Merge branch 'develop' into api

Author: luotao1

benchmark/fluid/fluid_benchmark.py

@@ -210,7 +210,7 @@ def train_parallel(avg_loss, infer_prog, optimizer, train_reader, test_reader,
     # generate fake:
     if args.use_fake_data:
         for var in feed_var_list:
-            v = startup_prog.global_block().clone_variable(var)
+            v = startup_prog.global_block()._clone_variable(var)
             var.persistable = True
             v.persistable = True
 

doc/fluid/design/modules/python_api.md

@@ -98,13 +98,13 @@ class Block(objects):
     def append_operator(self, ...):
         self.ops.append(Operator(self, ...))
 
-    def prepend_operator(self, ...): # Parameter's ctor prepands initialize operators.
+    def _prepend_operator(self, ...): # Parameter's ctor prepands initialize operators.
        self.ops.prepend(Operator(self, ...))
 ```
 
 `create_parameter` is necessary because parameters are global variables, defined in the global block, but can be created in some sub-blocks. For example, an FC layer in the step block of an RNN operator.
 
-`prepend_operator` is necessary because the constructor of `Parameter` needs to create the initialize (or load) operator of the parameter, and would like to put it in the *preamble* of the global block.
+`_prepend_operator` is necessary because the constructor of `Parameter` needs to create the initialize (or load) operator of the parameter, and would like to put it in the *preamble* of the global block.
 
 ### Operator
 

doc/fluid/howto/performance/error_clip.md

@@ -78,7 +78,7 @@ def error_clip_callback(block, context):
     op_desc = block.desc.op(block.desc.op_size() - 1)
     for grad_n in filter(lambda n: grad_to_var.has_key(n),
                          op_desc.output_arg_names()):
-        fwd_var = block.var_recursive(grad_to_var[grad_n])
+        fwd_var = block.__var_recursive(grad_to_var[grad_n])
         error_clip = getattr(fwd_var, "error_clip", None)
         if not (error_clip is None or isinstance(error_clip,
                                                  BaseErrorClipAttr)):

doc/v2/build_and_install/build_from_source_cn.rst

@@ -35,11 +35,16 @@ PaddlePaddle需要使用Docker环境完成编译，这样可以免去单独安
    # 2. 可选步骤：源码中构建用于编译PaddlePaddle的Docker镜像
    docker build -t paddle:dev .
    # 3. 执行下面的命令编译CPU-Only的二进制
-   docker run -it -v $PWD:/paddle -w /paddle -e "WITH_GPU=OFF" -e "WITH_TESTING=OFF" paddlepaddle/paddle_manylinux_devel:cuda8.0_cudnn5 ./paddle/scripts/paddle_build.sh build
+   docker run -it -v $PWD:/paddle -w /paddle -e "PYTHON_ABI=cp27-cp27mu" -e "WITH_GPU=OFF" -e "WITH_TESTING=OFF" paddlepaddle/paddle_manylinux_devel:cuda8.0_cudnn5 ./paddle/scripts/paddle_build.sh build
    # 4. 或者也可以使用为上述可选步骤构建的镜像（必须先执行第2步）
    docker run -it -v $PWD:/paddle -w /paddle -e "WITH_GPU=OFF" -e "WITH_TESTING=OFF" paddle:dev ./paddle/scripts/paddle_build.sh build
 
-注：上述命令把当前目录（源码树根目录）映射为 container 里的 :code:`/paddle` 目录。
+注：
+
+- 上述命令把当前目录（源码树根目录）映射为 container 里的 :code:`/paddle` 目录。
+
+- 如果您使用的是 manylinux 的镜像进行编译, 那么您需要通过环境变量 :code:`PYTHON_ABI` 来指定一个 `Python ABI <https://www.python.org/dev/peps/pep-0425/#id8>`__.
+PaddlePaddle目前支持的 Python ABI 有 :code:`cp27-cp27m` 和 :code:`cp27-cp27mu`.
 
 编译完成后会在build/python/dist目录下生成输出的whl包，可以选在在当前机器安装也可以拷贝到目标机器安装：
 

doc/v2/build_and_install/build_from_source_en.rst

@@ -36,13 +36,18 @@ If you don't wish to use docker，you need to install several compile dependenci
