Merge branch 'develop' into expand

Author: luotao1

CMakeLists.txt

@@ -138,12 +138,6 @@ else()
     set(THIRD_PARTY_BUILD_TYPE Release)
 endif()
 
-if(WITH_MKL)
-  option(MKL_SPLIT_GEMM "PaddlePaddle MKL gemm would split to small ones" OFF)
-  if (MKL_SPLIT_GEMM)
-    add_definitions(-DPADDLE_MKL_SPLIT_GEMM)
-  endif()
-endif()
 set(WITH_MKLML ${WITH_MKL})
 if (NOT DEFINED WITH_MKLDNN)
     if (WITH_MKL AND AVX2_FOUND)

cmake/external/mkldnn.cmake

@@ -54,7 +54,7 @@ ExternalProject_Add(
     ${EXTERNAL_PROJECT_LOG_ARGS}
     DEPENDS             ${MKLDNN_DEPENDS}
     GIT_REPOSITORY      "https://github.com/01org/mkl-dnn.git"
-    GIT_TAG             "a29d8487a63afca3d5b8c5bbdbb473cf8ccc6e51"
+    GIT_TAG             "64e03a1939e0d526aa8e9f2e3f7dc0ad8d372944"
     PREFIX              ${MKLDNN_SOURCES_DIR}
     UPDATE_COMMAND      ""
     CMAKE_ARGS          -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}

doc/fluid/design/dist_train/dist_train_nccl2.md

@@ -1,22 +1,22 @@
 # Distributed Training with NCCL2
 
 We design a pattern that can enable training with `ParallelExecutor` and
-using [NCCL2](https://developer.nvidia.com/nccl) as it's collective
+use [NCCL2](https://developer.nvidia.com/nccl) as it's collective
 communication library.
 
 In `ParallelExecutor` we can use `AllReduce` or `Reduce` and `Broadcast`
 to do multi GPU training. And if we initialize NCCL2 communicators as
 ranks in a distributed environment, we can simply run the `ParallelExecutor`
 as a distributed program! The only thing that may be different than in
 the single node version is that we need to broadcast the NCCL unique ID
-to all the nodes, and initialize communicators using that ID, so NCCL2
-will know each other as ranks.
+to all the nodes and initialize communicators using that ID, so NCCL2
+can know each other as ranks.
 
 To achieve this feature, we introduce a new operator: `gen_nccl_id` op,
 so we are ***not*** "bind to" running NCCL2 with MPI, we can run it in
-what ever platform you like.
+whatever platform you like.
 
-It have two running modes:
+It has two running modes:
 
 1. Generate and broadcast mode, which should be used on trainer 0;
 1. Listen and fetch mode, which should be used on trainers other than 0.
@@ -29,7 +29,7 @@ initialize NCCL communicator objects.
 <img src="src/ncc2_design.png">
 
 The above figure indicates the general process when training with NCCL2
-distributed. Each trainer have the number of communicators equal to the
+distributed. Each trainer has the number of communicators equal to the
 number of GPUs, but the ranks should match the global ranks number: here
 we have total 8 GPUs, so `nranks==8`, for each trainer, the ranks should
 be from 0 ~ 3 on trainer 0 and 4 ~ 7 on trainer 1.

