Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into...

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into add-program-cache-for-executor

doc/build_and_install/index_cn.rst

@@ -3,30 +3,54 @@
 
 .. _install_steps:
 
-安装流程
-++++++++
+PaddlePaddle针对不同的用户群体提供了多种安装方式。
 
-PaddlePaddle提供pip和Docker的安装方式：
+专注深度学习模型开发
+-----------------
+
+PaddlePaddle提供了多种python wheel包，可通过pip一键安装：
+
+.. toctree::
+	:maxdepth: 1
+
+	pip_install_cn.rst
+
+这是最便捷的安装方式，请根据机器配置和系统选择对应的安装包。
+
+关注底层框架
+----------
+
+PaddlePaddle提供了基于Docker的安装方式，请参照以下教程：
 
 .. toctree::
-   :maxdepth: 1
+	:maxdepth: 1
 
-   pip_install_cn.rst
-   docker_install_cn.rst
+	docker_install_cn.rst
 
-编译流程
-++++++++
+我们推荐在Docker中运行PaddlePaddle，该方式具有以下优势：
 
-..  warning::
+- 无需单独安装第三方依赖
+- 方便分享运行时环境，易于问题的复现
 
-    建议直接使用上述安装流程，方便快速安装。只有在遇到需要独立定制的二进制时才需要编译。
+对于有定制化二进制文件需求的用户，我们同样提供了从源码编译安装PaddlePaddle的方法：
 
-..  toctree::
+.. toctree::
     :maxdepth: 1
 
     build_from_source_cn.rst
 
-常见问题解答
-++++++++++
+.. warning::
+
+	需要提醒的是，这种安装方式会涉及到一些第三方库的下载、编译及安装，整个安装过程耗时较长。
+
+
+常见问题汇总
+-----------
+
+如果在安装过程中遇到了问题，请先尝试在下面的页面寻找答案：
+
+:ref:`常见问题解答 <install_faq>`
+
+如果问题没有得到解决，欢迎向PaddlePaddle社区反馈问题：
 
-`常见问题解答 <http://www.paddlepaddle.org/docs/develop/documentation/zh/faq/build_and_install/index_cn.html>`_
+`创建issue <https://github.com/PaddlePaddle/Paddle/issues/new>`_

doc/faq/build_and_install/index_cn.rst

+.. _install_faq:
+
 ###################
 编译安装与单元测试
 ###################

doc/howto/cmd_parameter/index_cn.rst

@@ -2,10 +2,25 @@
 
 命令行参数设置
 ===============
+深度学习算法的实现有着多样化的特点，运行环境、运行阶段、模型结构、训练策略等等这些都是常见的变化因素。PaddlePaddle支持用户灵活地设置各种命令行参数，以实现对模型训练或预测流程的控制。
+
+在这一部分，首先以几个实际场景为例，展示了部分命令行参数的使用:
 
 ..  toctree::
   :maxdepth: 1
 
   use_case_cn.md
+
+接着对所有参数的使用场合进行概述和分类:
+
+..  toctree::
+  :maxdepth: 1
+
   arguments_cn.md
+
+最后给出细节描述，详细解释这些参数的属性和意义:
+
+..  toctree::
+  :maxdepth: 1
+
   detail_introduction_cn.md

paddle/fluid/framework/dim.h

@@ -157,8 +157,7 @@ HOSTDEVICE int64_t& indexer<0>(Dim<0>& dim, int idx) {
   throw std::invalid_argument("Invalid index");
 #else
   PADDLE_ASSERT(false);
-#endif
-#if (defined __CUDA_ARCH__) && (CUDA_VERSION < 8000)
+#if CUDA_VERSION < 8000
   // On CUDA versions previous to 8.0, only __shared__ variables
   // could be declared as static in the device code.
   int64_t head = 0;
@@ -166,6 +165,7 @@ HOSTDEVICE int64_t& indexer<0>(Dim<0>& dim, int idx) {
   static int64_t head = 0;
 #endif
   return head;
+#endif
 }
 
 template <int D>
@@ -189,8 +189,7 @@ HOSTDEVICE int64_t indexer<0>(const Dim<0>& dim, int idx) {
   throw std::invalid_argument("Invalid index");
 #else
   PADDLE_ASSERT(false);
-#endif
-#if (defined __CUDA_ARCH__) && (CUDA_VERSION < 8000)
+#if CUDA_VERSION < 8000
   // On CUDA versions previous to 8.0, only __shared__ variables
   // could be declared as static in the device code.
   int64_t head = 0;
@@ -198,6 +197,7 @@ HOSTDEVICE int64_t indexer<0>(const Dim<0>& dim, int idx) {
   static int64_t head = 0;
 #endif
   return head;
+#endif
 }
 
