Merge branch 'release/1.3' into test/picked

test=release/1.3

Author: dzhwinter

paddle/fluid/API.spec

@@ -8,13 +8,13 @@ paddle.fluid.Program.parse_from_string ArgSpec(args=['binary_str'], varargs=None
 paddle.fluid.Program.to_string ArgSpec(args=['self', 'throw_on_error', 'with_details'], varargs=None, keywords=None, defaults=(False,))
 paddle.fluid.default_startup_program ArgSpec(args=[], varargs=None, keywords=None, defaults=None)
 paddle.fluid.default_main_program ArgSpec(args=[], varargs=None, keywords=None, defaults=None)
-paddle.fluid.program_guard ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
-paddle.fluid.name_scope ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
+paddle.fluid.program_guard ArgSpec(args=['main_program', 'startup_program'], varargs=None, keywords=None, defaults=(None,))
+paddle.fluid.name_scope ArgSpec(args=['prefix'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.Executor.__init__ ArgSpec(args=['self', 'place'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.Executor.close ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.Executor.run ArgSpec(args=['self', 'program', 'feed', 'fetch_list', 'feed_var_name', 'fetch_var_name', 'scope', 'return_numpy', 'use_program_cache'], varargs=None, keywords=None, defaults=(None, None, None, 'feed', 'fetch', None, True, False))
 paddle.fluid.global_scope ArgSpec(args=[], varargs=None, keywords=None, defaults=None)
-paddle.fluid.scope_guard ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
+paddle.fluid.scope_guard ArgSpec(args=['scope'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.DistributeTranspiler.__init__ ArgSpec(args=['self', 'config'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.DistributeTranspiler.get_pserver_program ArgSpec(args=['self', 'endpoint'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.DistributeTranspiler.get_pserver_programs ArgSpec(args=['self', 'endpoint'], varargs=None, keywords=None, defaults=None)
@@ -66,7 +66,7 @@ paddle.fluid.initializer.XavierInitializer.__init__ ArgSpec(args=['self', 'unifo
 paddle.fluid.initializer.BilinearInitializer.__init__ ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.initializer.MSRAInitializer.__init__ ArgSpec(args=['self', 'uniform', 'fan_in', 'seed'], varargs=None, keywords=None, defaults=(True, None, 0))
 paddle.fluid.initializer.force_init_on_cpu ArgSpec(args=[], varargs=None, keywords=None, defaults=None)
-paddle.fluid.initializer.init_on_cpu ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
+paddle.fluid.initializer.init_on_cpu ArgSpec(args=[], varargs=None, keywords=None, defaults=None)
 paddle.fluid.initializer.NumpyArrayInitializer.__init__ ArgSpec(args=['self', 'value'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.layers.fc ArgSpec(args=['input', 'size', 'num_flatten_dims', 'param_attr', 'bias_attr', 'act', 'is_test', 'name'], varargs=None, keywords=None, defaults=(1, None, None, None, False, None))
 paddle.fluid.layers.embedding ArgSpec(args=['input', 'size', 'is_sparse', 'is_distributed', 'padding_idx', 'param_attr', 'dtype'], varargs=None, keywords=None, defaults=(False, False, None, None, 'float32'))
@@ -229,7 +229,7 @@ paddle.fluid.layers.random_data_generator ArgSpec(args=['low', 'high', 'shapes',
 paddle.fluid.layers.py_reader ArgSpec(args=['capacity', 'shapes', 'dtypes', 'lod_levels', 'name', 'use_double_buffer'], varargs=None, keywords=None, defaults=(None, None, True))
 paddle.fluid.layers.create_py_reader_by_data ArgSpec(args=['capacity', 'feed_list', 'name', 'use_double_buffer'], varargs=None, keywords=None, defaults=(None, True))
 paddle.fluid.layers.Preprocessor.__init__ ArgSpec(args=['self', 'reader', 'name'], varargs=None, keywords=None, defaults=(None,))
-paddle.fluid.layers.Preprocessor.block ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
+paddle.fluid.layers.Preprocessor.block ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.layers.Preprocessor.inputs ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.layers.Preprocessor.outputs ArgSpec(args=['self'], varargs='outs', keywords=None, defaults=None)
 paddle.fluid.layers.load ArgSpec(args=['out', 'file_path', 'load_as_fp16'], varargs=None, keywords=None, defaults=(None,))
@@ -270,7 +270,7 @@ paddle.fluid.layers.IfElse.input ArgSpec(args=['self', 'x'], varargs=None, keywo
 paddle.fluid.layers.IfElse.output ArgSpec(args=['self'], varargs='outs', keywords=None, defaults=None)
 paddle.fluid.layers.IfElse.true_block ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.layers.DynamicRNN.__init__ ArgSpec(args=['self', 'name'], varargs=None, keywords=None, defaults=(None,))
-paddle.fluid.layers.DynamicRNN.block ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
+paddle.fluid.layers.DynamicRNN.block ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.layers.DynamicRNN.memory ArgSpec(args=['self', 'init', 'shape', 'value', 'need_reorder', 'dtype'], varargs=None, keywords=None, defaults=(None, None, 0.0, False, 'float32'))
 paddle.fluid.layers.DynamicRNN.output ArgSpec(args=['self'], varargs='outputs', keywords=None, defaults=None)
 paddle.fluid.layers.DynamicRNN.static_input ArgSpec(args=['self', 'x'], varargs=None, keywords=None, defaults=None)
@@ -346,12 +346,12 @@ paddle.fluid.contrib.StateCell.set_state ArgSpec(args=['self', 'state_name', 'st
