fix merge conflict

7 years ago · 8511bdb2d3
parent 1b59220d50 81f22bb2f6
commit 8511bdb2d3
1141 changed files with 21020 additions and 6047 deletions
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@ -23,7 +23,7 @@ repos:
    -   id: clang-format-with-version-check
        name: clang-format
        description: Format files with ClangFormat.
-        entry: bash ./.clang_format.hook -i
+        entry: bash ./tools/codestyle/clang_format.hook -i
        language: system
        files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx|proto)$
 -   repo: local
@ -52,7 +52,7 @@ repos:
    hooks:
    -   id: copyright_checker
        name: copyright_checker
-        entry: python ./.copyright.hook
+        entry: python ./tools/codestyle/copyright.hook
        language: system
        files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx|proto|py)$
        exclude: (?!.*third_party)^.*$ | (?!.*book)^.*$
--- a/AUTHORS.md
+++ b/AUTHORS.md
@ -22,6 +22,7 @@
 | jczaja | Jacek Czaja |
 | JiayiFeng | Jia-Yi Feng |
 | kbinias | Krzysztof Binias |
+| kexinzhao | Ke-Xin Zhao |
 | kuke | Yi-Bing Liu |
 | lcy-seso | Ying Cao |
 | lipeng-unisound | Peng Li |
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@ -61,7 +61,10 @@ option(EIGEN_USE_THREADS "Compile with multi-threaded Eigen"            OFF)
 option(WITH_ARM_FP16    "Use half precision support on armv8.2-a cpu"   OFF)
 option(WITH_FAST_BUNDLE_TEST    "Bundle tests that can be run in a single process together to reduce launch overhead"   OFF)
 option(WITH_CONTRIB     "Compile the third-party contributation"        OFF)
+option(REPLACE_ENFORCE_GLOG "Replace PADDLE_ENFORCE with glog/CHECK for better debug." OFF)
+option(WITH_ANAKIN      "Compile with Anakin library"                   OFF)
 option(WITH_GRPC     "Use grpc as the default rpc framework"            ${WITH_DISTRIBUTE})
+option(WITH_BRPC_RDMA     "Use brpc rdma as the rpc protocal"           OFF)

 # CMAKE_BUILD_TYPE
 if(NOT CMAKE_BUILD_TYPE)
@ -130,6 +133,10 @@ if (NOT DEFINED WITH_MKLDNN)
        set(WITH_MKLDNN OFF)
    endif()
 endif()
+
+if (REPLACE_ENFORCE_GLOG)
+  add_definitions("-DREPLACE_ENFORCE_GLOG")
+endif()
 ########################################################################################

 include(external/mklml)     # download mklml package
@ -152,12 +159,24 @@ include(external/cares)
 if(WITH_DISTRIBUTE)
    if(WITH_GRPC)
        include(external/grpc)
+        message(STATUS "Use grpc framework.")
    else()
+        message(STATUS "Use brpc framework.")
        include(external/leveldb)
        include(external/brpc)
    endif()
 endif()

+if(WITH_BRPC_RDMA)
+    message(STATUS "Use brpc with rdma.")
+    if(WITH_GRPC)
+        message(FATAL_ERROR "Can't use grpc with brpc rdma.")
+    endif()
+    if(NOT WITH_DISTRIBUTE)
+        message(FATAL_ERROR "Can't use brpc rdma in no distribute env.")
+    endif()
+endif()
+
 include(external/snappy)    # download snappy
 include(external/snappystream)
 include(external/threadpool)
@ -177,7 +196,7 @@ include(inference_lib)      # add paddle fluid inference libraries


 include_directories("${PADDLE_SOURCE_DIR}")
-include_directories("${PADDLE_SOURCE_DIR}/paddle/cuda/include")
+include_directories("${PADDLE_SOURCE_DIR}/paddle/legacy/cuda/include")
 include_directories("${CMAKE_CURRENT_BINARY_DIR}/proto")
 include_directories("${CMAKE_CURRENT_BINARY_DIR}/go/pserver/client/c")

@ -193,7 +212,10 @@ set(EXTERNAL_LIBS
 if(WITH_GPU)
    include(cuda)
    include(tensorrt)
-endif(WITH_GPU)
+    include(external/anakin)
+else()
+  set(WITH_ANAKIN OFF CACHE STRING "Anakin is valid only when GPU is set." FORCE)
+endif()

 if(WITH_AMD_GPU)
    find_package(HIP)
@ -218,7 +240,7 @@ add_subdirectory(proto)
 if(NOT MOBILE_INFERENCE AND NOT WITH_FLUID_ONLY)
    # "add_subdirectory(go)" should be placed after the following loine,
    # because it depends on paddle/optimizer.
-    add_subdirectory(paddle/optimizer)
+    add_subdirectory(paddle/legacy/optimizer)
 endif()

 # "add_subdirectory(paddle)" and "add_subdirectory(python)" should be
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@ -159,4 +159,4 @@ This will enable VLOG messages generated by `buddy_allocator.{h,cc}` and in the
 - verbose level 1: [framework](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/framework)
 - verbose level 3: [operators](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/operators)
 - verbose level 5: [memory](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/memory), [platform](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/platform)
- verbose level 7: [math](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/math)
+- verbose level 7: [math](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/legacy/math)
--- a/3
+++ b/3
@ -76,7 +76,8 @@ RUN easy_install -U pip && \
    pip install sphinx-rtd-theme==0.1.9 recommonmark

