pybind.cc 57.1 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
2 3 4 5 6

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

7
http://www.apache.org/licenses/LICENSE-2.0
8 9 10 11 12 13

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
L
lgone2000 已提交
14
#include <Python.h>
C
chengduoZH 已提交
15 16
#include <algorithm>
#include <map>
S
sneaxiy 已提交
17
#include <memory>
C
chengduoZH 已提交
18 19 20 21 22
#include <mutex>  // NOLINT // for call_once
#include <string>
#include <unordered_map>
#include <utility>
#include <vector>
23

Y
Yi Wang 已提交
24 25 26
#include "paddle/fluid/framework/executor.h"
#include "paddle/fluid/framework/feed_fetch_method.h"
#include "paddle/fluid/framework/framework.pb.h"
S
sneaxiy 已提交
27
#include "paddle/fluid/framework/garbage_collector.h"
28
#include "paddle/fluid/framework/ir/pass_builder.h"
Y
Yi Wang 已提交
29 30 31
#include "paddle/fluid/framework/lod_rank_table.h"
#include "paddle/fluid/framework/lod_tensor.h"
#include "paddle/fluid/framework/lod_tensor_array.h"
32
#include "paddle/fluid/framework/op_registry.h"
Y
Yu Yang 已提交
33
#include "paddle/fluid/framework/parallel_executor.h"
Y
Yi Wang 已提交
34
#include "paddle/fluid/framework/prune.h"
Y
Refine  
Yu Yang 已提交
35
#include "paddle/fluid/framework/reader.h"
S
sneaxiy 已提交
36
#include "paddle/fluid/framework/scope_pool.h"
Y
Yi Wang 已提交
37
#include "paddle/fluid/framework/selected_rows.h"
X
Xin Pan 已提交
38
#include "paddle/fluid/framework/version.h"
39
#include "paddle/fluid/imperative/layer.h"
M
minqiyang 已提交
40
#include "paddle/fluid/imperative/profiler.h"
Y
Refine  
Yu Yang 已提交
41
#include "paddle/fluid/memory/allocation/allocator_strategy.h"
42
#include "paddle/fluid/memory/allocation/legacy_allocator.h"
D
dzhwinter 已提交
43
#include "paddle/fluid/operators/activation_op.h"
S
sneaxiy 已提交
44
#include "paddle/fluid/operators/py_func_op.h"
S
sneaxiy 已提交
45
#include "paddle/fluid/operators/reader/lod_tensor_blocking_queue.h"
Y
Yu Yang 已提交
46
#include "paddle/fluid/platform/cpu_info.h"
Y
Yi Wang 已提交
47
#include "paddle/fluid/platform/enforce.h"
48
#include "paddle/fluid/platform/init.h"
Y
Yi Wang 已提交
49 50
#include "paddle/fluid/platform/place.h"
#include "paddle/fluid/platform/profiler.h"
W
Wang Guibao 已提交
51
#include "paddle/fluid/pybind/async_executor_py.h"
Y
Yi Wang 已提交
52 53
#include "paddle/fluid/pybind/const_value.h"
#include "paddle/fluid/pybind/exception.h"
D
dongdaxiang 已提交
54
#include "paddle/fluid/pybind/fleet_wrapper_py.h"
55
#include "paddle/fluid/pybind/imperative.h"
F
flame 已提交
56
#include "paddle/fluid/pybind/inference_api.h"
F
flame 已提交
57
#include "paddle/fluid/pybind/ir.h"
58 59
#include "paddle/fluid/pybind/protobuf.h"
#include "paddle/fluid/pybind/pybind.h"  // NOLINT
S
sneaxiy 已提交
60
#include "paddle/fluid/pybind/reader_py.h"
Y
Yu Yang 已提交
61
#include "paddle/fluid/pybind/recordio.h"
Y
Yi Wang 已提交
62
#include "paddle/fluid/pybind/tensor_py.h"
63
#include "paddle/fluid/string/to_string.h"
64

D
Dong Zhihong 已提交
65
#ifdef PADDLE_WITH_CUDA
P
peizhilin 已提交
66
#ifndef _WIN32
Y
Yi Wang 已提交
67
#include "paddle/fluid/operators/nccl/nccl_gpu_common.h"
P
peizhilin 已提交
68
#endif
Y
Yi Wang 已提交
69 70
#include "paddle/fluid/platform/cuda_profiler.h"
#include "paddle/fluid/platform/gpu_info.h"
D
Dong Zhihong 已提交
71 72
#endif

M
minqiyang 已提交
73 74
#include "pybind11/stl.h"

75 76 77 78
DEFINE_bool(reader_queue_speed_test_mode, false,
            "If set true, the queue.pop will only get data from queue but not "
            "remove the data from queue for speed testing");

Q
Qiao Longfei 已提交
79 80 81
// disable auto conversion to list in Python
PYBIND11_MAKE_OPAQUE(paddle::framework::LoDTensorArray);

82
namespace paddle {
83
namespace pybind {
84
bool IsCompiledWithCUDA() {
85
#ifndef PADDLE_WITH_CUDA
Q
qijun 已提交
86 87 88 89 90 91
  return false;
#else
  return true;
#endif
}

92 93 94 95 96 97 98 99
bool IsCompiledWithMKLDNN() {
#ifndef PADDLE_WITH_MKLDNN
  return false;
#else
  return true;
#endif
}

100 101 102 103 104 105 106 107
bool IsCompiledWithNGRAPH() {
#ifndef PADDLE_WITH_NGRAPH
  return false;
#else
  return true;
#endif
}

108
bool IsCompiledWithBrpc() {
109
#ifndef PADDLE_WITH_DISTRIBUTE
110 111
  return false;
#endif
112 113 114 115 116 117

#ifdef PADDLE_WITH_GRPC
  return false;
#endif

  return true;
118 119
}

Y
update  
Yancey1989 已提交
120
bool IsCompiledWithDIST() {
Y
Yancey1989 已提交
121
#ifdef PADDLE_WITH_DISTRIBUTE
Y
update  
Yancey1989 已提交
122 123 124 125 126 127
  return true;
#else
  return false;
#endif
}

S
sneaxiy 已提交
128 129 130 131 132 133 134 135 136 137
template <typename PlaceType1, typename PlaceType2>
static inline bool IsSamePlace(const PlaceType1 &p1, const PlaceType2 &p2) {
  return paddle::platform::Place(p1) == paddle::platform::Place(p2);
}

template <typename PlaceType>
static inline int PlaceIndex(const PlaceType &p) {
  return static_cast<int>(paddle::platform::Place(p).which());
}

138
PYBIND11_MODULE(core, m) {
Y
Yu Yang 已提交
139 140 141
  // Not used, just make sure cpu_info.cc is linked.
  paddle::platform::CpuTotalPhysicalMemory();

Y
Refine  
Yu Yang 已提交
142
  paddle::memory::allocation::UseAllocatorStrategyGFlag();
S
sneaxiy 已提交
143

144
  m.doc() = "C++ core of PaddlePaddle";
145

146 147 148 149
  // using framework in this function. Since it is inside a function, it will
  // not cause namespace pollution.
  using namespace paddle::framework;  // NOLINT

150
  BindException(&m);
Y
Yu Yang 已提交
151

S
sneaxiy 已提交
152
  m.def(
S
sneaxiy 已提交
153
      "_append_python_callable_object_and_return_id",
S
sneaxiy 已提交
154 155 156 157
      [](py::object py_obj) -> size_t {
        return paddle::operators::AppendPythonCallableObjectAndReturnId(py_obj);
      });

S
sneaxiy 已提交
158 159 160
  // NOTE(zjl): ctest would load environment variables at the beginning even
  // though we have not `import paddle.fluid as fluid`. So we add this API
  // to enable eager deletion mode in unittest.
S
sneaxiy 已提交
161
  m.def("_set_eager_deletion_mode", &paddle::framework::SetEagerDeletionMode);
S
sneaxiy 已提交
162

S
sneaxiy 已提交
163 164 165
  m.add_object("_cleanup",
               py::capsule([]() { ScopePool::Instance().Clear(); }));

166 167 168 169 170 171 172
  m.def("get_mem_usage", [](int device) {
    return memory::allocation::GPUMemMonitor.GetMemUsage(device);
  });

  m.def("print_mem_usage",
        []() { return memory::allocation::GPUMemMonitor.PrintMemUsage(); });

M
minqiyang 已提交
173
  m.def("start_imperative_gperf_profiler",
M
minqiyang 已提交
174 175
        []() { imperative::StartProfile(); });

M
minqiyang 已提交
176
  m.def("stop_imperative_gperf_profiler", []() { imperative::StopProfile(); });
M
minqiyang 已提交
177