 paddle.fluid.contrib.StateCell.state_updater ArgSpec(args=['self', 'updater'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.contrib.StateCell.update_states ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.contrib.TrainingDecoder.__init__ ArgSpec(args=['self', 'state_cell', 'name'], varargs=None, keywords=None, defaults=(None,))
-paddle.fluid.contrib.TrainingDecoder.block ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
+paddle.fluid.contrib.TrainingDecoder.block ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.contrib.TrainingDecoder.output ArgSpec(args=['self'], varargs='outputs', keywords=None, defaults=None)
 paddle.fluid.contrib.TrainingDecoder.static_input ArgSpec(args=['self', 'x'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.contrib.TrainingDecoder.step_input ArgSpec(args=['self', 'x'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.contrib.BeamSearchDecoder.__init__ ArgSpec(args=['self', 'state_cell', 'init_ids', 'init_scores', 'target_dict_dim', 'word_dim', 'input_var_dict', 'topk_size', 'sparse_emb', 'max_len', 'beam_size', 'end_id', 'name'], varargs=None, keywords=None, defaults=({}, 50, True, 100, 1, 1, None))
-paddle.fluid.contrib.BeamSearchDecoder.block ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
+paddle.fluid.contrib.BeamSearchDecoder.block ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.contrib.BeamSearchDecoder.decode ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.contrib.BeamSearchDecoder.early_stop ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.contrib.BeamSearchDecoder.read_array ArgSpec(args=['self', 'init', 'is_ids', 'is_scores'], varargs=None, keywords=None, defaults=(False, False))
@@ -456,7 +456,7 @@ paddle.fluid.optimizer.AdadeltaOptimizer.apply_gradients ArgSpec(args=['self', '
 paddle.fluid.optimizer.AdadeltaOptimizer.backward ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set', 'callbacks'], varargs=None, keywords=None, defaults=(None, None, None, None))
 paddle.fluid.optimizer.AdadeltaOptimizer.minimize ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set'], varargs=None, keywords=None, defaults=(None, None, None))
 paddle.fluid.optimizer.ModelAverage.__init__ ArgSpec(args=['self', 'average_window_rate', 'min_average_window', 'max_average_window', 'regularization', 'name'], varargs=None, keywords=None, defaults=(10000, 10000, None, None))
-paddle.fluid.optimizer.ModelAverage.apply ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
+paddle.fluid.optimizer.ModelAverage.apply ArgSpec(args=['self', 'executor', 'need_restore'], varargs=None, keywords=None, defaults=(True,))
 paddle.fluid.optimizer.ModelAverage.apply_gradients ArgSpec(args=['self', 'params_grads'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.optimizer.ModelAverage.backward ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set', 'callbacks'], varargs=None, keywords=None, defaults=(None, None, None, None))
 paddle.fluid.optimizer.ModelAverage.minimize ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set'], varargs=None, keywords=None, defaults=(None, None, None))
@@ -491,14 +491,14 @@ paddle.fluid.clip.ErrorClipByValue.__init__ ArgSpec(args=['self', 'max', 'min'],
 paddle.fluid.clip.GradientClipByValue.__init__ ArgSpec(args=['self', 'max', 'min'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.clip.GradientClipByNorm.__init__ ArgSpec(args=['self', 'clip_norm'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.clip.GradientClipByGlobalNorm.__init__ ArgSpec(args=['self', 'clip_norm', 'group_name'], varargs=None, keywords=None, defaults=('default_group',))
-paddle.fluid.profiler.cuda_profiler ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
+paddle.fluid.profiler.cuda_profiler ArgSpec(args=['output_file', 'output_mode', 'config'], varargs=None, keywords=None, defaults=(None, None))
 paddle.fluid.profiler.reset_profiler ArgSpec(args=[], varargs=None, keywords=None, defaults=None)
-paddle.fluid.profiler.profiler ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
+paddle.fluid.profiler.profiler ArgSpec(args=['state', 'sorted_key', 'profile_path'], varargs=None, keywords=None, defaults=(None, '/tmp/profile'))
 paddle.fluid.profiler.start_profiler ArgSpec(args=['state'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.profiler.stop_profiler ArgSpec(args=['sorted_key', 'profile_path'], varargs=None, keywords=None, defaults=(None, '/tmp/profile'))
 paddle.fluid.unique_name.generate ArgSpec(args=['key'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.unique_name.switch ArgSpec(args=['new_generator'], varargs=None, keywords=None, defaults=(None,))
-paddle.fluid.unique_name.guard ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
+paddle.fluid.unique_name.guard ArgSpec(args=['new_generator'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.recordio_writer.convert_reader_to_recordio_file ArgSpec(args=['filename', 'reader_creator', 'feeder', 'compressor', 'max_num_records', 'feed_order'], varargs=None, keywords=None, defaults=(Compressor.Snappy, 1000, None))
 paddle.fluid.recordio_writer.convert_reader_to_recordio_files ArgSpec(args=['filename', 'batch_per_file', 'reader_creator', 'feeder', 'compressor', 'max_num_records', 'feed_order'], varargs=None, keywords=None, defaults=(Compressor.Snappy, 1000, None))
 paddle.fluid.Scope Scope() -> paddle.fluid.core._Scope

paddle/fluid/inference/analysis/ir_passes/subgraph_detector.cc

@@ -460,77 +460,6 @@ inline bool CheckNodeIndegreeEquals(const Node &node, size_t n) {
   return node.inputs.size() == n;
 }
 
-NodesTSIterator::NodesTSIterator(const std::vector<Node *> &source) {
-  PADDLE_ENFORCE(!source.empty(),
-                 "Start points of topological sorting should not be empty!");
-  // CHECK all the inputs' in-degree is 0
-  for (auto *node : source) {
-    PADDLE_ENFORCE(CheckNodeIndegreeEquals(*node, 0));
-  }
-
-  std::unordered_set<Node *> visited;
-  std::unordered_set<Node *> to_visit{source.begin(), source.end()};
-
-  std::vector<Node *> inlink_visited;
-  while (!to_visit.empty()) {
-    std::vector<Node *> queue(to_visit.begin(), to_visit.end());
-    for (auto *p : queue) {
-      if (Agent(p).deleted()) {
-        visited.insert(p);
-        to_visit.erase(p);
-      }
-
-      inlink_visited.clear();
-
-      std::copy_if(p->inputs.begin(), p->inputs.end(),
-                   std::back_inserter(inlink_visited),
-                   [&](Node *x) -> bool { return visited.count(x) != 0; });
-
-      if (inlink_visited.size() == p->inputs.size()) {
-        sorted_.push_back(p);
-        for (auto *_ : p->outputs) {
-          if (!visited.count(_)) {
-            to_visit.insert(_);
-          }
-        }
-
-        to_visit.erase(p);
-        visited.insert(p);
-      }
-    }
-  }
-}
-
-NodesTSIterator::NodesTSIterator(const NodesTSIterator &other)
-    : sorted_(other.sorted_), cursor_(other.cursor_) {}
-
-Node &NodesTSIterator::operator*() {
-  PADDLE_ENFORCE_LT(cursor_, sorted_.size());
-  return *sorted_[cursor_];
-}
-
-NodesTSIterator &NodesTSIterator::operator++() {
-  if (++cursor_ >= sorted_.size()) {
-    sorted_.clear();
-    cursor_ = 0;
-  }
-  return *this;
-}
-NodesTSIterator &NodesTSIterator::operator=(const NodesTSIterator &other) {
-  cursor_ = other.cursor_;
-  sorted_ = other.sorted_;
-  return *this;
-}
-
-bool NodesTSIterator::operator==(const NodesTSIterator &other) {
-  return sorted_ == other.sorted_ && cursor_ == other.cursor_;
-}
-
-Node *NodesTSIterator::operator->() {
-  PADDLE_ENFORCE_LT(cursor_, sorted_.size());
-  return sorted_[cursor_];
-}
-
 }  // namespace analysis
 }  // namespace inference
 }  // namespace paddle

paddle/fluid/inference/analysis/ir_passes/subgraph_detector.h

@@ -30,6 +30,7 @@ namespace inference {
 namespace analysis {
 
 using framework::ir::Graph;
+using framework::ir::NodesTSIterator;
 
 const char kIsFunctionNode[] = "__is_function_node__";
 const char kFunctionNodeSubGraph[] = "__function_node_sub_graph__";
@@ -132,32 +133,6 @@ struct Agent {
   framework::ir::Node *x_;
 };
 
-// Topological sorting iterator on nodes.
-struct NodesTSIterator
-    : public std::iterator<std::forward_iterator_tag, framework::ir::Node *> {
-  NodesTSIterator() = default;
-  explicit NodesTSIterator(const std::vector<framework::ir::Node *> &source);
-  NodesTSIterator(NodesTSIterator &&other)
-      : sorted_(std::move(other.sorted_)), cursor_(other.cursor_) {
-    other.cursor_ = 0;
-  }
-  NodesTSIterator(const NodesTSIterator &other);
-
-  framework::ir::Node &operator*();
-  NodesTSIterator &operator++();
-  // TODO(Superjomn) current implementation just compare the first
-  // element, need to compare the graph and all the elements in the queue and
-  // set.
-  NodesTSIterator &operator=(const NodesTSIterator &other);
-  bool operator==(const NodesTSIterator &other);
-  bool operator!=(const NodesTSIterator &other) { return !(*this == other); }
-  framework::ir::Node *operator->();
-
- private:
-  std::vector<framework::ir::Node *> sorted_;
-  size_t cursor_{0};
-};
-
 // The nodes those have no input will be treated as start points.
 static std::vector<framework::ir::Node *> ExtractStartPoints(const Graph &g) {
   std::vector<framework::ir::Node *> result;

paddle/fluid/operators/fake_quantize_op.cc

@@ -21,26 +21,17 @@ limitations under the License. */
 namespace paddle {
 namespace operators {
 
-template <typename T, int MajorType = Eigen::RowMajor,
-          typename IndexType = Eigen::DenseIndex>
-using EigenVectorArrayMap =
-    Eigen::TensorMap<Eigen::Tensor<T, 1, MajorType, IndexType>>;
-
-template <typename T, int MajorType = Eigen::RowMajor,
-          typename IndexType = Eigen::DenseIndex>
-using ConstEigenVectorArrayMap =
-    Eigen::TensorMap<const Eigen::Tensor<T, 1, MajorType, IndexType>>;
+template <typename T>
+struct Compare {
+ public:
+  bool operator()(const T a, const T b) { return (std::abs(a) < std::abs(b)); }
+};
 
 template <typename T>
 struct FindAbsMaxFunctor<platform::CPUDeviceContext, T> {
   void operator()(const platform::CPUDeviceContext& ctx, const T* in,
                   const int num, T* out) {
-    Eigen::DSizes<Eigen::DenseIndex, 1> idim(num);
-    Eigen::DSizes<Eigen::DenseIndex, 1> odim(1);
-    Eigen::TensorMap<Eigen::Tensor<const T, 1, Eigen::RowMajor>> in_e(in, idim);
-    Eigen::TensorMap<Eigen::Tensor<T, 1, Eigen::RowMajor>> out_e(out, odim);
-
-    out_e = in_e.abs().maximum();
+    *out = *(std::max_element(in + 0, in + num, Compare<T>()));
   }
 };
 

paddle/fluid/operators/jit/gen/act.h

@@ -63,7 +63,6 @@ class VActFunc : public JitCode {
  public:
   explicit VActFunc(size_t code_size, void* code_ptr)
       : JitCode(code_size, code_ptr) {}
-  virtual const char* name() const = 0;
   virtual void genCode() = 0;
 
  protected:
@@ -269,7 +268,7 @@ class VActJitCode : public VActFunc {
     this->genCode();
   }
 
-  const char* name() const override {
+  std::string name() const override {
     std::string base = "VActJitCode";
     switch (type_) {
       case operand_type::RELU:
@@ -293,7 +292,7 @@ class VActJitCode : public VActFunc {
       default:
         break;
     }
-    return base.c_str();
+    return base;
   }
   void genCode() override;
 

paddle/fluid/operators/jit/gen/blas.h

@@ -41,7 +41,7 @@ class VXXJitCode : public JitCode {
     this->genCode();
   }
 
-  virtual const char* name() const {
+  std::string name() const override {
     std::string base = "VXXJitCode";
     if (scalar_index_ == 1) {
       base += "_Scalar";
@@ -62,7 +62,7 @@ class VXXJitCode : public JitCode {
     }
     base += (with_relu_ ? "_Relu" : "");
     base += "_D" + std::to_string(num_);
-    return base.c_str();
+    return base;
   }
   void genCode() override;
 

paddle/fluid/operators/jit/gen/gru.h

@@ -49,7 +49,7 @@ class GRUJitCode : public VActFunc {
     this->genCode();
   }
 
-  const char* name() const override {
+  std::string name() const override {
     std::string base = "GRUJitCode";
     if (id_ == 0) {
       base += "_H1";
@@ -81,7 +81,7 @@ class GRUJitCode : public VActFunc {
     };
     AddTypeStr(act_gate_);
     AddTypeStr(act_cand_);
-    return base.c_str();
+    return base;
   }
   void genCode() override;
 

paddle/fluid/operators/jit/gen/hopv.h

@@ -35,14 +35,14 @@ class HOPVJitCode : public JitCode {
     this->genCode();
   }
 
-  virtual const char* name() const {
+  std::string name() const override {
     std::string base = "VXXJitCode";
     if (type_ == operand_type::MAX) {
       base += "_MAX";
     } else {
       base += "_SUM";
     }
-    return base.c_str();
+    return base;
   }
   void genCode() override;
 

paddle/fluid/operators/jit/gen/jitcode.h

@@ -14,6 +14,7 @@
 
 #pragma once
 
+#include <string>
 #include <type_traits>
 #include "paddle/fluid/operators/jit/gen_base.h"
 #include "paddle/fluid/platform/cpu_info.h"
@@ -59,7 +60,7 @@ typedef enum {
 } operand_type;
 
 #define DECLARE_JIT_CODE(codename) \
-  const char* name() const override { return #codename; }
+  std::string name() const override { return #codename; }
 
 class JitCode : public GenBase, public Xbyak::CodeGenerator {
  public:
@@ -68,7 +69,6 @@ class JitCode : public GenBase, public Xbyak::CodeGenerator {
             (code_size % 4096 != 0 ? (code_size / 4096 + 1) * 4096 : code_size),
             code_ptr) {}
 
-  virtual const char* name() const = 0;
   virtual void genCode() = 0;
 
   size_t getSize() const override { return CodeGenerator::getSize(); }

paddle/fluid/operators/jit/gen/lstm.h

@@ -53,7 +53,7 @@ class LSTMJitCode : public VActFunc {
     this->genCode();
   }
 
-  const char* name() const override {
+  std::string name() const override {
     std::string base = "LSTMJitCode";
     if (use_peephole_) {
       base += "_Peephole";
@@ -85,7 +85,7 @@ class LSTMJitCode : public VActFunc {
     AddTypeStr(act_gate_);
     AddTypeStr(act_cand_);
     AddTypeStr(act_cell_);
-    return base.c_str();
+    return base;
   }
   void genCode() override;