    # 2. Optional: build development docker image from source
    docker build -t paddle:dev .
    # 3. Run the following command to build a CPU-Only binaries
-   docker run -it -v $PWD:/paddle -w /paddle -e "WITH_GPU=OFF" -e "WITH_TESTING=OFF" paddlepaddle/paddle_manylinux_devel:cuda8.0_cudnn5 ./paddle/scripts/paddle_build.sh build
+   docker run -it -v $PWD:/paddle -w /paddle -e "PYTHON_ABI=cp27-cp27mu" -e "WITH_GPU=OFF" -e "WITH_TESTING=OFF" paddlepaddle/paddle_manylinux_devel:cuda8.0_cudnn5 ./paddle/scripts/paddle_build.sh build
    # 4. Or, use your built Docker image to build PaddlePaddle (must run step 2)
    docker run -it -v $PWD:/paddle -w /paddle -e "WITH_GPU=OFF" -e "WITH_TESTING=OFF" paddle:dev ./paddle/scripts/paddle_build.sh build
 
-NOTE: The above command try to mount the current working directory (root directory of source code)
+NOTE: 
+
+- The above command try to mount the current working directory (root directory of source code)
 into :code:`/paddle` directory inside docker container.
 
+- You need to pass in the required environment variable :code:`PYTHON_ABI` to specify a `Python ABI <https://www.python.org/dev/peps/pep-0425/#id8>`__.
+Currently PaddlePaddle supported Python ABIs include :code:`cp27-cp27m` and :code:`cp27-cp27mu` .
+
 When the compile finishes, you can get the output whl package under
 build/python/dist, then you can choose to install the whl on local
 machine or copy it to the target machine.

paddle/contrib/float16/float16_transpiler.py

@@ -118,7 +118,7 @@ def _remove_unused_var(self):
 
         for var in self.block.vars.keys():
             if var not in args:
-                self.block.remove_var(var)
+                self.block._remove_var(var)
 
     def _modify_feed_fetch(self):
         '''
@@ -165,7 +165,7 @@ def find_op(var):
                     dtype=core.VarDesc.VarType.FP16,
                     shape=var.shape,
                     persistable=var.persistable)
-                self.block.insert_op(
+                self.block._insert_op(
                     i + 1,
                     type="cast",
                     inputs={"X": var},
@@ -188,7 +188,7 @@ def find_op(var):
                     persistable=var.persistable)
                 find_op(var)
                 var.op.rename_output(var_name, tmp_var_name)
-                self.block.insert_op(
+                self.block._insert_op(
                     i,
                     type="cast",
                     inputs={"X": tmp_var},
@@ -253,4 +253,4 @@ def _should_be_converted(var):
 
             # old var will be replaced by the fp16 var in program desc
             self.input_map[var.name] = fp16_var_name
-            self.block.remove_var(var.name)
+            self.block._remove_var(var.name)

paddle/fluid/pybind/protobuf.cc

@@ -145,14 +145,14 @@ void BindBlockDesc(pybind11::module *m) {
       .def_property_readonly("id", &pd::BlockDesc::ID)
       .def_property_readonly("parent", &pd::BlockDesc::Parent)
       .def("get_forward_block_idx", &pd::BlockDesc::ForwardBlockID)
-      .def("set_forward_block_idx", &pd::BlockDesc::SetForwardBlockID)
+      .def("_set_forward_block_idx", &pd::BlockDesc::SetForwardBlockID)
       .def("append_op", &pd::BlockDesc::AppendOp,
            pybind11::return_value_policy::reference)
-      .def("prepend_op", &pd::BlockDesc::PrependOp,
+      .def("_prepend_op", &pd::BlockDesc::PrependOp,
            pybind11::return_value_policy::reference)