M
minqiyang 已提交
178
  py::class_<imperative::VarBase>(m, "VarBase", R"DOC()DOC")
179 180 181 182 183 184 185 186
      .def(
          py::init<const std::string &, paddle::framework::proto::VarType::Type,
                   const std::vector<int64_t>, const paddle::platform::CPUPlace,
                   bool, bool>())
      .def(
          py::init<const std::string &, paddle::framework::proto::VarType::Type,
                   const std::vector<int64_t>,
                   const paddle::platform::CUDAPlace, bool, bool>())
187
      .def("_run_backward",
X
Xin Pan 已提交
188
           [](imperative::VarBase &self) { self.RunBackward(); })
M
minqiyang 已提交
189
      .def("_grad_name", &imperative::VarBase::GradName)
M
minqiyang 已提交
190
      .def("_grad_value", &imperative::VarBase::GradValue)
X
Xin Pan 已提交
191
      .def("_clear_gradient", &imperative::VarBase::ClearGradient)
M
minqiyang 已提交
192
      .def("_grad_ivar",
M
minqiyang 已提交
193
           [](const imperative::VarBase &self) { return self.grads_; },
M
minqiyang 已提交
194
           py::return_value_policy::reference)
M
minqiyang 已提交
195
      .def("_copy_to",
P
Paddle CI 已提交
196
           [](const imperative::VarBase &self, const platform::CPUPlace &place,
M
minqiyang 已提交
197 198 199 200 201
              bool blocking) {
             std::unique_ptr<imperative::VarBase> new_var =
                 self.NewVarBase(place, blocking);
             return new_var.release();
           },
P
Paddle CI 已提交
202
           py::return_value_policy::take_ownership)
M
minqiyang 已提交
203
      .def("_copy_to",
P
Paddle CI 已提交
204
           [](const imperative::VarBase &self, const platform::CUDAPlace &place,
M
minqiyang 已提交
205 206 207 208 209
              bool blocking) {
             std::unique_ptr<imperative::VarBase> new_var =
                 self.NewVarBase(place, blocking);
             return new_var.release();
           },
M
minqiyang 已提交
210
           py::return_value_policy::take_ownership)
M
minqiyang 已提交
211
      .def("value", [](const imperative::VarBase &self) { return self.var_; },
M
minqiyang 已提交
212
           py::return_value_policy::reference)
213 214 215
      .def_property("name", &imperative::VarBase::Name,
                    &imperative::VarBase::SetName)
      .def_property_readonly("shape", &imperative::VarBase::Shape)
M
minqiyang 已提交
216
      .def_property_readonly("dtype", &imperative::VarBase::DataType)
217 218 219 220
      .def_property("persistable", &imperative::VarBase::IsPersistable,
                    &imperative::VarBase::SetPersistable)
      .def_property("stop_gradient", &imperative::VarBase::IsStopGradient,
                    &imperative::VarBase::SetStopGradient);
221

222
  py::class_<imperative::OpBase, PyOpBase>(m, "OpBase", R"DOC()DOC")
223
      .def(py::init<const std::string &>())
224 225 226 227
      .def("register_backward_hooks",
           [](imperative::OpBase &self, const py::object &callable) {
             self.RegisterBackwardHooks(callable);
           })
M
minqiyang 已提交
228 229 230 231 232 233 234 235 236 237
      .def_property("_trace_id",
                    [](const imperative::OpBase &self) {
                      pybind11::gil_scoped_release release;
                      return self.trace_id_;
                    },
                    [](imperative::OpBase &self, int trace_id) {
                      pybind11::gil_scoped_release release;
                      self.trace_id_ = trace_id;
                    },
                    py::return_value_policy::reference)
X
Xin Pan 已提交
238 239 240 241 242 243
      .def_property(
          "forward_id",
          [](const imperative::OpBase &self) { return self.forward_id_; },
          [](imperative::OpBase &self, int forward_id) {
            self.forward_id_ = forward_id;
          },
X
Xin Pan 已提交
244
          py::return_value_policy::reference)
X
polish  
Xin Pan 已提交
245
      .def_property_readonly("type", &imperative::OpBase::Type)
X
Xin Pan 已提交
246 247 248 249 250 251
      .def_property(
          "backward_id",
          [](const imperative::OpBase &self) { return self.backward_id_; },
          [](imperative::OpBase &self, int backward_id) {
            self.backward_id_ = backward_id;
          },
252 253
          py::return_value_policy::reference);

X
Xin Pan 已提交
254
  py::class_<imperative::Layer, Layer /* <--- trampoline*/> layer(m, "Layer");
255
  layer.def(py::init<>())
X
Xin Pan 已提交
256 257 258
      .def("forward", [](imperative::Layer &self,
                         const std::vector<imperative::VarBase> &inputs) {
        return self.Forward(inputs);
X
Xin Pan 已提交
259
      });
X
Xin Pan 已提交
260

X
polish  
Xin Pan 已提交
261
  py::class_<imperative::PyLayer>(m, "PyLayer")
X
Xin Pan 已提交
262
      .def(py::init<>())
X
Xin Pan 已提交
263 264
      .def_static(
          "apply",
X
Xin Pan 已提交
265
          [](int func_id, const std::vector<imperative::VarBase *> &inputs)
X
Xin Pan 已提交
266
              -> std::vector<imperative::VarBase *> {
267 268 269 270 271 272 273 274 275 276 277
                auto ret_vars = imperative::PyLayer::Apply(func_id, inputs);
                std::vector<imperative::VarBase *> outputs;
                outputs.reserve(ret_vars.size());
                for (size_t i = 0U; i != ret_vars.size(); ++i) {
                  framework::Variable *v = ret_vars[i];
                  // TODO(minqiyang): use unique_name generator to set a name
                  outputs.emplace_back(
                      new imperative::VarBase("", v, nullptr, true));
                }

                return outputs;
X
Xin Pan 已提交
278 279
              },
          py::return_value_policy::take_ownership)
X
polish  
Xin Pan 已提交
280 281 282 283 284
      .def_static("register_func",
                  [](int func_id, const py::object &callable) {
                    imperative::PyLayer::RegisterFunc(func_id, callable);
                  })
      .def_static("num_funcs", &imperative::PyLayer::NumFuncs);
X
Xin Pan 已提交
285

286 287
  BindTracer(&m);

288 289 290
  py::class_<Tensor>(m, "Tensor", py::buffer_protocol())
      .def_buffer(
          [](Tensor &self) -> py::buffer_info { return CastToPyBuffer(self); })
S
sneaxiy 已提交
291 292
      .def("_is_initialized",
           [](const Tensor &self) { return self.IsInitialized(); })
Y
yuyang18 已提交
293
      .def("_get_dims",
294
           [](const Tensor &self) { return vectorize(self.dims()); })
Y
yuyang18 已提交
295
      .def("_set_dims",
Q
qijun 已提交
296
           [](Tensor &self, const std::vector<int64_t> &dim) {
Y
Yu Yang 已提交
297
             self.Resize(make_ddim(dim));
Y
Yu Yang 已提交
298
           })
Y
yuyang18 已提交
299
      .def("_set_layout",
D
dzhwinter 已提交
300 301 302
           [](Tensor &self, const std::string &layout) {
             self.set_layout(StringToDataLayout(layout));
           })
Y
yuyang18 已提交
303
      .def("_alloc_float",
D
dzhwinter 已提交
304
           [](Tensor &self, paddle::platform::CUDAPlace &place) {
Q
qijun 已提交
305
             self.mutable_data<float>(place);
Y
Yu Yang 已提交
306
           })
Y
yuyang18 已提交
307
      .def("_alloc_float",
Y
Yu Yang 已提交
308
           [](Tensor &self, paddle::platform::CPUPlace &place) {
Q
qijun 已提交
309
             self.mutable_data<float>(place);
Y
Yu Yang 已提交
310
           })
Y
yuyang18 已提交
311
      .def("_alloc_int",
Y
Yu Yang 已提交
312
           [](Tensor &self, paddle::platform::CPUPlace &place) {
Q
qijun 已提交
313
             self.mutable_data<int>(place);
Y
Yu Yang 已提交
314
           })
Y
yuyang18 已提交
315
      .def("_alloc_int",
D
dzhwinter 已提交
316
           [](Tensor &self, paddle::platform::CUDAPlace &place) {
Q
qijun 已提交
317
             self.mutable_data<int>(place);
Q
qijun 已提交
318
           })
Y
yuyang18 已提交
319
      .def("_alloc_int",
C
chengduoZH 已提交
320 321 322
           [](Tensor &self, paddle::platform::CUDAPinnedPlace &place) {
             self.mutable_data<int>(place);
           })
Y
yuyang18 已提交
323
      .def("_alloc_float",
C
chengduoZH 已提交
324 325 326
           [](Tensor &self, paddle::platform::CUDAPinnedPlace &place) {
             self.mutable_data<float>(place);
           })
Y
Yu Yang 已提交
327 328
      .def("set", PyCPUTensorSetFromArray<float>)
      .def("set", PyCPUTensorSetFromArray<int>)
329
      .def("set", PyCPUTensorSetFromArray<double>)
330
      .def("set", PyCPUTensorSetFromArray<int64_t>)
Y
Yu Yang 已提交
331
      .def("set", PyCPUTensorSetFromArray<bool>)
332
      .def("set", PyCPUTensorSetFromArray<uint16_t>)
F
fengjiayi 已提交
333
      .def("set", PyCPUTensorSetFromArray<uint8_t>)
Q
qingqing01 已提交
334
      .def("set", PyCPUTensorSetFromArray<int8_t>)
335
#ifdef PADDLE_WITH_CUDA
Y
Yu Yang 已提交
336 337
      .def("set", PyCUDATensorSetFromArray<float>)
      .def("set", PyCUDATensorSetFromArray<int>)
338
      .def("set", PyCUDATensorSetFromArray<double>)
339
      .def("set", PyCUDATensorSetFromArray<int64_t>)
Y
Yu Yang 已提交
340
      .def("set", PyCUDATensorSetFromArray<bool>)
341
      .def("set", PyCUDATensorSetFromArray<uint16_t>)
F
fengjiayi 已提交
342
      .def("set", PyCUDATensorSetFromArray<uint8_t>)
Q
qingqing01 已提交
343
      .def("set", PyCUDATensorSetFromArray<int8_t>)
C
chengduoZH 已提交
344 345 346 347 348 349
      .def("set", PyCUDAPinnedTensorSetFromArray<float>)
      .def("set", PyCUDAPinnedTensorSetFromArray<int>)
      .def("set", PyCUDAPinnedTensorSetFromArray<double>)
      .def("set", PyCUDAPinnedTensorSetFromArray<int64_t>)
      .def("set", PyCUDAPinnedTensorSetFromArray<bool>)
      .def("set", PyCUDAPinnedTensorSetFromArray<uint16_t>)
F
fengjiayi 已提交
350
      .def("set", PyCUDAPinnedTensorSetFromArray<uint8_t>)
Q
qingqing01 已提交
351
      .def("set", PyCUDAPinnedTensorSetFromArray<int8_t>)
Q
qijun 已提交
352
#endif
353
      .def("shape", [](Tensor &self) { return vectorize(self.dims()); })
Y
yuyang18 已提交
354 355 356 357
      .def("_set_float_element", TensorSetElement<float>)
      .def("_get_float_element", TensorGetElement<float>)
      .def("_set_double_element", TensorSetElement<double>)
      .def("_get_double_element", TensorGetElement<double>)
X
xuezhong 已提交
358
      .def("_place", [](Tensor &self) { return self.place(); })
W
wopeizl 已提交
359 360
      .def("_dtype", [](Tensor &self) { return self.type(); })
      .def("__getitem__", PySliceTensor, py::return_value_policy::reference);
Y
Yu Yang 已提交
361