 RUN pip install pre-commit 'ipython==5.3.0' && \
-    pip install 'ipykernel==4.6.0' 'jupyter==1.0.0'
+    pip install 'ipykernel==4.6.0' 'jupyter==1.0.0' && \
+    pip install opencv-python

 #For docstring checker
 RUN pip install pylint pytest astroid isort
--- a/README.md
+++ b/README.md
@ -4,7 +4,6 @@
 [![Build Status](https://travis-ci.org/PaddlePaddle/Paddle.svg?branch=develop)](https://travis-ci.org/PaddlePaddle/Paddle)
 [![Documentation Status](https://img.shields.io/badge/docs-latest-brightgreen.svg?style=flat)](http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/index_en.html)
 [![Documentation Status](https://img.shields.io/badge/中文文档-最新-brightgreen.svg)](http://www.paddlepaddle.org/docs/develop/documentation/zh/getstarted/index_cn.html)
-[![Coverage Status](https://coveralls.io/repos/github/PaddlePaddle/Paddle/badge.svg?branch=develop)](https://coveralls.io/github/PaddlePaddle/Paddle?branch=develop)
 [![Release](https://img.shields.io/github/release/PaddlePaddle/Paddle.svg)](https://github.com/PaddlePaddle/Paddle/releases)
 [![License](https://img.shields.io/badge/license-Apache%202-blue.svg)](LICENSE)

--- a/benchmark/fluid/Dockerfile
+++ b/benchmark/fluid/Dockerfile
@ -1,11 +1,18 @@
 FROM nvidia/cuda:9.0-cudnn7-devel-ubuntu16.04
+
+# Use UBUNTU_MIRROR can speed up apt-get speed.
+# ARG UBUNTU_MIRROR
+# RUN /bin/bash -c 'if [[ -n ${UBUNTU_MIRROR} ]]; then sed -i 's#http://archive.ubuntu.com/ubuntu#${UBUNTU_MIRROR}#g' /etc/apt/sources.list; fi'
+
 RUN apt-get update && apt-get install -y python python-pip iputils-ping libgtk2.0-dev wget vim net-tools iftop python-opencv
 RUN ln -s /usr/lib/x86_64-linux-gnu/libcudnn.so.7 /usr/lib/libcudnn.so && ln -s /usr/lib/x86_64-linux-gnu/libnccl.so.2 /usr/lib/libnccl.so
-RUN pip install -U pip
-RUN pip install -U kubernetes paddlepaddle

 # IMPORTANT:
 # Add "ENV http_proxy=http://ip:port" if your download is slow, and don't forget to unset it at runtime.
+# exmaple: unset http_proxy && unset https_proxy && python fluid_benchmark.py ...
+
+RUN pip install -U pip
+RUN pip install -U kubernetes paddlepaddle

 RUN sh -c 'echo "import paddle.v2 as paddle\npaddle.dataset.cifar.train10()\npaddle.dataset.flowers.fetch()" | python'
 RUN sh -c 'echo "import paddle.v2 as paddle\npaddle.dataset.mnist.train()\npaddle.dataset.mnist.test()\npaddle.dataset.imdb.fetch()" | python'
@ -14,9 +21,11 @@ RUN pip uninstall -y paddlepaddle && mkdir /workspace

 ADD https://raw.githubusercontent.com/PaddlePaddle/cloud/develop/docker/paddle_k8s /usr/bin
 ADD https://raw.githubusercontent.com/PaddlePaddle/cloud/develop/docker/k8s_tools.py /root
+RUN chmod +x /usr/bin/paddle_k8s

 ADD *.whl /
-RUN pip install /*.whl && rm -f /*.whl && chmod +x /usr/bin/paddle_k8s
+RUN pip install /*.whl && rm -f /*.whl 

 ENV LD_LIBRARY_PATH=/usr/local/lib
-ADD fluid_benchmark.py recordio_converter.py models/ /workspace/
+ADD fluid_benchmark.py recordio_converter.py args.py recordio_converter.py run.sh run_fluid_benchmark.sh /workspace/
+ADD models/ /workspace/models/
--- a/benchmark/fluid/args.py
+++ b/benchmark/fluid/args.py
@ -122,5 +122,9 @@ def parse_args():
        type=str,
        default="",
        help='Directory that contains all the training recordio files.')
+    parser.add_argument(
+        '--use_inference_transpiler',
+        action='store_true',
+        help='If set, uses inference transpiler to optimize the program.')
    args = parser.parse_args()
    return args
--- a/benchmark/fluid/fluid_benchmark.py
+++ b/benchmark/fluid/fluid_benchmark.py
@ -97,7 +97,7 @@ def dist_transpile(trainer_id, args):
        return train_program, fluid.default_startup_program()
    else:
        raise ValueError(
-            'TRAINING_ROLE environment variable must be either TRAINER or PSERVER'
+            'PADDLE_TRAINING_ROLE environment variable must be either TRAINER or PSERVER'
        )


@ -131,6 +131,11 @@ def train(avg_loss, infer_prog, optimizer, train_reader, test_reader, batch_acc,
    exe = fluid.Executor(place)
    exe.run(startup_prog)