doc/fluid/dev/new_op_cn.md

@@ -36,19 +36,19 @@
 <tbody>
 <tr>
 <td>OpProtoMake定义 </td>
-<td>`.cc`文件，Backward Op不需要定义OpProtoMake </td>
+<td>.cc 文件，Backward Op不需要定义OpProtoMake </td>
 </tr>
 <tr>
 <td>Op定义 </td>
-<td> `.cc`文件</td>
+<td> .cc 文件</td>
 </tr>
 <tr>
 <td>Kernel实现 </td>
-<td> CPU、CUDA共享Kernel实现在`.h`文件中，否则，CPU 实现在`.cc`文件中，CUDA 实现在`.cu`文件中。</td>
+<td> CPU、CUDA共享Kernel实现在.h 文件中，否则，CPU 实现在.cc 文件中，CUDA 实现在.cu 文件中。</td>
 </tr>
 <tr>
 <td>注册Op </td>
-<td> Op注册实现在`.cc`文件；Kernel注册CPU实现在`.cc`文件中，CUDA实现在`.cu`文件中</td>
+<td> Op注册实现在.cc 文件；Kernel注册CPU实现在.cc 文件中，CUDA实现在.cu 文件中</td>
 </tr>
 </tbody>
 </table>
@@ -391,7 +391,7 @@ PADDLE_ENFORCE(ctx->HasInput("X"), "");
 ```
 问题示例2 ：提示信息过于简单
 ```
-PADDLE_ENFORCE(i != nullptr, "I must be set"); // I是什么？
+PADDLE_ENFORCE(i != nullptr, "i must be set"); // i是什么？
 ```
 
 2. 在报错信息中使用开发人员定义的变量缩写，不易理解！

doc/fluid/howto/cluster/nccl2_rdma_training.md

@@ -1,12 +1,12 @@
 # Distributed Training with NCCL2 and RDMA
 
-When doing distributed multi-GPU training, network bandwith often becomes the
-bottle neck. We introduce a way to use NCCL2 to do such training job to
-achieve best performace.
+When doing distributed multi-GPU training, network bandwidth often becomes the
+bottleneck. We introduce a way to use NCCL2 to do such training job to
+achieve best performance.
 
-## Prepare Hardwares with RDMA and Multiple GPUs
+## Prepare Hardware with RDMA and Multiple GPUs
 
-I'm using two Linux servers each of them is installed with 8 GPUs and
+I'm using two Linux servers each of them installed with 8 GPUs and
 one 100Gb RDMA card.
 Base environment is:
 
@@ -25,15 +25,15 @@ In general, the steps including:
 1. Use docker to run tests and make sure GPUs and RDMA can work inside
    the container.
 
-I'll ommit section "Install GPU drivers" because we can find it easily
+I'll omit the section "Install GPU drivers" because we can find it easily
 somewhere else.
 
 ### Install RDMA drivers
 
 For my case, I've got two machines with device
 "Mellanox Technologies MT27700 Family [ConnectX-4]" installed. The OS was
 "CentOS 7.4" and I updated the kernel to version 4.4 so that docker can
-work with latest overlay2 filesystem.
+work with the latest overlay2 filesystem.
 
 ***NOTE: before you start, make sure you have a way to get a console
 of the server other than ssh because we may need to re-configure the
@@ -45,22 +45,22 @@ network device.***
 1. Run `./mlnxofedinstall --add-kernel-support` in the software package.
 1. Run `/etc/init.d/openibd restart` to make everything work, note that
    this operation may cause the network goes down if you are using this
-   RDMA device as default network device and use ssh to login the server.
+   RDMA device as default network device and use ssh to log in the server.
 1. Re-configure the network interface, for example:
    `ifconfig eth2 192.168.16.30/20 up`, then add routes if needed:
    `ip route add default via 192.168.16.1 dev eth2`.
 1. Do the same thing on the other node.
 1. Use `ping` to test if the two nodes have typical ICMP connection.
 1. Use either `udaddy` or `ib_write_bw` to test the network connection is
-   ready and have the desired bandwith.
+   ready and have the desired bandwidth.
 
 ### Prepare Docker Image to Run RDMA Programs
 
 1. Build a docker image using cuda base image like: `nvidia/cuda:8.0-cudnn5-devel-ubuntu16.04` and install paddlepaddle whl
    package in it.
 1. Start a docker container and mount GPU driver libs into it (you can
    skip this step if you are using nvidia-docker).
-1. Mount RDMA dirvers and libs into the docker image (see below section),
+1. Mount RDMA drivers and libs into the docker image (see below section),
    also `udaddy` and `ib_write_bw` if needed.
 1. Mount GPU devices and RDMA devices into the container using `--device`
    or just use privileged mode `--privileged`.

paddle/fluid/API.spec

@@ -162,6 +162,7 @@ paddle.fluid.layers.crop ArgSpec(args=['x', 'shape', 'offsets', 'name'], varargs
 paddle.fluid.layers.rank_loss ArgSpec(args=['label', 'left', 'right', 'name'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.layers.prelu ArgSpec(args=['x', 'mode', 'param_attr', 'name'], varargs=None, keywords=None, defaults=(None, None))
 paddle.fluid.layers.flatten ArgSpec(args=['x', 'axis', 'name'], varargs=None, keywords=None, defaults=(1, None))
+paddle.fluid.layers.stack ArgSpec(args=['x', 'axis'], varargs=None, keywords=None, defaults=(0,))
 paddle.fluid.layers.data ArgSpec(args=['name', 'shape', 'append_batch_size', 'dtype', 'lod_level', 'type', 'stop_gradient'], varargs=None, keywords=None, defaults=(True, 'float32', 0, VarType.LOD_TENSOR, True))
 paddle.fluid.layers.open_recordio_file ArgSpec(args=['filename', 'shapes', 'lod_levels', 'dtypes', 'pass_num', 'for_parallel'], varargs=None, keywords=None, defaults=(1, True))
 paddle.fluid.layers.open_files ArgSpec(args=['filenames', 'shapes', 'lod_levels', 'dtypes', 'thread_num', 'buffer_size', 'pass_num', 'is_test'], varargs=None, keywords=None, defaults=(None, None, 1, None))

paddle/fluid/framework/array.h

@@ -0,0 +1,48 @@
+// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+//
+// 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.
+
+#pragma once
+
+#include <cstdint>
+#include "paddle/fluid/platform/hostdevice.h"
+
+namespace paddle {
+namespace framework {
+template <typename T, size_t N>
+class Array {
+  static_assert(N > 0, "The size of array must be larger than 0");
+
+ public:
+  HOSTDEVICE Array() {}
+
+  HOSTDEVICE explicit Array(const T &val) {
+    for (size_t i = 0; i < N; ++i) data_[i] = val;
+  }
+
+  HOSTDEVICE const T *Get() const { return data_; }
+
+  HOSTDEVICE T *GetMutable() { return data_; }
+
+  HOSTDEVICE T &operator[](size_t index) { return data_[index]; }
+
+  HOSTDEVICE const T &operator[](size_t index) const { return data_[index]; }
+
+  HOSTDEVICE constexpr size_t size() const { return N; }
+
+ private:
+  T data_[N];
+};
+
+}  // namespace framework
+}  // namespace paddle

paddle/fluid/framework/op_proto_maker.cc

@@ -129,10 +129,6 @@ void OpProtoAndCheckerMaker::operator()(proto::OpProto* proto,
                                     "Optimized for variable")
       .SetDefault({});
 
-  AddAttr<std::vector<std::string>>(OpCreationCallstackAttrName(),
-                                    "Callstack for Op Creatation.")
-      .SetDefault({});
-
   Validate();
 }
 

paddle/fluid/framework/op_proto_maker.h

@@ -39,7 +39,6 @@ class OpProtoAndCheckerMaker {
  public:
   static const char *OpRoleAttrName() { return "op_role"; }
   static const char *OpRoleVarAttrName() { return "op_role_var"; }
-  static const char *OpCreationCallstackAttrName() { return "op_callstack"; }
 
   void operator()(proto::OpProto *proto, OpAttrChecker *attr_checker);
 

paddle/fluid/framework/operator.cc

@@ -11,17 +11,15 @@ 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. */
-#include "paddle/fluid/framework/operator.h"
+#include <gflags/gflags.h>
+#include <glog/logging.h>
+
 #include <algorithm>
-#include <sstream>
-#include <string>
-#include <vector>
-#include "gflags/gflags.h"
-#include "glog/logging.h"
+
 #include "paddle/fluid/framework/data_transform.h"
 #include "paddle/fluid/framework/executor.h"
 #include "paddle/fluid/framework/lod_tensor.h"
-#include "paddle/fluid/framework/op_proto_maker.h"
+#include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/shape_inference.h"
 #include "paddle/fluid/framework/var_type.h"
 #include "paddle/fluid/platform/profiler.h"
@@ -129,48 +127,19 @@ static LoD GetLoD(const Scope& scope, const std::string& name) {
 }
 
 void OperatorBase::Run(const Scope& scope, const platform::Place& place) {
-  try {
-    if (VLOG_IS_ON(4)) {
-      VLOG(4) << place << " " << DebugStringEx(&scope);
-    }
-    if (platform::is_gpu_place(place)) {
+  VLOG(4) << place << " " << DebugStringEx(&scope);
+  if (platform::is_gpu_place(place)) {
 #ifndef PADDLE_WITH_CUDA
-      PADDLE_THROW("Cannot run operator on place %s", place);
+    PADDLE_THROW("Cannot run operator on place %s", place);
 #else
-      auto dev_id = boost::get<platform::CUDAPlace>(place).device;
-      platform::SetDeviceId(dev_id);
+    auto dev_id = boost::get<platform::CUDAPlace>(place).device;
+    platform::SetDeviceId(dev_id);
 #endif
-    }
-    platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance();
-    platform::RecordEvent record_event(Type(), pool.Get(place));
-    RunImpl(scope, place);
-    if (VLOG_IS_ON(3)) {
-      VLOG(3) << place << " " << DebugStringEx(&scope);
-    }
-  } catch (platform::EnforceNotMet exception) {
-    if (Attrs().count("sub_block") != 0) {
-      throw exception;
-    }
-
-    auto& callstack = Attr<std::vector<std::string>>(
-        OpProtoAndCheckerMaker::OpCreationCallstackAttrName());
-
-    if (callstack.empty()) {
-      throw exception;
-    }
-    std::ostringstream sout;
-    sout << "Invoke operator " << Type() << " error.\n";
-    sout << "Python Callstacks: \n";
-    for (auto& line : callstack) {
-      sout << line;
-    }
-    sout << "C++ Callstacks: \n";
-    sout << exception.err_str_;
-    exception.err_str_ = sout.str();
-    throw exception;
-  } catch (...) {
-    std::rethrow_exception(std::current_exception());
   }
+  platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance();
+  platform::RecordEvent record_event(Type(), pool.Get(place));
+  RunImpl(scope, place);
+  VLOG(3) << place << " " << DebugStringEx(&scope);
 }
 
 bool OperatorBase::HasInputs(const std::string& name) const {
@@ -198,7 +167,7 @@ const std::vector<std::string>& OperatorBase::Inputs(
 }
 
 bool OperatorBase::HasOutputs(const std::string& name) const {
-  if (outputs_.end() != outputs_.find(name)) {
+  if (outputs_.find(name) != outputs_.end()) {
     return true;
   } else {
     return false;