X
Xin Pan 已提交
362 363 364 365 366 367 368 369 370 371 372 373 374
  py::class_<LoDTensor, Tensor>(m, "LoDTensor", R"DOC(
    LoDTensor is a Tensor with optional LoD information.

    np.array(lod_tensor) can convert LoDTensor to numpy array.
    lod_tensor.lod() can retrieve the LoD information.

    LoD is short for Level of Details and is usually used for varied sequence
    length. You can skip the following comment if you don't need optional LoD.

  For example:
     A LoDTensor X can look like the example below. It contains 2 sequences.
     The first has length 2 and the second has length 3, as described by x.lod.

X
fix doc  
Xin Pan 已提交
375
     The first tensor dimension 5=2+3 is calculated from LoD if it's available.
X
Xin Pan 已提交
376
     It means the total number of sequence element. In X, each element has 2
X
fix doc  
Xin Pan 已提交
377
     columns, hence [5, 2].
X
Xin Pan 已提交
378 379 380

      x.lod  = [[2, 3]]
      x.data = [[1, 2], [3, 4],
X
fix doc  
Xin Pan 已提交
381 382
                [5, 6], [7, 8], [9, 10]]
      x.shape = [5, 2]
X
Xin Pan 已提交
383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405

      LoD can have multiple levels (for example, a paragraph can have multiple
      sentences and a sentence can have multiple words). In the following
      LodTensor Y, the lod_level is 2. It means there are 2 sequence, the
      first sequence length is 2 (has 2 sub-sequences), the second one's
      length is 1. The first sequence's 2 sub-sequences have length 2 and 2,
      respectively. And the second sequence's 1 sub-sequence has length 3.

      y.lod = [[2 1], [2 2 3]]
      y.shape = [2+2+3, ...]

  Note:
      In above description, LoD is length-based. In Paddle internal
      implementation, lod is offset-based. Hence, internally,
      y.lod is represented as [[0, 2, 3], [0, 2, 4, 7]] (length-based
      equivlent would be [[2-0, 3-2], [2-0, 4-2, 7-4]]).

      Sometimes LoD is called recursive_sequence_length to be more
      self-explanatory. In this case, it must be length-based. Due to history
      reasons. when LoD is called lod in public API, it might be offset-based.
      Users should be careful about it.

        )DOC")
406 407
      .def_buffer(
          [](Tensor &self) -> py::buffer_info { return CastToPyBuffer(self); })
408 409 410 411 412 413 414 415 416 417 418 419 420 421
      .def("__init__",
           [](LoDTensor &instance, const std::vector<std::vector<size_t>>
                                       &recursive_sequence_lengths) {
             LoD new_lod;
             new_lod.reserve(recursive_sequence_lengths.size());
             std::copy(recursive_sequence_lengths.begin(),
                       recursive_sequence_lengths.end(),
                       std::back_inserter(new_lod));
             LoD new_offset_lod = ConvertToOffsetBasedLoD(new_lod);
             PADDLE_ENFORCE(
                 CheckLoD(new_offset_lod, -1),
                 "the provided recursive_sequence_lengths info is invalid");
             new (&instance) LoDTensor(new_offset_lod);
           })
Y
Yu Yang 已提交
422
      .def("__init__", [](LoDTensor &instance) { new (&instance) LoDTensor(); })
G
gongweibao 已提交
423 424 425 426 427
      // We implement offset based LOD in C++ while we use length based with
      // Python API. So we changed set_lod to set_recursive_sequence_lengths to
      // avoid misuse.
      // The discussion is here:
      // https://github.com/PaddlePaddle/Paddle/issues/10855
D
dangqingqing 已提交
428
      .def("set_lod",
429
           [](LoDTensor &self, const std::vector<std::vector<size_t>> &lod) {
430
             // the input lod is offset-based level-of-detail info
Y
Yu Yang 已提交
431
             LoD new_lod;
432 433
             new_lod.reserve(lod.size());
             std::copy(lod.begin(), lod.end(), std::back_inserter(new_lod));
434 435
             PADDLE_ENFORCE(CheckLoD(new_lod, vectorize(self.dims()).front()),
                            "the provided lod info is invalid");
436
             self.set_lod(new_lod);
S
sneaxiy 已提交
437 438 439 440 441 442 443
           },
           py::arg("lod"), R"DOC(
           Set LoD of the LoDTensor.

           Args:
               lod (List[List[int]]): the lod to be set.
           )DOC")
444 445 446 447 448 449 450 451 452 453 454 455 456 457 458
      .def("set_recursive_sequence_lengths",
           [](LoDTensor &self, const std::vector<std::vector<size_t>>
                                   &recursive_sequence_lengths) {
             // the input recursive_sequence_lengths is length-based
             // level-of-detail info
             LoD new_lod;
             new_lod.reserve(recursive_sequence_lengths.size());
             std::copy(recursive_sequence_lengths.begin(),
                       recursive_sequence_lengths.end(),
                       std::back_inserter(new_lod));
             LoD new_offset_lod = ConvertToOffsetBasedLoD(new_lod);
             PADDLE_ENFORCE(
                 CheckLoD(new_offset_lod, vectorize(self.dims()).front()),
                 "the provided recursive_sequence_lengths info is invalid");
             self.set_lod(new_offset_lod);
S
sneaxiy 已提交
459 460 461 462
           },
           py::arg("recursive_sequence_lengths"), R"DOC(
           Set LoD of the LoDTensor according to recursive sequence length.

S
sneaxiy 已提交
463
           For example, if recursive_sequence_lengths=[[2, 3]], meaning that
464 465
           there are two sequences with length 2 and 3 respectively, the
           corresponding lod would be [[0, 2, 2+3]], i.e, [[0, 2, 5]].
S
sneaxiy 已提交
466 467

           Args:
468
                recursive_sequence_lengths (List[List[int]]): sequence lengths.
S
sneaxiy 已提交
469
           )DOC")
470 471 472 473 474 475 476 477
      .def("lod",
           [](LoDTensor &self) -> std::vector<std::vector<size_t>> {
             // output the offset-based lod info
             LoD lod = self.lod();
             std::vector<std::vector<size_t>> new_lod;
             new_lod.reserve(lod.size());
             std::copy(lod.begin(), lod.end(), std::back_inserter(new_lod));
             return new_lod;
S
sneaxiy 已提交
478 479 480 481 482 483 484
           },
           R"DOC(
           Return the LoD of the LoDTensor.

           Returns:
               out (List[List[int]]): the lod of the LoDTensor.
           )DOC")
G
gongweibao 已提交
485
      // Set above comments of set_lod.
486 487 488 489 490 491 492 493
      .def("recursive_sequence_lengths",
           [](LoDTensor &self) -> std::vector<std::vector<size_t>> {
             // output the length-based lod info
             LoD lod = ConvertToLengthBasedLoD(self.lod());
             std::vector<std::vector<size_t>> new_lod;
             new_lod.reserve(lod.size());
             std::copy(lod.begin(), lod.end(), std::back_inserter(new_lod));
             return new_lod;
S
sneaxiy 已提交
494 495 496 497 498
           },
           R"DOC(
           Return the sequence length of the LoDTensor corresponding to LoD.

           Returns:
499
               out (List[List[int]): the sequence lengths.
S
sneaxiy 已提交
500 501 502 503 504 505 506 507 508 509 510 511
           )DOC")
      .def("has_valid_recursive_sequence_lengths",
           [](LoDTensor &self) -> bool {
             // Check that the lod info is valid and match the outermost
             // dimension of the LoDTensor data
             return CheckLoD(self.lod(), vectorize(self.dims()).front());
           },
           R"DOC(
           Check whether the lod of the LoDTensor is valid.

           Returns:
               out (bool): whether the lod is valid.
W
wopeizl 已提交
512 513 514 515 516 517 518
           )DOC")
      .def("__getitem__", PySliceTensor, py::return_value_policy::reference,
           R"DOC(
           Slice the original Tensor, and remove the LoD information.

           Returns:
               out (Tensor): new Tensor(NOT LoDTensor).
S
sneaxiy 已提交
519
           )DOC");
D
dangqingqing 已提交
520

Q
qijun 已提交
521 522 523 524 525 526 527 528 529 530 531
  py::class_<SelectedRows>(m, "SelectedRows")
      .def("__init__",
           [](SelectedRows &instance) { new (&instance) SelectedRows(); })
      .def("__init__",
           [](SelectedRows &instance, const std::vector<int64_t> rows,
              const int64_t &height) {
             new (&instance) SelectedRows(rows, height);
           })
      .def("get_tensor",
           [](SelectedRows &self) { return self.mutable_value(); },
           py::return_value_policy::reference)
532 533
      .def("numel",
           [](SelectedRows &self) -> int64_t { return self.value().numel(); })
Q
qijun 已提交
534 535
      .def("set_height", &SelectedRows::set_height)
      .def("height", &SelectedRows::height)
Q
qijun 已提交
536 537 538 539 540 541 542 543 544
      .def("set_rows",
           [](SelectedRows &self, std::vector<int64_t> rows) {
#ifndef PADDLE_WITH_CUDA
             self.set_rows(rows);
#else
        Vector<int64_t> new_rows(rows);
        self.set_rows(new_rows);
#endif
           })
545
      .def("sync_index", [](SelectedRows &instance) { instance.SyncIndex(); })
546
      .def("rows", [](SelectedRows &self) {
547 548 549 550 551
        auto rows = self.rows();
        std::vector<int64_t> new_rows;
        new_rows.reserve(rows.size());
        std::copy(rows.begin(), rows.end(), std::back_inserter(new_rows));
        return new_rows;
552
      });
Q
qijun 已提交
553

554
  py::class_<Variable>(m, "Variable", R"DOC(Variable Class.
555 556 557

All parameter, weight, gradient are variables in Paddle.
)DOC")
S
sneaxiy 已提交
558
      .def(py::init<>())
559
      .def("is_int", [](const Variable &var) { return var.IsType<int>(); })
560
      .def("set_int",
561 562
           [](Variable &var, int val) -> void { *var.GetMutable<int>() = val; })
      .def("get_int", [](const Variable &var) -> int { return var.Get<int>(); })
563 564 565 566 567 568 569
      .def("is_float", [](const Variable &var) { return var.IsType<float>(); })
      .def("set_float",
           [](Variable &var, float val) -> void {
             *var.GetMutable<float>() = val;
           })
      .def("get_float",
           [](const Variable &var) -> float { return var.Get<float>(); })
Y
Yu Yang 已提交
570
      .def("get_tensor",
571 572
           [](Variable &self) -> LoDTensor * {
             return self.GetMutable<LoDTensor>();
D
dangqingqing 已提交
573 574
           },
           py::return_value_policy::reference)
Y
Yu Yang 已提交
575 576 577
      .def("get_lod_rank_table",
           [](Variable &self) { return self.GetMutable<LoDRankTable>(); },
           py::return_value_policy::reference)
Q
qijun 已提交
578 579 580 581 582
      .def("get_selected_rows",
           [](Variable &self) -> SelectedRows * {
             return self.GetMutable<SelectedRows>();
           },
           py::return_value_policy::reference)
Y
Yu Yang 已提交
583 584 585
      .def("get_lod_tensor_array",
           [](Variable &self) { return self.GetMutable<LoDTensorArray>(); },
           py::return_value_policy::reference)
P
peizhilin 已提交
586
#if (defined(PADDLE_WITH_CUDA) && !defined(_WIN32))
D
Dong Zhihong 已提交
587 588 589 590 591
      .def("get_communicator",
           [](Variable &self) -> platform::Communicator * {
             return self.GetMutable<platform::Communicator>();
           },
           py::return_value_policy::reference)
Y
Yu Yang 已提交
592
#endif
Y
Refine  
Yu Yang 已提交
593 594 595 596 597
      .def("get_reader",
           [](Variable &self) -> framework::ReaderHolder * {
             PADDLE_ENFORCE(self.IsType<framework::ReaderHolder>());
             return self.GetMutable<framework::ReaderHolder>();
           },
W
wopeizl 已提交
598
           py::return_value_policy::reference);
599

S
sneaxiy 已提交
600
  BindReader(&m);
Y
Refine  
Yu Yang 已提交
601

S
sneaxiy 已提交
602 603 604 605
  using LoDTensorBlockingQueue =
      ::paddle::operators::reader::LoDTensorBlockingQueue;
  using LoDTensorBlockingQueueHolder =
      ::paddle::operators::reader::LoDTensorBlockingQueueHolder;
S
sneaxiy 已提交
606

S
sneaxiy 已提交
607 608
  py::class_<LoDTensorBlockingQueue, std::shared_ptr<LoDTensorBlockingQueue>>(
      m, "LoDTensorBlockingQueue", "")
S
sneaxiy 已提交
609
      .def("push",
S
sneaxiy 已提交
610
           [](LoDTensorBlockingQueue &self,
S
sneaxiy 已提交
611
              const std::vector<framework::LoDTensor> &lod_tensor_vec) {
S
sneaxiy 已提交
612
             pybind11::gil_scoped_release release;
S
sneaxiy 已提交
613
             return self.Push(lod_tensor_vec);
S
sneaxiy 已提交
614
           })
S
sneaxiy 已提交
615 616 617 618
      .def("size", &LoDTensorBlockingQueue::Size)
      .def("capacity", &LoDTensorBlockingQueue::Cap)
      .def("close", &LoDTensorBlockingQueue::Close)
      .def("is_closed", &LoDTensorBlockingQueue::IsClosed);
S
sneaxiy 已提交
619

S
sneaxiy 已提交
620
  m.def("init_lod_tensor_blocking_queue",
Q
Qiao Longfei 已提交
621 622 623 624 625 626
        [](Variable &var,
           size_t capacity) -> std::shared_ptr<LoDTensorBlockingQueue> {
          auto *holder = var.GetMutable<LoDTensorBlockingQueueHolder>();
          holder->InitOnce(capacity, FLAGS_reader_queue_speed_test_mode);
          return holder->GetQueue();
        },
S
sneaxiy 已提交
627
        py::return_value_policy::copy);
S
sneaxiy 已提交
628

S
sneaxiy 已提交
629
  py::class_<Scope>(m, "_Scope", R"DOC(
Q
Qiao Longfei 已提交
630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648
    Scope is an association of a name to Variable. All variables belong to Scope.

    Variables in a parent scope can be retrieved from local scope.

    You need to specify a scope to run a Net, i.e., `exe.Run(&scope)`.
    One net can run in different scopes and update different variable in the
    scope.

    You can create var in a scope and get it from the scope.

    Examples:
        .. code-block:: python

          # create tensor from a scope and set value to it.
          param = scope.var('Param').get_tensor()
          param_array = np.full((height, row_numel), 5.0).astype("float32")
          param.set(param_array, place)

        )DOC")
S
sneaxiy 已提交
649 650
      .def("_remove_from_pool",
           [](Scope &self) { ScopePool::Instance().Remove(&self); })
D
dongzhihong 已提交
651
      .def("var",
652
           [](Scope &self, const std::string &name) -> Variable * {
D
dongzhihong 已提交
653
             return self.Var(name);
Y
Yu Yang 已提交
654
           },
S
sneaxiy 已提交
655 656
           py::arg("name"),
           R"DOC(
657
           Find or create variable named :code:`name` in the current scope.
S
sneaxiy 已提交
658

659
           If the variable named :code:`name` does not exist in the
S
sneaxiy 已提交
660
           current scope, the variable would be created. Otherwise,
661
           return the existing variable.
S
sneaxiy 已提交
662 663

           Args:
664 665
               name (str): the variable name.

S
sneaxiy 已提交
666
           Returns:
667
               out (core.Variable): the found or created variable.
S
sneaxiy 已提交
668 669 670 671
           )DOC",
           py::return_value_policy::reference)
      .def("find_var", &Scope::FindVar, py::arg("name"),
           R"DOC(
672
           Find variable named :code:`name` in the current scope or
S
sneaxiy 已提交
673
           its parent scope. Return None if not found.
674

S
sneaxiy 已提交
675 676
           Args:
               name (str): the variable name.
677

S
sneaxiy 已提交
678
           Returns:
679
               out (core.Variable|None): the found variable or None.
S
sneaxiy 已提交
680
           )DOC",
681
           py::return_value_policy::reference)
682
      .def("new_scope", [](Scope &self) -> Scope * { return &self.NewScope(); },
S
sneaxiy 已提交
683 684 685 686 687 688
           R"DOC(
           Create a new sub-scope of the current scope.

