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

Author: JiayiFeng

CMakeLists.txt

@@ -39,6 +39,7 @@ option(WITH_GPU         "Compile PaddlePaddle with NVIDIA GPU"          ${CUDA_F
 option(WITH_AMD_GPU     "Compile PaddlePaddle with AMD GPU"             OFF)
 option(WITH_AVX         "Compile PaddlePaddle with AVX intrinsics"      ${AVX_FOUND})
 option(WITH_MKL         "Compile PaddlePaddle with MKL support."        ${AVX_FOUND})
+option(WITH_TENSORRT    "Compile PaddlePaddle with TensorRT support."   OFF)
 option(WITH_DSO         "Compile PaddlePaddle with dynamic linked CUDA" ON)
 option(WITH_TESTING     "Compile PaddlePaddle with unit testing"        OFF)
 option(WITH_SWIG_PY     "Compile PaddlePaddle with inference api"       ON)
@@ -181,6 +182,11 @@ if(WITH_GPU)
     include(cuda)
 endif(WITH_GPU)
 
+# TensorRT depends on GPU.
+if (NOT WITH_GPU)
+  set(WITH_TENSORRT OFF)
+endif()
+
 if(WITH_AMD_GPU)
     find_package(HIP)
     include(hip)

Dockerfile

@@ -45,6 +45,13 @@ ENV PATH=${PATH}:${GOROOT}/bin:${GOPATH}/bin
 # install glide
 RUN curl -s -q https://glide.sh/get | sh
 
+# Install TensorRT
+# The unnecessary files has been removed to make the library small.
+RUN wget -qO- http://paddlepaddledeps.bj.bcebos.com/TensorRT-4.0.0.3.Ubuntu-16.04.4.x86_64-gnu.cuda-8.0.cudnn7.0.tar.gz | \
+    tar -xz -C /usr/local && \
+    cp -rf /usr/local/TensorRT/include /usr && \
+    cp -rf /usr/local/TensorRT/lib /usr
+
 # git credential to skip password typing
 RUN git config --global credential.helper store
 

Dockerfile.android

@@ -27,7 +27,7 @@ RUN git config --global credential.helper store
 # Fix locales to en_US.UTF-8
 RUN localedef -i en_US -f UTF-8 en_US.UTF-8
 
-RUN pip install --upgrade pip && \
+RUN pip install --upgrade pip==9.0.3 && \
     pip install -U 'protobuf==3.1.0' && \
     pip install -U wheel sphinx && \
     pip install pre-commit

cmake/external/grpc.cmake

@@ -33,7 +33,7 @@ ExternalProject_Add(
     extern_grpc
     DEPENDS protobuf zlib
     GIT_REPOSITORY "https://github.com/grpc/grpc.git"
-    GIT_TAG "v1.11.x"
+    GIT_TAG "v1.10.x"
     PREFIX          ${GRPC_SOURCES_DIR}
     UPDATE_COMMAND  ""
     CONFIGURE_COMMAND ""

doc/CMakeLists.txt

@@ -3,7 +3,9 @@ add_custom_target(paddle_apis ALL
 
 add_custom_target(paddle_docs ALL
                   DEPENDS paddle_v2_docs paddle_v2_docs_cn
-                  paddle_fluid_docs paddle_fluid_docs_cn)
+                  paddle_fluid_docs paddle_fluid_docs_cn
+                  paddle_mobile_docs paddle_mobile_docs_cn)
 
 add_subdirectory(v2)
 add_subdirectory(fluid)
+add_subdirectory(mobile)

doc/fluid/api/layers.rst

@@ -473,6 +473,12 @@ multiplex
 ..  autofunction:: paddle.fluid.layers.multiplex
     :noindex:
 
+label_smooth
+------------
+
+..  autofunction:: paddle.fluid.layers.label_smooth
+    :noindex:
+
 ops
 ===
 

doc/fluid/design/concepts/parallel_executor.md

@@ -84,7 +84,7 @@ Running an operator can be asynchronized. There is a thread pool to execute an `
 
 ## Synchronize GPU Kernels
 
-The GPU is a non-blocking device. The different streams need be synchronized when switing streams. In current implementation, the synchronization based on the following algorithm:
+The GPU is a non-blocking device. The different streams need be synchronized when switching streams. In current implementation, the synchronization based on the following algorithm:
 
 1. `OpHandle` will record `DeviceContext` that it is used.
 2. In `OpHandle::Run`, if the `DeviceContext` of current operator is different from `DeviceContext` of any input variable, just wait the generate operator of this input variable.

doc/fluid/design/dist_train/README.md

@@ -0,0 +1,57 @@
+## Distributed training overview doc
+
+Currently Paddle Fluid use parameter server architecture to support distributed training.
+
+For synchronous and asynchronous training, the differences are mostly in the logic of parameter server. Now we have already support synchronous training.
+
+### Synchronous training
+
+The training process of synchronous training is:
+
+![synchronous distributed training](./src/sync_distributed_training.png)
+
+1. Pserver
+	1. set `barrier_condition_` to 0 and waits for trainers to send gradient.
+1. Trainer
+	1. Trainer read minibatch of data, run forward-backward with local parameter copy and get the gradients for parameters.
+	1. Trainer use split op to split all the gradient into blocks. The split method is determined at compile time.
+	1. Trainer use send_op to send all the split gradients to corresponding parameter server.
+	1. After trainer send all the gradients, it will send a `BATCH_BARRIER_MESSAGE` to all pservers.
+	1. Trainer call GetVariable to pserver and wait for `barrier_condition_` on pserver to be 1.
+1. Pserver
+   1. Pserver will count the number of `BATCH_BARRIER_MESSAGE`.
+	1. When the count of `BATCH_BARRIER_MESSAGE` is equal to the number of Trainer. Pserver thinks it received all gradient from all trainers.
+	1. Pserver will run the optimization block to optimize the parameters.
+	1. After optimization, pserver set `barrier_condition_` to 1.
+	1. Pserver wait for `FETCH_BARRIER_MESSAGE`.
+1. Trainer.
+	1. The trainer uses GetVariable to get all the parameters from pserver.
+	1. Trainer sends a `FETCH_BARRIER_MESSAGE` to each pserver.
+1. Pserver.
+	1. when the number of `FETCH_BARRIER_MESSAGE` reach the number of all trainers. Pserver think all the parameters have been got. it will go back to 1. to set `barrier_condition_` to 0.
+
+### Asynchronous training
+In the above process. There are two barriers for all trainers to synchronize with each other. In asynchronous training, these two barriers are not needed. The trainer can just send gradients to pserver and then get parameters back.
+
+The training process of asynchronous training can be:
+
+![asynchronous distributed training](./src/async_distributed_training.png)
+
+1. Pserver:
+	1. Each parameter has a queue to receive its gradient from trainers.
+	1. Each parameter has a thread to read data from the queue and run optimize block, using the gradient to optimize the parameter.
+	1. Using an independent thread to handle RPC call `GetVariable` for trainers to get parameters back.(Maybe here we should use a thread pool to speed up fetching the parameters.)
+
+1. Trainer:
+	1. Trainer read a batch of data. Run forward and backward with local parameter copy and get the gradients for parameters.
+	1. Trainer split all gradients to blocks and then send these gradient blocks to pservers(pserver will put them into the queue).
+	2. Trainer gets all parameters back from pserver.
+
+### Note:
+There are also some conditions that need to consider. For exmaple:
+
+1. If trainer needs to wait for the pserver to apply it's gradient and then get back the parameters back.
+1. If we need a lock between parameter update and parameter fetch.
+1. If one parameter must be on one server, or it can also be split and send to multiple parameter servers.
+
+The above architecture of asynchronous training can support different mode, we can have a detailed test in the future for these problems.

doc/fluid/design/dist_train/async_update.md

@@ -0,0 +1,58 @@
+# Design Doc: Asynchronous Update With Distributed Training
+
+## Background
+
+For the typical synchronous distributed training, some significant steps are as follows:
+
+1. A Trainer will compute the gradients and SEND them to the Parameter Server(PServer) nodes.
+1. After the PServer node received gradients came from all the Trainers, It will aggregate the
+gradient variables for the same parameter into one gradient variable and then apply the aggregated
+gradient to the respective parameter, finally using an optimize algorithms(SGD, Monument...)
+to update the parameters.
+1. The Trainer would wait for the PServers finished the optimize stage, and GET the parameters from PServer,
+so all the Trainers would get the same parameters.
+
+In the synchronously distributed training, there should be a `Barrier` to synchronise the
+parameters after the optimizing stage. The performance of a distributed training job would
+depend on the slowest node if there were hundreds or thousands of training nodes in a
+Job, the performance of synchronously distributed training might be very poor because of
+the slow node. So this design doc would introduce an approach to implement
+*asynchronously* distributed training in PaddlePaddle Fluid.
+
+## Design
+
+<img src="./src/async_update.png" width="600"/>
+
+As the figure above, we describe a global view of asynchronously update process and use
+the parameter `w1` as an example to introduce the steps:
+1. For each gradient variables, they may distribute on different GPU card and aggregate
+them while they are all calculated.
+1. Split the gradient variable into multiple blocks according to the number of PServer
+instances and then send them.
+1. PServer would run an `Optimize Block` using a specified optimize algorithm to update
+the specified parameter.
+1. The trainer will fetch latest parameter from PServer before running forward Op which depends
+on the specified parameter.
+1. Broadcast the received variable into multiple GPU cards and continue to run the next
+mini-batch.
+
+### Trainer
+
+- For the multiple devices distributed training, we need to aggregate the gradient
+variables which placed on different devices firstly and then schedule a `SendVars` Operator to
+send the gradient variables to the multiple PServer instances.
+- Schedule `FetchVars` operator to fetch the latest parameter from PServer before running
+the forward ops.
+- There could be a large number of gradient variables to be sent, so we need to use another
+thread pool(IO Threadpool) whose a number of the schedulable threads is larger than the
+computing thread pool to avoid competitive the thread resources with computing.
+
+### Parameter Server
+
+<img src="./src/async_pserver.png" width="750"/>
+
+- There should be multiple trainer instances want to optimize the same parameter at
+the same time, to avoid the racing, we need one `BlockingQueue` for each gradient
+variable to process them one by one.
+- We need a `Map` structure to map a gradient variable name to the `OptimizeBlock` which
+can optimize the respective parameter.