-      .def("insert_op", &pd::BlockDesc::InsertOp,
+      .def("_insert_op", &pd::BlockDesc::InsertOp,
            pybind11::return_value_policy::reference)
-      .def("remove_op", &pd::BlockDesc::RemoveOp)
+      .def("_remove_op", &pd::BlockDesc::RemoveOp)
       .def("var",
            [](pd::BlockDesc &self, pybind11::bytes byte_name) {
              std::string name = byte_name;
@@ -165,7 +165,7 @@ void BindBlockDesc(pybind11::module *m) {
              return self.HasVar(name);
            },
            pybind11::return_value_policy::reference)
-      .def("rename_var",
+      .def("_rename_var",
            [](pd::BlockDesc &self, const pybind11::bytes &byte_name,
               const pybind11::bytes &byte_name_new) {
              std::string name = byte_name;
@@ -189,7 +189,7 @@ void BindBlockDesc(pybind11::module *m) {
              return self.FindVarRecursive(name);
            },
            pybind11::return_value_policy::reference)
-      .def("remove_var",
+      .def("_remove_var",
            [](pd::BlockDesc &self, pybind11::bytes byte_name) {
              std::string name = byte_name;
              return self.RemoveVar(name);

python/paddle/fluid/__init__.py

@@ -56,6 +56,7 @@
 import recordio_writer
 import parallel_executor
 from parallel_executor import *
+from paddle.fluid.layers.math_op_patch import monkey_patch_variable
 
 Tensor = LoDTensor
 
@@ -138,5 +139,5 @@ def __bootstrap__():
 
 # TODO(panyx0718): Avoid doing complex initialization logic in __init__.py.
 # Consider paddle.init(args) or paddle.main(args)
-layers.monkey_patch_variable()
+monkey_patch_variable()
 __bootstrap__()

python/paddle/fluid/backward.py

@@ -328,7 +328,7 @@ def _append_backward_ops_(block,
         if op.has_attr("sub_block"):
             sub_block = program.block(op.block_attr("sub_block"))
             grad_sub_block = program.create_block()
-            grad_sub_block.set_forward_block_idx(sub_block.idx)
+            grad_sub_block._set_forward_block_idx(sub_block.idx)
             cb = _callback_lookup_(op)
             if cb is not None:
                 if callbacks is None:
@@ -571,7 +571,7 @@ def append_backward(loss, parameter_list=None, no_grad_set=None,
     _append_backward_vars_(root_block, fwd_op_num, grad_to_var, grad_info_map)
 
     program.current_block_idx = current_block_idx
-    program.sync_with_cpp()
+    program._sync_with_cpp()
     # FIXME(zcd): prevent loss.grad optimized by mem_opt.
     loss.block.var(_append_grad_suffix_(loss.name)).persistable = True
 
@@ -744,7 +744,7 @@ def calc_gradient(targets, inputs, target_gradients=None, no_grad_set=None):
     _rename_grad_(block, fwd_op_num, grad_to_var, target_grad_map)
 
     _append_backward_vars_(block, fwd_op_num, grad_to_var, grad_info_map)
-    prog.sync_with_cpp()
+    prog._sync_with_cpp()
 
     grad_vars = []
     for input_var in inputs:

python/paddle/fluid/clip.py

@@ -82,7 +82,7 @@ def error_clip_callback(block, context):
     op_desc = block.desc.op(block.desc.op_size() - 1)
     for grad_n in filter(lambda n: grad_to_var.has_key(n),
                          op_desc.output_arg_names()):
-        fwd_var = block.var_recursive(grad_to_var[grad_n])
+        fwd_var = block._var_recursive(grad_to_var[grad_n])
         error_clip = getattr(fwd_var, "error_clip", None)
         if not (error_clip is None or isinstance(error_clip,
                                                  BaseErrorClipAttr)):