           Returns:
               out (core._Scope): the created sub-scope.
           )DOC",
689
           py::return_value_policy::reference)
S
sneaxiy 已提交
690 691 692
      .def("drop_kids", &Scope::DropKids,
           R"DOC(
           Delete all sub-scopes of the current scope.
S
sneaxiy 已提交
693 694
           )DOC")
      .def("_kids", &Scope::kids);
695

S
sneaxiy 已提交
696 697 698 699 700 701
  m.def("Scope",
        []() -> Scope * {
          auto *s = new Scope();
          ScopePool::Instance().Insert(std::unique_ptr<Scope>(s));
          return s;
        },
S
sneaxiy 已提交
702 703
        R"DOC(
        Create a new scope.
704

S
sneaxiy 已提交
705 706 707
        Returns:
            out (core._Scope): the created scope.
        )DOC",
S
sneaxiy 已提交
708 709
        py::return_value_policy::reference);

Y
Yu Yang 已提交
710 711
  //! @note: Be careful! PyBind will return std::string as an unicode, not
  //! Python str. If you want a str object, you should cast them in Python.
Y
Yu Yang 已提交
712 713
  m.def("get_all_op_protos", []() -> std::vector<py::bytes> {
    std::vector<py::bytes> ret_values;
714 715 716 717 718 719 720 721 722 723
    for (auto &iter : OpInfoMap::Instance().map()) {
      auto &info = iter.second;
      if (info.HasOpProtoAndChecker()) {
        std::string str;
        PADDLE_ENFORCE(
            info.Proto().SerializeToString(&str),
            "Serialize OpProto Error. This could be a bug of Paddle.");
        ret_values.emplace_back(str);
      }
    }
Y
Yu Yang 已提交
724 725
    return ret_values;
  });
726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741
  m.def(
      "get_grad_op_desc", [](const OpDesc &op_desc,
                             const std::unordered_set<std::string> &no_grad_set,
                             const std::vector<BlockDesc *> &grad_sub_block) {
        std::unordered_map<std::string, std::string> grad_to_var;
        std::vector<std::unique_ptr<OpDesc>> grad_op_descs =
            framework::OpInfoMap::Instance()
                .Get(op_desc.Type())
                .GradOpMaker()(op_desc, no_grad_set, &grad_to_var,
                               grad_sub_block);
        std::vector<OpDesc *> grad_op_desc_ptrs(grad_op_descs.size());
        std::transform(grad_op_descs.begin(), grad_op_descs.end(),
                       grad_op_desc_ptrs.begin(),
                       [](std::unique_ptr<OpDesc> &p) { return p.release(); });
        return std::make_pair(grad_op_desc_ptrs, grad_to_var);
      });
Y
Yu Yang 已提交
742
  m.def("prune", [](const ProgramDesc &origin,
743
                    const std::vector<std::array<size_t, 2>> &targets) {
Y
Yu Yang 已提交
744
    ProgramDesc prog_with_targets(origin);
745
    for (const auto &t : targets) {
746
      prog_with_targets.MutableBlock(t[0])->Op(t[1])->SetIsTarget(true);
747
    }
748
    proto::ProgramDesc pruned_desc;
749
    Prune(*prog_with_targets.Proto(), &pruned_desc);
Y
Yu Yang 已提交
750
    return new ProgramDesc(pruned_desc);
751
  });
752 753 754 755
  m.def("empty_var_name",
        []() { return std::string(framework::kEmptyVarName); });
  m.def("grad_var_suffix",
        []() { return std::string(framework::kGradVarSuffix); });
756 757 758
  m.def_submodule(
       "var_names",
       "The module will return special predefined variable name in Paddle")
Y
Yi Wang 已提交
759 760
      .def("empty", []() { return kEmptyVarName; })
      .def("temp", []() { return kTempVarName; });
Q
qijun 已提交
761
  // clang-format off
Y
Yu Yang 已提交
762
  py::class_<paddle::platform::DeviceContext>(m, "DeviceContext")
Q
qijun 已提交
763 764
      .def_static("create",
                  [](paddle::platform::CPUPlace& place)
Q
qijun 已提交
765
                      -> paddle::platform::DeviceContext* {
Q
qijun 已提交
766 767 768
                    return new paddle::platform::CPUDeviceContext();
                  })
      .def_static("create",
D
dzhwinter 已提交
769
                  [](paddle::platform::CUDAPlace& place)
Q
qijun 已提交
770
                      -> paddle::platform::DeviceContext* {
771
#ifndef PADDLE_WITH_CUDA
D
dzhwinter 已提交
772
                    PADDLE_THROW("CUDAPlace is not supported in CPU device.");
Q
qijun 已提交
773
#else
Q
qijun 已提交
774
                    return new paddle::platform::CUDADeviceContext(place);
Q
qijun 已提交
775
#endif
C
chengduoZH 已提交
776 777 778 779 780 781 782 783 784 785 786
                  })
          .def_static("create",
                [](paddle::platform::CUDAPinnedPlace& place)
                        -> paddle::platform::DeviceContext* {
#ifndef PADDLE_WITH_CUDA
                  PADDLE_THROW(
                        "CUDAPinnedPlace is not supported in CPU device.");
#else
                  return new paddle::platform::CUDAPinnedDeviceContext(place);
#endif
                });;
D
Dong Zhihong 已提交
787
// clang-format on
P
peizhilin 已提交
788
#if (defined(PADDLE_WITH_CUDA) && !defined(_WIN32))
D
Dong Zhihong 已提交
789 790
  py::class_<platform::Communicator>(m, "Communicator").def(py::init<>());
#endif
791 792 793 794 795
  py::class_<platform::CUDAPlace>(m, "CUDAPlace", R"DOC(
    CUDAPlace is a descriptor of a device. It represents a GPU, and each CUDAPlace
    has a dev_id to indicate the number of cards represented by the current CUDAPlace.
    The memory of CUDAPlace with different dev_id is not accessible.
        )DOC")
S
sneaxiy 已提交
796 797 798 799 800 801 802 803 804 805 806 807
      .def("__init__",
           [](platform::CUDAPlace &self, int dev_id) {
#ifdef PADDLE_WITH_CUDA
             PADDLE_ENFORCE(
                 dev_id >= 0 && dev_id < platform::GetCUDADeviceCount(),
                 "Invalid CUDAPlace(%d), must inside [0, %d)", dev_id,
                 platform::GetCUDADeviceCount());
             new (&self) platform::CUDAPlace(dev_id);
#else
             PADDLE_THROW("Cannot use CUDAPlace in CPU only version");
#endif
           })
S
sneaxiy 已提交
808 809 810 811 812 813
      .def("_type", &PlaceIndex<platform::CUDAPlace>)
      .def("_equals", &IsSamePlace<platform::CUDAPlace, platform::Place>)
      .def("_equals", &IsSamePlace<platform::CUDAPlace, platform::CUDAPlace>)
      .def("_equals", &IsSamePlace<platform::CUDAPlace, platform::CPUPlace>)
      .def("_equals",
           &IsSamePlace<platform::CUDAPlace, platform::CUDAPinnedPlace>)
D
dzhwinter 已提交
814
      .def("__str__", string::to_string<const platform::CUDAPlace &>);
Q
qijun 已提交
815

816 817 818 819
  py::class_<paddle::platform::CPUPlace>(m, "CPUPlace", R"DOC(
    CPUPlace is a descriptor of a device. It represents a CPU, and the memory
    CPUPlace can be accessed by CPU.
        )DOC")
820
      .def(py::init<>())
S
sneaxiy 已提交
821 822 823 824 825 826
      .def("_type", &PlaceIndex<platform::CPUPlace>)
      .def("_equals", &IsSamePlace<platform::CPUPlace, platform::Place>)
      .def("_equals", &IsSamePlace<platform::CPUPlace, platform::CUDAPlace>)
      .def("_equals", &IsSamePlace<platform::CPUPlace, platform::CPUPlace>)
      .def("_equals",
           &IsSamePlace<platform::CPUPlace, platform::CUDAPinnedPlace>)
827
      .def("__str__", string::to_string<const platform::CPUPlace &>);
Y
Yu Yang 已提交
828

829 830 831 832
  py::class_<paddle::platform::CUDAPinnedPlace>(m, "CUDAPinnedPlace", R"DOC(
    CUDAPinnedPlace is a descriptor of a device. The memory of CUDAPinnedPlace
    can be accessed by GPU and CPU.
        )DOC")
S
sneaxiy 已提交
833
      .def("__init__",
S
sneaxiy 已提交
834
           [](platform::CUDAPinnedPlace &self) {
S
sneaxiy 已提交
835 836 837
#ifndef PADDLE_WITH_CUDA
             PADDLE_THROW("Cannot use CUDAPinnedPlace in CPU only version");
#endif
S
sneaxiy 已提交
838
             new (&self) platform::CUDAPinnedPlace();
S
sneaxiy 已提交
839
           })
S
sneaxiy 已提交
840 841 842 843 844 845 846 847
      .def("_type", &PlaceIndex<platform::CUDAPinnedPlace>)
      .def("_equals", &IsSamePlace<platform::CUDAPinnedPlace, platform::Place>)
      .def("_equals",
           &IsSamePlace<platform::CUDAPinnedPlace, platform::CUDAPlace>)
      .def("_equals",
           &IsSamePlace<platform::CUDAPinnedPlace, platform::CPUPlace>)
      .def("_equals",
           &IsSamePlace<platform::CUDAPinnedPlace, platform::CUDAPinnedPlace>)
C
chengduoZH 已提交
848 849
      .def("__str__", string::to_string<const platform::CUDAPinnedPlace &>);

Y
Yu Yang 已提交
850 851
  py::class_<platform::Place>(m, "Place")
      .def(py::init<>())
S
sneaxiy 已提交
852 853 854 855 856
      .def("_type", &PlaceIndex<platform::Place>)
      .def("_equals", &IsSamePlace<platform::Place, platform::Place>)
      .def("_equals", &IsSamePlace<platform::Place, platform::CUDAPlace>)
      .def("_equals", &IsSamePlace<platform::Place, platform::CPUPlace>)
      .def("_equals", &IsSamePlace<platform::Place, platform::CUDAPinnedPlace>)
X
xuezhong 已提交
857 858
      .def("is_gpu_place",
           [](platform::Place &self) { return platform::is_gpu_place(self); })
S
sneaxiy 已提交
859 860 861 862 863 864
      .def("is_cpu_place",
           [](platform::Place &self) { return platform::is_cpu_place(self); })
      .def("is_cuda_pinned_place",
           [](platform::Place &self) {
             return platform::is_cuda_pinned_place(self);
           })
X
xuezhong 已提交
865 866 867 868
      .def("gpu_device_id",
           [](platform::Place &self) {
             return boost::get<platform::CUDAPlace>(self).device;
           })
S
sneaxiy 已提交
869 870
      .def("set_place", [](platform::Place &self,
                           const platform::Place &other) { self = other; })
Y
Yu Yang 已提交
871 872 873 874 875
      .def("set_place",
           [](platform::Place &self, const platform::CPUPlace &cpu_place) {
             self = cpu_place;
           })
      .def("set_place",
D
dzhwinter 已提交
876
           [](platform::Place &self, const platform::CUDAPlace &gpu_place) {
Y
Yu Yang 已提交
877
             self = gpu_place;
C
chengduoZH 已提交
878 879
           })
      .def("set_place", [](platform::Place &self,
C
chengduoZH 已提交
880 881
                           const platform::CUDAPinnedPlace &cuda_pinned_place) {
        self = cuda_pinned_place;
C
chengduoZH 已提交
882
      });
Y
Yu Yang 已提交
883

