了，有点显得多余

hong19860320 · 2022-03-01T14:01:02Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/utility.h

+  }
+  return Shape(ov_shape);
+}
+// Add const ov::Node and return ov::Output<ov::Node>


为啥不直接在utility.h定义
// Convert C/C++ POD types to ElementType
template
ElementType GetElementType() {
NNADAPTER_LOG(FATAL) << "Unable to convert " << typeid(T).name() << " to ElementType";
return ov::element::f32;
}
template <>
ElementType GetElementType();
template <>
ElementType GetElementType<int8_t>();
template <>
ElementType GetElementType<int16_t>();
....

template
std::shared_ptr AddConstOutputNode(
std::vector<size_t> dimensions, std::vector values) {
auto const_node = std::make_shared<default_opset::Constant>(
GetElementType(), Shape(dimensions), values);
return std::make_shared(const_node->output(0));
}

hong19860320 · 2022-03-01T14:03:05Z

lite/kernels/nnadapter/converter/all.h

-    ConvertElementwise,
-    "huawei_ascend_npu,huawei_kirin_npu,imagination_nna,kunlunxin_xtcl");
+                   "kunlunxin_xtcl,cambricon_mlu,android_nnapi,intel_openvino");
+REGISTER_CONVERTER(elementwise_max,


这个文件好好检查下，不要误删其它硬件

hong19860320 · 2022-03-03T06:04:42Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/converter/all.h

+REGISTER_CONVERTER(SUB, ConvertElementwise)
+REGISTER_CONVERTER(TANH, ConvertUnaryActivations)
+
+#endif  // LITE_BACKENDS_NNADAPTER_NNADAPTER_SRC_DRIVER_INTEL_OPENVINO_CONVERTER_ALL_H_


NNADAPTER_DRIVER_INTEL_OPENVINO_CONVERTER_ALL_H

hong19860320 · 2022-03-03T06:06:11Z

lite/backends/nnadapter/nnadapter/src/driver/intel_openvino/converter/softmax.cc

+int ConvertSoftmax(Converter* converter, core::Operation* operation) {
+  SOFTMAX_OPERATION_EXTRACT_INPUTS_OUTPUTS
+
+  // Convert operand to OpenVINO Tensor


Tensor 第一个字母小写，其它地方都改一下

hong19860320

LGTM

NNAdapter support Intel OpenVINO

a25fdaf

csy0225 force-pushed the nnadapter_support_intel_openvino branch 2 times, most recently from 40f9e35 to da793e4 Compare March 1, 2022 12:31

NNAdapter support Intel OpenVINO

73daabf

csy0225 force-pushed the nnadapter_support_intel_openvino branch from da793e4 to 73daabf Compare March 1, 2022 12:36

hong19860320 reviewed Mar 1, 2022

View reviewed changes

csy0225 added 2 commits March 2, 2022 02:59

Add map_output_node define to decrease code line num

916791d

update comments

537f450

csy0225 force-pushed the nnadapter_support_intel_openvino branch 2 times, most recently from 558a882 to 8b5ea93 Compare March 2, 2022 09:26

using tensor=OutputNode, using operator=Node

4b887ee

csy0225 force-pushed the nnadapter_support_intel_openvino branch from 8b5ea93 to 2fae09d Compare March 3, 2022 06:03

hong19860320 reviewed Mar 3, 2022

View reviewed changes

hong19860320 changed the title ~~NNAdapter support Intel OpenVINO~~ [NNAdapter] Init support for OpenVINO Mar 3, 2022

hong19860320 changed the title ~~[NNAdapter] Init support for OpenVINO~~ [NNAdapter][IntelOpenVINO] Init support for OpenVINO Mar 3, 2022

merge develop into branch

1d39f14

csy0225 force-pushed the nnadapter_support_intel_openvino branch from 2fae09d to 1d39f14 Compare March 3, 2022 06:19

hong19860320 approved these changes Mar 3, 2022

View reviewed changes

hong19860320 merged commit afa3c18 into PaddlePaddle:develop Mar 3, 2022

Conversation

csy0225 commented Mar 1, 2022

Uh oh!

paddle-bot-old bot commented Mar 1, 2022

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

hong19860320 left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants