Merge branch 'develop' into fluid_infer

Author: luotao1

paddle/operators/compare_op.cc

@@ -39,6 +39,11 @@ N-dim tensor. X and Y could be any type.  The each element of the Out tensor is
 calculated by %s
 )DOC",
                                comment.type, comment.equation));
+    AddAttr<int>("axis",
+                 "(int, default -1). The start dimension index "
+                 "for broadcasting Y onto X.")
+        .SetDefault(-1)
+        .EqualGreaterThan(-1);
   }
 };
 
@@ -95,11 +100,5 @@ REGISTER_LOGICAL_OP(less_than, "Out = X < Y");
 REGISTER_LOGICAL_KERNEL(less_than, CPU, paddle::operators::LessThanFunctor);
 REGISTER_LOGICAL_OP(less_equal, "Out = X <= Y");
 REGISTER_LOGICAL_KERNEL(less_equal, CPU, paddle::operators::LessEqualFunctor);
-REGISTER_LOGICAL_OP(greater_than, "Out = X > Y");
-REGISTER_LOGICAL_KERNEL(greater_than, CPU,
-                        paddle::operators::GreaterThanFunctor);
-REGISTER_LOGICAL_OP(greater_equal, "Out = X >= Y");
-REGISTER_LOGICAL_KERNEL(greater_equal, CPU,
-                        paddle::operators::GreaterEqualFunctor);
 REGISTER_LOGICAL_OP(equal, "Out = X == Y");
 REGISTER_LOGICAL_KERNEL(equal, CPU, paddle::operators::EqualFunctor);

paddle/operators/compare_op.cu

@@ -16,8 +16,4 @@ limitations under the License. */
 
 REGISTER_LOGICAL_KERNEL(less_than, CUDA, paddle::operators::LessThanFunctor);
 REGISTER_LOGICAL_KERNEL(less_equal, CUDA, paddle::operators::LessEqualFunctor);
-REGISTER_LOGICAL_KERNEL(greater_than, CUDA,
-                        paddle::operators::GreaterThanFunctor);
-REGISTER_LOGICAL_KERNEL(greater_equal, CUDA,
-                        paddle::operators::GreaterEqualFunctor);
 REGISTER_LOGICAL_KERNEL(equal, CUDA, paddle::operators::EqualFunctor);

paddle/operators/compare_op.h

@@ -16,6 +16,7 @@ limitations under the License. */
 #include <math.h>
 #include <type_traits>
 #include "paddle/framework/op_registry.h"
+#include "paddle/operators/elementwise_op_function.h"
 #include "paddle/platform/transform.h"
 
 namespace paddle {
@@ -33,18 +34,6 @@ struct LessEqualFunctor {
   HOSTDEVICE bool operator()(const T& a, const T& b) const { return a <= b; }
 };
 
-template <typename T>
-struct GreaterThanFunctor {
-  using ELEM_TYPE = T;
-  HOSTDEVICE bool operator()(const T& a, const T& b) const { return a > b; }
-};
-
-template <typename T>
-struct GreaterEqualFunctor {
-  using ELEM_TYPE = T;
-  HOSTDEVICE bool operator()(const T& a, const T& b) const { return a >= b; }
-};
-
 template <typename T>
 struct EqualFunctor {
   using ELEM_TYPE = T;
@@ -65,14 +54,7 @@ class CompareOpKernel
  public:
   void Compute(const framework::ExecutionContext& context) const override {
     using T = typename Functor::ELEM_TYPE;
-    auto* x = context.Input<framework::Tensor>("X");
-    auto* y = context.Input<framework::Tensor>("Y");
-    auto* out = context.Output<framework::Tensor>("Out");
-    Functor binary_func;
-    platform::Transform<DeviceContext> trans;
-    trans(context.template device_context<DeviceContext>(), x->data<T>(),
-          x->data<T>() + x->numel(), y->data<T>(),
-          out->mutable_data<bool>(context.GetPlace()), binary_func);
+    ElementwiseComputeEx<Functor, DeviceContext, T, bool>(context);
   }
 };
 

paddle/operators/elementwise_op_function.h

@@ -176,14 +176,15 @@ class MidWiseTransformIterator<T, platform::CUDADeviceContext>
 };
 #endif
 
-template <typename Functor, typename T, typename DeviceContext>
+template <typename Functor, typename T, typename DeviceContext,
+          typename OutType = T>
 class TransformFunctor {
  public:
   TransformFunctor(const framework::Tensor* x, const framework::Tensor* y,
                    framework::Tensor* z, const DeviceContext& ctx, Functor func)
       : x_(x->data<T>()),
         y_(y->data<T>()),
-        z_(z->mutable_data<T>(ctx.GetPlace())),
+        z_(z->mutable_data<OutType>(ctx.GetPlace())),
         nx_(x->numel()),
         ctx_(ctx),
         func_(func) {}
@@ -208,7 +209,7 @@ class TransformFunctor {
  private:
   const T* x_;
   const T* y_;
-  T* z_;
+  OutType* z_;
   int64_t nx_;
   const DeviceContext& ctx_;
   Functor func_;
@@ -364,15 +365,16 @@ void ElementwiseGradCompute(const framework::ExecutionContext& ctx) {
   }
 }
 
-template <typename Functor, typename DeviceContext, typename T>
+template <typename Functor, typename DeviceContext, typename T,
+          typename OutType = T>
 void ElementwiseComputeEx(const framework::ExecutionContext& ctx) {
   using Tensor = framework::Tensor;
 
   auto* x = ctx.Input<Tensor>("X");
   auto* y = ctx.Input<Tensor>("Y");
   auto* z = ctx.Output<Tensor>("Out");
-  z->mutable_data<T>(ctx.GetPlace());
-  TransformFunctor<Functor, T, DeviceContext> functor(
+  z->mutable_data<OutType>(ctx.GetPlace());
+  TransformFunctor<Functor, T, DeviceContext, OutType> functor(
       x, y, z, ctx.template device_context<DeviceContext>(), Functor());
 
   auto x_dims = x->dims();

python/paddle/v2/fluid/layers/control_flow.py

@@ -11,13 +11,13 @@
 # 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.
+import contextlib
 
-from ..layer_helper import LayerHelper, unique_name
-from ..framework import Program, Variable, Operator
-from .. import core
+from layer_function_generator import autodoc
 from tensor import assign, fill_constant
-import contextlib
-from ..registry import autodoc
+from .. import core
+from ..framework import Program, Variable, Operator
+from ..layer_helper import LayerHelper, unique_name
 
 __all__ = [
     'split_lod_tensor',
@@ -1477,7 +1477,7 @@ def _assert_in_rnn_block_(self, method):
                 method))
 
 
-@autodoc
+@autodoc()
 def reorder_lod_tensor_by_rank(x, rank_table):
     helper = LayerHelper('reorder_lod_tensor_by_rank', **locals())
     helper.is_instance('x', Variable)

python/paddle/v2/fluid/layers/device.py

@@ -15,14 +15,14 @@
 All util layers.
 """
 
-from ..layer_helper import LayerHelper
+from layer_function_generator import autodoc
 from ..framework import unique_name
-from ..registry import autodoc
+from ..layer_helper import LayerHelper
 
 __all__ = ['get_places']
 
 
-@autodoc
+@autodoc()
 def get_places(device_count=None, device_type=None):
     helper = LayerHelper('get_places', **locals())
     out_places = helper.create_variable(name=unique_name(helper.name + ".out"))

python/paddle/v2/fluid/registry.py renamed to python/paddle/v2/fluid/layers/layer_function_generator.py

@@ -13,17 +13,19 @@
 # limitations under the License.
 import re
 import cStringIO
-import warnings
 import functools
-import inspect
+import warnings
+
+from .. import proto
 
-import proto.framework_pb2 as framework_pb2
-from framework import OpProtoHolder, Variable, Program, Operator
-from paddle.v2.fluid.layer_helper import LayerHelper, unique_name
+framework_pb2 = proto.framework_pb2
+
+from ..framework import OpProtoHolder, Variable
+from ..layer_helper import LayerHelper
 
 __all__ = [
     'deprecated',
-    'register_layer',
+    'generate_layer_fn',
     'autodoc',
 ]
 
@@ -96,7 +98,7 @@ def _type_to_str_(tp):
     return buf.getvalue()
 
 
-def register_layer(op_type):
+def generate_layer_fn(op_type):
     """Register the Python layer for an Operator.
 
     Args:
@@ -207,7 +209,10 @@ def func_wrapper(*args, **kwargs):
     return func_wrapper
 
 
-def autodoc(func):
-    func.__doc__ = _generate_doc_string_(OpProtoHolder.instance().get_op_proto(
-        func.__name__))
-    return func
+def autodoc(comment=""):
+    def __impl__(func):
+        func.__doc__ = _generate_doc_string_(OpProtoHolder.instance(
+        ).get_op_proto(func.__name__)) + comment
+        return func
+
+    return __impl__

python/paddle/v2/fluid/layers/ops.py

@@ -11,8 +11,7 @@
 # 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.
-
-from ..registry import register_layer
+from layer_function_generator import generate_layer_fn
 
 __activations__ = [
     'sigmoid',
@@ -53,4 +52,4 @@
 ] + __activations__
 
 for _OP in set(__all__):
-    globals()[_OP] = register_layer(_OP)
+    globals()[_OP] = generate_layer_fn(_OP)

python/paddle/v2/fluid/tests/book_memory_optimization/test_memopt_fit_a_line.py

@@ -1,3 +1,17 @@
+#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+# 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
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# 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.
+
 import numpy as np
 import paddle.v2 as paddle
 import paddle.v2.fluid as fluid

python/paddle/v2/fluid/tests/book_memory_optimization/test_memopt_image_classification_train.py

@@ -1,3 +1,17 @@
+#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+# 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
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# 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.
+
 from __future__ import print_function
 
 import sys