Y
Yu Yang 已提交
884 885 886
  py::class_<OperatorBase>(m, "Operator")
      .def_static("create",
                  [](py::bytes protobin) {
887
                    proto::OpDesc desc;
Y
Yu Yang 已提交
888 889 890 891 892
                    PADDLE_ENFORCE(desc.ParsePartialFromString(protobin),
                                   "Cannot parse user input to OpDesc");
                    PADDLE_ENFORCE(desc.IsInitialized(),
                                   "User OpDesc is not initialized, reason %s",
                                   desc.InitializationErrorString());
893
                    return OpRegistry::CreateOp(desc);
Y
Yu Yang 已提交
894
                  })
895
      .def("run",
896
           [](OperatorBase &self, const Scope &scope,
D
dzhwinter 已提交
897 898 899
              const platform::CPUPlace &place) { self.Run(scope, place); })
      .def("run",
           [](OperatorBase &self, const Scope &scope,
D
dzhwinter 已提交
900
              const platform::CUDAPlace &place) { self.Run(scope, place); })
C
chengduoZH 已提交
901 902 903 904 905
      .def("run",
           [](OperatorBase &self, const Scope &scope,
              const platform::CUDAPinnedPlace &place) {
             self.Run(scope, place);
           })
Y
Yu Yang 已提交
906 907 908 909 910 911 912
      .def("type",
           [](const OperatorBase &op) -> std::string { return op.Type(); })
      .def("outputs",
           [](const OperatorBase &op)
               -> std::map<std::string, std::vector<std::string>> {
                 return op.Outputs();
               })
Q
qijun 已提交
913 914
      .def("output_vars",
           [](const OperatorBase &op) { return op.OutputVars(true); })
Y
Yu Yang 已提交
915
      .def("inputs", [](const OperatorBase &op) { return op.Inputs(); })
Q
qijun 已提交
916
      .def("input_vars", [](const OperatorBase &op) { return op.InputVars(); })
Y
Yu Yang 已提交
917 918 919 920
      .def("__str__", &OperatorBase::DebugString)
      .def("no_intermediate_outputs",
           [](const OperatorBase &op) { return op.OutputVars(false); })
      .def("support_gpu", &OperatorBase::SupportGPU);
Y
Yu Yang 已提交
921

F
fengjiayi 已提交
922
  py::class_<framework::Executor>(m, "Executor")
D
dzhwinter 已提交
923
      .def(py::init<const platform::Place &>())
Y
Yancey1989 已提交
924
      .def("close", &Executor::Close)
S
sneaxiy 已提交
925
      .def("run", [](Executor &self, const ProgramDesc &prog, Scope *scope,
S
sneaxiy 已提交
926 927
                     int block_id, bool create_local_scope, bool create_vars,
                     const std::vector<std::string> &fetch_vars) {
S
sneaxiy 已提交
928
        pybind11::gil_scoped_release release;
S
sneaxiy 已提交
929 930
        self.Run(prog, scope, block_id, create_local_scope, create_vars,
                 fetch_vars);
S
sneaxiy 已提交
931
      });
S
sneaxiy 已提交
932

D
dzhwinter 已提交
933
  m.def("init_gflags", framework::InitGflags);
Y
Yang Yu 已提交
934
  m.def("init_glog", framework::InitGLOG);
935
  m.def("init_dgc", framework::InitDGC);
X
Xin Pan 已提交
936 937
  m.def("init_devices",
        [](bool init_p2p) { framework::InitDevices(init_p2p); });
938

939
  m.def("is_compiled_with_ngraph", IsCompiledWithNGRAPH);
940
  m.def("is_compiled_with_cuda", IsCompiledWithCUDA);
941
  m.def("is_compiled_with_mkldnn", IsCompiledWithMKLDNN);
942
  m.def("is_compiled_with_brpc", IsCompiledWithBrpc);
Y
update  
Yancey1989 已提交
943
  m.def("is_compiled_with_dist", IsCompiledWithDIST);
944 945 946 947 948 949
#ifdef PADDLE_WITH_CUDA
  m.def("is_float16_supported", [](const platform::CUDAPlace &place) -> bool {
    // Only GPUs with Compute Capability >= 53 support float16
    return platform::GetCUDAComputeCapability(place.device) >= 53;
  });
#endif
950

951
  m.def("set_feed_variable", framework::SetFeedVariable);
Q
qijun 已提交
952
  m.def("get_fetch_variable", framework::GetFetchVariable);
953
  m.def("get_variable_tensor", framework::GetVariableTensor);
Q
qijun 已提交
954

X
Xin Pan 已提交
955 956
  m.def("_is_program_version_supported", IsProgramVersionSupported);

957 958 959 960 961
  BindProgramDesc(&m);
  BindBlockDesc(&m);
  BindVarDsec(&m);
  BindOpDesc(&m);
  BindConstValue(&m);
Y
Yu Yang 已提交
962

Y
Yu Yang 已提交
963 964 965 966 967 968 969 970 971
  py::class_<framework::LoDRankTable>(m, "LodRankTable")
      .def("items", [](framework::LoDRankTable &table) {
        std::vector<std::pair<size_t, size_t>> res;
        for (auto &item : table.items()) {
          res.push_back({item.index, item.length});
        }
        return res;
      });

Y
Yu Yang 已提交
972
  py::class_<LoDTensorArray>(m, "LoDTensorArray")
S
sneaxiy 已提交
973 974
      .def("__init__",
           [](LoDTensorArray &instance) { new (&instance) LoDTensorArray(); })
Y
Yu Yang 已提交
975 976 977 978 979 980 981 982 983 984
      .def("__getitem__",
           [](LoDTensorArray &self, size_t i) { return &self.at(i); },
           py::return_value_policy::reference)
      .def("__len__", [](LoDTensorArray &self) { return self.size(); })
      .def("__setitem__",
           [](LoDTensorArray &self, size_t i, const LoDTensor &t) {
             PADDLE_ENFORCE_LT(i, self.size());
             self[i].ShareDataWith(t);
             self[i].set_lod(t.lod());
           })
S
sneaxiy 已提交
985 986 987 988 989 990 991
      .def("append",
           [](LoDTensorArray &self, const LoDTensor &t) {
             self.emplace_back();
             self.back().ShareDataWith(t);
             self.back().set_lod(t.lod());
           },
           py::arg("tensor"), "Append a LoDensor to LoDTensorArray.");
Y
Yu Yang 已提交
992

D
dzhwinter 已提交
993 994 995
  m.def("IsInplace",
        [](std::string op) -> bool { return operators::IsInplace(op); });

Y
Yu Yang 已提交
996
  m.def("op_support_gpu", OpSupportGPU);
D
Dong Zhihong 已提交
997
#ifdef PADDLE_WITH_CUDA
D
Dong Zhihong 已提交
998
  m.def("get_cuda_device_count", platform::GetCUDADeviceCount);
D
dangqingqing 已提交
999

P
peizhilin 已提交
1000
#ifndef _WIN32
D
dangqingqing 已提交
1001 1002 1003
  m.def("nvprof_init", platform::CudaProfilerInit);
  m.def("nvprof_start", platform::CudaProfilerStart);
  m.def("nvprof_stop", platform::CudaProfilerStop);
D
Dong Zhihong 已提交
1004
#endif
P
peizhilin 已提交
1005
#endif
Y
Yu Yang 已提交
1006

1007 1008 1009 1010
  py::enum_<platform::ProfilerState>(m, "ProfilerState", py::arithmetic())
      .value("kDisabled", platform::ProfilerState::kDisabled)
      .value("kCPU", platform::ProfilerState::kCPU)
      .value("kCUDA", platform::ProfilerState::kCUDA)
1011
      .value("kAll", platform::ProfilerState::kAll)
1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024
      .export_values();

  py::enum_<platform::EventSortingKey>(m, "EventSortingKey", py::arithmetic())
      .value("kDefault", platform::EventSortingKey::kDefault)
      .value("kCalls", platform::EventSortingKey::kCalls)
      .value("kTotal", platform::EventSortingKey::kTotal)
      .value("kMin", platform::EventSortingKey::kMin)
      .value("kMax", platform::EventSortingKey::kMax)
      .value("kAve", platform::EventSortingKey::kAve)
      .export_values();

  m.def("enable_profiler", platform::EnableProfiler);
  m.def("disable_profiler", platform::DisableProfiler);
X
Xin Pan 已提交
1025
  m.def("is_profiler_enabled", platform::IsProfileEnabled);
1026
  m.def("reset_profiler", platform::ResetProfiler);
1027
  m.def("get_pass", [](const std::string &pass_type) {
W
WangZhen 已提交
1028 1029 1030
    auto pass = framework::ir::PassRegistry::Instance().Get(pass_type);
    return std::shared_ptr<framework::ir::Pass>(std::move(pass));
  });
Y
Yu Yang 已提交
1031

