Use Python C-API to speed up dygraph trace (#17837)

* use python api to reduce python time cost, test=develop

* fix travis ci, test=develop

* fix Py_None error,test=develop

paddle/fluid/pybind/imperative.cc

@@ -14,11 +14,14 @@ limitations under the License. */
 
 #include "paddle/fluid/pybind/imperative.h"
 
+#include <Python.h>
 #include <pybind11/chrono.h>
 #include <pybind11/complex.h>
 #include <pybind11/functional.h>
 #include <pybind11/stl.h>
 #include <memory>
+#include <unordered_map>
+#include <utility>
 
 #include "paddle/fluid/framework/block_desc.h"
 #include "paddle/fluid/imperative/layer.h"
@@ -31,6 +34,8 @@ limitations under the License. */
 namespace paddle {
 namespace pybind {
 
+namespace py = ::pybind11;
+
 class Layer : public imperative::Layer {
  public:
   using imperative::Layer::Layer;  // Inherit constructors
@@ -51,10 +56,102 @@ class PYBIND11_HIDDEN PyOpBase : public imperative::OpBase {
   PyOpBase(const std::string &name) : OpBase(name) {}
 };
 
+// Function like obj.attr_name in Python.
+static PyObject *GetPythonAttribute(PyObject *obj, const char *attr_name) {
+  // NOTE(zjl): PyObject_GetAttrString would return nullptr when attr_name
+  // is not inside obj, but it would also set the error flag of Python.
+  // If the error flag is set in C++, C++ code would not raise Exception,
+  // but Python would raise Exception once C++ call ends.
+  // To avoid unexpected Exception raised in Python, we check whether
+  // attribute exists before calling PyObject_GetAttrString.
+  //
+  // Caution: PyObject_GetAttrString would increase reference count of PyObject.
+  // Developer should call Py_DECREF manually after the attribute is not used.
+  if (PyObject_HasAttrString(obj, attr_name)) {
+    return PyObject_GetAttrString(obj, attr_name);
+  } else {
+    return nullptr;
+  }
+}
+
+template <typename T>
+static T PyObjectCast(PyObject *obj) {
+  try {
+    return py::cast<T>(py::handle(obj));
+  } catch (py::cast_error &) {
+    PADDLE_THROW("Python object is not type of %s", typeid(T).name());
+  }
+}
+
+// NOTE(zjl): py::handle is a very light wrapper of PyObject *.
+// Unlike py::object, py::handle does not change reference count of PyObject *.
+static std::vector<std::shared_ptr<imperative::VarBase>>
+GetVarBaseListFromPyHandle(const py::handle &handle) {
+  PyObject *py_obj = handle.ptr();  // get underlying PyObject
+  // Python None is not nullptr in C++!
+  if (!py_obj || py_obj == Py_None) {
+    return {};
+  }
+
+  const char *kIVarField = "_ivar";
+  PyObject *py_ivar = GetPythonAttribute(py_obj, kIVarField);
+  std::vector<std::shared_ptr<imperative::VarBase>> result;
+
+  if (py_ivar) {  // Variable
+    result.emplace_back(
+        PyObjectCast<std::shared_ptr<imperative::VarBase>>(py_ivar));
+    Py_DECREF(py_ivar);
+  } else if (PyList_Check(py_obj)) {  // List of Variable
+    size_t len = PyList_GET_SIZE(py_obj);
+    result.reserve(len);
+    for (size_t i = 0; i < len; ++i) {
+      PyObject *py_ivar =
+          PyObject_GetAttrString(PyList_GET_ITEM(py_obj, i), kIVarField);
+      PADDLE_ENFORCE_NOT_NULL(py_ivar);
+      result.emplace_back(
+          PyObjectCast<std::shared_ptr<imperative::VarBase>>(py_ivar));
+      Py_DECREF(py_ivar);
+    }
+  } else if (PyTuple_Check(py_obj)) {  // Tuple of Variable
+    size_t len = PyTuple_GET_SIZE(py_obj);
+    result.reserve(len);
+    for (size_t i = 0; i < len; ++i) {
+      PyObject *py_ivar =
+          PyObject_GetAttrString(PyTuple_GET_ITEM(py_obj, i), kIVarField);
+      PADDLE_ENFORCE_NOT_NULL(py_ivar);
+      result.emplace_back(
+          PyObjectCast<std::shared_ptr<imperative::VarBase>>(py_ivar));
+      Py_DECREF(py_ivar);
+    }
+  } else {
+    PADDLE_THROW(
+        "unsupported type %s, must be Variable, List[Variable] or "
+        "tuple[Variable]",
+        py::str(handle));
+  }
+
+  PADDLE_ENFORCE(PyErr_Occurred() == nullptr,
+                 py::str(py::handle(PyErr_Occurred())));
+
+  return result;
+}
+
+using PyVarBaseMap = std::unordered_map<std::string, py::handle>;
+
+static imperative::VarBasePtrMap ConvertToVarBasePtrMap(
+    const PyVarBaseMap &map) {
+  imperative::VarBasePtrMap result;
+  for (auto &pair : map) {
+    auto var_vec = GetVarBaseListFromPyHandle(pair.second);
+    if (!var_vec.empty()) {
+      result.emplace(pair.first, std::move(var_vec));
+    }
+  }
+  return result;
+}
+
 // Bind Methods
 void BindImperative(pybind11::module *m_ptr) {
-  namespace py = ::pybind11;
-
   auto &m = *m_ptr;
 