+    # Use inference_transpiler to speedup
+    if args.use_inference_transpiler:
+        t = fluid.InferenceTranspiler()
+        t.transpile(infer_prog, place)
+
    if not args.use_reader_op:
        feed_var_list = [
            var for var in train_prog.global_block().vars.itervalues()
@ -180,7 +185,7 @@ def train(avg_loss, infer_prog, optimizer, train_reader, test_reader, batch_acc,
        print_train_time(start_time, time.time(), num_samples)
        print("Pass: %d, Loss: %f" % (pass_id, np.mean(train_losses))),
        # evaluation
-        if not args.no_test and batch_acc:
+        if not args.no_test and batch_acc and not args.use_reader_op:
            pass_test_acc = test(exe, infer_prog, test_reader, feeder,
                                 batch_acc)
            print(", Test Accuracy: %f" % pass_test_acc)
@ -264,8 +269,6 @@ def train_parallel(avg_loss, infer_prog, optimizer, train_reader, test_reader,
                    break
            else:
                loss, = exe.run([avg_loss.name], feed=feeder.feed(data))
-            if args.update_method == "pserver":
-                exe.bcast_params()
            if args.use_reader_op:
                num_samples += args.batch_size * args.gpus
            else:
@ -277,11 +280,12 @@ def train_parallel(avg_loss, infer_prog, optimizer, train_reader, test_reader,
            batch_id += 1

        print_train_time(start_time, time.time(), num_samples)
-        if not args.no_test and batch_acc:
+        if not args.no_test and batch_acc and not args.use_reader_op:
+            # we have not implement record io for test
+            # skip test when use args.use_reader_op
            test_acc = test(startup_exe, infer_prog, test_reader, feeder,
                            batch_acc)
            print("Pass: %d, Test Accuracy: %f\n" % (pass_id, test_acc))
-        exit(0)


 def print_arguments(args):
@ -300,9 +304,18 @@ def print_train_time(start_time, end_time, num_samples):
          (num_samples, train_elapsed, examples_per_sec))


+def print_paddle_envs():
+    print('----------- Configuration envs -----------')
+    for k in os.environ:
+        if "PADDLE_" in k:
+            print "ENV %s:%s" % (k, os.environ[k])
+    print('------------------------------------------------')
+
+
 def main():
    args = parse_args()
    print_arguments(args)
+    print_paddle_envs()

    # the unique trainer id, starting from 0, needed by trainer
    # only
--- a/benchmark/fluid/kube_gen_job.py
+++ b/benchmark/fluid/kube_gen_job.py
@ -17,6 +17,7 @@ import copy
 import argparse
 import random
 import os
+import copy
 from kube_templates import pserver, trainer, envs


@ -108,10 +109,9 @@ def gen_job():
    tn_container["ports"][0]["containerPort"] = spreadport

    envs.append({"name": "PADDLE_JOB_NAME", "value": args.jobname})
-    envs.append({"name": "TRAINERS", "value": str(args.trainers)})
-    envs.append({"name": "PSERVERS", "value": str(args.pservers)})
+    envs.append({"name": "PADDLE_TRAINERS", "value": str(args.trainers)})
+    envs.append({"name": "PADDLE_PSERVERS", "value": str(args.pservers)})
    envs.append({"name": "ENTRY", "value": args.entry})
-    envs.append({"name": "PADDLE_INIT_PORT", "value": str(args.port)})
    envs.append({"name": "PADDLE_PSERVER_PORT", "value": str(args.port)})
    # NOTE: these directories below are cluster specific, please modify
    # this settings before you run on your own cluster.
@ -166,17 +166,23 @@ def gen_job():
    tn["spec"]["template"]["spec"]["volumes"] = volumes
    tn_container["volumeMounts"] = volumeMounts

-    ps_container["env"] = envs
-    ps_container["env"].append({"name": "TRAINING_ROLE", "value": "PSERVER"})
+    ps_container["env"] = copy.deepcopy(envs)
+    ps_container["env"].append({
+        "name": "PADDLE_TRAINING_ROLE",
+        "value": "PSERVER"
+    })
    tn_container["env"] = envs
    if args.disttype == "pserver":
        tn_container["env"].append({
-            "name": "TRAINING_ROLE",
+            "name": "PADDLE_TRAINING_ROLE",
            "value": "TRAINER"
        })
    elif args.disttype == "nccl2" or args.disttype == "local":
        # NCCL2 have no training role, set to plain WORKER
-        tn_container["env"].append({"name": "TRAINING_ROLE", "value": "WORKER"})
+        tn_container["env"].append({
+            "name": "PADDLE_TRAINING_ROLE",
+            "value": "WORKER"
+        })

    os.mkdir(args.jobname)
    if args.disttype == "pserver":
--- a/benchmark/fluid/models/machine_translation.py
+++ b/benchmark/fluid/models/machine_translation.py
@ -173,21 +173,6 @@ def seq_to_seq_net(embedding_dim, encoder_size, decoder_size, source_dict_dim,
        return avg_cost, feeding_list


-def to_lodtensor(data, place):
-    seq_lens = [len(seq) for seq in data]
-    cur_len = 0
-    lod = [cur_len]
-    for l in seq_lens:
-        cur_len += l
-        lod.append(cur_len)
-    flattened_data = np.concatenate(data, axis=0).astype("int64")
-    flattened_data = flattened_data.reshape([len(flattened_data), 1])
-    lod_t = core.LoDTensor()
-    lod_t.set(flattened_data, place)
-    lod_t.set_lod([lod])
-    return lod_t, lod[-1]
-
-
 def lodtensor_to_ndarray(lod_tensor):
    dims = lod_tensor.get_dims()
    ndarray = np.zeros(shape=dims).astype('float32')
--- a/benchmark/fluid/models/resnet.py
+++ b/benchmark/fluid/models/resnet.py
@ -199,7 +199,10 @@ def get_model(args):
    batched_train_reader = paddle.batch(
        paddle.reader.shuffle(
            train_reader, buf_size=5120),
-        batch_size=args.batch_size * args.gpus)
-    batched_test_reader = paddle.batch(train_reader, batch_size=args.batch_size)
+        batch_size=args.batch_size * args.gpus,
+        drop_last=True)
+    batched_test_reader = paddle.batch(
+        train_reader, batch_size=args.batch_size, drop_last=True)

-    return avg_cost, inference_program, optimizer, batched_train_reader, batched_test_reader, batch_acc
+    return avg_cost, inference_program, optimizer, batched_train_reader,\
+                   batched_test_reader, batch_acc
--- a/benchmark/fluid/models/stacked_dynamic_lstm.py
+++ b/benchmark/fluid/models/stacked_dynamic_lstm.py
@ -125,18 +125,3 @@ def get_model(args):
        batch_size=args.batch_size)

    return loss, inference_program, adam, train_reader, test_reader, batch_acc
-
-
-def to_lodtensor(data, place):
-    seq_lens = [len(seq) for seq in data]
-    cur_len = 0
-    lod = [cur_len]
-    for l in seq_lens:
-        cur_len += l
-        lod.append(cur_len)
-    flattened_data = numpy.concatenate(data, axis=0).astype("int64")
-    flattened_data = flattened_data.reshape([len(flattened_data), 1])
-    res = fluid.LoDTensor()
-    res.set(flattened_data, place)
-    res.set_lod([lod])
-    return res
--- a/cmake/configure.cmake
+++ b/cmake/configure.cmake
@ -118,6 +118,10 @@ endif()
 set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${SIMD_FLAG}")
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${SIMD_FLAG}")

+if(WITH_DISTRIBUTE)
+  add_definitions(-DPADDLE_WITH_DISTRIBUTE)
+endif()
+
 if(WITH_GOLANG)
  # we need to symlink Paddle directory into GOPATH. If we
  # don't do it and we have code that depends on Paddle, go
@ -170,3 +174,7 @@ endif(WITH_GOLANG)
 if(WITH_GRPC)
    add_definitions(-DPADDLE_WITH_GRPC)
 endif(WITH_GRPC)
+
+if(WITH_BRPC_RDMA)
+    add_definitions(-DPADDLE_WITH_BRPC_RDMA)
+endif(WITH_BRPC_RDMA)
--- a/cmake/external/anakin.cmake
+++ b/cmake/external/anakin.cmake
@ -0,0 +1,44 @@
+if (NOT WITH_ANAKIN)
+  return()
+endif()
+
+set(ANAKIN_INSTALL_DIR "${THIRD_PARTY_PATH}/install/anakin" CACHE PATH
+  "Anakin install path." FORCE)
+set(ANAKIN_INCLUDE "${ANAKIN_INSTALL_DIR}" CACHE STRING "root of Anakin header files")
+set(ANAKIN_LIBRARY "${ANAKIN_INSTALL_DIR}" CACHE STRING "path of Anakin library")
+
+set(ANAKIN_COMPILE_EXTRA_FLAGS -Wno-error=unused-variable -Wno-error=format-extra-args -Wno-error=comment -Wno-error=format -Wno-error=switch -Wno-error=return-type -Wno-error=non-virtual-dtor -Wno-reorder -Wno-error=cpp)
+
+set(ANAKIN_LIBRARY_URL "https://github.com/pangge/Anakin/releases/download/3.0/anakin_release_simple.tar.gz")
+
+# A helper function used in Anakin, currently, to use it, one need to recursively include
+# nearly all the header files.
+function(fetch_include_recursively root_dir)
+    if (IS_DIRECTORY ${root_dir})
+        include_directories(${root_dir})
+    endif()
+
+    file(GLOB ALL_SUB RELATIVE ${root_dir} ${root_dir}/*)
+    foreach(sub ${ALL_SUB})
+        if (IS_DIRECTORY ${root_dir}/${sub})
+            fetch_include_recursively(${root_dir}/${sub})
+        endif()
+    endforeach()
+endfunction()
+
+if (NOT EXISTS "${ANAKIN_INSTALL_DIR}")
+    # download library
+    message(STATUS "Download Anakin library from ${ANAKIN_LIBRARY_URL}")
+    execute_process(COMMAND bash -c "mkdir -p ${ANAKIN_INSTALL_DIR}")
+    execute_process(COMMAND bash -c "rm -rf ${ANAKIN_INSTALL_DIR}/*")
+    execute_process(COMMAND bash -c "cd ${ANAKIN_INSTALL_DIR}; wget -q ${ANAKIN_LIBRARY_URL}")
+    execute_process(COMMAND bash -c "mkdir -p ${ANAKIN_INSTALL_DIR}")
+    execute_process(COMMAND bash -c "cd ${ANAKIN_INSTALL_DIR}; tar xzf anakin_release_simple.tar.gz")
+endif()
+
+if (WITH_ANAKIN)
+    message(STATUS "Anakin for inference is enabled")
+    message(STATUS "Anakin is set INCLUDE:${ANAKIN_INCLUDE} LIBRARY:${ANAKIN_LIBRARY}")
+    fetch_include_recursively(${ANAKIN_INCLUDE})
+    link_directories(${ANAKIN_LIBRARY})
+endif()
--- a/cmake/external/brpc.cmake
+++ b/cmake/external/brpc.cmake
@ -14,6 +14,15 @@

 INCLUDE(ExternalProject)

+find_library(SSL_LIBRARY NAMES ssl)
+ADD_LIBRARY(ssl SHARED IMPORTED GLOBAL)
+SET_PROPERTY(TARGET ssl PROPERTY IMPORTED_LOCATION ${SSL_LIBRARY})
+
+find_library(CRYPTO_LIBRARY NAMES crypto)
+ADD_LIBRARY(crypto SHARED IMPORTED GLOBAL)
+SET_PROPERTY(TARGET crypto PROPERTY IMPORTED_LOCATION ${CRYPTO_LIBRARY})
+
+
 SET(BRPC_SOURCES_DIR ${THIRD_PARTY_PATH}/brpc)
 SET(BRPC_INSTALL_DIR ${THIRD_PARTY_PATH}/install/brpc)
 SET(BRPC_INCLUDE_DIR "${BRPC_INSTALL_DIR}/include" CACHE PATH "brpc include directory." FORCE)
@ -22,14 +31,14 @@ SET(BRPC_LIBRARIES "${BRPC_INSTALL_DIR}/lib/libbrpc.a" CACHE FILEPATH "brpc libr
 INCLUDE_DIRECTORIES(${BRPC_INCLUDE_DIR})

 # Reference https://stackoverflow.com/questions/45414507/pass-a-list-of-prefix-paths-to-externalproject-add-in-cmake-args
-set(prefix_path "${THIRD_PARTY_PATH}/install/gflags|${THIRD_PARTY_PATH}/install/leveldb|${THIRD_PARTY_PATH}/install/snappy|${THIRD_PARTY_PATH}/install/gtest|${THIRD_PARTY_PATH}/install/protobuf")
+set(prefix_path "${THIRD_PARTY_PATH}/install/gflags|${THIRD_PARTY_PATH}/install/leveldb|${THIRD_PARTY_PATH}/install/snappy|${THIRD_PARTY_PATH}/install/gtest|${THIRD_PARTY_PATH}/install/protobuf|${THIRD_PARTY_PATH}/install/zlib")

 # If minimal .a is need, you can set  WITH_DEBUG_SYMBOLS=OFF
 ExternalProject_Add(
    extern_brpc
    ${EXTERNAL_PROJECT_LOG_ARGS}
-    GIT_REPOSITORY  "https://github.com/brpc/brpc"
-    GIT_TAG         "6d153dd7ff00f960ae6895c9c5fff0ce9f07aff2"
+    GIT_REPOSITORY  "https://github.com/gongweibao/brpc"
+    GIT_TAG         "7dc04defad1fd4173aae170c3fcbde131b65155a"
    PREFIX          ${BRPC_SOURCES_DIR}
    UPDATE_COMMAND  ""
    CMAKE_ARGS      -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
@ -42,6 +51,8 @@ ExternalProject_Add(
                    -DCMAKE_BUILD_TYPE=${THIRD_PARTY_BUILD_TYPE}
                    -DCMAKE_PREFIX_PATH=${prefix_path}
                    -DBRPC_WITH_GLOG=ON
+                    -DIOBUF_WITH_HUGE_BLOCK=ON
+                    -DBRPC_WITH_RDMA=${WITH_BRPC_RDMA}
                    ${EXTERNAL_OPTIONAL_ARGS}
    LIST_SEPARATOR |
    CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${BRPC_INSTALL_DIR}
@ -49,7 +60,7 @@ ExternalProject_Add(
                     -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON
                     -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE}
 )
-ADD_DEPENDENCIES(extern_brpc protobuf leveldb gflags glog gtest snappy)
+ADD_DEPENDENCIES(extern_brpc protobuf ssl crypto leveldb gflags glog gtest snappy)
 ADD_LIBRARY(brpc STATIC IMPORTED GLOBAL)
 SET_PROPERTY(TARGET brpc PROPERTY IMPORTED_LOCATION ${BRPC_LIBRARIES})
 ADD_DEPENDENCIES(brpc extern_brpc)
--- a/cmake/external/grpc.cmake
+++ b/cmake/external/grpc.cmake
@ -40,12 +40,12 @@ ExternalProject_Add(
    # NOTE(wuyi):
    # this package is generated by following steps:
    # 1. git clone -b v1.8.x https://github.com/grpc/grpc.git
-    # 2. submodule update --init
+    # 2. git submodule update --init
    # 3. keep only zlib, cares, protobuf, boringssl under "third_party",
    #    checkout and clean other dirs under third_party
    # 4. remove .git, and package the directory.
-    URL "http://paddlepaddledeps.bj.bcebos.com/grpc-v1.8.x.tar.gz"
-    URL_MD5  "c9c58ee7d0e8929a63155af6a2ecdbd0"
+    URL "http://paddlepaddledeps.bj.bcebos.com/grpc-v1.10.x.tar.gz"
+    URL_MD5  "1f268a2aff6759839dccd256adcc91cf"
    PREFIX          ${GRPC_SOURCES_DIR}
    UPDATE_COMMAND  ""
    CONFIGURE_COMMAND ""
--- a/cmake/external/mkldnn.cmake
+++ b/cmake/external/mkldnn.cmake
@ -45,7 +45,8 @@ IF(${CBLAS_PROVIDER} STREQUAL "MKLML")
 ELSE()
    MESSAGE(FATAL_ERROR "Should enable MKLML when build MKLDNN")
 ENDIF()
-SET(MKLDNN_FLAG "-Wno-error=strict-overflow -Wno-error=unused-result -Wno-unused-result")
+SET(MKLDNN_FLAG "-Wno-error=strict-overflow -Wno-error=unused-result")
+SET(MKLDNN_FLAG "${MKLDNN_FLAG} -Wno-unused-result -Wno-unused-value")
 SET(MKLDNN_CFLAG "${CMAKE_C_FLAGS} ${MKLDNN_FLAG}")
 SET(MKLDNN_CXXFLAG "${CMAKE_CXX_FLAGS} ${MKLDNN_FLAG}")
 ExternalProject_Add(
@ -53,7 +54,7 @@ ExternalProject_Add(
    ${EXTERNAL_PROJECT_LOG_ARGS}
    DEPENDS             ${MKLDNN_DEPENDS}
    GIT_REPOSITORY      "https://github.com/01org/mkl-dnn.git"
-    GIT_TAG             "db3424ad44901513c03a1ea31ccaacdf633fbe9f"
+    GIT_TAG             "a29d8487a63afca3d5b8c5bbdbb473cf8ccc6e51"
    PREFIX              ${MKLDNN_SOURCES_DIR}
    UPDATE_COMMAND      ""
    CMAKE_ARGS          -DCMAKE_INSTALL_PREFIX=${MKLDNN_INSTALL_DIR}
--- a/cmake/external/openblas.cmake
+++ b/cmake/external/openblas.cmake
@ -29,6 +29,8 @@ IF(NOT ${CBLAS_FOUND})
        "${CBLAS_INSTALL_DIR}/lib/${CMAKE_STATIC_LIBRARY_PREFIX}openblas${CMAKE_STATIC_LIBRARY_SUFFIX}"
        CACHE FILEPATH "openblas library." FORCE)

+    ADD_DEFINITIONS(-DPADDLE_USE_OPENBLAS)
+
    SET(OPENBLAS_CC "${CMAKE_C_COMPILER} -Wno-unused-but-set-variable -Wno-unused-variable")
    SET(OPENBLAS_COMMIT "v0.2.20")

@ -112,7 +114,12 @@ INCLUDE_DIRECTORIES(${CBLAS_INC_DIR})
 SET(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/cblas_dummy.c)
 FILE(WRITE ${dummyfile} "const char *dummy_cblas = \"${dummyfile}\";")
 ADD_LIBRARY(cblas STATIC ${dummyfile})
+
+IF("${CBLAS_PROVIDER}" STREQUAL "MKLML")
+  TARGET_LINK_LIBRARIES(cblas dynload_mklml)
+ELSE()
  TARGET_LINK_LIBRARIES(cblas ${CBLAS_LIBRARIES})
+ENDIF("${CBLAS_PROVIDER}" STREQUAL "MKLML")

 IF(NOT ${CBLAS_FOUND})
    ADD_DEPENDENCIES(cblas extern_openblas)
--- a/cmake/generic.cmake
+++ b/cmake/generic.cmake
@ -96,6 +96,20 @@ if(NOT APPLE AND NOT ANDROID)
    set(CMAKE_CXX_LINK_EXECUTABLE "${CMAKE_CXX_LINK_EXECUTABLE} -pthread -ldl -lrt")
 endif(NOT APPLE AND NOT ANDROID)

+set_property(GLOBAL PROPERTY FLUID_MODULES "")
+# find all fluid modules is used for paddle fluid static library
+# for building inference libs
+function(find_fluid_modules TARGET_NAME)
+  get_filename_component(__target_path ${TARGET_NAME} ABSOLUTE)
+  string(REGEX REPLACE "^${PADDLE_SOURCE_DIR}/" "" __target_path ${__target_path})
+  string(FIND "${__target_path}" "fluid" pos)
+  if(pos GREATER 1)
+    get_property(fluid_modules GLOBAL PROPERTY FLUID_MODULES)
+    set(fluid_modules ${fluid_modules} ${TARGET_NAME})
+    set_property(GLOBAL PROPERTY FLUID_MODULES "${fluid_modules}")
+  endif()
+endfunction(find_fluid_modules)
+
 function(merge_static_libs TARGET_NAME)
  set(libs ${ARGN})
  list(REMOVE_DUPLICATES libs)
@ -195,6 +209,15 @@ function(cc_library TARGET_NAME)
        list(REMOVE_ITEM cc_library_DEPS warpctc)
        add_dependencies(${TARGET_NAME} warpctc)
      endif()
+      # Only deps libmklml.so, not link
+      if("${cc_library_DEPS};" MATCHES "mklml;")
+        list(REMOVE_ITEM cc_library_DEPS mklml)
+        if(NOT "${TARGET_NAME}" MATCHES "dynload_mklml")
+          list(APPEND cc_library_DEPS dynload_mklml)
+        endif()
+        add_dependencies(${TARGET_NAME} mklml)
+        target_link_libraries(${TARGET_NAME} "-L${MKLML_LIB_DIR} -liomp5 -Wl,--as-needed")
+      endif()
      target_link_libraries(${TARGET_NAME} ${cc_library_DEPS})
      add_dependencies(${TARGET_NAME} ${cc_library_DEPS})
    endif()
@ -241,6 +264,7 @@ function(cc_test TARGET_NAME)
             WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
    if (${cc_test_SERIAL})
        set_property(TEST ${TARGET_NAME} PROPERTY SERIAL 1)
+    set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_init_allocated_mem=true)
    endif()
  endif()
 endfunction(cc_test)
@ -305,6 +329,7 @@ function(nv_test TARGET_NAME)
    add_test(${TARGET_NAME} ${TARGET_NAME})
    if (nv_test_SERIAL)
        set_property(TEST ${TARGET_NAME} PROPERTY SERIAL 1)
+    set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_init_allocated_mem=true)
    endif()
  endif()
 endfunction(nv_test)
@ -552,7 +577,7 @@ function(py_test TARGET_NAME)
    set(multiValueArgs SRCS DEPS ARGS ENVS)
    cmake_parse_arguments(py_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
    add_test(NAME ${TARGET_NAME}
-             COMMAND env PYTHONPATH=${PADDLE_BINARY_DIR}/python ${py_test_ENVS}
+             COMMAND env FLAGS_init_allocated_mem=true PYTHONPATH=${PADDLE_BINARY_DIR}/python ${py_test_ENVS}
             ${PYTHON_EXECUTABLE} -u ${py_test_SRCS} ${py_test_ARGS}
             WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
  endif()
--- a/cmake/inference_lib.cmake
+++ b/cmake/inference_lib.cmake
@ -12,19 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.

-set_property(GLOBAL PROPERTY FLUID_MODULES "")
-# find all fluid modules is used for paddle fluid static library
-function(find_fluid_modules TARGET_NAME)
-  get_filename_component(__target_path ${TARGET_NAME} ABSOLUTE)
-  string(REGEX REPLACE "^${PADDLE_SOURCE_DIR}/" "" __target_path ${__target_path})
-  string(FIND "${__target_path}" "fluid" pos)
-  if(pos GREATER 1)
-    get_property(fluid_modules GLOBAL PROPERTY FLUID_MODULES)
-    set(fluid_modules ${fluid_modules} ${TARGET_NAME})
-    set_property(GLOBAL PROPERTY FLUID_MODULES "${fluid_modules}")
-  endif()
-endfunction(find_fluid_modules)
-
 # make package for paddle fluid shared and static library
 function(copy TARGET)
    set(options "")
@ -149,8 +136,29 @@ copy(memory_lib
  DSTS ${dst_dir}/${module} ${dst_dir}/${module}/detail
 )