1032 1033
  py::class_<ir::Pass, std::shared_ptr<ir::Pass>> pass(m, "Pass");
  pass.def(py::init())
W
WangZhen 已提交
1034
      .def("has", &ir::Pass::Has)
1035 1036 1037
      .def("set_not_owned",
           [](ir::Pass &self, const std::string &attr_name, ProgramDesc &attr) {
             self.SetNotOwned<ProgramDesc>(attr_name, &attr);
W
WangZhen 已提交
1038
           })
1039
      .def(
1040
          "set",
1041 1042 1043
          [](ir::Pass &self, const std::string &name, const std::string &attr) {
            self.Set<std::string>(name, new std::string(attr));
          })
1044 1045
      .def("set", [](ir::Pass &self, const std::string &name,
                     int val) { self.Set<const int>(name, new int(val)); })
F
flame 已提交
1046 1047
      .def("type", &ir::Pass::Type)
      .def("apply", [](ir::Pass &self, std::shared_ptr<ir::Graph> graph) {
1048
        self.Apply(graph.get());
F
flame 已提交
1049
      });
1050

X
fix  
Xin Pan 已提交
1051 1052
  py::class_<ir::PassBuilder, std::shared_ptr<ir::PassBuilder>> pb(
      m, "PassBuilder");
1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066
  pb.def(py::init())
      .def("append_pass",
           [](ir::PassBuilder &self,
              const std::string &pass_type) -> std::shared_ptr<ir::Pass> {
             return self.AppendPass(pass_type);
           })
      .def("all_passes", [](ir::PassBuilder &self) { return self.AllPasses(); })
      .def("insert_pass",
           [](ir::PassBuilder &self, size_t idx, const std::string &pass_type) {
             return self.InsertPass(idx, pass_type);
           })
      .def("remove_pass",
           [](ir::PassBuilder &self, size_t idx) { self.RemovePass(idx); });

Y
yuyang18 已提交
1067
  // -- python binds for parallel executor.
X
Xin Pan 已提交
1068

Y
yuyang18 已提交
1069
  py::class_<ParallelExecutor> pe(m, "ParallelExecutor");
C
chengduo 已提交
1070 1071 1072 1073
  py::class_<ExecutionStrategy> exec_strategy(pe, "ExecutionStrategy", R"DOC(
    ExecutionStrategy allows the user to more preciously control how to run
    the program in ParallelExecutor by setting the property.

C
chengduo 已提交
1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084
    Examples:
        .. code-block:: python

          exec_strategy = fluid.ExecutionStrategy()
          exec_strategy.num_threads = 4

          train_exe = fluid.ParallelExecutor(use_cuda=True,
                                             loss_name=loss.name,
                                             exec_strategy=exec_strategy)

          train_loss, = train_exe.run([loss.name], feed=feed_dict)
C
chengduo 已提交
1085 1086 1087

        )DOC");

Y
yuyang18 已提交
1088
  exec_strategy.def(py::init())
Y
yuyang18 已提交
1089 1090 1091 1092 1093
      .def_property(
          "num_threads",
          [](const ExecutionStrategy &self) { return self.num_threads_; },
          [](ExecutionStrategy &self, size_t num_threads) {
            self.num_threads_ = num_threads;
C
chengduo 已提交
1094 1095 1096 1097 1098 1099 1100 1101 1102 1103
          },
          R"DOC(The type is INT, num_threads represents the size of thread pool that
            used to run the operators of the current program in ParallelExecutor.
            If :math:`num\_threads=1`, all the operators will execute one by one,
            but the order maybe difference between iterations.
            If it is not set, it will be set in ParallelExecutor according to the
            device type and device count, for GPU, :math:`num\_threads=device\_count*4`, for CPU,
            :math:`num\_threads=CPU\_NUM*4`, the explanation of:math:`CPU\_NUM` is in ParallelExecutor.
            if it is not set, ParallelExecutor will get the cpu count by calling
            `multiprocessing.cpu_count()`. Default 0.)DOC")
Y
yuyang18 已提交
1104
      .def_property(
1105 1106 1107 1108
          "use_cuda",
          [](const ExecutionStrategy &self) { return self.use_cuda_; },
          [](ExecutionStrategy &self, bool use_cuda) {
            self.use_cuda_ = use_cuda;
C
chengduo 已提交
1109 1110 1111 1112
          })  // FIXME(chengduo): Doesn't add doc for 'use_cuda', use_cuda may
      // make user confuse, because ParallelExecutor has a parameter named
      // 'use_cuda' too, in current implementation, ParallelExecutor's
      // 'use_cuda' will rewrite ExecutionStrategy's 'use_cuda'.
Y
yuyang18 已提交
1113 1114 1115 1116 1117
      .def_property(
          "allow_op_delay",
          [](const ExecutionStrategy &self) { return self.allow_op_delay_; },
          [](ExecutionStrategy &self, bool allow_op_delay) {
            self.allow_op_delay_ = allow_op_delay;
C
chengduo 已提交
1118 1119 1120 1121
          },
          R"DOC(The type is BOOL, allow_op_delay represents whether to delay the
                communication operators to run, it may make the execution faster.
                Note that in some models, allow_op_delay may cause program hang. Default False.)DOC")
Y
yuyang18 已提交
1122 1123 1124 1125 1126 1127 1128
      .def_property(
          "num_iteration_per_drop_scope",
          [](const ExecutionStrategy &self) {
            return self.num_iteration_per_drop_scope_;
          },
          [](ExecutionStrategy &self, size_t num_iteration_per_drop_scope) {
            self.num_iteration_per_drop_scope_ = num_iteration_per_drop_scope;
C
chengduo 已提交
1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139
          },
          R"DOC(The type is INT, num_iteration_per_drop_scope indicates how
                many iterations to clean up the temp variables which
                is generated during execution. It may make the execution faster,
                because the temp variable's shape maybe the same between two iterations. Default 100.

                NOTES:
                    1. If you fetch data when calling the 'run', the ParallelExecutor
                       will clean up the temp variables at the end of the current iteration.
                    2. In some NLP model, it may cause the GPU memory is insufficient,
                       in this case, you should reduce `num_iteration_per_drop_scope`.
1140 1141 1142 1143 1144 1145
              )DOC")
      .def_property("_dry_run",
                    [](const ExecutionStrategy &self) { return self.dry_run_; },
                    [](ExecutionStrategy &self, bool dry_run) {
                      self.dry_run_ = dry_run;
                    });
C
chengduo 已提交
1146

Y
yuyang18 已提交
1147
  exec_strategy.def_property(
Y
yuyang18 已提交
1148 1149 1150 1151 1152 1153 1154
      "use_experimental_executor",
      [](const ExecutionStrategy &self) {
        return self.type_ == ExecutionStrategy::kExperimental;
      },
      [](ExecutionStrategy &self, bool experimental) {
        self.type_ = experimental ? ExecutionStrategy::kExperimental
                                  : ExecutionStrategy::kDefault;
Y
yuyang18 已提交
1155 1156
      });

C
chengduo 已提交
1157 1158 1159 1160
  py::class_<BuildStrategy> build_strategy(pe, "BuildStrategy", R"DOC(
    BuildStrategy allows the user to more preciously control how to
    build the SSA Graph in ParallelExecutor by setting the property.

C
chengduo 已提交
1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171
    Examples:
        .. code-block:: python

          build_strategy = fluid.BuildStrategy()
          build_strategy.reduce_strategy = fluid.BuildStrategy.ReduceStrategy.Reduce

          train_exe = fluid.ParallelExecutor(use_cuda=True,
                                             loss_name=loss.name,
                                             build_strategy=build_strategy)

          train_loss, = train_exe.run([loss.name], feed=feed_dict)
C
chengduo 已提交
1172
)DOC");
Y
yuyang18 已提交
1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188

  py::enum_<BuildStrategy::ReduceStrategy>(build_strategy, "ReduceStrategy")
      .value("Reduce", BuildStrategy::ReduceStrategy::kReduce)
      .value("AllReduce", BuildStrategy::ReduceStrategy::kAllReduce);
  py::enum_<BuildStrategy::GradientScaleStrategy>(build_strategy,
                                                  "GradientScaleStrategy")
      .value("CoeffNumDevice",
             BuildStrategy::GradientScaleStrategy::kCoeffNumDevice)
      .value("One", BuildStrategy::GradientScaleStrategy::kOne)
      .value("Customized", BuildStrategy::GradientScaleStrategy::kCustomized);