   py::class_<imperative::detail::BackwardStrategy> backward_strategy(
@@ -145,31 +242,41 @@ void BindImperative(pybind11::module *m_ptr) {
              return self.Forward(inputs);
            });
 
-  py::class_<imperative::Tracer>(*m, "Tracer", "")
+  // NOTE(zjl): Tracer use PyVarBaseMap as its parameter but not VarBasePtrMap.
+  // We call Python C-API to convert PyVarBaseMap to VarBasePtrMap, instead
+  // making conversion in Python code. This speed up Tracer.trace() about 6%
+  // in ptb model and make time cost in Python to be nearly zero.
+  py::class_<imperative::Tracer>(m, "Tracer", "")
       .def("__init__",
            [](imperative::Tracer &self, framework::BlockDesc *root_block) {
              new (&self) imperative::Tracer(root_block);
            })
       .def("trace",
            [](imperative::Tracer &self, imperative::OpBase *op,
-              const imperative::VarBasePtrMap &inputs,
-              imperative::VarBasePtrMap *outputs,
+              const PyVarBaseMap &inputs, const PyVarBaseMap &outputs,
               framework::AttributeMap attrs_map,
               const platform::CPUPlace expected_place,
               const bool stop_gradient = false) {
+             auto ins = ConvertToVarBasePtrMap(inputs);
+             auto outs = ConvertToVarBasePtrMap(outputs);
+             {
                py::gil_scoped_release release;
-             self.Trace(op, inputs, outputs, attrs_map, expected_place,
+               self.Trace(op, std::move(ins), &outs, attrs_map, expected_place,
                           stop_gradient);
+             }
            })
       .def("trace", [](imperative::Tracer &self, imperative::OpBase *op,
-                       const imperative::VarBasePtrMap &inputs,
-                       imperative::VarBasePtrMap *outputs,
+                       const PyVarBaseMap &inputs, const PyVarBaseMap &outputs,
                        framework::AttributeMap attrs_map,
                        const platform::CUDAPlace expected_place,
                        const bool stop_gradient = false) {
+        auto ins = ConvertToVarBasePtrMap(inputs);
+        auto outs = ConvertToVarBasePtrMap(outputs);
+        {
           py::gil_scoped_release release;
-        self.Trace(op, inputs, outputs, attrs_map, expected_place,
+          self.Trace(op, std::move(ins), &outs, attrs_map, expected_place,
                      stop_gradient);
+        }
       });
 
   // define parallel context

python/paddle/fluid/dygraph/tracer.py

@@ -52,27 +52,10 @@ class Tracer(core.Tracer):
         self._trace_id = 0
 
     def trace_op(self, op, inputs, outputs, stop_gradient=False):
-        # TODO(hy): previous version will cause memory failed
-        inps = defaultdict(list)
-        for k, vars in six.iteritems(inputs):
-            if isinstance(vars, framework.Variable):
-                inps[k].append(vars._ivar)
-            elif isinstance(vars, list) or isinstance(vars, tuple):
-                for var in vars:
-                    inps[k].append(var._ivar)
-
-        outs = defaultdict(list)
-        for k, vars in six.iteritems(outputs):
-            if isinstance(vars, framework.Variable):
-                outs[k].append(vars._ivar)
-            elif isinstance(vars, list) or isinstance(vars, tuple):
-                for var in vars:
-                    outs[k].append(var._ivar)
-
         # record op's trace id
         op.iop._trace_id = self._trace_id
 
-        self.trace(op.iop, inps, outs, op.attrs,
+        self.trace(op.iop, inputs, outputs, op.attrs,
                    framework._current_expected_place(), stop_gradient)
 
         if not stop_gradient and self._train_mode: