add python share_data interface (#41626)

* add python share_data interface

* Update inference_api.cc

* Update inference_api.cc

* add python share_data interface

paddle/fluid/inference/api/paddle_api.h

@@ -36,6 +36,7 @@ namespace paddle {
 
 using PaddleDType = paddle_infer::DataType;
 using PaddlePlace = paddle_infer::PlaceType;
+using PaddleDataLayout = paddle_infer::DataLayout;
 
 /// \brief Memory manager for PaddleTensor.
 ///

paddle/fluid/pybind/inference_api.cc

@@ -76,6 +76,7 @@ using paddle::NativeConfig;
 using paddle::NativePaddlePredictor;
 using paddle::PaddleBuf;
 using paddle::PaddleDType;
+using paddle::PaddleDataLayout;
 using paddle::PaddlePassBuilder;
 using paddle::PaddlePlace;
 using paddle::PaddlePredictor;
@@ -85,6 +86,7 @@ using paddle::ZeroCopyTensor;
 
 namespace {
 void BindPaddleDType(py::module *m);
+void BindPaddleDataLayout(py::module *m);
 void BindPaddleBuf(py::module *m);
 void BindPaddleTensor(py::module *m);
 void BindPaddlePlace(py::module *m);
@@ -211,6 +213,34 @@ void PaddleInferTensorCreate(
   tensor.CopyFromCpu(static_cast<const T *>(data.data()));
 }
 
+paddle_infer::PlaceType ToPaddleInferPlace(
+    phi::AllocationType allocation_type) {
+  if (allocation_type == phi::AllocationType::CPU) {
+    return paddle_infer::PlaceType::kCPU;
+  } else if (allocation_type == phi::AllocationType::GPU) {
+    return paddle_infer::PlaceType::kGPU;
+  } else {
+    return paddle_infer::PlaceType::kCPU;
+  }
+}
+
+void PaddleInferShareExternalData(paddle_infer::Tensor &tensor,  // NOLINT
+                                  framework::Tensor input_tensor) {
+  std::vector<int> shape;
+  for (int i = 0; i < input_tensor.dims().size(); ++i) {
+    shape.push_back(input_tensor.dims()[i]);
+  }
+  if (input_tensor.dtype() == phi::DataType::FLOAT32) {
+    tensor.ShareExternalData(
+        static_cast<float *>(input_tensor.data()), shape,
+        ToPaddleInferPlace(input_tensor.place().GetType()));
+  } else if (input_tensor.dtype() == phi::DataType::FLOAT16) {
+    tensor.ShareExternalData(
+        static_cast<paddle::platform::float16 *>(input_tensor.data()), shape,
+        ToPaddleInferPlace(input_tensor.place().GetType()));
+  }
+}
+
 /// \brief Experimental interface.
 /// Create the Strings tensor from data.
 /// \param tensor The tensor will be created and
@@ -327,6 +357,7 @@ void CopyPaddleInferTensor(paddle_infer::Tensor &dst,  // NOLINT
 
 void BindInferenceApi(py::module *m) {
   BindPaddleDType(m);
+  BindPaddleDataLayout(m);
   BindPaddleBuf(m);
   BindPaddleTensor(m);
   BindPaddlePlace(m);
@@ -372,6 +403,14 @@ void BindPaddleDType(py::module *m) {
       .value("INT32", PaddleDType::INT32);
 }
 
+void BindPaddleDataLayout(py::module *m) {
+  py::enum_<PaddleDataLayout>(*m, "PaddleDataLayout")
+      .value("UNK", PaddleDataLayout::kUNK)
+      .value("Any", PaddleDataLayout::kAny)
+      .value("NHWC", PaddleDataLayout::kNHWC)
+      .value("NCHW", PaddleDataLayout::kNCHW);
+}
+
 void BindPaddleBuf(py::module *m) {
   py::class_<PaddleBuf>(*m, "PaddleBuf")
       .def(py::init<size_t>())
@@ -817,6 +856,7 @@ void BindPaddleInferTensor(py::module *m) {
       .def("copy_from_cpu_bind",
            &PaddleInferTensorCreate<paddle_infer::float16>)
       .def("copy_from_cpu_bind", &PaddleInferStringTensorCreate)
+      .def("share_external_data_bind", &PaddleInferShareExternalData)
       .def("copy_to_cpu", &PaddleInferTensorToNumpy)
       .def("shape", &paddle_infer::Tensor::shape)
       .def("set_lod", &paddle_infer::Tensor::SetLoD)

python/paddle/fluid/inference/wrapper.py

@@ -14,6 +14,7 @@
 
 from ..core import AnalysisConfig, PaddleDType, PaddlePlace
 from ..core import PaddleInferPredictor, PaddleInferTensor
+from .. import core
 
 import numpy as np
 
@@ -39,4 +40,16 @@ def tensor_copy_from_cpu(self, data):
         )
 
 
+def tensor_share_external_data(self, data):
+    '''
+    Support input type check based on tensor.share_external_data.
+    '''
+    if isinstance(data, core.LoDTensor):
+        self.share_external_data_bind(data)
+    else:
+        raise TypeError(
+            "In share_external_data, we only support LoDTensor data type.")
+
+
 Tensor.copy_from_cpu = tensor_copy_from_cpu
+Tensor.share_external_data = tensor_share_external_data