  build_strategy.def(py::init())
      .def_property(
          "reduce_strategy",
          [](const BuildStrategy &self) { return self.reduce_; },
          [](BuildStrategy &self, BuildStrategy::ReduceStrategy strategy) {
X
Xin Pan 已提交
1189
            PADDLE_ENFORCE(!self.IsFinalized(), "BuildStrategy is finlaized.");
Y
yuyang18 已提交
1190
            self.reduce_ = strategy;
C
chengduo 已提交
1191 1192 1193 1194 1195 1196 1197
          },
          R"DOC(The type is STR, there are two reduce strategies in ParallelExecutor,
                  'AllReduce' and 'Reduce'. If you want that all the parameters'
                  optimization are done on all devices independently, you should choose 'AllReduce';
                  if you choose 'Reduce', all the parameters' optimization will be evenly distributed
                  to different devices, and then broadcast the optimized parameter to other devices.
                  In some models, `Reduce` is faster. Default 'AllReduce'. )DOC")
Y
yuyang18 已提交
1198 1199 1200 1201 1202
      .def_property(
          "gradient_scale_strategy",
          [](const BuildStrategy &self) { return self.gradient_scale_; },
          [](BuildStrategy &self,
             BuildStrategy::GradientScaleStrategy strategy) {
X
Xin Pan 已提交
1203
            PADDLE_ENFORCE(!self.IsFinalized(), "BuildStrategy is finlaized.");
Y
yuyang18 已提交
1204
            self.gradient_scale_ = strategy;
C
chengduo 已提交
1205 1206 1207 1208 1209 1210
          },
          R"DOC(The type is STR, there are three ways of defining :math:`loss@grad` in
                   ParallelExecutor, 'CoeffNumDevice', 'One' and 'Customized'. By default,
                   ParallelExecutor sets the :math:`loss@grad` according to the number of devices.
                   If you want to customize :math:`loss@grad`, you can choose 'Customized'.
                   Default 'CoeffNumDevice'.)DOC")
Y
yuyang18 已提交
1211 1212 1213 1214
      .def_property(
          "debug_graphviz_path",
          [](const BuildStrategy &self) { return self.debug_graphviz_path_; },
          [](BuildStrategy &self, const std::string &path) {
X
Xin Pan 已提交
1215
            PADDLE_ENFORCE(!self.IsFinalized(), "BuildStrategy is finlaized.");
Y
yuyang18 已提交
1216
            self.debug_graphviz_path_ = path;
C
chengduo 已提交
1217 1218 1219 1220
          },
          R"DOC(The type is STR, debug_graphviz_path indicate the path that
                    writing the SSA Graph to file in the form of graphviz, you.
                    It is useful for debugging. Default "")DOC")
S
sneaxiy 已提交
1221 1222 1223 1224 1225 1226
      .def_property(
          "enable_sequential_execution",
          [](const BuildStrategy &self) {
            return self.enable_sequential_execution_;
          },
          [](BuildStrategy &self, bool b) {
X
Xin Pan 已提交
1227
            PADDLE_ENFORCE(!self.IsFinalized(), "BuildStrategy is finlaized.");
S
sneaxiy 已提交
1228 1229 1230 1231 1232 1233 1234 1235 1236
            self.enable_sequential_execution_ = b;
          },
          R"DOC(The type is BOOL. If set True, the execution order of ops would be the same as what is in the program. Default False.)DOC")
      .def_property(
          "remove_unnecessary_lock",
          [](const BuildStrategy &self) {
            return self.remove_unnecessary_lock_;
          },
          [](BuildStrategy &self, bool b) {
X
Xin Pan 已提交
1237
            PADDLE_ENFORCE(!self.IsFinalized(), "BuildStrategy is finlaized.");
S
sneaxiy 已提交
1238 1239
            self.remove_unnecessary_lock_ = b;
          },
S
sneaxiy 已提交
1240
          R"DOC(The type is BOOL. If set True, some locks in GPU ops would be released and ParallelExecutor would run faster. Default True.)DOC")
1241 1242 1243 1244 1245 1246
      .def_property(
          "num_trainers",
          [](const BuildStrategy &self) { return self.num_trainers_; },
          [](BuildStrategy &self, int num_trainers) {
            self.num_trainers_ = num_trainers;
          })
1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258
      .def_property(
          "trainers_endpoints",
          [](const BuildStrategy &self) { return self.trainers_endpoints_; },
          [](BuildStrategy &self,
             const std::vector<std::string> &trainers_endpoints) {
            self.trainers_endpoints_ = trainers_endpoints;
          })
      .def_property("trainer_id",
                    [](const BuildStrategy &self) { return self.trainer_id_; },
                    [](BuildStrategy &self, int trainer_id) {
                      self.trainer_id_ = trainer_id;
                    })
C
chengduo 已提交
1259 1260 1261 1262 1263 1264
      .def_property(
          "fuse_elewise_add_act_ops",
          [](const BuildStrategy &self) {
            return self.fuse_elewise_add_act_ops_;
          },
          [](BuildStrategy &self, bool b) {
X
Xin Pan 已提交
1265
            PADDLE_ENFORCE(!self.IsFinalized(), "BuildStrategy is finlaized.");
C
chengduo 已提交
1266 1267 1268 1269 1270
            self.fuse_elewise_add_act_ops_ = b;
          },
          R"DOC(The type is BOOL, fuse_elewise_add_act_ops indicate whether
                     to fuse elementwise_add_op and activation_op,
                     it may make the execution faster. Default False)DOC")
1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284
      .def_property(
          "fuse_relu_depthwise_conv",
          [](const BuildStrategy &self) {
            return self.fuse_relu_depthwise_conv_;
          },
          [](BuildStrategy &self, bool b) {
            PADDLE_ENFORCE(!self.IsFinalized(), "BuildStrategy is finlaized.");
            self.fuse_relu_depthwise_conv_ = b;
          },
          R"DOC(The type is BOOL, fuse_relu_depthwise_conv indicate whether
                      to fuse relu and depthwise_conv2d,
                      it will save GPU memory and may make the execution faster.
                      This options is only available in GPU devices.
                      Default False)DOC")
C
chengduo 已提交
1285 1286 1287 1288 1289 1290 1291 1292 1293
      .def_property("fuse_all_optimizer_ops",
                    [](const BuildStrategy &self) {
                      return self.fuse_all_optimizer_ops_;
                    },
                    [](BuildStrategy &self, bool b) {
                      PADDLE_ENFORCE(!self.IsFinalized(),
                                     "BuildStrategy is finlaized.");
                      self.fuse_all_optimizer_ops_ = b;
                    })
Q
qingqing01 已提交
1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308
      .def_property(
          "sync_batch_norm",
          [](const BuildStrategy &self) { return self.sync_batch_norm_; },
          [](BuildStrategy &self, bool b) {
            PADDLE_ENFORCE(!self.IsFinalized(), "BuildStrategy is finlaized.");
            self.sync_batch_norm_ = b;
          },
          R"DOC(The type is BOOL, sync_batch_norm indicates whether to use
                synchronous batch normalization which synchronizes the mean
                and variance through multi-devices in training phase.

                Current implementation doesn't support FP16 training and CPU.
                And only synchronous on one machine, not all machines.

                Default False)DOC")
D
dzhwinter 已提交
1309 1310 1311 1312
      .def_property(
          "memory_optimize",
          [](const BuildStrategy &self) { return self.memory_optimize_; },
          [](BuildStrategy &self, bool b) { self.memory_optimize_ = b; })
1313 1314 1315 1316
      .def_property(
          "is_distribution",
          [](const BuildStrategy &self) { return self.is_distribution_; },
          [](BuildStrategy &self, bool b) { self.is_distribution_ = b; })
D
dzhwinter 已提交
1317
      .def_property(
D
dzhwinter 已提交
1318 1319 1320
          "enable_inplace",
          [](const BuildStrategy &self) { return self.enable_inplace_; },
          [](BuildStrategy &self, bool b) { self.enable_inplace_ = b; })
C
chengduo 已提交
1321 1322 1323 1324
      .def_property(
          "fuse_all_reduce_ops",
          [](const BuildStrategy &self) { return self.fuse_all_reduce_ops_; },
          [](BuildStrategy &self, bool b) { self.fuse_all_reduce_ops_ = b; })
1325
      .def("_finalize_strategy_and_create_passes",
X
fix  
Xin Pan 已提交
1326
           [](BuildStrategy &self) -> std::shared_ptr<ir::PassBuilder> {
1327 1328 1329 1330 1331
             return self.CreatePassesFromStrategy(true);
           },
           R"DOC(Allow user to customized passes. Normally model-specific
                optimization passes should be defined in this way. BuildStrategy
                cannot be updated after being finalized.)DOC");
Y
yuyang18 已提交
1332 1333

  pe.def(py::init<const std::vector<platform::Place> &,
Y
Yan Xu 已提交
1334
                  const std::vector<std::string> &, const std::string &,
X
Xin Pan 已提交
1335
                  Scope *, std::vector<Scope *> &, const ExecutionStrategy &,
X
Xin Pan 已提交
1336
                  const BuildStrategy &, ir::Graph *>())
Y
Yu Yang 已提交
1337 1338 1339 1340
      // NOTE: even we return a vec<Scope*>* to Python use reference policy.
      // We still cannot get local_scope from this vector, since the element
      // of vec<Scope*> will be freed by Python GC. We can only return Scope*
      // one by one and mark them as reference.
1341 1342 1343 1344 1345
      .def("local_scopes",
           [](ParallelExecutor &self) -> std::vector<Scope *> * {
             return &self.GetLocalScopes();
           },
           py::return_value_policy::reference)
Y
Yu Yang 已提交
1346 1347 1348 1349
      .def("feed_tensors_into_local_scopes",
           &ParallelExecutor::FeedTensorsIntoLocalScopes)
      .def("feed_and_split_tensor_into_local_scopes",
           &ParallelExecutor::FeedAndSplitTensorIntoLocalScopes)
S
sneaxiy 已提交
1350 1351 1352 1353 1354 1355
      .def("run", [](ParallelExecutor &self,
                     const std::vector<std::string> &fetch_tensors,
                     const std::string &fetched_var_name) {
        pybind11::gil_scoped_release release;
        self.Run(fetch_tensors, fetched_var_name);
      });
Y
Yu Yang 已提交
1356

1357
  BindRecordIOWriter(&m);
W
Wang Guibao 已提交
1358
  BindAsyncExecutor(&m);
F
flame 已提交
1359 1360
  BindGraph(&m);
  BindNode(&m);
F
flame 已提交
1361
  BindInferenceApi(&m);
L
Luo Tao 已提交
1362
}
1363
}  // namespace pybind
1364
}  // namespace paddle