+set(inference_deps paddle_fluid_shared paddle_fluid)
+
+if(WITH_CONTRIB)
+    message(STATUS "installing contrib")
+    set(contrib_dst_dir "${FLUID_INSTALL_DIR}/contrib/inference")
+    if (WITH_ANAKIN AND WITH_GPU)
+        copy(contrib_anakin_inference_lib DEPS paddle_inference_api inference_anakin_api
+            SRCS
+            ${PADDLE_BINARY_DIR}/paddle/contrib/inference/libinference_anakin_api* # compiled anakin api
+            ${PADDLE_BINARY_DIR}/third_party/install/anakin/*.tar.gz # anakin release
+            DSTS ${contrib_dst_dir}/anakin ${contrib_dst_dir}/anakin)
+        list(APPEND inference_deps contrib_anakin_inference_lib)
+   endif()
+
+  copy(contrib_inference_lib DEPS paddle_inference_api paddle_inference_api_shared
+        SRCS ${PADDLE_SOURCE_DIR}/paddle/contrib/inference/paddle_inference_api.h
+        ${PADDLE_BINARY_DIR}/paddle/contrib/inference/libpaddle_inference_api*
+        DSTS ${contrib_dst_dir} ${contrib_dst_dir})
+  list(APPEND inference_deps contrib_inference_lib)
+endif()
+
 set(module "inference")
-copy(inference_lib DEPS paddle_fluid_shared paddle_fluid
+copy(inference_lib DEPS ${inference_deps}
  SRCS ${src_dir}/${module}/*.h ${PADDLE_BINARY_DIR}/paddle/fluid/inference/libpaddle_fluid.*
  DSTS ${dst_dir}/${module} ${dst_dir}/${module}
 )
--- a/doc/about/about_us.rst
+++ b/doc/about/about_us.rst
@ -0,0 +1,53 @@
+=========
+关于我们
+=========
+
+什么是PaddlePaddle
+--------------------
+
+- PaddlePaddle是百度自主研发并开源的深度学习框架，它能够让开发者和企业安全、快速地实现自己的AI想法
+
+- 项目团队汇聚了全球顶级的深度学习科学家，致力于为开发者和企业提供最好的深度学习研发体验
+
+- 框架具有易学、易用、安全、高效四大特性，是最适合中国开发者和企业的深度学习工具
+
+PaddlePaddle的技术特色
+-------------------------
+
+- 新一代深度学习框架： PaddlePaddle是基于“深度学习编程语言”的新一代深度学习框架，在保证性能的同时，极大的提升了框架对模型的表达能力，能够描述任意潜在可能出现的模型
+
+- 对大规模计算更加友好：经过百度内多种大规模计算业务的打磨，PaddlePaddle在分布式计算上表现优异，基于EDL技术能够节约大量计算资源，同时也能支持大规模稀疏模型的训练
+
+- 提供可视化的深度学习：通过Visual DL可以帮助开发者方便的观测训练整体趋势、数据样本质量和中间结果、参数分布和变化趋势、以及模型的结构，帮助开发者更便捷的完成编程过程
+
+提供基于PaddlePaddle的教育体系
+--------------------------------
+
+- 深度学习课程：百度与中国市场顶级的教育、培训机构共同开发了深度学习精品课程以及学习教材，帮助开发者从零掌握深度学习
+
+- 深度学习实训：对于目的是科研和学习的用户，PaddlePaddle提供了无需安装、线上运行的开发环境，并提供算法、算力、数据支持
+
+- 线下培训：提供丰富、高质量的线下教育活动，如青年教师培训、线下实战营、沙龙等多种形式的培训和交流
+
+
+提供基于PaddlePaddle的AI服务
+------------------------------
+
+- EadyDL：可以帮助零算法基础的企业快速完成一个深度学习任务，只需少量的数据即可得到优质的模型
+
+- AI市场：提供标准化的AI 能力、产品的交易机制，帮助企业快速找到所需，有效开展AI业务
+
+- 深度学习竞赛： PaddlePaddle汇聚顶尖深度学习开发者，企业可以发布自己的商业问题，通过竞赛方式快速找到最优的解决方案
+
+你对PaddlePaddle有任何的问题都可以通过以下方式联系到我们
+-----------------------------------------------------------
+
+- 学习/使用问题：可以在 `PaddlePaddle开源社区 <https://github.com/PaddlePaddle/Paddle/issues>`_，以及 `PaddlePaddle中文社区 <http://ai.baidu.com/forum/topic/list/168>`_ 向我们反馈
+
+- 对PaddlePaddle框架发展的建议：可发送邮件至Paddle-better@baidu.com
+
+我们期待与你一起打造世界顶级深度学习框架，共同推动AI技术的进步
+
+
+
+PaddlePaddle团队
--- a/doc/fluid/api/average.rst
+++ b/doc/fluid/api/average.rst
@ -0,0 +1,16 @@
+..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
+    !DO NOT EDIT THIS FILE MANUALLY!
+
+=============
+fluid.average
+=============
+
+.. _api_fluid_average_WeightedAverage:
+
+WeightedAverage
+---------------
+
+..  autoclass:: paddle.fluid.average.WeightedAverage
+    :members:
+    :noindex:
+
--- a/doc/fluid/api/backward.rst
+++ b/doc/fluid/api/backward.rst
@ -0,0 +1,23 @@
+..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
+    !DO NOT EDIT THIS FILE MANUALLY!
+
+==============
+fluid.backward
+==============
+
+.. _api_fluid_backward_append_backward:
+
+append_backward
+---------------
+
+..  autofunction:: paddle.fluid.backward.append_backward
+    :noindex:
+
+.. _api_fluid_backward_calc_gradient:
+
+calc_gradient
+-------------
+
+..  autofunction:: paddle.fluid.backward.calc_gradient
+    :noindex:
+
--- a/doc/fluid/api/clip.rst
+++ b/doc/fluid/api/clip.rst
@ -1,9 +1,11 @@
 ..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
    !DO NOT EDIT THIS FILE MANUALLY!

-====
-clip
-====
+==========
+fluid.clip
+==========
+
+.. _api_fluid_clip_ErrorClipByValue:

 ErrorClipByValue
 ----------------
@ -12,6 +14,8 @@ ErrorClipByValue
    :members:
    :noindex:

+.. _api_fluid_clip_GradientClipByValue:
+
 GradientClipByValue
 -------------------

@ -19,6 +23,8 @@ GradientClipByValue
    :members:
    :noindex:

+.. _api_fluid_clip_GradientClipByNorm:
+
 GradientClipByNorm
 ------------------

@ -26,6 +32,8 @@ GradientClipByNorm
    :members:
    :noindex:

+.. _api_fluid_clip_GradientClipByGlobalNorm:
+
 GradientClipByGlobalNorm
 ------------------------

@ -33,15 +41,3 @@ GradientClipByGlobalNorm
    :members:
    :noindex:

-append_gradient_clip_ops
------------------------
-
-..  autofunction:: paddle.fluid.clip.append_gradient_clip_ops
-    :noindex:
-
-error_clip_callback
-------------------
-
-..  autofunction:: paddle.fluid.clip.error_clip_callback
-    :noindex:
-
--- a/Show More
+++ b/Show More