 }  // namespace

paddle/fluid/operators/mine_hard_examples_op.cc

@@ -247,7 +247,7 @@ class MineHardExamplesOp : public framework::OperatorWithKernel {
       const framework::ExecutionContext& ctx) const override {
     return framework::OpKernelType(
         framework::ToDataType(ctx.Input<framework::Tensor>("ClsLoss")->type()),
-        ctx.device_context());
+        platform::CPUPlace());
   }
 };
 

python/paddle/fluid/layers/detection.py

@@ -54,11 +54,17 @@ def detection_output(loc,
                      score_threshold=0.01,
                      nms_eta=1.0):
     """
-    **Detection Output Layer**
+    **Detection Output Layer for Single Shot Multibox Detector (SSD).**
 
-    This layer applies the NMS to the output of network and computes the
-    predict bounding box location. The output's shape of this layer could
-    be zero if there is no valid bounding box.
+    This operation is to get the detection results by performing following
+    two steps:
+    
+    1. Decode input bounding box predictions according to the prior boxes.
+    2. Get the final detection results by applying multi-class non maximum
+       suppression (NMS).
+
+    Please note, this operation doesn't clip the final output bounding boxes
+    to the image window.
 
     Args:
         loc(Variable): A 3-D Tensor with shape [N, M, 4] represents the
@@ -91,7 +97,15 @@ def detection_output(loc,
         nms_eta(float): The parameter for adaptive NMS.
 
     Returns:
-        The detected bounding boxes which are a Tensor.
+        Variable: The detection outputs is a LoDTensor with shape [No, 6].
+            Each row has six values: [label, confidence, xmin, ymin, xmax, ymax].
+            `No` is the total number of detections in this mini-batch. For each
+            instance, the offsets in first dimension are called LoD, the offset
+            number is N + 1, N is the batch size. The i-th image has
+            `LoD[i + 1] - LoD[i]` detected results, if it is 0, the i-th image
+            has no detected results. If all images have not detected results,
+            all the elements in LoD are 0, and output tensor only contains one
+            value, which is -1.
 
     Examples:
         .. code-block:: python

python/paddle/fluid/layers/nn.py

@@ -3206,7 +3206,7 @@ def one_hot(input, depth):
     operator.
 
     Args:
-        input(Tensor/LodTensor):  A Tensor/LodTensor of indices, last dimension must be 1.
+        input(variable):  A Tensor/LodTensor of indices, last dimension must be 1.
         depth(scalar): an interger defining the depth of the one hot dimension.
 
     Returns:

python/paddle/fluid/layers/tensor.py

@@ -362,3 +362,75 @@ def zeros(shape, dtype, force_cpu=False):
           data = fluid.layers.zeros(shape=[1], dtype='int64')
     """
     return fill_constant(value=0.0, **locals())
+
+
+def save(x, file_path, overwrite=True):
+    """
+    Saves a variable as a file.
+
+    Args:
+        x(variable): The Tensor/LoDTensor to be saved.
+        file_path(str): The file path where the variable will be saved.
+        overwrite(bool): Whether or not cover the given file when it has already 
+            existed. If it's set 'False' and the file is existed, a runtime 
+            error will be thrown. 
+    """
+    helper = LayerHelper("save", **locals())
+    helper.append_op(
+        type="save",
+        inputs={"input": x},
+        outputs={},
+        args={"file_path": file_path,
+              "overwrite": overwrite})
+
+
+def save_combine(x, file_path, overwrite=True):
+    """
+    Saves a list of variables into a single file.
+
+    Args:
+        x(list): A list of Tensor/LoDTensor to be saved together in a single file.
+        file_path(str): The file path where variables will be saved.
+        overwrite(bool): Whether or not cover the given file when it has already 
+            existed. If it's set 'False' and the file is existed, a runtime 
+            error will be thrown. 
+    """
+    helper = LayerHelper("save_combine", **locals())
+    helper.append_op(
+        type="save_combine",
+        inputs={"input": x},
+        outputs={},
+        args={"file_path": file_path,
+              "overwrite": overwrite})
+
+
+def load(out, file_path):
+    """
+    Loads a variable from a given file.
+
+    Args:
+        out(variable): The variable to be read from the disk file.
+        file_path(str): The path of the disk file.
+    """
+    helper = LayerHelper("load", **locals())
+    helper.append_op(
+        type="load",
+        inputs={},
+        output={"Out": out},
+        args={"file_path": file_path})
+
+
+def load_combine(out, file_path):
+    """
+    Loads a list of vairables from a single file.
+
+    Args:
+        out(list): The list of variables to be read from the disk file.
+        file_path(str): The path of the disk file.
+    """
+    helper = LayerHelper("load_combine", **locals())
+    helper.append_op(
+        type="load_combine",
+        inputs={},
+        output={"Out": out},
+        args={"file_path": file_path})